Letter pubs.acs.org/JPCL
Modeling Coordination-Directed Self-Assembly of M2L4 Nanocapsule Featuring Competitive Guest Encapsulation Yang Jiang,† Haiyang Zhang,‡ Ziheng Cui,† and Tianwei Tan*,† †
Beijing Key Lab of Bioprocess, College of Life Science and Technology, Beijing University of Chemical Technology, Beijing 100029, China ‡ Department of Biological Science and Engineering, School of Chemistry and Biological Engineering, University of Science and Technology Beijing, 100083 Beijing, China S Supporting Information *
ABSTRACT: Exploring the mechanism of self-assembly and guest encapsulation of nanocapsules is highly imperative for the design of sophisticated molecular containers and multistimuli-responsive functional materials. Here we present a molecular dynamics simulation protocol with implicit solvent and simulated annealing techniques to investigate the self-assembly and competitive guest (C60 and C70 fullerenes) encapsulation of a M2L4 nanocapsule that is self-assembled by the coordination of mercury cations and bent bidentate ligands. Stepwise formation of the nanocapsule and competitive fullerene encapsulation during dynamic structural changes in the self-assembly were detected successfully. Such processes were driven by coordination bonding and π−π stacking and obey the minimum total potential energy principle. Potential of mean force calculations for guest binding to the M2L4 nanocapsule explained the mechanism underlying the competitive encapsulations of C60 and C70. This work helps design new functional nanomaterials capable of guest encapsulation and release.
T
calculations for large-scale assembly simulations. The success of simulating the nanocage M6L8 and M12L24 self-assembly motivates us to investigate the mechanism of guest encapsulation within nanocapsules during the self-assembly process, which is yet largely unknown. In this work, we propose another simulation protocol for self-assembly simulations of the M2L4 nanocapsule designed by Kishi et al.3 and unveil the mechanism for the competitive guest encapsulation of C60 and C70 by M2L4 during dynamic self-assembly. A cationic dummy atom model (Figure 1c) that was extensively applied in simulating the self-assembly of metal− organic materials7−9,13 was used to model Hg2+ ions in the M2L4 nanocapsule. The protocol for metal force field development using a dummy atom model in our previous work14,15 was used to generate the proper parameters of Hg2+. Since the crystal structure of M2L4 is not resolved in the experiment, the crystal structure (Figure 1b) 3 of M 2L2 metallacycle was used as a reference during the force field development. The optimized Hg2+ model (Rmin/2 = 0.5366 Å and ε = 213.2399 kcal/mol for MC, and both equal 0 for D) reproduces both the Hg−N distance and the N−Hg−N angle in M2L2 and further predicts the rational structure of the M2L4 nanocapsule (Figure 1d). The general AMBER force field (GAFF)16 was adopted to model the bent bidentate ligand, fullerene C60, and fullerene C70. Several torsion angle restraints were performed on the ligands to avoid unusual conformational
he nanocapsule has been a hot topic during the past decade because of its outstanding function as a molecular container.1 Unlike covalently bonded structures, coordinationdriven assemblies allow reversibly changing their structures under specific conditions, and are therefore more useful in practical applications, especially for assemblies capable of guest encapsulation.2−4 Recently, a transformable coordination capsule/tube featuring selective fullerene binding was designed by Yoshizawa et al.3,5,6 The reported M2L4 nanocapsule was assembled from two Hg2+ ions and four bent bidentate ligands (Figure 1a) in acetonitrile and can encapsulate both fullerene C60 and C70. Interestingly, addition of C60 to a M2L4 ⊃ C70 complex released C70 and reassembled a M2L4 ⊃ C60 complex. Moreover, a M2L4 nanocapsule can be transformed to a M2L2 metallacycle (Figure 1b) via adding two extra Hg2+ ions; adding two extra ligands to M2L2 produces M2L4. During the transformation between M2L4 and M2L2, fullerenes can be released and encapsulated by M2L4 reversibly. Unveiling the mechanism underlying such dynamic structural self-assembly and guest encapsulation is highly desirable for the design of molecular containers and multistimuli-responsive functional materials.3 Computational study has been proven to play an important role in exploring the mechanism of coordination-directed selfassembly nanocages.7−9 Yoneya et al.7,8 reported a molecular dynamics (MD) simulation protocol to study the self-assembly of Fujita’s nanocages M6L8 and M12L24. They adopted a coarsegrained solvent model combining Langevin dynamics (LD),10 the generalized reaction-field method,11 and the Weeks− Chandler−Andersen (WCA) potential12 to speed up the © 2017 American Chemical Society
