Spontaneous Domain Formation in Spherically-Confined Elastic Filaments

Tine Curk,1, 2, ∗ James Daniel Farrell,1, ∗ Jure Dobnikar,1, 3, 4, † and Rudolf Podgornik5, 1, ‡ 1Institute of Physics, Chinese Academy of Sciences, Beijing 100190, China. 2Faculty of Chemistry and Chemical Engineering, University of Maribor, 2000 Maribor, Slovenia 3Songshan Lake Materials Laboratory, Dongguan, Guangdong 523808, China. 4Department of Chemistry, University of Cambridge, Cambridge, CB2 1EW United Kingdom. 5School of Physical Sciences and Kavli Institute for Theoretical Sciences, University of Chinese Academy of Sciences, Beijing 100190, China. (Dated: July 2, 2019) Although the free energy of a genome packing into a virus is dominated by DNA-DNA interactions, ordering of the DNA inside the capsid is elasticity–driven, suggesting general solutions with DNA organized into spool-like domains. Using analytical calculations and computer simulations of a long elastic filament confined to a spherical container, we show that the ground state is not a single spool as assumed hitherto, but an ordering mosaic of multiple homogeneously-ordered domains. At low densities, we observe concentric spools, while at higher densities, other morphologies emerge, which resemble topological links. We discuss our results in the context of metallic wires, viral DNA, and flexible polymers.

Introduction. Spatial organization of constrained elas- liquid-crystalline order of compacted DNA was found to tic filaments underlies a range of packing problems from play a crucial role promoting the formation of knots [19]. macroscopic wires in spherical cavities [1, 2] to genomes The slow dynamics of stiff chains in confinement in bacteriophages [3, 4]. A theoretical interpretation implies that the configurations observed in the non- of single molecule experiments on DNA packaging in equilibrium simulations crucially depend on the applied viruses [5] suggests single-domain spool-like structures protocol, e.g., on how the external force is applied in with axial symmetry as a prevailing morphology. How- packaging-motor driven DNA encapsidation in bacterio- ever, recent experiments report a range of different struc- phages [14]. It is therefore not surprising that, even in tures with inhomogeneously-ordered genomes, defects, cases where the models are very similar, a wide variety and phase transitions between the packings. Cryo- of behaviours is reported. Here, we focus on the na- electron microscopy of bacteriophages [6, 7] at vari- ture of the ground-state packing configurations. While ous physicochemical conditions indicates a range of pos- the present analysis thus answers a different question sible morphologies of DNA within the capsid: from from previous works, its results should nevertheless pro- isotropically disordered, cholesteric liquid crystalline or- vide a well-defined background scenario with which non- dering with a disordered core, to uniform, concentric equilibrium configurations can be compared and assessed. spools with local hexagonal order, reminiscent of the Theoretical Model. We consider a spherical enclosure nested spools first proposed in [8]. Measurements of C with radius R0 (volume V0), and a hard semi-flexible DNA dynamics during packaging [9], and intermittent cylinder with cross-section σ, length L, and a persistence ejection dynamics [10] suggest multi-domain structures. length lp = Kc/kBT , where Kc is the bending rigidity. Similarly, experiments and simulations of quasi-one- The chain cannot overlap with itself nor with the confin- dimensional elastoplastic filaments forced into spherical ing wall and it is torsionally relaxed; its elastic energy is confinement [1, 2], as well as the theoretical description of 1 R L 2 described by the integral Eel = 2 Kc 0 ds/R (s), with elastic filaments within the continuum polymer nematic R(s) the local radius of curvature. theory [11, 12], indicate a multidomain structures with a We explore the limit of a rigid, long, thin chain, where mixture of local disorder and nematic order. the elastic energy dominates over the configurational en- tropy. In this limit, the chain can be described by the Coarse-grained simulations of confined semiflexible

