Self-Assembly of Nanocubic Molecular Capsules via Solvent-Guided

Oct 1, 2018 - Diabatic States at Construction (DAC) through Generalized Singular Value Decomposition. The Journal of Physical Chemistry Letters...
0 downloads 0 Views 3MB Size
Letter Cite This: J. Phys. Chem. Lett. 2018, 9, 6082−6088

pubs.acs.org/JPCL

Self-Assembly of Nanocubic Molecular Capsules via Solvent-Guided Formation of Rectangular Blocks Takeshi Yamamoto,*,† Hadi Arefi,† Sudhanshu Shanker,† Hirofumi Sato,*,‡ and Shuichi Hiraoka*,¶ †

Department of Chemistry, Graduate School of Science, Kyoto University, Kyoto 606-8502, Japan Department of Molecular Engineering, Graduate School of Engineering, Kyoto University, Kyoto 615-8510, Japan ¶ Department of Basic Science, Graduate School of Arts and Science, The University of Tokyo, Tokyo 153-8902, Japan Downloaded via KAOHSIUNG MEDICAL UNIV on October 6, 2018 at 07:34:56 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.



S Supporting Information *

ABSTRACT: We investigate the mechanism underlying the self-assembly of gearshaped amphiphilic molecules into a highly ordered nanocubic capsule (“nanocube”) in aqueous methanol. Simulation results show that the solvent molecules play a significant role in the assembly process by directing the primitive intermediates to orthogonal/rectangular shapes, thus creating appropriate building blocks for cubic assembly while avoiding off-pathway stacked aggregates. Freeenergy analyses reveal that the interplay of the direct intermonomer interaction and the solvent-mediated repulsion between large aromatic cores (via preferential solvation of methanol on hydrophobic surfaces) leads to the strong trend for perpendicular binding of monomers and hence the solvent-guided formation of rectangular blocks. Furthermore, we report the self-assembly simulation of the nanocube using replica exchange with solute tempering and demonstrate that the simulation can predict a highly ordered nanocapsule structure, assembly intermediates, and encapsulated molecules, which helps promote computer-aided design of functional molecular self-assemblies in explicit solvent.

M

Chart 1. GSA Molecule 1 (left) and Experimental Structure of the Nanocubic Capsule 16 (right) Obtained from the SelfAssembly of Six Neutral GSA Monomers in Aqueous Methanol10,11

olecular containers with a confined nanosized space have been receiving ongoing attention for applications in molecular recognition, transport, and reactions.1−6 A variety of well-defined structures (cages and capsules) have been developed using highly directional interactions such as coordination bonds. This is in contrast to amphiphile-based self-assembly, in which less directional interactions such as van der Waals (vdW) and hydrophobic effects play central roles.7−9 Amphiphile-based molecular containers (e.g., micelles and vesicles) often exhibit a larger dispersity in the number of components and are typically less ordered than coordination assemblies. Gear-shaped amphiphilic (GSA) molecules, 1, provide an interesting exception to those systems that selfassemble into a highly ordered hexameric nanocubic capsule (called a “nanocube”)10−12 without strong directional bonds (Chart 1). The GSA monomer 1 has a hexaphenylbenzene core, in which three hydrophilic pyridyl groups and three hydrophobic methyl groups are introduced. The GSA monomers are molecularly dissolved in pure methanol, while they spontaneously form the nanocube in aqueous methanol.10,11 Previous studies have demonstrated that the vdW, πstacking, and CH−π interactions play a crucial role in the stability of the nanocube, particularly through meshing of the indented molecular surfaces of GSA monomers.13,14 Very recently, it was also reported that the water-soluble nanocubes (obtained from the self-assembly of cationic GSA molecules in © XXXX American Chemical Society

pure water) exhibit a high thermal stability beyond the boiling point of water.14 Understanding the self-assembly process of molecular systems remains a significant challenge for computer simulation.15−18 For discrete capsule systems, molecular dynamics (MD) simulations based on implicit-solvent models have been performed successfully to study the formation of Received: August 24, 2018 Accepted: October 1, 2018 Published: October 1, 2018 6082