Received: March 30, 2017 Accepted: April 21, 2017 Published: April 22, 2017 2082
DOI: 10.1021/acs.jpclett.7b00773 J. Phys. Chem. Lett. 2017, 8, 2082−2086
Letter
The Journal of Physical Chemistry Letters
Figure 1. (a) Molecular structure of the bent bidentate ligand in M2L2 and M2L4. (b) Crystal structure of M2L2 metallacycle (CCDC 971934), of which TfO− anions and acetonitrile molecules are omitted for clarity and Hg2+ ions are represented by a cationic dummy atom model. (c) Charge of the metal core (MC) is −0.8 e and for each dummy atom (D) is +0.7 e. MC and D atoms are coplanar with a MC−D bond of 0.9 Å. (d) Predicted structure of M2L4 nanocapsule by the self-assembly simulations in this work. Figure 2. (a) Potential energies and structures of a single M2L4 selfassembly in different stages. The red line indicates the stage of simulated annealing. (b) Total coordination number (CN) of Hg2+ to the main-chain nitrogen atoms (blue) and the side-chain oxygen atoms (red) in the single M2L4 self-assembly system versus the simulation time. The simulated annealing is marked by an orange stripe. (c) The number of complete self-assembled M2L4 (Ncage) during the assembly simulation of four M2L4 nanocapsules. Periodic annealing steps are marked by orange stripes. (d) Total coordination number (CN) of Hg2+ to the main-chain nitrogen atoms (blue) and the side-chain oxygen atoms (red) in the four M2L4 self-assembly system versus the simulation time.
changes during the simulations. The solvation effect of acetonitrile was modeled using the generalized Born/surface area (GBSA) model17,18 combined with the LD simulation.10 The anion effect of TfO− was neglected, similar to the treatment by Yoneya et al.7,8 All the molecules were placed randomly (see Figure S4) in the simulated systems and restrained within a spherical boundary to prevent them from escaping far away in the nonperiodic space and to increase the contact between metal and ligand. To speed up the conformational search and mimic the effect of stirring, we applied the simulated annealing to the simulated systems with several cycles. The target temperature of the simulated annealing for M2L4 nanocapsule self-assembly is set to 1000 K, at which the nanocapsule can maintain the ligand binding completely. For M2L4 nanocapsule with guest encapsulation, the target temperature is 1100 K to promote the guest encapsulation. For the self-assembly and guest encapsulation processes of a single nanocapsule, the reference temperature was set to be the same as in the experiment (298 K). For multiple nanocapsules, the reference temperature is 700 K for self-assembly and 800 K for guest encapsulation, and these temperature levels allow reproducing the dynamic properties of the ligand in the explicit solvent system and preventing the ligand molecules from gathering speedily. All the trajectories were produced using the AMBER15 suite.19 More details on the simulation setup protocol are given in the Supporting Information. We first focused on a simple system that only contained two Hg2+ ions and four ligands (a single M2L4 nanocapsule). The particles were randomly placed and restrained within 60 Å around one Hg2+ ion. During the 500 ns simulation at 298 K, the particles assembled to different structures in different stages (Figure 2a, stage 1−4) and maintained a final structure for about 450 ns, which is inferred as an intermediate of the selfassembly nanocapsule. Each Hg2+ ion in the intermediate has a
