NIH Public Access

Author Manuscript
Topol Appl. Author manuscript; available in PMC 2009 November 17.
Published in final edited form as: Topol Appl. 2007 April 1; 154(7): 13811397. doi:10.1016/j.topol.2006.05.010.

NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript

Random state transitions of knots: a first step towards modeling unknotting by type II topoisomerases
Xia Hua2, Diana Nguyen3, Barath Raghavan4, Javier Arsuaga5, and Mariel Vazquez1
Xia Hua: h0399923@graduate.hku.hk; Diana Nguyen: diananguyen@ucla.edu; Barath Raghavan: barath@cs.ucsd.edu; Javier Arsuaga: jarsuaga@sfsu.edu; Mariel Vazquez: mariel@math.sfsu.edu
1 2 3 4

Mathematics Department, San Francisco State University, San Francisco, CA Mathematics Department, The University of Hong Kong, Hong Kong David Geffen School of Medicine, UCLA, Los Angeles, CA

Computer Science and Engineering Department, University of California at San Diego, San Diego, CA
5

Mathematics Department and Center for Computing in the Life Sciences, San Francisco State University, San Francisco, CA

Abstract
Type II topoisomerases are enzymes that change the topology of DNA by performing strand-passage. In particular, they unknot knotted DNA very efficiently. Motivated by this experimental observation, we investigate transition probabilities between knots. We use the BFACF algorithm to generate ensembles of polygons in Z3 of fixed knot type. We introduce a novel strand-passage algorithm which generates a Markov chain in knot space. The entries of the corresponding transition probability matrix determine state-transitions in knot space and can track the evolution of different knots after repeated strand-passage events. We outline future applications of this work to DNA unknotting.

Keywords Topoisomerase II; DNA knots; polymer models; BFACF algorithm; random knotting

1 Introduction
Biological systems pose a wealth of problems that give rise to very interesting mathematical questions. We are interested in the process of DNA unknotting by type II topoisomerases (topo II). Motivated by this biological problem, we study state transitions within knot space for all prime knots with 8 or fewer crossings and fixed length. We model each knot of type K and length L as a polygonal chain in the simple cubic lattice (Z3). We explore the space of configurations of type K and length L e using the Monte Carlo BFACF algorithm, first proposed by Berg and Foerster [4], and Argao de Carvalho, Caracciolo and Frlich [2]. The BFACF algorithm generates a reducible Markov Chain whose ergodicity

Correspondence to: Mariel Vazquez, mariel@math.sfsu.edu. Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Hua et al.

Page 2

classes are the knot types [30][36]. We generate a list of Dowker-Thistlethwaite (DT) codes [18][26] for each knot type K by taking multiple projections in random directions for each knotted configuration (section 4). We perform a random-strand-passage simulation on the obtained lists of DT codes, and at each step we use the HOMFLY polynomial for knot identification (section 5). For fixed L and e this process results in a transition probability matrix PLe which models random transitions in knot space. We generate PLe for several values of L and e, and find their corresponding steady state distribution vectors. We compare our results to those of Flammini et al. [21] and use our simulation to verify and improve, results on minimal number of sticks needed to realize a knot in Z3 [35]. Finally, in section 7 we outline limitations of our method and future directions.

NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript

2 Biological motivation
Certain enzymes such as site-specific recombinases and topoisomerases are able to change the topology of circular DNA molecules by changing their knotting, linking and supercoiling levels [37,63,11,27,29,61,48]. The mechanisms of several site-specific recombinases have been successfully studied using knot theory and low-dimensional topology to investigate the topological changes performed by these enzymes [19,13,55,8]. We here study state-transitions within knot space, a problem motivated by the action of DNA unknotting by type II topoisomerases (topo II). Random cyclization of linear DNA results in DNA circles that are knotted with some probability. This probability depends on the length and effective helical diameter of the DNA chain [45,44,28]. For instance, for DNA molecules 10.5 kilobases long free in solution, the knotting probability is about 0.03, with a predominance of trefoils. Furthermore, Arsuaga et al. [3] showed that knots are the most likely configuration of DNA molecules in confined volumes. Confinement of DNA inside P4 viral capsids increased the knotting probability to 0.95. However, in vitro studies have shown that DNA knots are known to inhibit important cellular processes such as transcription and chromatin assembly [40,32]. These findings suggest that DNA knots are undesirable for the cell [46,16]. The cell is able to solve these topological obstructions (i.e. DNA knots) by means of some enzymes called Type II topoisomerases. The molecular mechanism of topo II is fairly well understood [37,39,5,6,49]: topo II traps two segments of the DNA molecule, cuts one of the segments, performs a strand-passage and reseals the broken segment. It has been shown that topo II unknots and unlinks DNA very efficiently [43], i.e. it does it beyond expected steady-state levels obtained by performing random strandpassage. A number of models have been proposed to explain the efficiency of topology simplification by topo II [7,21,38,52,53,60,62,50]. Our long-term goal is to find an accurate model for the mechanism of unknotting by topo II. The first step towards this goal is to model DNA unknotting by random strand-passage. In this paper we set the theoretical framework for such modeling by studying the effects of random strand-passage on polygonal knots with crossing number Cr(K) < 9 [31,21]. Flammini et al. [21] investigated transition probabilities within knot space by modeling single phantom crossings using a modified crank-shaft algorithm on freely-jointed polygonal chains R3. In our approach we represent knots as polygonal chains in the simple cubic lattice (Z3) and strand-passage is done at the Dowker code level. A comparison with [21] is done in section 6. Traditionally, circular DNA molecules have been modeled in the computer as polygonal chains and their behavior in solution has been accurately simulated using Monte Carlo simulations in 3-space [58,56,59,57,54,22,42,15]. We here consider those models in the simple cubic lattice. We work in Z3 because it has a number of advantages over R3. There is extensive analytical work that investigates the knotting behavior of lattice chains. There is a remarkably solid
Topol Appl. Author manuscript; available in PMC 2009 November 17.

Hua et al.

Page 3

mathematical foundation for the development of algorithms that generate equilibrium distributions of self-avoiding polygons in Z3 [30]. Also, polygons in Z3 inherently have volume exclusion (which can be used to model the effective diameter of DNA). However a simplistic lattice model is not well-suited for DNA studies. In order to apply our results to DNA we will need to adjust our theoretical model as proposed in [54]; we leave a detailed biological application of our work to a future publication. In section 7, we discuss steps needed to achieve accurate modeling of DNA in solution as a polygonal chain in Z3, including excluded volume effects, as well as future topological biases as those suggested by fig. 1 to approximate topo II action.

NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript

3 Basic Definitions
A knot is a proper embedding of a circle in 3-space. Here we let the knot, K be the image of the circle under the embedding. Two knots K1 and K2 are topologically equivalent if they are related by an ambient isotopy. A knot diagram is a 2-dimensional projection of the knot where there are only finitely many multiple points, all multiple points are double and can be resolved into over and undercrossings. Knots can be studied through their knot diagrams [41]. Polygonal knot diagrams have the extra constraint that no vertex is mapped to a double point. The DT code of a knot diagram is an integer vector from which the knot diagram and the knot type can be recovered [18]. The DT code of the diagram is obtained by selecting a starting point and an orientation. The knot diagram is followed along the orientation and each crossing is labeled with consecutive numbers. At the end of the process, each crossing is labeled with one even and one odd number. The numbers are stored in a 2 n matrix where the first row lists the odd numbers in lexicographical order and the second row lists the corresponding even numbers. This second row forms the DT code with the convention that an entry is negative if it was assigned to a crossing while traversing the knot on the understrand. The DT code representation is used in computational studies due to its simplicity. In particular, DT codes can be used to compute most polynomial invariants of knots and links (e.g.[20,26]). Furthermore, note that a change in sign in one of the DT code entries corresponds to a change of crossing in the knot diagram (from and undercrossing to an overcrossing), and therefore to a strand-passage in the 3-dimensional configuration. This key feature of the DT code, added to its computational advantages, make it particularly attractive for our study. We aim to generate lists of DT codes for each knot type which will be representative of a polygonal chain as it samples its state space. To this end we first generate equilibrium distributions of knotted polygons using Markov chains. Here we introduce Markov chains and leave the details of our work to sections 46. A discrete time Markov process is a family of random variables {Xn: n } where the Xn take values in a subset of S . S is called the state space and each element of S, Xn, is called a state, or configuration, of the process. A Markov process is an homogeneous Markov chain if P(Xn = i | X0, X1 Xn1) = P(Xn = i | Xn1) and P (Xn = i | Xn1 = j) = P (X1 = i | X0 = j) = pij. The pijs are called transition probabilities and the square matrix whose entries are the transition probabilities is called the transition matrix which we denote by P. The evolution of the Markov chain is given by the n-step transition matrix P(Xn = i | P0 = j). By the ChapmanKolmogrov equations P(Xn = i | P0 = j) = Pn (i.e. the evolution of the chain is determined by repeatedly multiplying P with itself). If we denote P(Xn = i) by , and , then it is a well known fact that un = u0Pn. Therefore the evolution of the chain is determined by P and u0. In other words, Markov chains are determined by the probabilities of their states [25].

Topol Appl. Author manuscript; available in PMC 2009 November 17.

Hua et al.

Page 4

A state j is accessible from state i if pij > 0. Accessibility defines an equivalence relation. A Markov Chain is said to be irreducible is there is only one equivalence class. Otherwise it is said to be reducible. The period of a state d(i) is the . A Markov chain is aperiodic if the gcd{d(i) | i S} = 1. A state is positive recurrent if the probability of going back to the same state on a finite number of iterations is non-zero. A Markov chain is irreducible if for some n and all i, j. A Markov chain is ergodic if it is irreducible, positive recurrent, and aperiodic. ) exists. A vector u is called stationary distribution For ergodic Markov chains the ( of the chain if u = {ui | i S} is such that ui 0 and j(uj) = 1 and uj = uipij. If the state space S is finite, then the stationary distribution can be obtained analytically. In this case u is obtained by computing the left eigenvector of P corresponding to the largest eigenvalue, which is 1 (all the other real eigenvalues are strictly smaller than 1).

NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript

4 Knots in Z3 and the BFACF Algorithm

The simple cubic lattice Z3 is a graph whose vertices are the points in R3 with integer coordinates, and whose edges connect each pair of vertices, which are a unit distance apart. A self-avoiding polygon in Z3 (SAP) is a connected subgraph of the lattice with all vertices of degree 2. SAPs in Z3 may be knotted. Knotting of SAPs has been extensively studied both from a theoretical and a computational point of view [47,51]. In our study, we generate SAPs in Z3 using the BFACF algorithm [4,2]. The BFACF algorithm was first applied to self-avoiding walks in [1,2] and has been used to measure a number of properties of polygons in Z3 such as the writhe of knots and links, the radius of gyration of specific knot types, and the minimal number of sticks needed to realize a knot type in Z3 [3335]. The BFACF algorithm is a dynamic Monte Carlo method that operates on paths in Zd and samples along the realization of a Markov Chain [30]. In 3-D, the Monte Carlo method generates a Markov Chain that is reducible and whose ergodicity classes are the knot types [30,36]. More specifically the algorithm samples from the Gibbs distribution with one adjustable parameter z, the fugacity per bond, where 0 z zc, and where is the connective constant for Z3. The connective constant is estimated at = 4.6834066 0.0002 in the cubic lattice, and therefore 0 z 0.2134. We apply BFACF to unrooted SAPs in Z3 as explained in [30,35], and produce a set of 3-dimensional conformations representative of each knot type with 8 or less crossings. The Markov Chain is generated as follows. BFACF selects an edge at a random location and performs one of the three elementary moves shown in fig. 2. Move 1 changes the length of the polygon by (2), move 2 by (2), and move 3 leaves the length fixed. Each of these moves is selected with a probability given by p(2), p(2), p(0), respectively. For each edge there are four possible lattice directions to choose from, and not all three moves are possible. If the probability of the four possibilities adds up to q < 1 then we choose to leave the walk unchanged with probability 1 q. The BFACF algorithm is well defined and is reversible with respect to the Gibbs distribution if and only if the following constraints are satisfied:

Topol Appl. Author manuscript; available in PMC 2009 November 17.

Hua et al.