DOI: 10.1021/acs.jpclett.8b02624 J. Phys. Chem. Lett. 2018, 9, 6082−6088

Letter

The Journal of Physical Chemistry Letters

Figure 1. Calculated free energies for nanocube formation. The individual terms in eq 1 are plotted for the formation of the cluster state Cn consisting of n monomers (n = 1, ..., 6; see the lower left). (a) Gas-phase association energy ΔE obtained with the rev-vdW-DF2 and PBE+D2 methods (labeled as QM1 and QM2) and the OPLS-AA force field (labeled as MM). ΔG(tr/rot) is the translational/rotational contribution to ΔGconf. (b,c) Solvation term ΔGsolv and its nonpolar (NP) and electrostatic (ES) components for pure methanol and pure water. (d) Comparison of the different energy components in eq 1 for nanocube formation. (e) Free energy of aggregation ΔGa as a function of xM (the mole fraction of methanol). Solid and dashed lines are the results obtained using the experimental and DFT-optimized solute geometries, respectively. The experimental value of ΔGa (−26 kcal/mol) is also shown.11 All energies are in kcal/mol.

coordination-driven nanocages19,20 and the mechanism of competitive guest encapsulation.21 In those systems, coordination bonds play the role of “designed” interactions for directing well-defined structures.22 For the nanocube, on the other hand, the vdW/π-stacking interactions between the terminal aromatic groups should correspond to the designed interactions. However, the gear-shaped monomer 1 is a planar molecule with a large aromatic core, which gives rise to multipoint vdW interactions between the molecular surfaces. If the GSA monomers are stacked with each other, the selfassembly may proceed to off-pathway aggregates such as supramolecular fibers. In this respect, we are interested in the role of solvent molecules, namely, whether they play an active role in regulating the competition between on- and off-pathway assembly processes. Our interest above is partly motivated by the recent experimental observation that the discrete nature of solvent molecules affects the structure of various selfassembling systems rather significantly23−26 in a way beyond the semimacroscopic picture of solvent effects.27,28 As such, in this Letter, we computationally investigate the self-assembly mechanism of the nanocube in aqueous methanol (Chart 1)10,11 with a particular focus on the solvent effect at the molecular level. Furthermore, we perform self-assembly simulation of the nanocube using replica exchange with solute tempering,29−36 (i.e., a type of enhanced sampling method),37−39 based on our recent study that it is quite effective for the self-assembly of a model system in explicit solvent.36 Here, we demonstrate that the simulation can reproduce the highly ordered nanocube structure from scratch while also providing

atomistic insight into assembly intermediates and encapsulated molecules. We first investigated the driving force for nanocube assembly by performing all-atom free-energy calculations. Specifically, we calculated the free energy of aggregation ΔGa for a series of partial cluster states taken from the experimental nanocube structure [labeled as Cn (n = 1, ..., 6); see Figure 1]. The ΔGa for the formation of Cn was calculated as ΔGa(n , xM) = ΔE(n) + ΔGsolv (n , xM) + ΔGconf (n)

(1)

Here, ΔE(n) and ΔGsolv(n,xM) are the gas-phase association energy and the change in the solvation free energy (SFE), respectively. Note that the ΔGsolv depends on the mole fraction of methanol xM in solution. We evaluated ΔE and ΔGsolv, respectively, using density functional theory (DFT) and MD calculations with the Bennett Acceptance Ratio method.40 ΔGconf is the configurational entropic term, which is evaluated using the rigid-rotor harmonic-oscillator approximation. Additional computational details of eq 1 are provided in the Supporting Information (SI). Accurate evaluation of the solute association energy (ΔE in eq 1) is challenging because of the large system size and the crucial importance of vdW interactions. Here we calculated ΔE by using dispersion-corrected density functional theory (DFT)41 together with a converged plane-wave basis set. Figure 1a displays the calculated values of ΔE for the formation of the cluster states Cn. It is observed that the value of ΔE increases rapidly with n and becomes −185 and −205 kcal/mol for the formation of C6 (the nanocube) using the rev-vdW-DF242 and PBE+D243 functionals, respectively. 6083