complete coordination state (Figure 2b) with a coordination number (CN) of 4, but one bonds to the side-chain oxygen atoms instead of the main-chain nitrogen atoms. This intermediate belongs to a suboptimum structure, and it is difficult to cross the energy barrier in the simulation. The resulting intermediate may be ascribed to the lack of explicit solvent−metal ion interactions in the implicit solvent simulations as done in this work. Then we increased the temperature to 1000 K. During the annealing, the Hg2+ ion that bonded to the side-chain oxygen atoms changed to bond the main-chain nitrogen atoms (Figure 2a, Stage 5; Figure 2b). The potential energy then reached the global minima, and the nanocapsule was correctly assembled (Figure 2a, Stage 6), indicating that the self-assembly process obeys the minimum total potential energy principle. Followed by the successful in silico self-assembly of a single M2L4 nanocapsule, we further turn to a multiple self-assembly system containing eight Hg2+ ions and 16 ligands (four M2L4 nanocapsules). All the particles were randomly placed and restrained within 150 Å around one Hg2+. As expected, four M2L4 nanocapsules formed correctly during the 600 ns simulation with periodic annealing (see Figure 2c and Supporting Movie S1). The self-assembly of multiple nanocapsules is more complex than that of the single 2083
DOI: 10.1021/acs.jpclett.7b00773 J. Phys. Chem. Lett. 2017, 8, 2082−2086
Letter
The Journal of Physical Chemistry Letters nanocapsule because of the enhanced ligand exchange. As shown in Figure 2d, the intermediate structure observed in the single nanocapsule simulation was still formed during the selfassembly of multiple nanocapsules, and more complex intermediate structures were detected (see Figure S5). Interestingly, the M2L2 metallacycle and a M2L3 nanocapsule were observed as well in the self-assembly simulation (Figure S6), indicating a stepwise assembly from M2L2 metallacycle to M2L3 nanocapsule and finally to M2L4 nanocapsule. As reported previously,3 the M 2L 4 nanocapsule can encapsulate both C60 and C70. Using our proposed simulation protocol, we then simulated the self-assembly processes of a single M2L4 nanocapsule with C60 and C70. As shown in Figure S7 and Figure S8, the ligands gathered fast around the guest molecules due to extensive and strong π−π stacking interactions. Due to the fitted shape of the guest to the M2L4 cavity, the presence of guests makes the self-assembly of the complexes easier (less intermediate stages observed) than that without guests. Without simulated annealing, interestingly, the M2L4 ⊃ C60 complex can be successfully obtained, and the equilibrated structure was presented in Figure 3a. However, the M2L4 ⊃ C70 complex (Figure 3a) may fall into an intermediate structure (see Figure S8, stage 2) before annealing. On the other hand, using the proposed simulation protocol, the encapsulation of the guests into a single M2L4 nanocapsule was reproduced in silico as well (see Figures S9 and S10). The potential energies of these systems decreased after the encapsulation, indicating that the encapsulation process obeys the minimum total potential energy principle as well. Free energy profiles for releasing the guests from the M2L4 nanocapsule were then calculated using the adaptive steered molecular dynamics (ASMD) simulation.20 The reaction coordinate was defined as the distance between geometric centers of the guests and the nanocapsule (Figure 3b). Potential of mean force (PMF) profiles for guest release are shown in Figure 3c. PMF was set to zero at large distances where the guest and the nanocapsule were separated completely. The binding free energy of C60 (−79.68 kcal/ mol) is more negative than C70 (−71.48 kcal/mol), indicating a stronger binding affinity for the M2L4 ⊃ C60 complex than M2L4 ⊃ C70. During the guest release, the system crossed two energy barriers (P1 and P2) and underwent five stages marked in Figure 3c (corresponding structures given in Figure 3d). The difference in the free energy change (ΔΔG) between the two guests at different stages was calculated by
Figure 3. (a) Predicted structures of M2L4 ⊃ C60 and M2L4 ⊃ C70 complexes. C60 is shown in orange and C70 is in mauve. (b) Definition of the reaction coordinate used in the adaptive steered molecular dynamics (ASMD) simulation. The shape and size of the nanocapsule and the fullerenes can be measured according to the reaction coordinate. (c) Potential of mean force (PMF) versus the reaction coordinate. PMF for M2L4 ⊃ C60 is shown in orange and for M2L4 ⊃ C70 in mauve. The wells (Wi) and peaks (Pi) in PMF profiles are marked, and the difference in the free energy change (ΔΔG) between two guests versus different stages is calculated by eq 1 and shown in the subfigure. The largest ΔΔG is marked by a red pentagram. (d) The structures of M2L4 ⊃ C60 at the wells (Wi) and peaks (Pi) in panel c. The structures for M2L4 ⊃ C70 are similar to M2L4 ⊃ C60 (not shown here). (e) Potential energies and structures of the system versus different stages during the simulated annealing of M2L4 ⊃ C70 attacked by C60.
