Csaba Daday, Carles Curutchet, Adalgisa Sinicropi, Benedetta Mennucci, and Claudia Filippi,
J. Chem. Theory Comput. 2015, 11, 4825 – 4839
Contributed by Tobias Schwabe
Modeling enzymichromism, the sprectral tuning of a chromophore by its protein surroundings, is a formidable challenge for computational chemists. It involves methods from molecular dynamics simulations to excited state multi-reference computations in a QM/MM framework (or even more advanced embedding schemes).
Obviously, in each step of the modeling one has to choose from several methods and different applications. As a consequence, no standard protocol how to address the problem has been agreed upon so far. Daday et al. now presented a thorough analysis of possible routes and identified key ingredients for success in enzymichromism modeling in the paper highlighted here. As a test case, they picked the green fluorescent protein (GFP) and the electronic excitation of its chromophore in the protonated (A) and deprotonated (B) form. Two main issues are covered: What is the effect of using an optimized protein structure vs. snapshots from a MD run (and average) and how should the protein environment be modeled?
Let's look at the MD results first: For an average of 50 snapshots, the excitation energies are 3.36 ± 0.1 eV (A form) and 3.07 ± 0.1 eV (B form) at the CAM-B3LYP/6-31+G*/MM level of theory, but the energy spread is 0.5 eV in both cases. Nevertheless, the average is close to the value of an annealed structure (representative for an optimized crystal structure), which the authors also computed: 3.38 eV (A form), 3.11 eV (B form). It is hard to tell if this finding might be transferable to other protein systems or if this good agreement between ensemble average and optimized structure is just a lucky match. The energy spread might be a caveat.
In any case, the obtained values are blue-shifted in comparison to experiment which is not a failure of CAM-B3LYP (alone). This has been checked by recalculating the results with CASPT2/MM. Therefore, the authors also tried different embedding schemes instead of static point charges: state-specific induced dipoles, linear response methods including induced dipoles, and even frozen density embedding (Those readers who want to learn more about the methods and their differences, should refer to [1]). The bottom line: frozen density embedding is not a significant improvement over state-specific induced dipoles, but combining the latter with the effects of linear response in a polarizable embedding yields very good results in comparison. This is in accordance with our previous study of the problem for which CC2 in a polarizable embedding (PERI-CC2) has been applied.[2] There, a very good agreement between PERI-CC2 and all-QM computations has been demonstrated (for smaller cluster models, of course).
The study highlighted here emphasizes the importance of coupling induced dipoles to the transition moments of the excitation which is not done in all methods which allow for polarization in the classical region of a QM/MM approach. These findings also concern the even broader field of solvation models like PCM or COSMO. Everyone interested in environmental effects on electronic excitations (and other dynamical properties) should make oneself familiar with the differences of state-specific and linear response treated polarization effects. Often, they are not pointed out clearly in the literature which hampers a good interpretation of computed results and their assessment in comparison to experiment.
[1] A. S. P. Gomes, C. R. Jacob, Quantum-chemical embedding methods for treating local electronic excitations in complex chemical systems, Annu. Rep. Prog. Chem.,Scet. C:Phys. Chem. 2012, 108, 222–277 (http://dx.doi.org/10.1039/c2pc90007f)
[2] T. Schwabe, M. T. P. Beerepoot, M. T. P., J. M. H. Olsen, J. Kongsted, Analysis of computational models for an accurate study of electronic excitations in GFP Phys. Chem. Chem. Phys. 2015, 17, 2582–2588. (http://dx.doi.org/10.1039/C4CP04524F)
J. Chem. Theory Comput. 2015, 11, 4825 – 4839
Contributed by Tobias Schwabe
Modeling enzymichromism, the sprectral tuning of a chromophore by its protein surroundings, is a formidable challenge for computational chemists. It involves methods from molecular dynamics simulations to excited state multi-reference computations in a QM/MM framework (or even more advanced embedding schemes).
Obviously, in each step of the modeling one has to choose from several methods and different applications. As a consequence, no standard protocol how to address the problem has been agreed upon so far. Daday et al. now presented a thorough analysis of possible routes and identified key ingredients for success in enzymichromism modeling in the paper highlighted here. As a test case, they picked the green fluorescent protein (GFP) and the electronic excitation of its chromophore in the protonated (A) and deprotonated (B) form. Two main issues are covered: What is the effect of using an optimized protein structure vs. snapshots from a MD run (and average) and how should the protein environment be modeled?
Let's look at the MD results first: For an average of 50 snapshots, the excitation energies are 3.36 ± 0.1 eV (A form) and 3.07 ± 0.1 eV (B form) at the CAM-B3LYP/6-31+G*/MM level of theory, but the energy spread is 0.5 eV in both cases. Nevertheless, the average is close to the value of an annealed structure (representative for an optimized crystal structure), which the authors also computed: 3.38 eV (A form), 3.11 eV (B form). It is hard to tell if this finding might be transferable to other protein systems or if this good agreement between ensemble average and optimized structure is just a lucky match. The energy spread might be a caveat.
In any case, the obtained values are blue-shifted in comparison to experiment which is not a failure of CAM-B3LYP (alone). This has been checked by recalculating the results with CASPT2/MM. Therefore, the authors also tried different embedding schemes instead of static point charges: state-specific induced dipoles, linear response methods including induced dipoles, and even frozen density embedding (Those readers who want to learn more about the methods and their differences, should refer to [1]). The bottom line: frozen density embedding is not a significant improvement over state-specific induced dipoles, but combining the latter with the effects of linear response in a polarizable embedding yields very good results in comparison. This is in accordance with our previous study of the problem for which CC2 in a polarizable embedding (PERI-CC2) has been applied.[2] There, a very good agreement between PERI-CC2 and all-QM computations has been demonstrated (for smaller cluster models, of course).
The study highlighted here emphasizes the importance of coupling induced dipoles to the transition moments of the excitation which is not done in all methods which allow for polarization in the classical region of a QM/MM approach. These findings also concern the even broader field of solvation models like PCM or COSMO. Everyone interested in environmental effects on electronic excitations (and other dynamical properties) should make oneself familiar with the differences of state-specific and linear response treated polarization effects. Often, they are not pointed out clearly in the literature which hampers a good interpretation of computed results and their assessment in comparison to experiment.
[1] A. S. P. Gomes, C. R. Jacob, Quantum-chemical embedding methods for treating local electronic excitations in complex chemical systems, Annu. Rep. Prog. Chem.,Scet. C:Phys. Chem. 2012, 108, 222–277 (http://dx.doi.org/10.1039/c2pc90007f)
[2] T. Schwabe, M. T. P. Beerepoot, M. T. P., J. M. H. Olsen, J. Kongsted, Analysis of computational models for an accurate study of electronic excitations in GFP Phys. Chem. Chem. Phys. 2015, 17, 2582–2588. (http://dx.doi.org/10.1039/C4CP04524F)
No comments:
Post a Comment