Page 5

Furthermore, large values of p(2), p(0), and p(+2) would minimize the probability of null transitions (1 q) and thus improve the algorithms efficiency. It was shown by Caracciolo, Pelissetto, and Sokal that the following choice is close to optimal [10]. Where z is an adjustable parameter called fugacity parameter. Varying z results in a change in the mean length of the sampled conformations. It is proved in [30] that in 3-dimensions the best choices for the probabilities are:

NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript

The algorithm changes the length of the polygon as illustrated in fig. 3. Varying the fugacity parameter z results in a change in the mean length of the sample confirmation. The mean length is finite for z < z0 [30]. In our simulations, z [0.06, 0.1975] corresponds to an expected mean length of 56 and varies with the knot type. Similarly z [0.023, 0.185] corresponds to 44, z [0.1535, 0.2048] corresponds to 75 and z [0.1825, 0.2081] corresponds to 100. In fig. 3, we can observe that the variance of the chain lengths is very large. Since our future goal is to simulate DNA chains of fixed length, we restricted our sampling to chains contained in intervals [Le, L+e]. We computed the standard deviation for each mean length and each knot type. We chose values for e that were roughly one standard deviation. More specifically we sampled from the following intervals: [40,48], [52,60], [70, 80], and [96, 104]. In BFACF, consecutive conformations along the Markov Chain are highly correlated. This correlation diminishes by decreasing the sampling frequency between samples, reaching a minimum for uncorrelated samples. Correlation along the Markov Chain for the chains length was computed using time-series analysis methods [30]. Auto-correlation analysis determined the sampling frequency. Once the correct number of iterations Nk and the sampling frequency Fk were determined for each knot type K, the BFACF algorithm was run Nk times and configurations were sampled every Fk iterations. Of the sampled configurations, only those within the [L e, L + e] target length were saved. For example for lengths 56 4 and the K = 61 knot, we used Nk = 100 million iterations, and Fk = 3 * 5000. For all knots used in this study, a Nk with 100million Nk 300million iterations, and a frequency Fk with 5000 Fk 55 * 5000 iterations were used, and we caimed for an integrated auto-correlation time of 0.5. After binning for lengths in the [Le, L+e] range, the number of sampled configurations ranged from 1061 (for K = 812, N812 = 300 million, F812 = 55 * 5000) to 16226 (for K = 52, N52 = 200 million, F52= 1 * 5000). The BFACF simulations on unknotted chains were particularly challenging as we had to raise the z value very close to the critical value z0 = 0.2134 in an effort to approach the expected mean length. This happened due to the fact that the minimum number of segments needed to realize an unknot in Z3 is 4 and during BFACF runs 4-segment configurations appeared repeatedly which kept the mean length from reaching its target. However, since we were only interested in samples within the range [Le, L+e], we set another lower bound for the length higher that 4, but still considerably smaller than the mean length. To ensure that our simulations were not substantially biased, we varied the lower bound between 20 and 35 and kept all other settings unchanged. We obtained similar knot distributions after performing one strand-passage on the unknots. For example, after performing one strandpassage on the sample of unknots with mean length 56 we observed the following: for the sample with lower bound 20, 88.43% remained unknotted and 10.16% became trefoils; whereas for the sample with lower bound 35, 88.42% remained unknotted and 10.27% became trefoils.

Topol Appl. Author manuscript; available in PMC 2009 November 17.

Hua et al.

Page 6

5 DT Code Representations
The BFACF algorithm produced a list of chain confirmations for each prime knot K with length in the interval [L e, L + e] and 8 or less crossings. Each knotted polygon sampled along the realization of the Markov Chain was projected in 50100 random directions. The corresponding knot diagrams and associated DT codes were computed. We thus obtained a list of DT codes for each knot type K. For non-trivial knots, the lists of DT codes were further processed as follows. A high percentage of Reidemeister moves of types I and II (RI, RII; illustrated in fig. 4 as 1 and 2, respectively) were identified and simplified. The simplification is motivated by our future goal of modeling circular DNA chains. In the experiments that we wish to model the DNA is usually in relaxed form and therefore presents a limited number of RI loops. For the RII moves, it can always be assumed that a rigid rotation in 3-space may remove the RII move in the projection without altering the configuration. Failure to remove RI and RII moves resulted in DT codes with many entries, which may not represent the reality of the knotted DNA. After simplification, the set of DT codes were stored for each knot. In this way, proper weights were assigned to each of the DT codes associated to the configurations along the realization of the Markov Chain defined by the BCFAF algorithm. Fig. 5 shows the weights for different DT code lengths associated to a polygonal knot of type 61 and length 56 4 segments in Z3. Removing RI and RII moves for unknotted configurations results in a set of projections with no crossings which do not realistically represent the 3-dimensional configuration. Therefore, in the case of the unknot we opted to not remove RI or RII moves. The lists of DT codes produced by BFACF for a knot K may include some DT codes of length much larger than the crossing number of K. This is true even after removing RI and RII moves. For each knot type K, the probability of occurrence of large DT codes decreases with the number of entries (i.e. with the crossing number of the chosen projection; see fig. 5). For large DT codes 1-crossing changes rarely dramatically increase the crossing number of the resulting knot. In future work we expect to expand our algorithm to knots with 9 crossings. The study for 10 or more crossings becomes combinatorially intractable. We therefore choose to restrict our state space to the set of all prime knots with 8 or less crossings. Occasionally one strandpassage (described in the next section) results in a knot with crossing number greater than 9, in such case we reject the configuration.

NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript

6 Strand-Passage Simulation
Once a representative sample of DT codes for each knot type K (and length range L e) was found, we proceeded to estimate the transition probabilities by single strand-passage between knot types. Transitions between states in knot space define a Markov Chain, whose behavior is given by the transition probability matrix PLe as described below. For each knot type K, we computed the transition probabilities after one strand-passage as follows. We selected 30,000 samples (or 60,000 depending on the knot) at random from the list of DT codes for K. A single strand-passage was performed at a crossing selected at random on each DT code by simply changing the sign of the crossing. Each strand-passage led to a new knot type K2, which was identified by computing the HOMFLY polynomial [23] with an implementation based on the algorithm of Gouesbet et al. [24]. The HOMFLY polynomial distinguishes between a knot and its mirror image. For each length and knot type we repeated the simulation taking 30,000 random samples several times and computed mean and standard deviation for each transition probability. The standard deviation was typically 10% of the mean. Occasionally we increased the sample size, however shorter the sample sizes correspond

Topol Appl. Author manuscript; available in PMC 2009 November 17.

Hua et al.

Page 7

to faster HOMFLY calculations which have a great impact on the overall efficiency of the simulation.

NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript

Our working set consisted of all prime knots with 8 or less crossings. We rejected all strandpassages taking a knot K to a knot outside the working set, which happened only a small percentage of the time for small knots. For example, knots with 6 or less crossings resulted in a HOMFLY polynomial outside the working set < 1.1% of the time for mean length L = 44, < 3% for L = 75, and < 4% for L = 100. The percentages were always < 20% for 7-crossing knots, and < 30% for all 8-crossing knots except 85, 810, and 815 (and their mirror images). Large values may be indicative of the presence of a knot outside the working set whose HOMFLY polynomial coincides with a knot in the set. For short chains (e.g. [40,48] and [52,60]) the rejection is justified as we know that at those lengths very few knots with more than 9 crossings can be tied in Z3 (cf.table 1) [35]. As the mean length increases, the rejection rate increases consistent with the fact that the probability of reaching a higher crossing knot is larger. The strand-passage simulation was repeated for each knot type thus providing all the entries to the matrix PLe of transition probabilities for the Markov Chain. Table 2 shows a modified P1005 where a knot and its mirror image were combined. Knots are indicated by their knot type, appended by the letter a (average) for chiral pairs. The one-step transitions and their probabilities (as computed in Table 2) are illustrated in Fig. 6. Figure 7 shows the evolution of the twist knot 61 after 1, 2, 3, 4 and 8 single strand-passages. The data corresponds to mean length L = 100. The first column shows the distribution of knots, as represented by their crossing number, after one strand-passage: 26% of the knots went to the unknot, 44% to the figure-8 knot (41), 21% to 6-crossing knots and 9% to 8-crossing knots. No knots were converted into trefoils or 5-crossing knots after one strand-passage. The 2nd (3th, 4th) column shows the knot distribution after 2 strand-passages (3, 4, respectively). The fifth column illustrates the distribution after 8 strand-passages. Notice that after 8 strandpassages most of the original knots have been unknotted. In figure 8 we see the behavior of these transitions as the length of the chain changes. On the left we observe that after one strandpassage, the larger the chain the higher is the probability of transitioning into a knot other than the unknot. The absence of 8-crossing knots from the L = 44 column is due to the fact that it is very hard to ties any of such knots in Z3 with 4048 segments. On the right we can observe that after 8 strand-passages most of the chains have been unknotted, and that the knotting probability increases with the chains length. The transition probability matrix (before combining a knot with its mirror image) is a stochastic matrix: it has non-negative entries, and the row sums are equal to one. Given a stochastic matrix P, let Pn be the matrix of n-step transition probabilities where pij(n) is a probability of going from knot i to knot j after n single strand-passages. By applying the Chapman-Kolmogorov equations we compute the stationary distribution of the Markov Chain correpoding to P by finding the limit of Pn as n goes to infinity. Analytically, this can be achieved by computing the eigenvector corresponding to the largest eigenvalue for PLe, which in this case is 1. The stationary distribution is illustrated in fig. 9. Intuitively, this means that after repeated strandpassages the Markov Chain reaches equilibrium. In figure 10 we show a comparison in the steady-state knot probabilities for prime knots with 6 or less crossings, and for polygons of varying lengths. In concordance with other studies of random knotting (e.g. [3,59]) we observe in the figure that the probability of the trefoil (not shown) was higher than that of the figure-8 knot, which was higher than that of the 5-crossing, and the 6-cossing knots. Noteworthy the probability of 52 is higher than that of 51. This is also in qualitative agreement with experimental results of random DNA knotting (e.g. [45,44,42]).

Topol Appl. Author manuscript; available in PMC 2009 November 17.

Hua et al.

Page 8

In biology, the stationary state is obtained for DNA knots by performing random closure of DNA molecules in solution.

NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript

We compared our results with published data of Flammini et al. [21]. In [21] knots with 6 or less crossings were modeled as freely-jointed polygonal chains in R3, and strand-passage was simulated in R3. In order to compare our data, we counted all transitions into a knot with 6 or more crossings in a column labeled > 6. The first observation is that the diagonal values in our matrix were approximately 10 times smaller than those in [21], ours being lower. For example for mean length 100 the probability of going from 31 to 31 was 0.04, while in [21] it was 0.34; for 41 we obtained 0.045 versus 0.26 in [21]; for 51 we obtained 0.021 versus 0.23 in [21]. This discrepancy can be attributed to the fact that we remove RI moves prior to performing strandpassage. Intuitively a knot with RI loops will have a much higher probability to stay in the same knot type as one for which RI moves have been removed. Preliminary runs performed without removing RI moves prior to strand-passage gave the expected 10-fold increase in the diagonal values. Despite the discrepancies we observed a remarkable agreement in the 1-step transition probabilities of certain knots with length in [96, 104]. Fig. 11 shows the comparison for the 61 knot. The 1-step transition probabilities for knots with 6 crossings all showed remarkable agreement with the data of [21]. This suggests that despite their qualitative differences (R3 vs Z3; strand-passage in 3 dimensions vs strand-passage at the DT code level), both models are able to consistently descrbe transitions of certain polymer chains by single strand-passage. The differences are reflected in full at the steady-state. Greater discrepancies arise when comparing n-step transitions as n goes to infinity. The knotting probabilities at the steady state in our lattice model are much higher than those detected experimentally, and those estimated using polymer models in R3. This is partially explained by the removal of Reidemeister moves used in our model, and by the inherent rigidity of the lattice. This problems can be overcome (work in preparation) by developing a model for DNA analogous to that of [54].

7 Applications to Minimal Lattice Knots and the Strand-passage Metric

In order to validate our implementation of BFACF, we computed the length of minimal lattice knots for each knot type with 8 crossings or less in Z3 and compared them to [34]. We reproduced, and sometimes improved, the lower bounds of [34] (table 1). Our results mostly agree with those previously given in [35]. In a few instances (i.e. 83, 88, 810, 812) the minimum number of segments is less than the one previously estimated (table 2 and [35]. The strand-passage metric, defined by Darcy [12], computes the minimal number of strandpassage steps (over all projections) needed to go from one knot type to another. One interesting question is how often the paths determined by the strand-passage metric are traveled when random strand-passage is performed on knot populations. Preliminary results show that with our method we can remove some ambiguities from the existing strand-passage metric table. For example we see 2-step transitions between knots 41 and 51, while in the original table it was unclear whether the strand-passage metric between these two knots was two or three. Likewise we observe 3-step transitions between 41 and 71, versus 3 4 on the table. We leave detailed analysis of the strand-passage metric to a future publication.