DOI: 10.1021/acs.jpclett.8b02624 J. Phys. Chem. Lett. 2018, 9, 6082−6088

Letter

The Journal of Physical Chemistry Letters

Figure 2. Time evolution of a dimer in different environments (vacuum, pure methanol, the mixed solvent with xM = 0.5, and pure water) obtained from MD simulations. (a) Center-of-mass distance d between the two monomers. (b) Angle θ formed by the molecular planes of the two monomers, with 0 and 90° corresponding to the stacked and orthogonal configurations, respectively. (c) Distribution of the dimer angle θ for the bound state with d < 1.2 Å obtained from longer (1000 ns) MD simulations. (d) (i) Stacked dimer used as the initial state. (ii) Typical L-shaped dimer observed in the mixed solvent and pure water. (iii) Typical twisted/stacked dimer in the gas phase. (iv) Solvent spatial distribution function (SDF) calculated for the C2 state in the mixed solvent. The isodensity surfaces at g(r) = 2.5 are plotted for the methanol oxygen atoms (in yellow) and water oxygen atoms (in cyan). (v) Typical rectangular, half-box-shaped trimer observed in the mixed solvent and pure water. See the SI for additional figures.

The large negative value of ΔE arises mainly from the vdW/πstacking interactions between the monomers. For comparison, we also evaluated ΔE with a molecular-mechanics force field (here the OPLS-AA model).44 The calculated value is found to agree reasonably well with the DFT results (Figure 1a). On the basis of this agreement, we used the above force field for subsequent MD simulations. The solvation term ΔGsolv is responsible for the solvent dependence of the aggregation free energy. Figure 1b displays the calculated values of ΔGsolv(n) for the formation of Cn in pure methanol (xM = 1). Importantly, the ΔGsolv increases very rapidly with n and becomes 83 kcal/mol for nanocube formation. This indicates that methanol disfavors aggregation and has a strong effect of decomposing the nanocube into isolated monomers. To obtain more insight, we decompose ΔGsolv as NP ES ΔGsolv = ΔGsolv + ΔGsolv

methanol strongly prefers an open/extended structure of the system to increase the solvent-accessible surface area (SASA). The situation changes drastically for pure water (xM = 0). ΔGsolv for nanocube formation in water (17 kcal/mol) is much smaller than that in methanol (83 kcal/mol). The large reduction in ΔGsolv is essentially due to the decrease in the nonpolar term and is physically attributed to the weaker solute−solvent vdW interactions and to the hydrophobic effect of water. By contrast, the electrostatic term remains large (24 kcal/mol). Two factors are probably responsible for this: The first is the water−pyridyl H-bonds that prefer isolated monomers, and the second is the water−benzene interactions (such as OH−π)45−48 that favor greater SASA. As a result, the total ΔGsolv in pure water remains positive, suggesting that water also prefers a more open/extended state of the system (though this effect is weaker than that in pure methanol). Figure 1d compares all of the free-energy components for nanocube formation. As noted above, the solvation term is rather unfavorable in both pure methanol and water. Another unfavorable contribution comes from the configurational term (ΔGconf), which is quite large (110 kcal/mol) because of the significant loss of translational/rotational entropy. For the aggregation to occur, these unfavorable terms must be compensated by the large negative value of ΔE, making the total ΔGa negative. Nanocube formation is thus driven by direct intermonomer (vdW/π-stacking) interaction, in qualitative agreement with the experimental interpretation.10,11 We repeated the above calculation for different values of xM (the mole fraction of methanol). The calculated ΔGa(xM) decreases

(2)

Here, ΔGNP solv represents the nonpolar contribution to ΔGsolv including cavitation and vdW solvation, while ΔGES solv accounts for the solute−solvent electrostatic interactions (see the SI for details). Figure 1b shows that both terms are unfavorable for aggregation, with the nonpolar term (59 kcal/mol) accounting for 70% of the total SFE change. The large value of ΔGNP solv is mainly attributed to the solute−solvent vdW interactions that favor greater exposure of the solute surfaces. The electrostatic term also disfavors aggregation because the solute−solvent Hbonds prefer isolated monomers (because of the greater exposure of the pyridyl groups). Overall, it turns out that 6084

DOI: 10.1021/acs.jpclett.8b02624 J. Phys. Chem. Lett. 2018, 9, 6082−6088

Letter

The Journal of Physical Chemistry Letters

Figure 3. Self-assembly simulation of a hexamer in the mixed solvent (xM = 0.5). (a) Evolution of the number of orthogonal monomer pairs in the system, NL, obtained from cMD and REST simulations. The abscissa for the REST data represents MD time per replica. (b) Solute structure at different times obtained from the REST simulation. A typical trapped structure observed in the cMD simulation is also shown. (c) Snapshot of a nanocube at 1200 ns containing seven methanol molecules in the internal cavity (where one monomer is not shown for visual clarity).

