cond-mat0508288/art.tex
1: \documentclass[pre, aps, showpacs]{revtex4}
2: \usepackage[dvips]{graphicx}
3: \usepackage{mathrsfs}
4: \usepackage{latexsym}
5: %
6: \begin{document} 
7: %
8: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 
9: %
10: \title{Protein-Mediated DNA~Loops: Effects of Protein Bridge Size and Kinks} 
11: %
12: \author{Nicolas~\textsc{Douarche} and Simona~\textsc{Cocco}} 
13: %
14: \affiliation{CNRS -- Laboratoire de Physique Statistique de l'ENS, 
15: 24~rue Lhomond, 75005~Paris, France} 
16: %
17: \begin{abstract}
18: This paper focuses on the probability that a portion of~DNA closes on itself 
19: through thermal fluctuations. We investigate the dependence of this probability 
20: upon the size~$r$ of a protein bridge and/or the presence of a kink at half 
21: DNA~length. The~DNA is modeled by the Worm-Like Chain model, and the probability 
22: of loop formation is calculated in two ways: exact numerical evaluation of the 
23: constrained path integral and the extension of the \textsc{Shimada} 
24: and~\textsc{Yamakawa} saddle point approximation. For example, we find that the 
25: looping free energy of a 100~base~pairs~DNA decreases from 24~$k_\mathrm{B}T$ to 
26: 13~$k_\mathrm{B}T$ when the loop is closed by a protein of $r=10$~nm length. 
27: It further decreases to 5~$k_\mathrm{B}T$ when the loop has a kink of 
28: $120^\circ$ at half-length. 
29: \end{abstract}
30: % 
31: \pacs{36.20.Hb, 46.70.Hg, 82.35.Pq, 87.14.Gg, 87.15.Aa, 87.15.La}
32: %
33: \maketitle
34: %
35: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
36: %
37: \section{Introduction: DNA Loops in Gene Transcription Regulation} 
38: %
39: Gene expression is regulated by a wide variety of mechanisms. These activation 
40: as well as repression phenomena may occur at every expression steps~(translation, 
41: transcription, etc) and do involve interactions between several biological 
42: molecules~(DNA, RNAs, proteins, etc). For instance proteins bound on specific 
43: DNA~sequences may turn~on/off genes transcription by interacting with each other. 
44: By bringing closer those proteins DNA~looping can ease their 
45: interactions~\cite{rippe95, focus, finzi, lia}. Such looping events may be 
46: observed over a wide range of lengths spreading from hundreds to thousands base 
47: pairs~(bp). The looping probability has been firstly measured by the cyclization 
48: of DNA~segments in solution with cohesive ends. Once the loop is formed proteins 
49: called ligases stabilize it~\cite{shore, du}. It is then possible to count the 
50: circular~DNAs with respect to the linear ones. The loop formation mediated by 
51: proteins have also been experimentally studied. Two examples are the loops 
52: formed by the~LacR or the~GalR transcriptional repressors~\cite{brenowitz}. 
53: Two units of such proteins bind at two specific positions along a same~DNA and 
54: associate to form a complex when the binding sites come in contact. The 
55: formation of such loops has been recently studied using micromanipulation 
56: experiments on a single DNA~molecule~\cite{finzi, lia}. The study of the~GalR 
57: mediated loop has shed light on the role of a third protein called~HU that 
58: sharply bends~(\emph{i.e.} kinks) the~DNA at half-length. \\
59: % 
60: The DNA~loop probability depends mainly on its length and flexibility. 
61: Long~DNAs~(typically longer than 1500~bp) essentially behave as Gaussian 
62: Polymers~(GP)~\cite{yamakawa, kleinert}: the cyclization cost is mainly of an 
63: entropic nature~\cite{hanke}. On the contrary, for small lengths DNA~cyclization 
64: is difficult mainly because of the bending energy cost. The computation of the 
65: elastic energy for the Worm-Like Chain~(WLC) model~\cite{kratky, yamakawa, kleinert} 
66: can be analytically performed~\cite{shimada}; numerical methods have also been 
67: employed when electrostatic properties are included~\cite{balaeff}. At 
68: intermediate length scales~(from about 500~bp) elastic rigidity and entropic 
69: loss are both important. Several approximations have been developed to study 
70: this lengths range~\cite{focus, daniels, gobush, wilhelm}, among them the 
71: calculations of the fluctuations around the lowest bending energy configurations 
72: performed by~\textsc{Shimada} and~\textsc{Yamakawa}~\cite{shimada}. Numerical 
73: approaches have also been developed: Monte~Carlo~\cite{podtelezhnikov} and brownian 
74: dynamics based simulations~\cite{rippe95, merlitz, rippe00} as well as numerical 
75: calculations of the~WLC path integral under the closed ends constraint~\cite{yan03,yan04}. 
76: This last method allowed~\textsc{Yan}, \textsc{Kawamura}~and~\textsc{Marko} to 
77: study the elastic response of~DNA subject to permanent or thermally excited bendings 
78: caused by binding proteins~(such as~HU) or  inhomogeneities along the DNA~double 
79: helix~\cite{yan04,yan05}. All these studies do provide a better understanding of the 
80: underlying regulation phenomena despite their overall complexity. \\ 
81: % 
82: In this paper we study two processes that turned out to be important in 
83: DNA~looping. Namely the size of the protein complex clamping the 
84: loop~\cite{rippe95, merlitz}, acting as a bridge between the two DNA~ends, and 
85: mechanisms implying DNA~stiffness loss which are taken into account in an 
86: effective way by kinking the~WLC at half-contour length~\cite{rippe95, merlitz, %
87: rippe00, sankararaman}. In section~\ref{s:defmeth} we define the model 
88: and the methods: we describe the numerical approach~(\S~\ref{secnum}) and the 
89: analytical Saddle Point Approximation~(SPA, \S~\ref{s:sp1}) to calculate the 
90: $r$-dependent closure factor and the looping free energy. In section~\ref{s:size} 
91: we compare the numerical and~SPA  results with previous experimental and 
92: theoretical results. In section~\ref{s:kink} we extend the numerical and~SPA 
93: approaches to a kinked loop; we discuss our results and we propose a simple 
94: formula that accounts for both the protein bridge and kink 
95: effects~(\S~\ref{ss:kink}). We conclude (section~\ref{s:conclusion}) by sketching 
96: how to include omitted DNA~properties which may also play an important role in 
97: its closure such as twist rigidity or electrostactic interactions. 
98: % 
99: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 
100: %
101: \section{Definitions and Methods}\label{s:defmeth}
102: % 
103: We use the well known Worm-Like Chain~(WLC) model~\cite{kratky, yamakawa, kleinert}. 
104: The DNA~polymer is described as an \emph{inextensible} continuous differentiable 
105: curve of contour length~$L$, with unit tangent 
106: vector~$\vec t(s)$~($0 \le s \le L$). The polymer is characterized by the 
107: persistence length~$A$ beyond which tangent vectors lose their alignment: 
108: $\langle \vec t(s) \cdot \vec t(s') \rangle = %
109: \exp{\! \left(-\vert s-s' \vert/A\right)}$. The energy of a configuration 
110: of the polymer stretched under an external force~$f \, \vec e_z$ reads 
111: %
112: \begin{equation} 
113: E \! \left[\vec t ; L, f\right] = \frac{1}{2} \frac{A}{\beta} \int_0^L 
114: \! \left[\frac{\mathrm{d}\vec t(s)}{\mathrm{d}s}\right]^2 \!\!\! \mathrm{d}s 
115: - f \int_0^L \!\! \vec e_z \cdot \vec t(s) \, \mathrm{d}s 
116: \end{equation} 
117: %
118: where we use $\beta=1/k_\mathrm{B}T$. No twist elasticity nor extensibility 
119: will be considered. The partition function is 
120: %
121: \begin{equation}\label{defz}
122: Z(L, f) = \int \!\! \mathscr{D}\vec t \; 
123: \exp{\! \left\{-\beta E \! \left[\vec t ; L, f\right] \right\}}. 
124: \end{equation}
125: %
126: Notice that summation over all initial and final tangent vectors orientations, 
127: $\vec t(0)$ and~$\vec t(L)$, is implicitly understood in this path integral. \\ 
128: % 
129: In this paper we are interested in the formation of a loop in a DNA~molecule, 
130: and the Probability Density Functions~(PDFs) of end-to-end distances play an 
131: important role. The quantities under study are denoted by~$Q$,~$S$,~$P$ and~$J$ 
132: respectively and defined as follows: 
133: % 
134: %%%%%%%%%%%%%%%%%%
135: %
136: \begin{itemize}
137: %
138: \item the end-to-end extension~$\vec r = (x, y, z)$~PDF at zero force, 
139: % 
140: \begin{equation}\label{Q} 
141: Q \! \left(\vec r , L\right) = \frac{1}{Z(L, f=0)} \, 
142: \int \!\! \mathscr{D}\vec t \,\,\, 
143: \delta \! \left[\int_0^L \!\! \vec t(s) \, \mathrm{d}s - \vec r \right] 
144: \exp{\! \left\{-\beta E \! \left[\vec t ; L, f=0\right]\right\}}. 