8 Conclusions and Future Directions

Motivated by the problem of DNA unknotting by topoII we have here created a framework based in knot theory and Monte Carlo simulations to study state transitions in knot space. In our study the state space is the set of all polygonal chains Z3 of length L e and knot type K, where all K are prime knots with 8 or less crossings. We performed simulations for lengths in the intervals [40,48], [52,60], [70, 80], and [96, 104]. We determine the transition probabilities

Topol Appl. Author manuscript; available in PMC 2009 November 17.

Hua et al.

Page 9

between states. Our method uses the BFACF algorithm to generate ensembles of polygons with identical knot type and mean length L. For each K we generate a list of configurations of type K and length in the chosen interval, representative of the knot moving in space. By taking random projections, we produce a list of DT codes for each knot. We introduce a new algorithm to simulate strand-passage, which produces a table of transition probabilities (table 1). Our strand-passage algorithm is very fast but somewhat simplistic. For instance, accuracy of the results depends on the faithfulness of the list of DT-codes to represent each knot type (guaranteed in this case by the properties of the BFACF algorithm). Ensuring that the transition from the 3-dimensional object to the DT representation keeps the conformational information is a challenge. Intuitively two strands that are far apart in space will produce a crossing in a projection with low probability, and therefore by taking a sufficient number of 3-dimensional conformations for each knot type, and by taking sufficient random projections, we guarantee that DT-codes with such crossings have very small weight in the strand-passage simulation. This approach serves as a framework to model the non-random unknotting behavior of topo II. The representation of the computational data shown in Figure 7 can be compared directly to the experimental results of J. Mann (Zechiedrich lab, Baylor College of Medicine) where a population of n-crossing twist knots (for fixed n) is incubated with topo II and the products are harvested after a single strand-passage. We aim to compare our simulation results with their data when it becomes available. In order to do this our model needs to be improved to realistically simulate DNA. Our first challenge is to simulate long enough polygons in Z3 with the bending properties of DNA [54] and reproduce computational and experimental data on random knotting probabilities [45,28]. The stationary state in our simulations currently results in knotting probabilities considerably larger than those experimentally observed for DNA in solution. The knotting probabilities that we observe ranged from 0.12 for mean length 44, to 0.18 for mean length 100. As the length increases the knotting probability increases which indicates a much higher level of knotting that that observed when modeling random knotting in R3, or that for DNA in solution (0.97 for polygonal chains of 33 Kuhn lengths, or 10.5kb DNA plasmids) [45,44,28].

NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript

Acknowledgments
The authors would like to thank the referee for a careful revision of our manuscript which led to major improvements. We are particularly grateful to Miki Suga and Janella Slaga for their contributions to this project, to G. Gouesbet for providing the algorithm and code for computing the HOM-FLY polynomial, and to S. Whittington for making the auto-correlation code available to us. We acknowledge Kathleen Phu for her help type-setting the manuscript and Mark Contois for his help with table 2. We also thank R. Sachs, D. W. Sumners, J. Roca, J. Mann, S. Whittington, C. Soteros and A. Stasiak for their valuable comments. Research was supported through NSF grant DMS 9971169 and SFSU Center for Computing in the Life Sciences grant program (J.A.). M.V. was supported through NASA NSCOR04-0014-0017, a Focus-Avengoa research award and NIH-RIMI grant NMD000262 (M.V.). B.R. was partially supported by a NSF graduate research fellowship. Other support came from a URAP summer stipend (D.N.) and NIH R01-GM68423 (X.H.).

References
1. Argao de Carvalho C, Caracciolo S. A new Monte Carlo approach to the critical properties of selfavoiding random walks. J Phys (Paris) 1983;44:323331. 2. Argao de Carvalho C, Caracciolo S, Frlich J. Polymers and g4 theory in four dimensions. Nucl Phys B 1983;215:209248. 3. Arsuaga J, Vazquez M, Trigueros S, Sumners de W, Roca J. Knotting probability of DNA molecules confined in restricted volumes: DNA knotting in phage capsids. Proc Natl Acad Sci U S A 2002;99:53735377. [PubMed: 11959991] 4. Berg B, Foerster D. Random paths and random surfaces on a digital computer. Phys Lett B 1981;106 (4):323326.

Topol Appl. Author manuscript; available in PMC 2009 November 17.

Hua et al.