monotonically with decreasing xM (Figure 1e). At a certain amount of water, ΔGa becomes sufficiently large to induce a monomer−aggregate transition. Experimentally, cube formation occurs in methanol and water in a 3:1 (v/v) ratio, which corresponds to xM = 0.57. The experimental value of ΔGa (−26 kcal/mol) is found to be near the lower end of our calculated results. We think that this level of agreement is satisfactory considering the large magnitude of different terms and their strong cancellation. [It should be noted that although the magnitude of ΔGa in pure water is very large, this condition is not realized experimentally because of the insolubility of the GSA monomer (1) and the nanocube (16).10,11] We next investigated the self-assembly pathway of the nanocube 16. To obtain basic insight, we first performed MD simulation of dimers (12) and trimers (13) in different environments. The initial state was chosen as a stacked state, which corresponds to a local energy minimum in the gas phase. Figure 2 displays the time evolution of the distance d between two monomers and the angle θ formed by their molecular planes obtained from the MD simulation of a dimer. In pure methanol, the initially stacked dimer dissociates rapidly within 20 ns and remains in a molecularly dispersed state. The rapid dissociation is induced mainly by the strong vdW solvation of methanol. In a mixed solvent (xM = 0.5), the dimer remains in a bound state with d < 1.2 nm because of the increased effective attraction between the monomers. The lifetime of the bound state is approximately 100−300 ns (see Figure S1 for longer simulations). Importantly, the dimer is found to adopt an L-shaped structure with the angle fluctuating around 80° (Figure 2b,c). In this structure, the terminal pyridyl groups play the role of a “hinge” for orthogonal binding. The physical reason for the L-shaped dimer can be inferred from the solvent SDFs in the mixed solvent (Figures 2d and S4). The SDFs

show that the large aromatic (hexaphenylbenzene) cores are solvated preferentially by methanol, while the terminal pyridyl groups are solvated by water. The preferential solvation of methanol likely exerts a solvent-mediated force for opening the dimer because methanol strongly favors an open/extended state of the system. To gain more insight, we performed additional free-energy calculations for L-shaped versus face-toface stacked dimers (Table S2). The results show that the stacked dimer in solution is rather unstable because of the unfavorable solvation term, indicating that the solventmediated interaction is repulsive between the molecular planes of GSA 1. Taken together, these results suggest that the trend for the perpendicular binding of monomers arises from the interplay of the direct vdW attraction between the indented molecular surfaces of the GSA monomers and the solventmediated repulsion between the large aromatic cores of GSA. Interestingly, the dimer in pure water also exhibits a strong tendency to adopt L-shaped structures with the angle distribution centered at ∼80°. Here as well, we attribute the perpendicular binding of monomers to the interplay of the direct and solvent-mediated interactions (with the latter mainly due to the electrostatic solvation). In addition, we performed MD calculations of the trimer in different environments and observed that the trimer has a similar tendency to adopt a rectangular, half-box shape both in the mixed solvent and in pure water (see Figure S3 and the MD trajectory in the SI). The behaviors of the dimer and trimer are qualitatively different in the gas phase. The dimer tends to adopt a twisted stacked structure with an angle distribution centered at ∼5° (Figure 2c). The predominance of the stacked structure is due to the multipoint vdW interactions between the hexaphenylbenzene core of the GSA monomers (we note that stacked states are slightly lower in energy than L-shaped states; see 6085

DOI: 10.1021/acs.jpclett.8b02624 J. Phys. Chem. Lett. 2018, 9, 6082−6088

Letter

The Journal of Physical Chemistry Letters Table S2). The trimer in the gas phase exhibits more heterogeneous structures including partial stacks (Figure S3). These results indicate that the solvent prefers the rectangularshaped intermediates, which is advantageous for the cubic assembly. We now proceed to the self-assembly simulation of the nanocube from the monomers in the mixed solvent (xM = 0.5). The initial state was chosen as a molecularly dispersed state (Figure S6). In the conventional MD (cMD) simulation, six monomers rapidly formed a single aggregate and continued to make local rearrangements of the monomers. To monitor the evolution of intermediates of the nanocube, we calculated the number of approximately L-shaped monomer pairs in the system, NL. The latter is defined as NL = ∑i