A Multiscale Simulation Framework for Elucidating Photochemical Structure–Activity Relationships of Photoswitchable Ligands in Complex Biomolecular Environments

Predicting the photochemical structure–activity
relationships
(photo-SAR) of photoswitchable ligands in complex biomolecular environments
remains a great challenge due to intricate protein–ligand interactions,
strong electron correlation in multiple electronic states, coupled
nuclear and electronic dynamics, and protein conformational flexibility.
To bridge this gap, here we develop a unified multiscale simulation
framework that integrates first-principles nonadiabatic dynamics,
excited-state enhanced sampling, and ground-state alchemical free-energy
calculations. We applied this approach to photostatins (PSTs), a class
of photoswitchable tubulin inhibitors with promising light-regulated
anticancer bioactivity, and validated our predictions against extensive
experimental data, including ultrafast time-resolved crystallography,
absorption spectra, and isomer-dependent bioactivity assays. Our simulations
reveal, for the first time, that nonradiative decay rates correlate
directly with equilibrium excited-state free-energy surfaces, which
are modulated by substituents, protein electrostatics, and steric
confinement. Specifically, protein electrostatic fields accelerate
excited-state relaxation, whereas steric constraints oppose it. The
balance of these factors determines the trend of excited-state dynamics
across PST derivatives. Our results further show that the photoisomerization
quantum yield depends on (1) the directional alignment of torsional
motions with nonadiabatic coupling vectors during nonradiative decay,
and (2) the propensity for backward ground-state isomerization, both
of which are shaped by protein–ligand interactions. Finally,
among the free-energy methods tested, thermodynamic integration most
accurately captures subtle substituent effects on the contrast in
binding affinities between isomers, a critical metric for minimizing
their off-target effects in the dark-adapted state. This work establishes
a robust computational platform for accurately predicting photodynamics
and light-responsive binding affinities of photoswitchable ligands
in biomolecular systems, while also providing novel mechanistic insights
that can facilitate their rational design in biological and biomedical
applications.