arXiv:1907.00469v1 [cond-mat.soft] 30 Jun 2019 continuum theory of polymer nematics [11, 12], where polymers mimicking DNA in viral shells have been per- the polymer is described by a nematic-order vector field formed at various levels of detail. DNA interactions and or local polymer “current” t(r) [12] with ˆt = t/t the unit entropic contributions need to be considered to explain tangent vector (director) to the chain. The total amount the energetics of confinement [13, 14], but the conforma- of material is V = R t dr, with t ≡ |t| the local poly- tion of the confined DNA is dictated by elasticity [15]. p C mer density. Due to the excluded volume of the chain, Packing into icosahedral capsids was shown to have a t is bounded by the densest (hexagonal) packing of rigid strong stochastic component [16], and structures without √ m π 3 spool-like symmetry have been reported, including multi- disks in two dimensions: (t = 6 ≈ 0.907) [20]. The domain spool packing, observed when DNA is pushed ratio between the actual amount of polymer in the√ capsid 3 3σ2 into the capsid through a portal [17]. The effect of elas- and its theoretical maximum, ν ≡ Vp/(tmV0) = 3 L, 8πR0 tic kinks on packing has been discussed in [18], while local is a key dimensionless parameter in the model. 2

The polymer current is only non-zero inside the capsid (a) 1 spool (b) 2 spools (c) 3 spools (d) Hopf link and has to be solenoidal (∇ · t = 0) except at the poly- mer end-points (the contribution of which vanishes in the long-chain limit). The lowest non-trivial order in the ex- pansion of the elastic free energy of the divergence-free z polymer current is given by [11, 12] (e) y Z 2 x 2Kc   1 ˆ ˆ E, 2 spools Eel = 2 t t × (∇ × t) dr , (1) πσ V E, 3 spools

where twist deformations (ˆt · (∇ ׈t)) are by assumption 1s n 0.9 excluded. The problem is to find the solenoidal field t(r) 0 0.2 0.4 0.6 0.8 11 E / within C which minimizes Eel. 0.8 Assuming axial symmetry, the solution is an spool [20]. w 0.6 w1, 2 s. However, relaxing the symmetry constraints might lead 0.8 w , 3 s. 1 0.4 to more complex structures such as those seen in experi- w2, 3 s. Elastic energy ments [7, 9], theory [11, 12], and simulations [15, 16, 18]. 0.2 Partitioning Indeed, as we show later, a single-domain spool is al- 0 most never the lowest elastic energy configuration. For 0.7 0 0.2 0.4 0.6 0.8 1 solutions containing more than one domain, t(r) is not Packing fraction n continuous in space, but it can still be solenoidal if at discontinuity interfaces S, the local field on both sides FIG. 1. Theoretical model. Visualization of the trial struc- is perpendicular to the normal, t · n(S) = 0. Multiple- tures: single spool (a) with polymer-occupied volume blue, domain structures should therefore be considered. Their double (b) and triple (c) spools (only the inner spools shown advantage is flexibility in arranging the material so as in green and orange, respectively), and the Hopf link (d). to avoid highly curved regions close to the capsid center, The structures shown are the solutions with optimal parti- tioning w at ν = 0.9. (e): The optimal ratio of elastic en- which cannot be done in single spool structures. 2s,3s 1s ergies Eel /Eel as a function of ν for double spools (green With this in mind, we assume that the field is parti- circles) and triple spools (red diamonds). Inset: The optimal Pnd tioned into nd domains, ν = 1 νi, with relative sizes partitioning as a function of ν for double spools (w1, green wi ≡ νi/ν. For nd ≤ 3, we construct candidate struc- stars) and triple spools (w1,2, red triangles). tures and search for the optimum partitioning {mi} at several packing fractions 0 < ν < 1. Inspired by obser- q vations [1, 2, 15–17, 21], we consider four classes of trial 2/3 R0 1 − ν1 is composed of two small hemi-toroidal structures depicted in Fig. 1: single spools (a), double t caps of combined length proportional to ν2, connected spools (b) comprising two nested tori with perpendicular s t s by straight sections (ν2): ν2 = ν2 + ν2. The elastic en- axes, triply nested spools (c) with perpendicular axes, ergy is localized in the hemispheres and is determined by and—deviating from the nested-spool paradigm—Hopf νt alone (see section SI(B)): links (d), with a symmetric partitioning into two equiva- 2 