145: \end{equation} 
146: % 
147: In the absence of force $Q$~depends on its argument~$\vec r$ through its 
148: modulus~$r = \vert \vec r \vert$ only, and we may introduce the radial~PDF 
149: % 
150: \begin{equation}
151: S(r, L) = 4 \pi r^2 \; Q \! \left(r, L\right). 
152: \end{equation}
153: %
154: \item The $z$~extension~PDF reads 
155: %
156: \begin{equation}\label{P} 
157: P(z, L) = \int_{-L}^L \!\!\!\!\! \mathrm{d}x 
158: \int_{-L}^L \!\!\!\!\! \mathrm{d}y \,\, Q\left[(x, y, z), L\right]. 
159: \end{equation}
160: %
161: In the absence of external force, notice the~$x$ or~$y$ extensions~PDFs are 
162: given by~$P$ too. Interestingly, the radial and $z$~extensions~PDF are related 
163: to each other through the useful identity~\cite{samuel} 
164: %
165: \begin{equation}\label{relasp}
166: S(r, L) = -2r \; \frac{\mathrm{d}P}{\mathrm{d}z}(r, L). 
167: \end{equation}
168: % 
169: \item The cyclization factor 
170: %
171: \begin{equation}
172: J(L) = Q(0, L) 
173: \end{equation}
174: %
175: defined as the density of probability that the two ends of the~DNA are in 
176: contact with one another. 
177: % 
178: \item The $r$-dependent closure factor is 
179: % 
180: \begin{equation}\label{defjr}
181: J(r, L) = \left[\int_0^{r} \!\!\! S(r', L) \; \mathrm{d}r'\right]/
182: \left(\frac{4}{3} \pi r^3\right). 
183: \end{equation} 
184: % 
185: It gives the density probability for the two ends of the chain to stay within 
186: a sphere of radius~$r$. It is easy to check that~$J(r, L) \to J(L)$ 
187: when~$r \to 0$. For experimental convenience, units used for~$J(L)$ and~$J(r, L)$ 
188: are moles~per~liter: $1~\mathrm{nm}^{-3} \approx 1.66~\mathrm{mol}\cdot\mathrm{L}^{-1} %
189: \equiv 1.66~\mathrm{M}$. In these units $J(r, L)$~gives directly the concentration 
190: of one binding site in proximity of the other. 
191: % 
192: \item Finally we consider the looping free energy cost 
193: % 
194: \begin{equation}\label{defdeltag} 
195: \beta \Delta G(r, L) = - \ln{\! \left[J(r, L) \times \frac{4}{3} \pi r^3\right]}. 
196: \end{equation}
197: % 
198: Note that this definition does not include the details of the geometry nor 
199: the affinities of the DNA/protein and protein/protein interactions. We actually 
200: assume all the sphere of radius~$r$ to be the reacting volume, \emph{i.e.} that 
201: the loop will form if the DNA~ends happen to be in this sphere.
202: % 
203: \end{itemize}
204: %
205: %%%%%%%%%%%%%%%% 
206: %
207: Despite intense studies of the~WLC model no exact analytical expression is 
208: known for~$Q$ and the quantities of interest here, namely~$J$ and~$S$. However 
209: approximations expanding from the two limiting regimes 
210: (entropic~\cite{daniels, gobush} and elastic~\cite{wilhelm, shimada}) along with 
211: exact numerical computations are available. Hereafter, we have resorted to numerical 
212: as well as approximate analytical techniques~(SPA) for calculating the cyclization 
213: factor~$J$ and the probability of almost closed DNA~configurations. 
214: %
215: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 
216: % 
217: \subsection{Numerical Calculation of the Probabilities~$P$,~$S$ and~$J$}\label{secnum} 
218: % 
219: Our starting point for the calculation of the $z$~extension~PDF is the 
220: \textsc{Fourier}~representation of the \textsc{Dirac}~$\delta$-function 
221: in~(\ref{Q}) and~(\ref{P}), 
222: % 
223: \begin{equation}\label{ft} 
224: P(z, L) = \int_{-\infty}^{+\infty} \frac{\mathrm{d}k}{2\pi} \; 
225: e^{+ i k z} \; Z \! \left(L, f = -i k/\beta\right). 
226: \end{equation} 
227: % 
228: At fixed momentum~$k$ we are left with the calculation of the partition 
229: function~$Z(L, f)$ at (imaginary) force $f=-i k/\beta$. The path 
230: integral~(\ref{defz}) defining~$Z$ is interpreted as the evolution operator 
231: of a quantum system, the rigid rotator under an external imaginary field 
232: % 
233: \begin{equation}\label{defz2}
234: Z(L, f) = \langle \mathrm{final} \vert 
235: \exp{\! \left[-L/A \times \widehat{H}(f)\right]} 
236: \vert \mathrm{initial} \rangle. 
237: \end{equation}
238: % 
239: The entries of its hamiltonian~$\widehat H$ are easily expressed in the 
240: spherical harmonics~$\vert \ell, m \rangle$ basis: 
241: $\langle \ell, m \vert \widehat{H}(f) \vert \ell', m' \rangle = %
242: H_{\ell, \ell'}(f) \; \delta_{m, m'}$ with 
243: % 
244: \begin{equation}\label{Hlm}
245: H_{\ell, \ell'}(f) = \frac{\ell(\ell+1)}{2} \, \delta_{\ell,\ell'} 
246: - \beta A f \, \frac{\ell \, \delta_{\ell', \ell-1} + 
247: \ell' \, \delta_{\ell, \ell'-1}}{\sqrt{(2\ell+1)(2\ell'+1)}}. 
248: \end{equation}
249: % 
250: The entries of~$\widehat{H}$ do not depend on the azimuthal number~$m$ due 
251: to cylindrical symmetry around the force axis. Finally, integration over all 
252: initial and final orientations for the tangent vectors at the ends of the 
253: polymer chain selects~$\vert \mathrm{initial} \rangle = %
254: \vert \mathrm{final} \rangle = \vert 0, 0 \rangle$. \\ 
255: % 
256: A recent paper~\cite{samuel} used~\textsf{Mathematica} to compute the vacuum 
257: amplitude~(\ref{defz2}) through a direct matrix exponentiation. We have instead 
258: used the~\textsf{Expokit} library~\cite{expokit} since it proves to be more 
259: accurate and faster for intensive numerical calculations. We truncated 
260: hamiltonian~(\ref{Hlm}) in a way the outcome is insensitive to the cut-off on 
261: the harmonics order. The~(inverse) \textsc{Fourier} transform~(\ref{ft}) is then 
262: handled by a Fast \textsc{Fourier} Transform~(FFT) algorithm~\cite{nr}. This 
263: task is in particular facilitated thanks to the inextensibility constraint which 
264: makes the distribution bandwidth limited. 
265: % 
266: \begin{figure} [!h]
267: \includegraphics{Pz.eps} 
268: \caption{Numerical computation of the $z$~extension~PDF over a wide 
269: range of contour lengths~$L$. As expected the agreement with the~SPA prediction 
270: (see section~\ref{s:sp1}) improves as~$L$ decreases (upper bound~$L/A\lesssim 1$). 
271: The long~WLC behavior is caught by the GP approximation~\cite{yamakawa, kleinert} 
272: as soon as, say~$L/A\gtrsim 5$.} 
273: \label{f:Pz} 
274: \end{figure}
275: %
276: Our results for the $z$~extension~PDF are shown in Fig.~\ref{f:Pz}. The 
277: cross-over from the rigid elastic regime~$(L/A = 0.1, 0.5, 1)$ to the flexible 
278: entropic regime~$(L/A = 5, 10, 15)$ is clearly visible. Using~(\ref{relasp}) 
279: then gives us access to~$S$, the radial extension~PDF~(see Fig.~\ref{f:Sr}). 
280: The most probable value for the distance $r_\star$, switches from full 
281: extension~$r_\star \lesssim L$ for contour lengths~$L\lesssim A$ (elastic 
282: dominated regime) to the~GP most probable extension~$\sqrt{4LA/3}$ for longer 
283: ones~$L\gg A$ (entropic dominated regime). Also note that~$S$ always goes 
284: continuously to zero in the~$r\to L$ limit due to WLC~inextensibility. 
285: We have finally calculated the $r$-dependent closure factor~$J$ according 
286: to~(\ref{defjr}) by numerical integration of~$S$, and the looping free energy 
287: cost~$\Delta G$ defined in~(\ref{defdeltag}). \\
288: %
289: \begin{figure} [!h]
290: \includegraphics{Sr.eps}
291: \caption{Numerical computation of~$S$, the radial $r$~extension~PDF. 