Page 10

5. Berger JM, Gamblin SJ, Harrison SC, Wang JC. Structure and mechanism of DNA topoisomerase II. Nature 1996;379:225232. [PubMed: 8538787] 6. Berger JM, Wang JC. Recent developments in DNA topoisomerase II structure and mechanism. Curr Opin Struct Biol 1996;6:8490. [PubMed: 8696977] 7. Buck GR, Zechiedrich EL. DNA disentangling by type-2 topoisomerases. J Mol Biol 2004;340:933 939. [PubMed: 15236957] 8. Buck D, Verjovsky Marcotte C. Tangle solutions for a family of DNA-rearranging proteins. Math Proc Cambridge Philos Soc 2005;139(1):5980. 9. Calvo JA. Geometric knot spaces and polygonal isotopy. J Knot Theory Ramifications 2001;10:245 267. 10. Caracciolo S, Pelissetto A, Sokal AD. A general limitation on Monte Carlo algorithms of Metropolis type. Phys Rev Lett 1994;72(2):179182. [PubMed: 10056079] 11. Champloux JJ. DNA topoisomerases: structures, function, and mechanism. Annu Rev Biochem 2001;70:369413. [PubMed: 11395412] 12. Darcy I. Rational Tangle Distances on Knots and Links. Math Proc Cambridge Philos Sco 2000;128:497510. 13. Darcy I. Biological distances on DNA knots and links: applications to XER recombination. Knots in Hellas 1998;2DelphiJ Knot Theory Ramifications 2001;10(2):269294. 14. Darcy I. Solving unoriented tangle equations involving 4-plats. J Knot Theory Ramifications. to appear 15. Deguchi T, Tsurasaki K. A statistical study of random knotting using the Vassiliev invariants, random knotting and linking. J Knot Theory Ramifications 1994;3:321353. 16. Deibler RW, Rahmati S, Zechiedrich EL. Topoisomerase IV, alone, unknots DNA in E. coli. Genes Dev 2001;15:748761. [PubMed: 11274059] 17. Diao Y. Minimal knotted polygons on the cubic lattice. J Knot Theory Ramifications 1993;2:413 425. 18. Dowker CH, Thistlethwaite MB. Classification of Knot Projections. Topol Appl 1983;16:1931. 19. Ernst C, Sumners DW. A calculus for rational tangles: applications to DNA recombination. Math Proc Cambridge Philos Soc 1990;108(3):489515. 20. Ewing, B.; Millett, KC., editors. Progress in knot theory and related topics. Vol. 56. Travaux en Cours. Paris: Hermann; 1997. Computational algorithms and the complexity of link polynomials. 21. Flammini A, Maritan A, Stasiak A. Simulations of action of DNA topoisomerases to investigate boundaries and shapes of spaces of knots. Biophys J 2004;87:29682975. [PubMed: 15326026] 22. Frank-Kamenetskii MD, Lukashin AV, Vologodskii AV. Statistical mechanics and topology of polymer chains. Nature 1975;258:398402. [PubMed: 1196369] 23. Freyd P, Yetter D, Hoste J, Lickorish WBR, Millett K, Oceanu A. A New Polynomial Invariant of Knots and Links. Bull Amer Math Soc 1985;12:239246. 24. Gouesbet G, Meunier-Guttin-Cluzel S, Letellier C. Computer evaluation of Homfly polynomials by using Gauss codes, with a skein-template algorithm. Appl Math Comput 1999;105(23):271289. 25. Grimmett, GR.; Stirzaker, DR. Probability and Random Processes. Oxford Science; Oxford: 1982. 26. Hoste J, Thistlethwaite M, Weeks J. The first 1,701,935 knots. Math Intelligencer 1998;20:3348. 27. Hsieh T. Knotting of the circular duplex DNA by type II DNA topoisomerase from Drosophilia melanogaster. J Biol Chem 1983;258:84138420. [PubMed: 6305983] 28. Klenin KV, Vologodskii AV, Anshelevich VV, Dykhne AM, Frank-Kamenetskii MD. Effect of excluded volume on topological properties of circular DNA. J Biomol Struct Dyn 1988;5 5:1173 1185. [PubMed: 3271506] 29. Liu LF, Davis JL, Calendar R. Novel topologically knotted DNA from bacteriophage P4 capsids: studies with DNA topoisomerases. Nucleic Acids Res 1981;9:39793989. [PubMed: 6272191] 30. Madras, N.; Slade, G. Probability and its Applications. Birkhauser; Boston: 1996. The Self-Avoiding Walk. 31. Millett, KC. Knots in Hellas 98. Vol. 24. World Scientific; 1998. Monte Carlo explorations of polygonal knot spaces.

NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript

Topol Appl. Author manuscript; available in PMC 2009 November 17.

Hua et al.

Page 11

32. Portugal J, Rodriguez-Campos A. T7 RNA polymerase cannot transcribe through a highly knotted DNA template. Nucleic Acids Res 1996;24:48904894. [PubMed: 9016657] 33. Van Rensburg EJ, Orlandini E, Sumners DW, Tesi MC, Whittington SG. The writhe of knots in the cubic lattice. J Knot Theory Ramifications 1997;6:3144. 34. Van Rensburg EJ, Promislow SD. Minimal knots in the cubic lattice. J Knot Theory Ramifications 1995;4:115130. 35. van Rensburg, EJ. Ideal Knots. Vol. 19. World Scientific; Singapore: 1998. Minimal lattice knots. 36. van Rensburg EJ, Whittington SG. The BFACF algorithm and knotted polygons. J Phys A: Math Gen 1991;24:55535567. 37. Roca J. The mechanisms of DNA topoisomerases. Trends Biochem Sci 1995;20:156160. [PubMed: 7770916] 38. Roca J. Varying levels of positive and negative supercoiling differently affect the efficiency with which topoisomerase II catenates and decatenates DNA. J Mol Biol 2001;305:441450. [PubMed: 11152602] 39. Roca J, Berger JM, Harrison SC, Wang JC. DNA transport by a type II topoisomerase: direct evidence for a two-gate mechanism. Proc Natl Acad Sci U S A 1996;93:40574062. [PubMed: 8633016] 40. Rodriguez-Campos A. DNA knotting abolishes in vitro chromatin assembly. J Biol Chem 1996;271:1415014155. [PubMed: 8662864] 41. Rolfsen D. Knots and Links. Publish or Parish. 1976 42. Rybenkov VV, Cozzarelli NR, Vologodskii AV. Probability of DNA knotting and the effective diameter of the DNA double helix. Proc Natl Acad Sci U S A 1993;90:53075311. [PubMed: 8506378] 43. Rybenkov VV, Ullsperger C, Vologodskii AV, Cozzarelli NR. Simplification of DNA topology below equilibrium values by type II topoisomerases. Science 1997;277:690693. [PubMed: 9235892] 44. Shaw SY, Wang JC. DNA knot formation in aqueous solutions. J Knot Theory Ramifications 1994;3:287298. 45. Shaw SY, Wang JC. Knotting of a DNA chain during ring closure. Science 1993;260:533536. [PubMed: 8475384] 46. Shishido K, Ishii S, Komiyama N. The presence of the region on pBR322 that encodes resistance to tetracycline is responsible for high levels of plasmid DNA knotting in Escherichia coli DNA topoisomerase I deletion mutant. Nucleic Acids Res 1989;17:97499759. [PubMed: 2557587] 47. Soteros CE, Sumners DW, Whittington SG. Entanglement complexity of graphs in Z3. Math Proc Cambridge Philos Soc 1992;111 48. Stark WM, Sherratt DJ, Boocock MR. Site-specific recombination by Tn3 resolvase: topological changes in the forward and reverse reactions. Cell 1989;58:779790. [PubMed: 2548736] 49. Stasiak A. DNA topology: feeling the pulse of a topoisomerase. Curr Biol 2000;10:R526528. [PubMed: 10898993] 50. Stone MD, Bryant Z, Crisona NJ, Smith SB, Vologodskii A, Bustamante C, Cozzarelli NR. Chirality sensing by Escherichia coli topoisomerase IV and the mechanism of type II topoisomerases. Proc Natl Acad Sci U S A 2003 Jul 22;100(15):86548659. [PubMed: 12857958] 51. Sumners DW, Whittington SG. Knots in Self-Avoiding Walks. J Phys A: Math Gen 1988;21:1689 1694. 52. Szafron, M. Masters thesis. University of Saskatoon; 2000. Monte Carlo Simulations of Strandpassage in Unknotted Self-Avoiding Polygons. 53. Trigueros S, Salceda J, Bermudez I, Fernandez X, Roca J. Asymmetric removal of supercoils suggests how topoisomerase II simplifies DNA topology. J Mol Biol 2004;335:723731. [PubMed: 14687569] 54. Tesi MC, Janse Van Rensburg EJ, Orlandini E, Sumners DW, Whittington SG. Knotting and supercoiling in circular DNA: A model incorporating the effect of added salt. Physical Review E. 1994 55. Vazquez M, Sumners DW. Tangle analysis of Gin site-specific recombination. Math Proc Cambridge Philos Soc 2004;136(3):565582.

NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript

Topol Appl. Author manuscript; available in PMC 2009 November 17.

Hua et al.

Page 12

56. Vologodskii AV, Anshelevich VV, Lukashin AV, Frank-Kamenetskii MD. Statistical mechanics of supercoils and the torsional stiffness of the DNA double helix. Nature 1979;280:294298. [PubMed: 460401] 57. Vologodskii AV, Cozzarelli NR. Monte Carlo analysis of the conformation of DNA catenanes. J Mol Biol 1993;232:11301140. [PubMed: 8371271] 58. Vologodskii AV, Frank-Kamenetskii MD. Modeling supercoiled DNA. Methods Enzymol 1992;211:467480. [PubMed: 1406321] 59. Vologodskii AV, Levene SD, Klenin K, Frank-Kamenetskii MD, Cozzarelli NR. Conformational and thermodynamic properties of supercoiled DNA. J Mol biol 1992;227:12241243. [PubMed: 1433295] 60. Vologodskii AV, Zhang W, Rybenkov VV, Podtelezhnikov AA, Subramanian D, Griffth JD, Cozzarelli NR. Mechanism of topology simplification by type II DNA topoisomerases. Proc Natl Acad Sci U S A 2001;98:30453049. [PubMed: 11248029] 61. Wasserman SR, Cozzarelli NR. Biochemical topology: applications to DNA recombination and replication. Science 1986;232:951960. [PubMed: 3010458] 62. Yan J, Magnasco MO, Marko JF. A kinetic proofreading mechanism for disentanglement of DNA by topoisomerases. Nature 1999;401:932935. [PubMed: 10553912] 63. Zechiedrich EL, Khodursky AB, Cozzarelli NR. Topoisomerase IV, not gyrase, decatenates products of site-specific recombination in Escherichia coli. Genes Dev 1997;11:25802592. [PubMed: 9334322]

NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript

Topol Appl. Author manuscript; available in PMC 2009 November 17.

Hua et al.

Page 13

NIH-PA Author Manuscript NIH-PA Author Manuscript

Figure 1.

Strand-passage on twist knots.

NIH-PA Author Manuscript

Topol Appl. Author manuscript; available in PMC 2009 November 17.

Hua et al.

Page 14

NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript

Topol Appl. Author manuscript; available in PMC 2009 November 17.

Figure 2.

BFACF moves.

Hua et al.

Page 15

NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript

Topol Appl. Author manuscript; available in PMC 2009 November 17.

Figure 3.

Conformation lengths for the 61 knot after BFACF runs. The fugacity parameter z was chosen so that the mean chain length would be 56. The band bounded by red lines determine the range of sampling ([52,60]).

Hua et al.

Page 16

NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript

Topol Appl. Author manuscript; available in PMC 2009 November 17.

Figure 4.

Reidemeister Moves I and II denoted as 1 and 2, respectively.

Hua et al.

Page 17

NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript

Topol Appl. Author manuscript; available in PMC 2009 November 17.

Figure 5.

Distribution of the lengths of the Dowker codes corresponding to the 61 knot. The data was obtained by performing the BFACF algorithm with z= 0.153, for 100 million iterations, sampling every 15,000 iterations, binning for chain lengths in [52,60]. The BFACF was followed by 100 projections in random directions, and RI and RII move removal.

Hua et al.

Page 18

NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript

Topol Appl. Author manuscript; available in PMC 2009 November 17.

Figure 6.

Directed graph illustrating a modified transition matrix P where the results for a knot and its mirror image were averaged. The nodes correspond to a knot type and its mirror image, the edges correspond to one strand-passage between two states. The data presented here correspond to those on table 2. The knot length was in the interval [96, 104]. Only prime knots with 7 or less crossings are shown, and they are organized by crossing number along the y axis and by writhe alon the x-axis.

Hua et al.

Page 19

NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript

Topol Appl. Author manuscript; available in PMC 2009 November 17.

Figure 7.

Unknotting of the knot 61 with length in the interval [96, 104] by random strand-passage.

Hua et al.

Page 20

NIH-PA Author Manuscript

Figure 8.

One-step transitions (left) and unknotting after 8 steps (right) of the 61 knot as a function of the chains length.

NIH-PA Author Manuscript NIH-PA Author Manuscript

Topol Appl. Author manuscript; available in PMC 2009 November 17.

Hua et al.

Page 21

NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript

Topol Appl. Author manuscript; available in PMC 2009 November 17.

Figure 9.

Knotting distribution at the steady state for knots with length in [96, 104].

Hua et al.

Page 22

NIH-PA Author Manuscript NIH-PA Author Manuscript

Figure 10.

Knotting distributions at the steady state as a function of chain length.

NIH-PA Author Manuscript

Topol Appl. Author manuscript; available in PMC 2009 November 17.

Hua et al.

Page 23

NIH-PA Author Manuscript NIH-PA Author Manuscript

Figure 11.

Comparison of the knot 61 transitions with results of [21].

NIH-PA Author Manuscript

Topol Appl. Author manuscript; available in PMC 2009 November 17.

Table 1

Minimum length of knots in

Z3; Comparison with van Rensburgs results [34] (see text). Inprovements are highlighted in bold.