lent, perpendicular, topologically-linked tori. We assume 2s 1s 1s t Eel (ν1, ν2,R0) = Eel (ν1,R0) + Eel (ν2,R1). (3) that the field density is tm within the defined domains 2s 1s and zero elsewhere and approximated (1) by a sum of in- Interestingly, at a given ν, the ratio Eel /Eel is com- tegrals over independent domains. We neglect the short pletely determined by the partitioning of the material segments connecting the domains, arguing that this con- w1. We calculate the optimal w1 for several values of ν tribution is negligible in the thin, long chain limit. by minimizing (3). The calculation for triple spools is 1s p 2 2 For a single spool, ˆt = tm (−y, x, 0) / x + y , we analogous: ν = ν1 +ν2 +ν3, and the elastic energy a sum have [5]: over two outer spools (3) and an innermost spool with an 3s elaborate shape (see section SI(C)): Eel (ν1, ν2, ν3,R0) = 2 " # 3rdspool π KR 1/3 2s 1s √ 0 1/3 1 1+ν Eel (ν1, ν2,R0) + Eel (ν1, ν2, ν3,R0). The optimal Eel (ν, R0) = log −ν + 2 ln 1−ν1/3 . (2) 3 structure is obtained by minimizing over w1 and w2. Fig. 1e) shows the ratio of the elastic energies as a The energy diverges for ν → 1, due to diverging curva- function of ν for optimal double- and triple-spool config- ture in the center of the capsid. The Hopf link comprises urations. We note that both are always lower in energy two perpendicular spools with a partitioning dictated by than a single spool, and that the triple spool wins over hl 1s symmetry (w1 = 0.5) and Eel (ν, R0) ≈ 2Eel (ν/2,R0). the double spool. Adding a fourth spool in an analogous In the double spool, the outer domain (ν1) is the same manner does not reduce the energy further (see online SI as above, and the inner spool (ν2) with radius R1 = for details). 3

Our best Hopf link structure (Fig. S4) can have lower Shanno (l-BFGS) algorithm [29, 30] as implemented in elastic energy than the single spool only at extremely low the pele package [31], converging the root-mean-squared −3  ν and even then by a tiny margin, but never wins over gradient to a tolerance of 1 × 10 σ . We further op- double or triple spools. timized the energy of selected structures with a combi- The robust conclusion within our analytical model is nation of random displacement, reptation, and l-BFGS that in the long, thin, and hard cylinder limit, multiple minimization cycles (see algorithm S1). domains are favored over a single-domain. In principle, Our theoretical analysis suggests that the elastic en- this may differ if we consider spools with non-orthogonal ergy of a polymer packed into a sphere can be minimized axes, or any other, better arrangements not included in by segregating into up to three distinct spools. Chain seg- our theoretical considerations. Moving away from this ments that belong to the same spool lie in almost parallel limit (i.e., finite σ, repulsive interactions other than ex- planes. As such, we can characterize the degree to which cluded volume), the volume accessible to the polymer two beads belong to the same spool domain by compar- will be modified, which can result in a domain-wall en- ing the binormal vectors at their positions. The binormal 0 00 ergy penalty, disfavoring multi-domain arrangements. At to the chain is given by bˆ = ˆt × ˆn = r (s)×r (s) , where finite temperature, entropic terms due to conformational |r0 (s)×r00 (s)| fluctuations of the chain also become important. We will r is the position of the particle, s is the arc length along ˆ address all of these issues with computer simulations. the chain, t is the unit tangent vector, and ˆn is the unit Simulation Model. We consider a long elastic filament normal vector. A simple way to visualize the structure of a compacted chain is to plot a correlation plot of the of diameter σ confined to a sphere with radius R0 = 5σ. We model the elastic filament as a sequence of N soft, angles between binormal vectors at each pair of particle repulsive beads interacting with each other and the con- centers (e.g. Fig. 2). We identify the principal domains fining wall via the Weeks–Chandler–Anderson (WCA) by searching for cliques using the networkx package [32]. repulsion [22]. We use standard LJ units, expressing The size of a domain Di is denoted as mi = wiN. The energies in  and lengths in σ. The