292: The outcomes of the numerical calculations are tested against exactly 
293: known values for the first even 
294: moments~$\langle r^{2n}\rangle$ (inset)~\cite{yamakawa, kleinert}. The shape 
295: of~$S$ compares very well to the widely used~\textsc{Wilhelm} and~\textsc{Frey}~(W\&F) 
296: expansion~\cite{wilhelm}, valid up to~$L/A\lesssim 2$. Similar tests were 
297: achieved with other popular approximation schemes~\cite{focus} (data not shown).} 
298: \label{f:Sr} 
299: \end{figure}
300: % 
301: Let us now discuss the numerical errors that could be important when calculating 
302: probabilities of rare events. Main sources of error are the hamiltonian~(\ref{Hlm}) 
303: truncation and the integration step to compute the~$r$ dependent closure 
304: factor~(\ref{defjr}). As mentionned above the cut-off on the harmonic order was 
305: systematically choosed in order convergence is observed. We have used $\ell = 50$ 
306: after having verified that the result is unchanged for $\ell = 100$. Concerning 
307: integration, limitation comes from the number of available data in the 
308: range~$0\le r'\le r$ which is directly related to the~$k$ sampling of the partition 
309: function~$Z$. For~$r=1$~nm the numerical integration is still reliable, but 
310: decreasing further~$r$ turns out to be critical. Other potential sources of error 
311: are negligeable. Indeed bandwidth limitation of~$P(z ; L)$ prevents any 
312: aliasing~\cite{nr} during~FFT~(\ref{ft}) and the derivative of $P$~(\ref{relasp}) 
313: can actually be skipped by an integration by part of (\ref{defjr}) to compute~$J$. 
314: Further hypothetical errors would then come from the~\textsf{Expokit} 
315: library~\cite{expokit} itself but its routines were coded to compute accurately 
316: matrix exponentials over a broad range of matrices~\cite{moler}. We have checked 
317: the numerical precision of our method by the comparison with the exact values for 
318: the first even moments  of $S(r)$ (Fig.~\ref{f:Sr} Inset); moreover as shown in 
319: Fig.\ref{f:Sr} $S(r)$ agrees with Wilhelm and Frey expansion for small~$L/A$ 
320: values; finally we will see in section~\ref{s:size} that the numerically 
321: calculated cyclization factor~$J(L)$ (\ref{defjr}) is in agreement with the 
322: \textsc{Shimada}-\textsc{Yamakawa} and Gaussian approximation results for 
323: respectively small and large~$L$ (Fig.~\ref{jyagaus}). 
324: %
325: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 
326: %
327: \subsection{Saddle Point Approximation for~$J$}\label{s:sp1}
328: % 
329: In addition to the exact numerical calculations detailed above, we have carried 
330: out approximate calculations based on a saddle point estimate of the partition 
331: function~(\ref{defz}). We follow the~\textsc{Shimada} and~\textsc{Yamakawa} 
332: calculation for the saddle point configuration~\cite{shimada}, extending it for 
333: an opened~DNA. The saddle point configuration for a closed loop is shown in 
334: Fig.~\ref{jyagaus} inset: it is a \emph{planar loop}. The tangent vector at 
335: position~$s$ along the chain is characterized by its angle~$\theta(s)$ with 
336: respect to the end-to-end extension~$\vec r$. The optimal configuration is 
337: symmetric with respect to the perpendicular~$\vec e_\perp$-axis while the 
338: half-length angle is~$\theta(L/2) = 180^\circ$. The initial angle~$\theta(0)$ 
339: is chosen by the minimization of the bending energy of the chain. The 
340: optimization gives rise to the following condition on the 
341: parameter~$x = \cos{\! \left[\theta(0)/4\right]}$ 
342: %
343: \begin{equation}
344: \left(1 + r/L\right) \; \widehat{K} \! \left(x^2\right) = 
345: 2 \, \widehat{E} \! \left(x^2\right), 
346: \end{equation}
347: %
348: where~$\widehat{K} \! \left(x^2\right) = K\!\left(\frac{\pi}{2}, x^2\right)$ 
349: and~$\widehat{E} \! \left(x^2\right) = E\left(\frac{\pi}{2}, x^2\right)$ are 
350: the complete elliptical integrals of the first and second kinds 
351: respectively~\cite{abramowitz}. The corresponding elastic energy is 
352: % 
353: \begin{equation}\label{f*} 
354: \beta \Delta E(r, L) = \frac {4}{L/A} \; \widehat{K}^2 \! \left(x^2\right) 
355: \times \left(2 x^2 - 1 + r/L\right). 
356: \end{equation}
357: %
358: The end-to-end extension~PDF is then approximated as 
359: % 
360: \begin{equation}\label{kc}
361: Q(r, L) = C(r, L) \; \exp{\! \left[-\beta \Delta E(r, L)\right]}. 
362: \end{equation}
363: %
364: The prefactor~$C(r, L)$ should be calculated by taking into account quadratic 
365: fluctuations to the saddle point configuration. Since such calculation is quite 
366: involved we will actually only extend the~\textsc{Shimada} and~\textsc{Yamakawa} 
367: results which was computed considering fluctuations to a closed loop. 
368: In~$\mathrm{M} = \mathrm{mol}\cdot\mathrm{L}^{-1}$ units this reads 
369: %
370: \begin{equation}\label{c} 
371: C_\mathrm{SY}(L) = \frac{1.66}{A^3} \times \frac{112.04}{(L/A)^5} \, 
372: \exp{\! \left(0.246 \times L/A\right)}. 
373: \end{equation}
374: % 
375: For this factor to fit the correct fluctuations~(to the opened loop) we have 
376: to consider the fluctuations to a fake closed loop of similar bending energy. 
377: Such a loop may be obtained by considering the optimal configuration of a loop 
378: of contour length~$(L + 2r)$ instead of~$L$, as shown in Fig.~\ref{f:J}~(bottom, 
379: inset). The choice of the factor~$2r$ derives from the following geometrical 
380: considerations: the closed saddle point configuration has an initial 
381: angle~$\theta(0) = 49.2^\circ$, the~$\Delta L$ closing the loop could be 
382: calculated for each value of~$r$ by requiring 
383: % 
384: \begin{equation}
385: \int_0^{\Delta L} \!\!\!\!\!\! \cos{\theta(s)} \; \mathrm{d}s = r. 
386: \end{equation}
387: % 
388: The angle~$\theta(s)$ increases slightly on the first part of the 
389: trajectory~$\theta(s) > \theta(0)$ and~$\Delta L > 1.53 \; r$. We have 
390: chosen~$\Delta L = 2r$ as an approximate value, this approximation has the 
391: advantage that it can be directly put in the fluctuation 
392: expression~$C(r, L) \approx C_\mathrm{SY}(L + 2r)$ in equation~(\ref{kc}) 
393: to obtain: 
394: %
395: \begin{equation}\label{qsp}
396: Q(r, L) = C_\mathrm{SY}(L + 2r) \; \exp{\! \left[-\beta \Delta E(r, L)\right]}
397: \end{equation}
398: %
399: where~$\Delta E$ is given in formula~(\ref{f*}) and~$C_\mathrm{SY}$ 
400: is given in formula~(\ref{c}). The validity of this approximation for the 
401: fluctuations prefactor was checked out by comparing~$J(r , L)$ obtained 
402: from~$Q(r, L)$ through formula~(\ref{defjr}), with the numerical results. 
403: The good agreement shown in Fig.~\ref{f:J} allows to obtain a semi-analytical 
404: formula for the loop probability with a finite interacting volume, which is 
405: valid for molecules of up to 2~kb~(kilo base pairs). 
406: % 
407: \begin{figure}
408: \includegraphics[height=12cm, angle=-90]{jyagaus.eps}
409: \caption{Cyclization factor as a function of the DNA~length with: the gaussian 
410: model~(gray line); the WLC~model with the~\textsc{Shimada} and~\textsc{Yamakawa} 
411: formula~(dotted black line); the WLC~model with the numerical calculation~(full 
412: black line). The most probable length is 500~bp. Inset: the lowest bending energy 
413: configuration of a closed loop.} 
414: \label{jyagaus}
415: \end{figure}
416: % 
417: \begin{figure}
418: \begin{tabular} {rl}
419: \includegraphics[height=8cm, angle=-90]{j180num.eps} & 
420: \includegraphics[height=8cm, angle=-90]{dg180num.eps} \\ 
421: \includegraphics[height=8cm, angle=-90]{j180col.eps} & 
422: \includegraphics[height=8cm, angle=-90]{dg180col.eps} 
423: \end{tabular}
424: \caption{Closure factor~(left panel) and free energy~(right panel) with a 
425: protein bridge of sizes: $r = 1$~nm~(dashed line); $5$~nm~(dotted line); 
426: $10$~nm~(full line). The error bars are shown when they are larger than the 
427: symbol sizes. Top: numerical calculation of the constrained path integral. 