ΔΔG(A → B) = [PMFC60(A) − PMFC60(B)] − [PMFC70(A) − PMFC70(B)]
(1)
where A and B are the well values (Wi) or the peak values (Pi) in the PMF (Figure 3c). The first stage W1 → P1 makes the greatest contribution to the difference in binding free energies between C60 and C70 (ΔΔG = −13.59 kcal/mol). In the stage of W1 → P1, M2L4 was broken partially, resulting in a channel at the side faces of two ligands, and the guests appear to escape in part from the nanocapsule (P1 → W2). The ellipsoidal shape and large volume of C70 (see Figure 3b) makes it much easier to pass the stage W1 → P1 than C60 and thus makes the M2L4 ⊃ C70 complex more unstable. For W2 → P2, the guests were in the process of crossing the channel, followed by the expulsion of the guests and the closure of the channel in the last two
stages. The positive values of ΔΔG in stages W2 → P2, P2 → W3, and W3 → P3 indicate that it is more difficult for C70 to cross the channel than C60, but such difference is insignificant (about 2 kcal/mol) compared with the first stage. That is, compared with C70, it is more difficult for C60 to expand the channel on the side face of the M2L4 nanocapsule without any external force, but once the channel formed, C60 would be much easier encapsulated within the nanocapsule. The M2L4 ⊃ C70 complex is therefore vulnerable to C60 attack. As shown in Figure 3e, the M2L4 ⊃ C70 complex was under attack of C60 and 2084
DOI: 10.1021/acs.jpclett.7b00773 J. Phys. Chem. Lett. 2017, 8, 2082−2086
Letter
The Journal of Physical Chemistry Letters finally exchanged the guest. At the high temperature (1100 K) of the simulated annealing (mimicking the stir effect), the M2L4 nanocapsule became unstable, and one of the nitrogen atoms was dissociated from the Hg2+ ion, producing a channel at the side face. The included guest C70 was then expulsed, and a new guest C60 was finally encapsulated. On the contrary, the M2L4 ⊃ C60 complex was very stable in the presence of C70 (see Figure S11). Finally, competitive guest encapsulation within multiple M2L4 nanocapsules was studied using the simulation protocol stated above. Four M2L4 nanocapsules and eight fullerene molecules (four C60 and four C70) were randomly placed and restrained within 160 Å around one of the C60 molecules. As expected, the fullerenes were competitively encapsulated into the M2L4 nanocapsules (Figure 4a,b) and the M2L4 nano-
the complexes. The fullerenes can also be encapsulated into the intermediate M2L3 nanocapsule (see Figure S13 and Table 1) but rarely encapsulated into the M2L2 metallacycle (see Figure S14 and Table 1), indicating that the encapsulation capability and selectivity to the fullerenes also depend on the assembly structure itself. In summary, coordination-directed self-assembly of a M2L4 nanocapsule and competitive encapsulation of C60 and C70 fullerenes were successfully modeled, in good agreement with previously reported experiments.3,21−24 The driving force of the nanocapsule self-assembly and fullerene encapsulation originates from coordination bonding and π stacking, and such processes obey the minimum total potential energy principle. The encapsulation capability and selectivity to the fullerenes highly depend on the shape and volume of guest molecules, as well as the assembled nanostructure. The M2L4 and M2L3 nanocapsules have a high selectivity to the spherical and small fullerene C60, and the M2L2 metallacycle has a slightly high selectivity to the ellipsoidal and large fullerene C70. The release of the fullerenes during the dynamic structure transformation in response to the surrounding environment is due to the low encapsulation capability of the open structure and the weakened π stacking interaction of the M2L2 metallacycle. Self-assembly and host behavior of another M2L4 nanocapsules with varied metal ions and panels21−23 can probably be readily modeled as well with the proposed simulation protocols. This work offers an efficient computational approach to investigate the self-assembly and guest encapsulation of nanocapsules, and will thus promote the design of new molecular containers and multistimuli-responsive functional materials.
■
ASSOCIATED CONTENT
S Supporting Information *
The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jpclett.7b00773.
Figure 4. Number of (a) M2L4 ⊃ C60 complex, (b) M2L4 ⊃ C70 complex, and (c) complete M2L4 nanocapsules as a function of the simulation time. Periodic annealing is marked by orange stripes.