beads are bonded largest domain, D1, is the largest set of beads such that 2 by a two-body stretching term Vb = 16 Γ (r − 1) and a the angle between the binormal to the chain for every pair ◦ th three-body bending term Vθ = Γ (1 + cos θ), where r is of beads is less than 30 (see Fig. S5). The n largest the distance between two consecutive beads, and θ the domain, Dn is then the largest such set disjoint with the Sn−1 angle between consecutive bond vectors. The ratio 16 union of all larger domains, i=1 Di. follows the exact result for elastic rods [23], and the pref- Results. Analysis of the partitioning at different con- ∗ ∗ actor Γ determines the chain rigidity, Γ = T lp, with ditions reveals the following picture. In the lowest- ∗ ∗ lp the dimensionless persistence length and T ≡ kBT/ temperature replicas we observe spool morphologies for ∗ ∗ the reduced simulation temperature. Changing T or lp all chain lengths (see Fig. 2). The ground state of a short has the same effect on the elastic energy; therefore, by chain (N = 100) is a single spool whose ends curve inward fixing Γ, we can simulate polymers with different rigid- toward the sphere center (Fig. 2a). Simulations of 150- ity by changing the simulation temperature [24]. In our and 200-bead chains show the emergence of double-spool simulations, we fix Γ = 1000, so at low temperatures we structures (Fig. 2b-c), while the variety of low-energy ∗ are modeling fairly rigid chains (at T = 1, lp/σ = 1000) morphologies sharply increases at N = 200. This trend ∗ and at T ≈ 40 (lp/σ ≈ 25) the parameters correspond is illustrated by histograms of m1 (see Fig. S6). A dou- to DNA. ble spool is the putative ground state structure for 200 We performed NVT replica-exchange molecular dy- beads, but other morphologies, resembling topological namics simulations [25, 26] of chains of up to N = 250 links, are also observed (Fig. 3a-b) [33]. The spool-like beads using the LAMMPS package [27]. We simulated structures are consistent with our theoretical assump- 128 replicas with geometrically-distributed temperatures tions (Fig.1a-c). We also observe a hierarchical mass dis- ∗ n T (n) = 1.045 regulated by a Nos´e–Hoover thermostat tribution m1  m2  m3 characteristic of the optimal for up to 2 × 109 time steps, attempting exchanges be- analytic solutions, distinguishing spools from other mor- 3 tween replicas every 10 steps, and dumping snapshots phologies, though quantitative comparison of wi between every 106 steps. We employ a variable time step which theory and simulations is difficult due to the sensitivity is dynamically determined such that no bead moves by of the partitioning to the model parameters. more than 0.005σ per time step [28]. The polymer pack- Upon increasing the polymer length to 250 beads, −3 ing fraction is ν = 3N(R0 − 0.5) /16tm, where the a triple spool (Fig. 2d) becomes more favourable accessible capsid volume is rescaled due to finite chain than double spools, but non-spool-like morphologies thickness. Thus, for 100 ≤ N ≤ 250, 0.23 ≤ ν ≤ 0.57. are more commonly observed. The structure with For larger packing fractions (we tried up to N = 350, the lowest elastic energy (Vb + Vθ) is a Hopf link-like ν = 0.8), slow kinetics prevents equilibration. morphology (Fig. 3c), some 36 lower in total energy We a posteriori minimized the energy of each snapshot than the triple spool. The putative ground state, another using the limited-memory Broyden–Fletcher–Goldfarb– ≈ 59 lower in total energy, resembles three linked rings 4

(a) (b) (c) (d) (a) (b) (c) (d)

FIG. 2. Spool morphologies at N = 100 (a), 150 (b), 200 FIG. 3. Link motifs. N = 200: Hopf link (a) (c), and 250 (d). Structures are drawn as tubes connecting and link L6n1 (b); N = 250: Hopf link (c) and bonded monomers, and coloured according the to direction of link L6n1 (d). (d) is the putative ground state for bˆ. Putative ground-state structures are denoted by asterisks. N = 250. Bottom row: binormal correlation/domain Bottom: binormal angle correlation (upper-right) combined decomposition plots. From (a-d): (m1, m2, m3) = with domain decomposition (lower-left). (m1, m2, m3) = (91, 73, 16) , (61, 60, 50) , (89, 80, 29) , (83, 60, 55). (84, 13, 3)–(107, 35, 8)–(114, 73, 10)–(115, 77, 34) from (a-d).