428: The~$r = 1$~nm curve coincides with the~$r = 0$ cyclization factor. Bottom: 
429: closure factor obtained by the extension of the~\textsc{Shimada} 
430: and~\textsc{Yamakawa} calculation, that includes~$r$. Theoretical results 
431: are in very good agreement with Monte Carlo simulations obtained 
432: by~\textsc{Podtelezhnikov} and~\textsc{Vologodskii} ~(filled circle, 
433: $\bullet$)~\cite{podtelezhnikov} and in fair agreement with brownian dynamics 
434: simulations obtained by~\textsc{Langowski} and~\emph{al.}~(empty 
435: square, $\Box$)~\cite{rippe95, merlitz}. Inset of the bottom left panel: 
436: lowest bending energy configurations for~100~bp and~$r=0$~(bottom) 
437: or~$r=10$~nm~(top), the closure of the~$r=10$~nm configuration is shown by 
438: a thin line.} 
439: \label{f:J}
440: \end{figure}
441: % 
442: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
443: %
444: \section{Results: the effect of a Protein Bridge}\label{s:size} 
445: %
446: In Fig.~\ref{jyagaus} we show the cyclization factor~$J(L)$ for lengths~$L$ 
447: up to~3~kb. The most probable loop length is of about~$L_\star = 500$~bp that 
448: is~$L_\star/A\approx 3.5$~\cite{shimada, rippe95}. Both shorter and longer cyclized 
449: lengths are less probable: stiffness makes difficult the bending of smaller polymers 
450: while entropy makes longer polymers ends unlikely to meet. The numerical cyclization 
451: factor is compared with the saddle point calculation of \textsc{Shimada} 
452: and~\textsc{Yamakawa}~\cite{shimada} and with the gaussian polymer~(GP) model. The 
453: former is in good agreement with the numerical results for lengths smaller than 
454: about~1.5~kb while the latter works well for loops larger than~2.5~kb. \\
455: % 
456: The effects of the finite size~$r$ of the protein bridge are displayed 
457: in~Fig.~\ref{f:J} which shows~$J(r, L)$~(on the left) and~$\Delta G(r, L)$~(on 
458: the right) for lengths~$L$ ranging from~75~bp~(\emph{i.e.}~25~nm) 
459: to~300~bp~(\emph{i.e.}~100~nm) and~$r$ respectively ranging from 1~nm to 10~nm. 
460: The numerical results (on the top ) are in good agreement with the SPA~results~(on 
461: the bottom). Note that for small lengths~$L$, $J(r, L)$~has a peak for 
462: the~$L \approx r$ event corresponding to rigid rod-like configurations. 
463: Fig.~\ref{f:J}  does not show this peak, occurring for~$r = 10$~nm at 
464: about~$L = 30$~bp~(or~$L=10$~nm) because we focus on the cyclisation events. 
465: As shown in this figure $r$~values ranging from~1~nm to~10~nm make no difference 
466: for the $r$-dependent closure factor, considering contour lengths~$L$ larger 
467: than~300~bp~(or~100~nm). The cyclization factor~$J(L)$ is evaluated as the 
468: limit~$r\to 0$ of~$J(r, L)$. In practice convergence is reached as soon as~$r$ 
469: is about one order of magnitude smaller than~$L$. In the range~$L > 75$~bp~(25~nm) 
470: $J(5~\mathrm{nm}, 150~\mathrm{bp})$ converges 
471: to~$J(1~\mathrm{nm}, 150~\mathrm{bp})$. On the other hand~$J(r, L)$ is 
472: considerably different from~$J(L)$ when~$r$ is of the same order of magnitude 
473: than~$L$. For example for loops of $L = 100$~bp~(34~nm) an end-to-end extension 
474: of~10~nm increases by two orders of magnitude the closure factor. Therefore 
475: proteins of size~$\approx 10$~nm are expected to produce drastic enhancements in 
476: looping short DNA~sequences. \\
477: % 
478: In terms of energetics~(see Fig.~\ref{f:J}, right) cycling a~100~bp DNA~sequence 
479: costs~25~$k_\mathrm{B}T$ when the loop ends are required to stay within a sphere 
480: of radius~$r = 1$~nm. This cost decreases to~$13~k_\mathrm{B}T$ if the sphere has 
481: the typical protein size of~$r=10$~nm. For loop lengths larger than~300~bp the 
482: only difference in the three curves of Fig.~\ref{f:J}~(right), is a free energy 
483: shift due to the difference in the reacting volume. For instance, a reaction 
484: radius of~10~nm decreases the looping free energy 
485: of~$3 \times \ln(10) \approx 7~k_\mathrm{B}T$ with respect to a~1~nm reaction 
486: radius. \\
487: % 
488: Our results for the closure factor and the looping free energy are compared in 
489: Fig.~\ref{f:J} with the Monte Carlo~(MC) and Brownian Dynamics~(BD) simulations 
490: results, obtained respectively by~\textsc{Podtelezhnikov} 
491: and~\textsc{Vologodskii}~\cite{podtelezhnikov}~(shown in the figure with filled 
492: circles, $\bullet$) and~\textsc{Langowski} 
493: and~\emph{al.}~\cite{rippe95, merlitz}~(displayed in the figure with empty 
494: squares, $\Box$). Numerical and SPA~data~(that for~$r=1$~nm converge to 
495: the~\textsc{Shimada} and~\textsc{Yamakawa} curve) are in better agreement with 
496: the~MC data than with the~BD ones; indeed numerical, SPA~and~MC data are obtained 
497: with a simpler model than BD~ones, which does not include twist rigidity and 
498: electrostatic effects. \\ 
499: % 
500: Considerations about Lac~operon repression energetics will help us illustrating 
501: our results and compare them with other previous results. Expression of proteins 
502: enabling bacteria~\emph{E.~Coli} to perform the lactose metabolism can be prevented 
503: at the transcriptional level by cycling two different sequences~\cite{finzi}, 
504: the smallest one including the operon promoter. 
505: Let~$\mathrm{O}_1\mathrm{O}_3 = 76$~bp and~$\mathrm{O}_1\mathrm{O}_2 = 385$~bp 
506: denote these two resulting DNA~loops where the so-called 
507: operators~$\mathrm{O}_{1, 2, 3}$ actually are the small specific 
508: DNA~sequences~(10~bp or so) at which the tetrameric repressor protein~LacR can 
509: bind thus clamping the desired loop. Notice both processes are needed for efficient 
510: repression: despite~$\mathrm{O}_1\mathrm{O}_3$ contains the operon promoter its 
511: cyclization is much less probable to occur than the~$\mathrm{O}_1\mathrm{O}_2$ 
512: one~(see cyclization factor~$J(L)$ in Fig.~\ref{jyagaus}). The LacR~size is estimated 
513: from its cristallized stucture to~$r\approx 13$~nm~\cite{brenowitz}. In the \emph{in 
514: vitro} experiments many parameters are under control among which the operators sequence 
515: and location. The distance between the two operators, that defines the length of the 
516: DNA~loop has been fixed in~\cite{finzi, lia} to about~100~bp. Our results are in good 
517: agreement with the experimental measured stability of a~114~bp DNA~loop mediated by a 
518: LacR~protein, obtained by \textsc{Brenowitz} and~\emph{al.} in~1991~\cite{brenowitz}. 