Knot Types van Rensburg Our Algorithm Knot Types van Rensburg Our Algorithm 31 24 24 86 50 50 41 30 30 87 48 48 51 34 34 88 52 50 52 36 36 89 50 50 61 40 40 810 52 50 62 40 40 811 52 50 63 40 40 812 54 52 71 44 44 813 50 50 72 46 46 814 54 50 73 44 44 815 52 52 74 46 44 816 52 50 75 46 46 817 52 52 76 46 46 818 52 52 77 44 44 819 42 42 81 50 50 820 46 46 82 50 50 821 46 46 83 52 48 91 54 54 84 50 50 92 56 56 85 50 50 101 60

NIH-PA Author Manuscript

Hua et al. Page 24

NIH-PA Author Manuscript

NIH-PA Author Manuscript

Topol Appl. Author manuscript; available in PMC 2009 November 17.

Table 2
51a 52a 61a 62a 63 71a 72a 73a 74a 75a 76a 77a 81a 82a 83 84a 85a 86a 87a 88a 89 810a 811a 812 813a 814a 815a 816a 817a 818a 819a 820a 821a

knot typeunknot

31a

41

unknot 31a 41 51a 52a 61a 62a 63 71a 72a 73a 74a 75a 76a 77a 81a 82a 83 84a 85a 86a 87a 88a 89 810a 811a 812 813a 814a 815a 816a 817a 818a 819a 820a 821a

0.852 0.780.842 00.3120.2660.168 0.33 00.242 0 0 00.1530.3250.214 0 0 0 0 00.137 00.266 00.141 00.2550.137 0 00.269 0 00.2580.515 0.061 0.04 00.3180.206 00.1270.245 0 0 00.0790.1190.118 0.08 00.072 0 0 00.1030.0720.173 0 0 0 0 00.071 00.1310.1350.492 0 0 0 0.022 00.045 0 00.4430.339 0 0 0 0 0 00.2440.488 0 0 00.131 0 0 0 0 0 0 00.4370.1330.216 00.137 0 0 0 0 0 00.027 00.0210.061 0 0 00.3420.0050.1560.004 0.12 0 0 00.108 0 0 0 00.104 0 00.134 0 00.008 00.168 0 0 00.4440.012 0.04 0.0020.035 00.0980.018 0 0 0 00.2460.2010.3210.1540.121 0 0 0 0 0 0 0 00.106 00.0880.108 00.1010.1110.2720.069 0 0 00.195 0 0 00.027 0 00.0510.045 0 0 0 0 0 0 0 00.258 00.3470.141 0 0.14 0 0 0 00.1090.221 0 0 0 0 0 0 0 0 0 00.0040.024 0 00.0540.013 0 0 0 0 0 0 0 0 00.217 00.1810.3660.141 0 00.282 00.139 0 00.106 0 00.221 0 0 00.181 00.007 0 0 0 0 00.024 0 0 0 0 0 0 0 0 0 0 0 0 00.3650.287 00.357 0 00.267 0 00.445 0 0 00.232 0 0 0 00.016 0 0 0 00.016 00.047 0 0.03 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00.011 0 0 0 00.0130.034 00.0420.028 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00.016 0.01 0 0 00.0870.0320.0140.084 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00.009 0 0 0 0 00.0480.012 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00.0160.008 0 0 00.0540.052 0 00.012 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00.035 0 0 00.0010.002 0 0.01 0 0 0 0 0.03 0 0 00.011 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00.023 0 0 00.002 0 0 0 0 0 0 0 0 0 0 00.011 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.01 0 0 0 0 0 0 0 0 00.009 00.062 0 00.039 0 0 0 00.0250.052 0 0 0 0 0 0 0 0 0 0 0 00.002 0 00.007 0 0 0 0 0 0 0 0 0 0.01 0 0 00.023 0 0 0 00.039 0 00.026 0 0 0 0 0 0 0 0 0 0 0 00.011 0 0 0 0 0 0 0 0 00.053 00.0190.075 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00.001 0 00.0060.005 0 0 0 0 0 0 0 0 0 00.0820.009 00.021 0 00.076 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00.004 0 0 0 0 0 0 0 0 0 0 0 00.0130.024 0 0 0 0 0 0 0 0 0 0 0 0 0 00.007 0 0 0 0 00.0070.008 0 0 0 0 0 0 0 00.0530.023 00.0270.0820.009 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00.001 0 0 0 0.01 0 0 0 0 0 0 0 0 0 0 0 0 00.0090.021 0 0 0 00.035 0 0 0 0 0 0 0 0 0 0 0 00.001 0 00.015 0 0 0 0 00.003 0 0 0 0 0 0 00.0260.034 00.044 0 00.011 0 0 0 0 0 00.007 0 0 0 0 0 0 0 0.01 0 0 0 0 0 0 0 0 0 0 00.078 0 0 0 0 0.02 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00.0010.001 0 00.007 0 0 0 0 0 0 0 0 0 0 0 0 0 00.024 00.013 0 0 0 0 0 0 0 0 00.005 0 0 0 0 00.0010.0060.007 0 0 0 0 0 0 0 00.0210.045 0 0 0 0 0 0 0 0 0.01 0 0 0 0 0 0 0 0 0 0 0 00.001 0 00.014 0 0 0 0 0 0 0 0 00.051 0 0 0 0 0 0 0 0 0 00.018 0 0 0 0 0 0 0 0 0 0 0 0 00.001 0 00.009 0 0 0 0 0 0 0 0 0 0 0 0 00.038 0 0 0 0 00.009 0 0 0 0 0 0 0 0 0 0 0 00.001 00.006 0 0 0 0 0 0 0 0 00.025 0 0 0 0 0 0 0 0 0 0 0 0.01 0 0 0 0 0 0 0 0 0 00.001 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00.014 0 0 00.002 0 0 0 0 0 0 0 0 00.004 0 0 0 0 0 00.002 0 0 0 0 0 0 0 0 0 0 0 00.008 0 00.009 0 0 0 00.001 0 0 0 0 0 00.002 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.01 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00.008 0 0 0 0 0 0 0.01 0 0 0 0 0 0 0 00.022 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00.046 0 0 00.019 0 0 0 0 0 00.006 0 00.033 0 0 0 0 00.021 0 0 0 0 0 0 0 0 0 00.042 0 0 0 0 0 0 0 0 00.013 0 0 0 00.001 0 00.017 0 0 0 0 0 0 0 0 0 0 0 0 0.04 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00.013

NIH-PA Author Manuscript

Hua et al. Page 25

NIH-PA Author Manuscript

NIH-PA Author Manuscript

Topol Appl. Author manuscript; available in PMC 2009 November 17.