avoid this protocol dependence by employing the replica (L6n1 in Dowker-Thistlethwaite notation [34]; Figs. 3d, exchange method; as such, our low-temperature struc- S7). For the dependence of the lowest energies of these tures may describe kinetically-inaccessible ground states motifs on chain length, see Fig. S8. Twisted packings of their systems. Furthermore, in the case of metallic have been proposed as models of toroidal aggregates wires, the confinement size was varied, with a focus on of long chiral molecules such as DNA [35, 36]; even in relatively large containers. In our work, we fixed the en- the absence of intrinsic twist, such packings were shown closure size to R0 = 5σ, which is at the very lowest end in [11] to be preferred at high densities, as they can fill of their parameter range, where little data is available the sphere without disclinations or voids. Space-filling about the observed structures. Further work is needed considerations may also drive the transition to twisted, to explore the packing as a function of the capsid size. link motifs we see here. The degree of binormal ordering decreases with tem- perature (Fig. 4). The distribution of largest domain sizes m1 for a chain of 250 beads is visualized as a heat map for temperatures 1 ≤ T ∗ ≤ 270. spool structures ∗ (m1 & 100) are rarely observed beyond T = 10. For 10 < T ∗ < 40, there is less order, but there are clear indi- cations of link motifs. At high temperatures (T ∗ > 100), while patches of order are retained near the boundary wall—due to local hexagonal packing—large ordered do- mains seldom emerge. Discussion & Conclusions. According to our results, at low temperature, spherical confinement is sufficient to order the majority of the filament into tight spools. We predict a variety of multi-domain morphologies such as nested spools and links of two or three components com- peting in the low temperature regime. Their energetic ordering may be sensitive to the interplay between elastic and repulsive forces and other microscopic details of the FIG. 4. Temperature dependence. The probability distri- system. Our low-temperature results can be compared ∗ bution of largest cluster size m1 as a function of T for simu- with previous experiments and simulations of packing lations with chain length N = 250. The persistence length is ∗ ∗ ∗ metallic wires into spheres [1, 2], where ring-like coiling lp = 1000/T . The yellow curve is the mean value hm1i(T ). was observed, reminiscent of our nested spools. However, The snapshots are representative configurations at the stated those systems are in the athermal limit, and the struc- temperatures: triple spool and L6n1 link (T ∗ = 1), partially- ∗ tures adopted by wires that are pushed into the system ordered Hopf link (T ≈ 40), and disordered random coil ∗ with large forces are protocol-dependent. In our work, we (T ≈ 270). 5

On the other hand, the chosen capsid size is appro- D. E. Smith, Proc. Natl. Acad. Sci. U.S.A. 111, 8345 priate for modeling DNA packing into viral capsids. For (2014). DNA, σ ≈ 2 nm, which means that the viral capsid cor- [10] N. Chiaruttini, M. de Frutos, E. Augarde, P. Boulanger, responding to our model would be about 20 nm in di- L. Letellier, and V. Viasnoff, Biophys. J. 99, 447 (2010). [11] H. Shin and G. M. Grason, EPL 96, 36007 (2011). ameter, which is at the small end of the range for viral [12] D. Svenˇsek,G. Veble, and R. Podgornik, Phys. Rev. E capsids in nature (20 - 200 nm). Given a persistence 82, 011708 (2010). length of 50 nm, the DNA regime is described by simu- [13] J. Kindt, S. Tzlil, A. Ben-Shaul, and W. M. Gelbart, lation temperatures T ∗ ≈ 40. The structures we observe Proc. Natl. Acad. Sci. U.S.A. 98, 13671 (2001). in this regime are only partially ordered, retaining only [14] Y.