519: By measuring the proportion of looped complexes present in a solution with respect to 
520: the unlooped molecules they obtained a looping free enegy 
521: of~$20.3 \pm 0.3~k_\mathrm{B}T$ to wich they associated a closure factor 
522: of~$8 \; 10^{-10}$~M. From Fig.~\ref{f:J}, the closure factor of a loop of~114~bp with 
523: a protein bridge of~$r = 10$~nm is~$J(10~\mathrm{nm}, 114~\mathrm{bp}) = 10^{-9}$~M to 
524: which we associate, from formula~(\ref{defdeltag}) a cyclization free 
525: enegy~$\Delta G(10~\mathrm{nm}, 114~\mathrm{bp}) = 12~k_\mathrm{B}T$. Note that the 
526: very good agreement between the closure factor contrasts with the bad agreement 
527: for the cyclization free energy. The latter could have been calculated considering 
528: a different reaction volume or it could also include the competition with 
529: configurations that do not allow the formation of a loop (see Fig.~2 
530: of~\cite{brenowitz}). To explain the high value found for the closure factor 
531: \textsc{Brenowitz} and~\emph{al.} already included the size of the protein in 
532: the analysis of their results by comparing their~$J$ result with the value 
533: expected for the cyclization probability of a free~DNA when the length of the 
534: protein is included in the size of the loop. \\
535: % 
536: Another result on DNA~loop mediated by LacR protein has been obtained by 
537: \textsc{Balaeff} and~\emph{al.}~\cite{balaeff} who have numerically calculated 
538: the elastic energy of the $\mathrm{O}_1\mathrm{O}_3$~loop from a WLC~model 
539: also including: the twist rigidity, the short range electrostatic repulsion 
540: and the details of the LacR/DNA complex crystal structure. The elastic energy 
541: is estimated to~23~$k_\mathrm{B}T$ in~\cite{balaeff}, of which~$81 \%$ (that is 
542: $18~k_\mathrm{B}T$) due to the bending and~$19\%$ due to the unwinding. The 
543: bending energy of~$18~k_\mathrm{B}T$ is to be compared to the saddle point 
544: energy~(\ref{f*}) of $15~k_\mathrm{B}T$ for a~75~bp loop with an end-to-end 
545: extension of~$r = 10$~nm. The corresponding free energy of loop formation 
546: obtained by~(\ref{qsp}) and after integration over the reacting 
547: volume~(\ref{defdeltag}) 
548: is~$\Delta G(10~\mathrm{nm}, 75~\mathrm{bp}) = 14~k_\mathrm{B}T$ (see 
549: Fig.~\ref{f:J}, bottom/right). For a~400~bp loop since LacR is 
550: only~$\approx 10 \%$ of the loop length, its size is expected to play a less 
551: important role. Indeed the cyclization probability does not depend 
552: on~$r < 10$~nm and the free energy of forming such a loop decreases 
553: from~$15~k_\mathrm{B}T$ for~$r=1$~nm to~$8~k_\mathrm{B}T$ for~$r=10$~nm only 
554: because the reaction volume increases by a 
555: factor~$3 \times \ln{(10)} \approx 7~k_\mathrm{B}T$. 
556: %
557: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
558: %
559: \section{Effect of the Presence of a Kink}\label{s:kink} 
560: % 
561: The previous protein mediated DNA~looping modelization (section~\ref{s:size}) 
562: assumes that the only intervening proteins are the ones clamping the loop ends 
563: (\emph{e.g.}, the Lac~operon repressor~LacR). Actually regulation phenomena 
564: involve several proteins which may bind along the whole~DNA. Indeed naked~DNA 
565: situations barely exist \emph{in vivo}. For instance single molecule 
566: manipulations~\cite{lia} have shown that efficient Gal~operon repression needs 
567: a stiffness loss of the 113~bp DNA portion to be looped. The HU~protein produces 
568: such loss by kinking the sequence. 
569: %
570: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 
571: % 
572: \subsection{Numerical Calculation of~$J$ for a Bridged and Kinked Loop}
573: %
574: Such stiffness loss may be taken into account in an effective way by kinking 
575: the~WLC at half-length~$L/2$. Let us call $\theta _-$ and $\theta_+$ the angles
576: of the~DNA just before and after the kink respectively. We assume that the kink 
577: plane is vertical and choose it to define the origin of the azimuthal 
578: angle~$\phi$: $\phi_- = \phi_+ = 0$. Using the quantum language of 
579: section~\ref{secnum}, we replace the calculation of the evolution 
580: operator~$Z(L, f)$ (\ref{defz2}) with its kinked counterpart
581: %
582: \begin{eqnarray}
583: Z_\mathrm{kinked}(L, f) & = & 
584: \langle \mathrm{final} \vert 
585: \exp{\! \left[-\frac{L}{2A} \, \widehat{H}(f)\right]} 
586: \vert \theta _+, \phi_+ \rangle 
587: \times
588: \langle \theta_-, \phi_- \vert 
589: \exp{\! \left[-\frac{L}{2A} \, \widehat{H}(f)\right]} 
590: \vert \mathrm{initial} \rangle 
591: \nonumber \\ 
592:  & = & 
593: \sum_{\ell, \ell'} 
594: \langle 0, 0 \vert 
595: \exp{\! \left[-\frac{L}{2A} \, \widehat{H}(f)\right]} 
596: \vert \ell, 0 \rangle 
597: \, 
598: \langle \ell', 0 \vert 
599: \exp{\! \left[-\frac{L}{2A} \, \widehat{H}(f)\right]} 
600: \vert 0, 0 \rangle 
601: \times Y_\ell^0 (\theta_+, 0) \, Y_{\ell'}^0 (\theta_-, 0) 
602: \end{eqnarray}
603: % 
604: where we have used the change of basis from angles to spherical harmonics 
605: % 
606: \begin{equation}
607: \vert \theta_\pm, \phi_\pm = 0 \rangle = 
608: \sum_{\ell \ge 0} Y_\ell^0(\theta_\pm, 0) \vert \ell, 0 \rangle. 
609: \end{equation}
610: % 
611: Although calculations are a little bit more involved the evaluation scheme 
612: for the cyclization factor~$J$ remains unchanged in its 
613: principle~(section~\ref{secnum}). We now have ``$\vert \ell\ne 0,0 \rangle$ 
614: elements'' corresponding to particular orientations arriving at~($-$) and 
615: leaving from~($+$) $s = L/2$. The kink angle~$\kappa$ is intuitively defined 
616: from these WLC~tangent vectors at half-length (in spherical coordinates) 
617: through 
618: % 
619: \begin{equation}
620: \kappa = \theta_+ - \theta_-. 
621: \end{equation}
622: %
623: \begin{figure}
624: \begin{tabular} {@{\hspace{-2.5cm}}r@{\hspace{-8.4cm}}c%
625: @{\hspace{-8.4cm}}c@{\hspace{-8.4cm}}l} 
626: \includegraphics[height=9cm]{Cspa-0.3-180-x2.eps} & 
627: \includegraphics[height=9cm]{Cspa-0.3-150-x2.eps} & 
628: \includegraphics[height=9cm]{Cspa-0.3-120-x2.eps} & 
629: \includegraphics[height=9cm]{Cspa-0.3-090-x2.eps} 
630: \end{tabular}
631: \caption{Lowest bending energy configurations for a~100~bp DNA~loop with an 
632: end-to-end extension~$r = 10$~nm and kinks~$\kappa = 90^\circ$, $120^\circ$, 
633: $150^\circ$, $180^\circ$ respectively. The gray configurations are the closed 
634: loops  used in the calculation of the 
635: fluctuation.} 
636: \label{confkink}
637: \end{figure}
638: % 
639: \begin{figure}
640: \begin{tabular} {rl} 
641: \includegraphics[height=8cm, angle=-90]{jr10num.eps} &
642: \includegraphics[height=8cm, angle=-90]{dgr10num.eps} \\
643: \includegraphics[height=8cm, angle=-90]{jr10appcol.eps} &
644: \includegraphics[height=8cm, angle=-90]{dgr10appcol.eps}
645: \end{tabular}
646: \caption{Closure factor (left) and free energy for a loop with~$r = 10$~nm 
647: and a kink~$\kappa$ in the middle of the chain: $\kappa=180^\circ$~(full lines); 
648: $\kappa = 150^\circ$~(dotted lines); $\kappa = 120^\circ$~(dashed line); 
649: $\kappa = 90^\circ$~(long dashed lines). Top: numerical calculation of the 
650: constrained path integral. Bottom: extension of the~\textsc{Shimada} 
651: and~\textsc{Yamakawa} calculation~(black lines) and approximate formula~(gray 
652: line) given in the text~(\ref{for}). Numerical results are in very good 
653: agreement with the SPA~approximation and the approximate formula~(\ref{for}).
654: Inset: brownian dynamics simulations point obtained by the~\textsc{Langowski} 
655: and collaborators~(empty square, $\Box$)~\cite{rippe95, merlitz} , fitted by a 
656: simple formula by~\textsc{Rippe}~\cite{rippe00}.} 
657: \label{jr10}
658: \end{figure}
659: % 
660: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
661: % 
662: \subsection{Saddle Point Approximation of~$J$ for a Bridged and Kinked Loop} 
663: % 
664: The saddle-point calculation of section~\ref{s:sp1} can be straightforwardly 
665: extended to the case of a kinked loop. In Fig.~\ref{confkink} we show the 
666: configurations with the lowest bending energy for a~100~bp DNA~loop with an 
667: end-to-end extension~$r=10$~nm. The kink is accounted for by a bending angle 
668: in the middle of the chain~$\kappa = 2 \times \theta(L/2) - \pi$ \emph{a priori} 
669: different from the previous trivial value~$\kappa=\pi$, ranging from~$150^\circ$ 
670: to~$90^\circ$. We introduce the phase~$\psi = \arcsin{\! \left[\sin{\! %
671: \left(\frac{\pi-\kappa}{4}\right)}/x\right]}$. The parameter~$x$ is now obtained 
672: from equation 
673: %
674: \begin{equation}
675: \left(1 + r/L\right) \; 
676: \left[\widehat{K} \! \left(x^2\right) - K \! \left(\psi, x^2\right)\right] = 
677: 2 \left[\widehat{E} \! \left(x^2\right) - E \! \left(\psi, x^2\right)\right] 
678: \end{equation}
679: %
680: and the total bending energy is 
681: %
682: \begin{equation}\label{delk}
683: \beta \Delta E(r, L, \kappa) = \frac{4}{L/A} \left[\widehat{K} \! \left(x^2\right) 
684: - K \! \left(\psi, x^2\right)\right]^2 \left(2 x^2 - 1 + r/L\right). 