Computation details, supporting figures, and force field parameters (PDF) Movie of multiple M2L4 nanocapsule self-assembly (AVI) Movie of fullerenes competitive encapsulation (AVI)
capsules underwent structure transformations (Figure 4c) during the 600 ns simulation with periodic annealing, as shown in Supporting Movie S2. Although the C70 fullerenes were encapsulated several times, the encapsulation probability of fullerene C60 is much higher than that of C70 (Table 1),
■
Corresponding Author
Table 1. Encapsulation Probability (P) of Fullerene C60 and C70 and the Selectivity of Different Assemblies to the Fullerene Guest
a
assembly
P(C60)a
P(C70)a
P(C60) + P(C70)
selectivityb
M2L2 metallacycle M2L3 nanocapsule M2L4 nanocapsule
0.053 0.659 0.942
0.090 0.130 0.004
0.143 0.789 0.946
−0.260 0.671 0.992
AUTHOR INFORMATION
*E-mail:
[email protected]. ORCID
Yang Jiang: 0000-0003-1100-9177 Haiyang Zhang: 0000-0002-2410-7078 Tianwei Tan: 0000-0002-9471-8202 Notes
The authors declare no competing financial interest.
b
Calculated by eq S1. Calculated by eq S2. A value closer to 1 indicates a stronger selectivity to C60 for encapsulation than C70 and a value closer to −1 indicates a stronger selectivity to C70 encapsulation.
■
ACKNOWLEDGMENTS All the simulations were supported by CHEMCLOUDCOMPUTING. This work was supported by the National Basic Research Program of China (973 program) (2013CB733600), the National Nature Science Foundation of China (21436002, 21390202, 21606016), and Beijing Natural Science Foundation (5174036) as well as by the China Postdoctoral Science Foundation (2015M580993).
indicating that the encapsulation capability and selectivity to the fullerenes highly depend on the shape and volume of guest molecules. Complex assemblies were also formed during the fullerene competition and encapsulation (see Figure S12) due to the aggregation of fullerenes. We therefore forced the fullerenes to disperse in the space after 350 ns to disassemble 2085
DOI: 10.1021/acs.jpclett.7b00773 J. Phys. Chem. Lett. 2017, 8, 2082−2086
Letter
The Journal of Physical Chemistry Letters
■
(20) Ozer, G.; Quirk, S.; Hernandez, R. Adaptive Steered Molecular Dynamics: Validation of the Selection Criterion and Benchmarking Energetics in Vacuum. J. Chem. Phys. 2012, 136, 215104. (21) Li, Z.; Kishi, N.; Hasegawa, K.; Akita, M.; Yoshizawa, M. Highly Fluorescent M2L4 Molecular Capsules with Anthracene Shells. Chem. Commun. 2011, 47, 8605−8607. (22) Kishi, N.; Li, Z.; Yoza, K.; Akita, M.; Yoshizawa, M. An M2L4 Molecular Capsule with an Anthracene Shell: Encapsulation of Large Guests up to 1 nm. J. Am. Chem. Soc. 2011, 133, 11438−11441. (23) Li, Z.; Kishi, N.; Yoza, K.; Akita, M.; Yoshizawa, M. Isostructural M2L4 Molecular Capsules with Anthracene Shells: Synthesis, Crystal Structures, and Fluorescent Properties. Chem. - Eur. J. 2012, 18, 8358− 8365. (24) Kishi, N.; Akita, M.; Kamiya, M.; Hayashi, S.; Hsu, H.-F.; Yoshizawa, M. Facile Catch and Release of Fullerenes Using a Photoresponsive Molecular Tube. J. Am. Chem. Soc. 2013, 135, 12976−12979.