-J. Chen, D. Wu, W. Gelbart, C. M. Knobler, some features of low-temperature motifs. Approaching R. Phillips, and W. K. Kegel, Phys. Rev. X 8, 021029 the high-temperature limit (the flexible-chain regime), (2018). [15] A. Petrov and S. C. Harvey, Biophys. J. 95, 497 (2008). we eventually observe little structure, in agreement with [16] C. Forrey and M. Muthukumar, Biophys. J. 91, 25 previous studies [37]. (2006). Projecting our model on real systems, we acknowledge [17] D. C. Rapaport, Phys. Rev. E 94, 030401(R) (2016). that system-specific details may influence packing motifs. [18] C. G. Myers and B. M. Pettitt, J. Comp. Chem. 38, 1191 Specific DNA–capsid interactions enhance the ordering of (2017). DNA in viral capsids via built-in folding pathways [38]. [19] D. Marenduzzo, E. Orlandini, A. Stasiak, D. W. Structures incompatible with the I symmetry of many Sumners, L. Tubiana, and C. Micheletti, h Proc. Natl. Acad. Sci. U.S.A. 106, 22269 (2009). viral capsids (e.g., the orthogonal Hopf link with D2d [20] R. Podgornik, A. Aksoyoglu, S. Yasar, D. Svensek, and symmetry) might be disfavoured. Moreover, the kinetic V. Parsegian, J. Phys. Chem. B 120, 6051 (2016). pathway of motor-driven insertion likely selects spool- [21] D. Svensek and R. Podgornik, EPL 100, 66005 (2012). like morphologies [17] even if they are metastable. Sim- [22] J. D. Weeks, D. Chandler, and H. C. Andersen, ilarly, capsid ejection dynamics might depend crucially J. Chem. Phys. 54, 5237 (1971). on the topology of the DNA structure, with knotted [39] [23] R. Podgornik, P. L. Hansen, and V. A. Parsegian, or linked morphologies possibly obstructing the ejection. J. Chem. Phys. 113, 9343 (2000). [24] This is not entirely true, since the LJ term does not scale Acknowledgements. We acknowledge enlightening dis- with Γ. However, in the rigid limit the effect of repulsive cussions with Arman Boromand, Daan Frenkel, Erika interactions on the persistence length is negligible. Eiser, Erik Luijten, and William Gelbart. The work [25] U. H. Hansmann, Chem. Phys. Lett. 281, 140 (1997). was supported by the EU’s Horizon 2020 Program [26] Y. Sugita and Y. Okamoto, Chem. Phys. Lett. 314, 141 through grants ETN 674979-NANOTRANS and FET- (1999). OPEN 766972-NANOPHLOW, by the ARRS grant Z1- [27] S. Plimpton, J. Comp. Phys. 117, 1 (1995). 9170, by the 1000-Talents Program of the Chinese For- [28] The time step is thus smaller than in standard LJ simu- lations because we need to resolve the time scale of bond eign Experts Bureau, and by the Chinese National Sci- vibrations. ence Foundation through grants 11874398, 11850410443, [29] J. Nocedal, Math. Comput. 35, 773 (1980). and 21850410459. [30] D. C. Liu and J. Nocedal, Math. Program. 45, 503 (1989). [31] J. Stevenson, V. R¨uhle, S. Martiniani, et al., “Python energy landscape explorer,” https://github.com/pele-python/pele. [32] A. Hagberg, P. Swart, and D. S Chult, Exploring network ∗ These authors contributed equally to the manuscript structure, dynamics, and function using NetworkX, Tech. † [email protected] Rep. (Los Alamos National Lab.(LANL), Los Alamos, ‡ [email protected] NM (United States), 2008). [1] N. Stoop, J. Najafi, F. K. Wittel, M. Habibi, and H. J. [33] We stress that these structures only resemble topologi- Herrmann, Phys. Rev. Lett. 106, 214102 (2011). cal links, in a superficial sense; the simulated chains are [2] M. R. Shaebani, J. Najafi, A. Farnudi, D. Bonn, and not closed, nor, in general, are the chain segments that M. Habibi, Nat. Commun. 8, 15568 (2017). constitute the ‘components’ of the links contiguous. [3] C. M. Knobler and W. M. Gelbart, [34] H. Doll and J. Hoste, Math. Comput. 57, 747 (1991). Annu. Rev. Phys. Chem. 60, 367 (2009). [35] I. Kuli´c,D. Andrienko, and M. Deserno, EPL 67, 418 [4] N. Keller, Z. T. Berndsen, P. J. Jardine, and D. E. Smith, (2004). Phys. Rev. E 95, 052408 (2017). [36] J. Charvolin and J.-F. Sadoc, Eur. Phys. J. E 25, 335 [5] P. K. Purohit, J. Kondev, and R. Phillips, (2008). Proc. Natl. Acad. Sci. U.S.A. 100, 3173 (2003). [37] A. Cacciuto and E. Luijten, Nano Lett. 6, 901 (2006). [6] A. Leforestier and F. Livolant, J. Mol. Biol. 396, 384 [38] R. Twarock, G. Leonov, and P. G. Stockley, Nat. Comm. (2010). 9, 2021 (2018). [7] A. Leforestier, J. Biol. Phys 39, 201 (2013). [39] D. Marenduzzo, C. Micheletti, E. Orlandini, and D. W. [8] S. B. Hall and J. A. Schellman, Biopolymers 21, 2011 Sumners, Proc. Natl. Acad. Sci. U.S.A. (2013). (1982). [9] Z. T. Berndsen, N. Keller, S. Grimes, P. J. Jardine, and