685: \end{equation}
686: %
687: In analogy with~(\ref{kc}) we obtain the end-to-end extension~$\vec r$~PDF 
688: %
689: \begin{equation}\label{jrlk}
690: Q(r, L, \kappa) = C_\mathrm{SY}(L + 2r) \, 
691: \exp{\! \left[-\beta \Delta E(r, L, \kappa)\right]}, 
692: \end{equation}
693: %
694: where~$C_\mathrm{SY}$ is given in~(\ref{c}), from wich we 
695: calculate~$J(r, L, \kappa)$ through formula~(\ref{defjr}). Note that~(\ref{jrlk}) 
696: reduces to the loop probability given in~\cite{sankararaman} for a closed and 
697: kinked~DNA. Again the good agreement obtained with numerical results allows us 
698: to establish a semi-analytical formula for the loop probability with a finite 
699: interacting volume and kink. 
700: %
701: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 
702: %
703: \subsection{Results for a Kinked and Bridged Loop}\label{ss:kink}
704: % 
705: In Fig.~\ref{jr10} results for the looping probability density~(left) and free 
706: energy (right) are shown for a typical end-to-end extension~$r=10$~nm and 
707: kinks~$\kappa = 150^\circ$, $120^\circ$, $90^\circ$. The curve with no 
708: kink~($\kappa = 180^\circ$) is also shown for comparison. The numerical results 
709: on the top of the figure are in very good agreement with
710:  the extension of the~\textsc{Shimada} 
711: and~\textsc{Yamakawa} saddle point approach on the bottom of the figure~(black 
712: lines). In Fig.~\ref{jr10}~(inset) we show the results obtained by BD~simulations 
713: by~\textsc{Langowski} and~\emph{al.}~\cite{rippe95, merlitz} 
714: for~$J(10~\mathrm{nm}, L, \kappa)$~(empty squares, $\Box$) fitted 
715: by~\textsc{Rippe}~\cite{rippe00} with a simple formula containing one fitting 
716: parameter for each curve. The curves in the inset of Fig.~\ref{jr10} reproduce the 
717: same  behavior with~$L$ and~$\kappa$ of our numerical~(top of Fig.~\ref{jr10}) or 
718: SPA~curves~(bottom of Fig.~\ref{jr10}). The numerical gap with BD~results increases 
719: for large kinks: at~$\kappa =90^\circ$ our closure factor is ten times larger than 
720: the closure factor obtained by BD~simulations. The electrostatic and twist rigidity effect 
721: could indeed play a more important role when the chain is kinked. Note that the 
722: lengths range of the numerical results are from~100~bp to~500~bp, 
723: while the lengths range of the saddle point results is from~75~bp to 1500~bp. 
724: The lengths range of the~SPA is limited by the validity of the approximation 
725: $\approx 1500$~bp shown in Fig.~\ref{jyagaus}. As an example the closure 
726: factor~$J(10~\mathrm{nm}, 113~\mathrm{bp}, \kappa)$ of a~113~bp fragment, 
727: obtained by the numerical calculations, increases from the value of~$10^{-9}$~M 
728: for a non kinked loop~($\kappa=180^\circ$) to~$4 \, 10^{-8}$~M, $2 \, 10^{-6}$~M 
729: and~$4 \, 10^{-5}$~M for respectively~$\kappa = 150^\circ$, $120^\circ$ 
730: and~$90^\circ$. The corresponding looping free energy is 
731: $\Delta G(10~\mathrm{nm}, 113~\mathrm{bp}, \kappa) = 13~k_\mathrm{B}T$, 
732: $9~k_\mathrm{B}T$, $5~k_\mathrm{B}T$ and $2.5~k_\mathrm{B}T$ 
733: for~$\kappa = 180^\circ$, $150^\circ$, $120^\circ$ and $90^\circ$ respectively. 
734: With the saddle point approach we find similar results: 
735: $J(10~\mathrm{nm}, 113~\mathrm{bp}, \kappa) = 2 \, 10^{-9}$~M, $10^{-7}$~M, 
736: $5 \, 10^{-6}$~M and $8 \, 10^{-5}$~M while 
737: $\Delta G(10~\mathrm{nm}, 113~\mathrm{bp}, \kappa) = 11~k_\mathrm{B}T$, 
738: $8~k_\mathrm{B}T$, $4.4~k_\mathrm{B}T$ and $1.7~k_\mathrm{B}T$. As it is shown 
739: in Fig.~\ref{jr10} the presence of the kink become irrelevant for DNA~segment 
740: larger than about~1500~bp. The example of a 113~bp DNA~segment has been chosen 
741: to compare the results with the single molecule experiments on the~GalR mediated 
742: loop of an~$\approx 113$~bp DNA~portion between the two operators. From the 
743: kinetics of the loop formation mediated by the~GalR and~HU proteins, \textsc{Lia} 
744: and~\emph{al.}~\cite{lia} have deduced a looping free energy of~$12~k_\mathrm{B}T$, 
745: that with respect to our values should correspond to a kink angle of more 
746: than~$150^\circ$ or to an end-to-end extension smaller than~$r= 10$~nm. Another 
747: significant change in the  cyclization probability is that the stiffness loss 
748: induced by~$\kappa$ reduces the most probable loop length from~500~bp for a non 
749: kinked~DNA to 340~bp and 190~bp~(from the numerical calculation) or to 300~bp 
750: and 150~bp (from SPA) for kinks of respectively $150^\circ$ and $120^\circ$. 
751: Note that for a kinked loop with an end-to-end extension~$r$ the minimal 
752: length~$L_0$ corresponding to the rigid rod-like configuration fulfills the 
753: relation~$L_0 \sin(\kappa/2) = r$ and therefore it is of~$\approx 42$~bp 
754: for~$\kappa = 90^\circ$ instead of~$\approx 30$~bp for~$\kappa = 180^\circ$. 
755: for~$\kappa = 90^\circ$ the most probable loop length is the rigid \emph{kinked} 
756: rod-like configuration of the two half-DNA portions. To catch both kink and protein 
757: bridge effects in a simple formula, we have calculated the cyclization factor with 
758: the extension of~\textsc{Shimada} and~\textsc{Yamakawa} formula for a kinked closed 
759: loop of length~$(L + 2r)$. This approach is similar to what was suggested en~1991 by 
760: \textsc{Brenowitz} and~\emph{al.}~\cite{brenowitz} to interpret their experimental 
761: data, \emph{i.e.} to directly consider the protein as part of the length of the loop. 
762: A linear fit~\cite{nr} of the bending energy~(\ref{delk}) for the optimal 
763: closed configuration~($r = 0$) in the presence of a kink (expressed in 
764: degrees):~$\beta \Delta E(r = 0, L, \kappa) \approx \left(-7.1 %
765: + 0.1155 \, \kappa\right)/(L/A)$, is shown in Fig.~\ref{eangoli}. It gives 
766: the following approximated formula for the closure factor as a function of the 
767: protein size~$r$, the length~$L$ of the~DNA, and the kink angle~$\kappa$ 
768: %
769: \begin{equation}\label{for}
770: J_\mathrm{approx} (L, r, \kappa) = C_\mathrm{SY}(L + 2r) \; 
771: \exp{\! \left[\frac{7.1 - 0.1155 \, \kappa}{(L+2r)/A}\right]}
772: \end{equation}
773: %
774: where~$C_\mathrm{SY}$ is given in~(\ref{c}). Notice the integration 
775: step~(\ref{defjr}) has been skipped since it does not make any significant 
776: difference. Formula~(\ref{for}) allows us to obtain a simple prediction for 
777: the loop probability in presence of a kink in the middle 
778: of the sequence and a finite separation between the extremities. As shown in 
779: Fig.~\ref{jr10} this formula~(gray lines) is in good agreement 
780: with the loop probability obtained with the exact calculation of the saddle point 
781: energy of the open configuration~(full line). In particular Fig.~\ref{jr10} 
782: shows that for kink angles in the range~$90^\circ < \kappa < 150^\circ$ this 
783: simple formula works remarkably well for lengths~$L$ larger than about $5r$, 
784: that is 150~bp for $r=10$~nm. For smaller lengths the optimal configuration 
785: is more a rigid rod-like and cannot be approximated by a closed loop. Similar 
786: simple formulas that includes a kink angle~$\kappa$ and a finite end-to-end 
787: distance~$r$ in an effective way have also been written down by~\textsc{Rippe} 
788: or~\textsc{Ringrose} to fit their brownian dynamics simulation~\cite{rippe00} 
789: or experimental data~\cite{ringrose}, but these formulas contain a parameter 
790: that must be fitted for each values of~$r$ and~$\kappa$ from the data points
791: (Fig.~\ref{jr10}). 