REFERENCES
(1) Zarra, S.; Wood, D. M.; Roberts, D. A.; Nitschke, J. R. Molecular Containers in Complex Chemical Systems. Chem. Soc. Rev. 2015, 44, 419−432. (2) Cook, T. R.; Stang, P. J. Recent Developments in the Preparation and Chemistry of Metallacycles and Metallacages via Coordination. Chem. Rev. 2015, 115, 7001−7045. (3) Kishi, N.; Akita, M.; Yoshizawa, M. Selective Host−Guest Interactions of a Transformable Coordination Capsule/Tube with Fullerenes. Angew. Chem., Int. Ed. 2014, 53, 3604−3607. (4) Debata, N. B.; Tripathy, D.; Chand, D. K. Self-Assembled Coordination Complexes from Various Palladium (II) Components and Bidentate or Polydentate Ligands. Coord. Chem. Rev. 2012, 256, 1831−1945. (5) Wang, W.; Wang, Y.-X.; Yang, H.-B. Supramolecular Transformations within Discrete Coordination-Driven Supramolecular Architectures. Chem. Soc. Rev. 2016, 45, 2656−2693. (6) Meng, W.; Breiner, B.; Rissanen, K.; Thoburn, J. D.; Clegg, J. K.; Nitschke, J. R. A Self-Assembled M8L6 Cubic Cage that Selectively Encapsulates Large Aromatic Guests. Angew. Chem., Int. Ed. 2011, 50, 3479−3483. (7) Yoneya, M.; Yamaguchi, T.; Sato, S.; Fujita, M. Simulation of Metal-Ligand Self-Assembly into Spherical Complex M6L8. J. Am. Chem. Soc. 2012, 134, 14401−14407. (8) Yoneya, M.; Tsuzuki, S.; Yamaguchi, T.; Sato, S.; Fujita, M. Coordination-Directed Self-Assembly of M12L24 Nanocage: Effects of Kinetic Trapping on the Assembly Process. ACS Nano 2014, 8, 1290− 1296. (9) Zhukhovitskiy, A. V.; Zhong, M.; Keeler, E. G.; Michaelis, V. K.; Sun, J. E. P.; Hore, M. J. A.; Pochan, D. J.; Griffin, R. G.; Willard, A. P.; Johnson, J. A. Highly Branched and Loop-Rich Gels via Formation of Metal−Organic Cages Linked by Polymers. Nat. Chem. 2016, 8, 33− 41. (10) Yun-yu, S.; Lu, W.; van Gunsteren, W. F. On the Approximation of Solvent Effects on the Conformation and Dynamics of Cyclosporin A by Stochastic Dynamics Simulation Techniques. Mol. Simul. 1988, 1, 369−383. (11) Tironi, I. G.; Sperb, R.; Smith, P. E.; van Gunsteren, W. F. V. A Generalized Reaction Field Method for Molecular Dynamics Simulations. J. Chem. Phys. 1995, 102, 5451−5459. (12) Weeks, J. D.; Chandler, D.; Andersen, H. C. The Role of Repulsive Forces in Determining the Equilibrium Structure of Simple Liquids. J. Chem. Phys. 1971, 54, 5237−5247. (13) Yoneya, M.; Tsuzuki, S.; Aoyagi, M. Simulation of MetalOrganic Framework Self-Assembly. Phys. Chem. Chem. Phys. 2015, 17, 8649−8652. (14) Jiang, Y.; Zhang, H.; Tan, T. Rational Design of MethodologyIndependent Metal Parameters Using a Nonbonded Dummy Model. J. Chem. Theory Comput. 2016, 12, 3250−3260. (15) Jiang, Y.; Zhang, H.; Feng, W.; Tan, T. Refined Dummy Atom Model of Mg2+ by Simple Parameter Screening Strategy with Revised Experimental Solvation Free Energy. J. Chem. Inf. Model. 2015, 55, 2575−2586. (16) Wang, J.; Wolf, R. M.; Caldwell, J. W.; Kollman, P. A.; Case, D. A. Development and Testing of a General AMBER Force Field. J. Comput. Chem. 2004, 25, 1157−1174. (17) Hawkins, G. D.; Cramer, C. J.; Truhlar, D. G. Parametrized Models of Aqueous Free Energies of Solvation Based on Pairwise Descreening of Solute Atomic Charges from a Dielectric Medium. J. Phys. Chem. 1996, 100, 19824−19839. (18) Tsui, V.; Case, D. A. Theory and Applications of the Generalized Born Solvation Model in Macromolecular Simulations. Biopolymers 2000, 56, 275−291. (19) Case, D. A.; Berryman, J. T.; Betz, R. M.; Cerutti, D. S.; Cheatham, T. E., III; Darden, T. A.; Duke, R. E.; Giese, T. J.; Gohlke, H.; Goetz, A. W.; et al. AMBER, version 15; University of California: San Francisco, 2015. 2086
DOI: 10.1021/acs.jpclett.7b00773 J. Phys. Chem. Lett. 2017, 8, 2082−2086