792: %
793: \begin{figure}
794: \includegraphics[height=12cm, angle=-90]{eangoli.eps}
795: \caption{The ``$\times$'' points: saddle point energy in units of~$L/A$~(that 
796: is~$\varepsilon = \Delta E \times L/A$) for the saddle point 
797: configurations~(also displayed in the figure) with kink angles 
798: of~$\kappa = 90^\circ$, $120^\circ$, $150^\circ$, $180^\circ$. Dotted line: 
799: linear interpolation used in formula~(\ref{for}).}
800: \label{eangoli}
801: \end{figure}
802: %
803: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 
804: %
805: \section{Conclusion}\label{s:conclusion} 
806: %
807: We performed both numerical and analytical calculations of the closure factor~$J$, 
808: even in the presence of a protein bridge and of a protein-mediated kink. More 
809: precisely we have numerically calculated the path integral of the WLC~polymer 
810: model under the constraints of a fixed end-to-end distance~$r$ and a kink~$\kappa$ 
811: in the middle of the DNA~portion. Analytically we have extended 
812: the~\textsc{Shimada} and~\textsc{Yamakawa} saddle point approximation~\cite{shimada} 
813: to the case of a bridged and kinked loop. We have seen that the formation of DNA 
814: loops is significantly sensitive to the size of the protein bridge when this 
815: size~$r$ is more than~$10\%$ of the loop length~$L$, that is~300~bp~(or~100~nm) 
816: for a typical protein bridge size of~$r=10$~nm. To give an example, the closure 
817: factor for a 100~bp DNA~segment increases 
818: from~$J(100~\mathrm{bp}, 0) \approx 10^{-11}$~M 
819: to~$J(100~\mathrm{bp}, 10~\mathrm{nm}) \approx 10^{-9}$~M. Correspondingly, 
820: looping free energy decreases from~$\Delta G(100~\mathrm{bp}, 0) = 24~k_\mathrm{B}T$ 
821: to~$\Delta G(100~\mathrm{bp}, 10~\mathrm{nm}) = 13~k_\mathrm{B}T$. A kink ranging 
822: from~$150^\circ$ to~$90^\circ$ produces a significant increase of~$J$ for 
823: DNA~fragments of lengths up to about~2500~bp. For instance the closure factor for 
824: a 100~bp DNA~segment with a protein bridge~$r = 10$~nm and a kink of~$90^\circ$ 
825: is~$J(100~\mathrm{bp}, 10~\mathrm{nm}, 90^\circ) \approx 10^{-4}$~M, and the 
826: corresponding looping free energy~$\Delta G(100~\mathrm{bp}, 10~\mathrm{nm}, %
827: 90^\circ) \approx 2~k_\mathrm{B}T$. A kink also changes the most probable loop 
828: length from 500~bp~(no kink) to about 175~bp for a kink of~$120^\circ$, 
829: going to the rigid \emph{kinked} rod-like configuration for smaller~$\kappa$ values. 
830: This is an interesting mechanism because the loop lengths implied in \emph{in vivo} 
831: DNA~processing by proteins spread on a large lengths range. Our results were 
832: compared to previous analytical approximations~(in particular the results of the 
833: gaussian model, the~\textsc{Wilhelm} and~\textsc{Frey} expansion~\cite{wilhelm} 
834: and the~\textsc{Shimada} and~\textsc{Yamakawa} formula~\cite{shimada}) and 
835: numerical calculations~(in particular the Monte~Carlo simulations data obtained 
836: by~\textsc{Podtelezhnikov} and~\textsc{Vologodskii}~\cite{podtelezhnikov}, 
837: and the brownian dynamics simulations data obtained by \textsc{Langowski} 
838: and~\emph{al.}~\cite{rippe95, merlitz, rippe00})  as well as 
839: experimental results~\cite{brenowitz, finzi, lia} (\emph{e.g.}, the ones obtained 
840: by \textsc{Lia} and~\emph{al.} on the looping dynamics mediated by the~Gal and~HU 
841: proteins). Finally a simple formula~(\ref{for}) including both the protein bribge 
842: and kink effects has been proposed. This formula has the advantage of not containing 
843: adjustable parameters with respect to the existing formulas that include both these 
844: effects~\cite{rippe00}. \\ 
845: % 
846: Still many effects omitted in this work can be included without significant changes 
847: in the numerical algorithm. The kink we considered is actually permanent~(that is 
848: not thermally excited), site specific~(at half-length) and rigid~($\kappa$~fixed). 
849: Although this rigidity seems relevant to most protein bindings to~DNA at first 
850: glance~\cite{popov}, it was pointed out in~\textsc{Yan} and~\emph{al.} 
851: works~\cite{yan03, yan05}, kinks may also be semi-flexible~(exhibiting higher or 
852: lower rigidities than the bare~DNA) or even fully flexible~\cite{yan04, wiggins}. 
853: For instance, the HU/DNA~complex was recently observed to be very flexible under 
854: specific experimental conditions~\cite{noort}. Flexible hinges were also stated to 
855: occur along the~DNA due to the opening of small denaturation 
856: bubbles~\cite{yan04, ranjith}, such as the one needed by~HU to fit in the double 
857: helix~\cite{lia}. Such flexibility could be taken into account~\cite{popov} in our 
858: model. This kind of defects could also be thermally activated, occuring at multiple 
859: non-specific sites along the~DNA~\cite{yan03, yan05,wiggins, popov, chakrabarti}. 
860: Both effects may be included in our model. Note that using effective persistence 
861: lengths could turn out convenient, despite these inform little about the kink 
862: properties~(number, location, rigidity, etc). Actually this would be equivalent to 
863: study DNA~stiffness loss due to sequence effects~\cite{nelson} by cutting~WLC in 
864: different stiff fragments, depending on the~CG or~AT content of the whole sequence 
865: to cyclize. The same approach may allow an approximative study of the~DNA 
866: polyelectrolyte nature too~\cite{zandi}. Otherwise electrostatic potential has to 
867: be included in WLC~energy. Twist elasticity leads to slight modifications of the 
868: quantum analog we used although requiring some care~\cite{bouchiat}. This is 
869: expected to play an important role in looping, especially when specific alignment 
870: of the loop extremities are required. Finally cyclization dynamics could be modeled 
871: using a simple two states model where~DNA is ``closed'' or ``opened'', that is 
872: cyclized or not. Such study relies on the~(statics) cyclization factor we computed 
873: in this article~\cite{jun}. Direct comparison to experimental lifetimes measures 
874: would be possible~\cite{finzi, lia}. 
875: %
876: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 
877: % 
878: \begin{acknowledgments}
879: The authors would like to thank D.~\textsc{Chatenay}, R.~\textsc{Monasson} 
880: and F.~\textsc{Thalmann} for useful discussions and a critical reading of 
881: the manuscript. 
882: \end{acknowledgments}
883: %
884: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 
885: %
886: \begin{thebibliography}{}
887: %
888: \bibitem{shore} D.~\textsc{Shore}, J.~\textsc{Langowski}, 
889: and R.~L.~\textsc{Baldwin}, 
890: PNAS~\textbf{78}, 4833~(1981). 
891: % 
892: \bibitem{du} Q.~\textsc{Du}, C.~\textsc{Smith}, 
893: N.~\textsc{Shiffeldrim}, M.~\textsc{Vologodskaia}, 
894: and A.~\textsc{Vologodskii}, 
895: PNAS~\textbf{102}, 5397~(2005). 
896: %
897: \bibitem{brenowitz} M.~\textsc{Brenowitz}, A.~\textsc{Pickar}, 
898: and E.~\textsc{Jamison}, 
899: Biochemistry~\textbf{30}, 5986~(1991). 
900: %
901: \bibitem{finzi} L.~\textsc{Finzi}, and J.~\textsc{Gelles}, 
902: Science~\textbf{267}, 378~(1995). 
903: % 
904: \bibitem{lia} G.~\textsc{Lia}, D.~\textsc{Bensimon}, 
905: V.~\textsc{Croquette}, J.-F.~\textsc{Allemand}, 
906: D.~\textsc{Dunlap}, D.~E.~A.~\textsc{Lewis}, 
907: S.~\textsc{Adhya}, and L.~\textsc{Finzi}, 
908: PNAS~\textbf{100}, 11373~(2003). 
909: % 
910: \bibitem{focus} J.-F.~\textsc{Allemand}, S.~\textsc{Cocco}, 
911: N.~\textsc{Douarche}, and G.~\textsc{Lia}, 
912: Eur.~Phys.~J.~B (to be published). 
913: %
914: \bibitem{kratky} O.~\textsc{Kratky}, and G.~\textsc{Porod}, 
915: Rec.~Trav.~Chim.~\textbf{68}, 1106~(1949). 
916: % 
917: \bibitem{yamakawa} H.~\textsc{Yamakawa}, 
918: \emph{Helical Wormlike Chains in Polymer Solutions} 
919: (Springer-Verlag, Berlin, 1997). 
920: %
921: \bibitem{kleinert} H.~\textsc{Kleinert}, 
922: \emph{Path Integrals in Quantum Mechanics, Statistics, 
923: Polymer Physics, and Financial Markets} 
924: (World Scientific Publishing Co., Singapore, 2004). 
925: %
926: \bibitem{daniels} H.~E.~\textsc{Daniels}, 
927: Proc.~Roy.~Soc.~Edinburgh~A~\textbf{63}, 290~(1952). 
928: %
929: \bibitem{gobush} W.~\textsc{Gobush}, H.~\textsc{Yamakawa}, 
930: W.~H.~\textsc{Stockmayer}, and W.~S.~\textsc{Magee}, 
931: J.~Chem.~Phys.~\textbf{57}, 2839~(1972). 
932: %
933: \bibitem{hanke} A.~\textsc{Hanke}, and R.~\textsc{Metzler}, 
934: Biophys.~J.~\textbf{85}, 167~(2003). 
935: %
936: \bibitem{balaeff} A.~\textsc{Balaeff}, L.~\textsc{Mahadevan}, 
937: and K.~\textsc{Schulten}, 
938: Phys.~Rev.~Lett.~\textbf{83}, 4900~(1999). 
939: %
940: \bibitem{wilhelm} J.~\textsc{Wilhelm} and E.~\textsc{Frey}, 
941: Phys.~Rev.~Lett.~\textbf{77}, 2581~(1996). 
942: %
943: \bibitem{shimada} J.~\textsc{Shimada}, and H.~\textsc{Yamakawa}, 
944: Macromolecules~\textbf{17}, 689~(1984). 
945: 
946: \bibitem{podtelezhnikov} A.~A.~\textsc{Podtelezhnikov}, and A.~V.~\textsc{Vologodskii}, 
947: Macromolecules~\textbf{33}, 2767~(2000).
948: %
949: \bibitem{rippe95} K.~\textsc{Rippe}, P.~H.~\textsc{von Hippel}, 
950: and J.~\textsc{Langowski}, 
951: Trends~Biochem.~Sci.~\textbf{20}, 500~(1995). 
952: %
953: \bibitem{merlitz} H.~\textsc{Merlitz}, K.~\textsc{Rippe}, 
954: K.~V.~\textsc{Klenin}, and J.~\textsc{Langowski}, 
955: Biophys.~J.~\textbf{74}, 773~(1998). 
956: %
957: \bibitem{rippe00} K.~\textsc{Rippe}, 
958: Trends~Biochem.~Sci.~\textbf{26}, 733~(2001). 
959: 
960: \bibitem{yan03} J.~\textsc{Yan}, and J.~F.~\textsc{Marko}, 
961: Phys.~Rev.~E~\textbf{68}, 011905~(2003). 
962: %
963: \bibitem{yan04} J.~\textsc{Yan}, and J.~F.~\textsc{Marko}, 
964: Phys.~Rev.~Lett.~\textbf{93}, 108108~(2004). 
965: %
966: \bibitem{yan05} J.~\textsc{Yan}, R.~\textsc{Kawamura}, 
967: and J.~F.~\textsc{Marko}, 
968: Phys.~Rev.~E~\textbf{71}, 061905~(2005). 
969: %
970: \bibitem{sankararaman} S.~\textsc{Sankararaman}, and J.~F.~\textsc{Marko}, 
971: Phys.~Rev.~E~\textbf{71}, 021911~(2005). 
972: %
973: \bibitem{samuel} J.~\textsc{Samuel}, and S.~\textsc{Sinha}, 
974: Phys.~Rev.~E~\textbf{66}, 050801(R)~(2002). 
975: %
976: \bibitem{expokit} R.~B.~\textsc{Sidje}, 
977: ACM~Trans.~Math.~Softw.~\textbf{24}, 130~(1998). 
978: The \textsf{Expokit} library is available for download at URL 
979: \texttt{http://www.maths.uq.edu.au/expokit}. 
980: %
981: \bibitem{nr} W.~H.~\textsc{Press}, S.~A.~\textsc{Teukolsky}, 
982: W.~T.~\textsc{Vetterling}, and B.~P.~\textsc{Flannery}, 
983: \emph{Numerical Recipes in C: the Art of Scientific Computing} 
984: (Cambridge University Press, 1992).
985: % 
986: \bibitem{moler} C.~\textsc{Moler}, and C.~van~\textsc{Loan}, 
987: SIAM Review~\textbf{45}, 3~(2003). 
988: % 
989: \bibitem{abramowitz} M.~\textsc{Abramowitz}, and I.~A.~\textsc{Stegun}, 
990: \emph{Handbook of mathematical functions} 
991: (Dover Publications Inc., New York, 1970). The elliptic integrals of 
992: first and second kinds respectively read in \textsc{Legendre} 
993: definition~$K\!\left(\psi, x^2\right) = \int_0^{\psi} %
994: \frac{\mathrm{d}y}{\sqrt{1-x^2 \sin^2{\!y}}}$ \linebreak 
995: and~$E\!\left(\psi, x^2\right) %
996: = \int_0^{\psi} \sqrt{1-x^2 \sin^2{\!y}} \,\, \mathrm{d}y$. 
997: %
998: \bibitem{ringrose} L.~\textsc{Ringrose}, S.~\textsc{Chabanis}, 
999: P.-O.~\textsc{Angrand}, C.~\textsc{Woodroofe}, 
1000: and A.~F.~\textsc{Stewart}, 
1001: EMBO~J.~\textbf{18}, 6630~(1999). 
1002: % 
1003: \bibitem{popov} Y.~O.~\textsc{Popov}, and A.~V.~\textsc{Tkachenko}, 
1004: Phys.~Rev.~E~\textbf{71}, 051905~(2005). 
1005: %
1006: \bibitem{wiggins} P.~A.~\textsc{Wiggins}, R.~\textsc{Phillips}, 
1007: and P.~C.~\textsc{Nelson}, 
1008: Phys.~Rev.~E~\textbf{71}, 021909~(2005). 
1009: %
1010: \bibitem{noort} J.~V.~\textsc{Noort}, S.~\textsc{Verbrugge}, 
1011: N.~\textsc{Goosen}, C.~\textsc{Dekker}, and R.~T.~\textsc{Dame}, 
1012: PNAS~\textbf{101}, 6969~(2004). 
1013: %
1014: \bibitem{ranjith} P.~\textsc{Ranjith}, P.~B.~\textsc{Sunil}~\textsc{Kumar}, 
1015: and G.~I.~\textsc{Menon}, 
1016: Phys.~Rev.~Lett.~\textbf{94}, 138102~(2005). 
1017: %
1018: \bibitem{chakrabarti} B.~\textsc{Chakrabarti}, and A.~J.~\textsc{Levine}, 
1019: Phys.~Rev.~E~\textbf{71}, 031905~(2005). 
1020: %
1021: \bibitem{nelson} P.~\textsc{Nelson}, 
1022: Phys.~Rev.~Lett.~\textbf{80}, 5810~(1998).
1023: %
1024: \bibitem{zandi} R.~\textsc{Zandi}, J.~\textsc{Rudnick}, 
1025: and R.~\textsc{Golestanian}, 
1026: Phys.~Rev.~E~\textbf{67}, 021803~(2003). 
1027: % 
1028: \bibitem{bouchiat} C.~\textsc{Bouchiat}, and M.~\textsc{M\'ezard} 
1029: Phys.~Rev.~Lett.~\textbf{80}, 1556~(1998); 
1030: V.~\textsc{Rossetto}, and A.~C.~\textsc{Maggs}, 
1031: \emph{ibid.}~\textbf{88}, 089801~(2002) [comment]; 
1032: C.~\textsc{Bouchiat}, and M.~\textsc{M\'ezard}, 
1033: \emph{ibid.}~\textbf{88}, 089802~(2002) [reply]. 
1034: %
1035: \bibitem{jun} S.~\textsc{Jun}, J.~\textsc{Bechhoefer}, 
1036: and B.-Y.~\textsc{Ha}, 
1037: Europhys.~Lett.~\textbf{64}, 420~(2003). 
1038: %
1039: \end{thebibliography}
1040: %
1041: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 
1042: %
1043: \end{document}
1044: