Elucidating Chiral Complexation in Diffusion-NMR with Multiscale Simulations and Machine Learning

Enantiomers play a crucial role in pharmaceuticals, agrochemicals, and food systems due to their distinct physicochemical and biological properties; however, assigning absolute configuration remains a persistent challenge for both experimental and theoretical methods.1 Matrix-assisted diffusion-ordered spectroscopy (DOSY), using chiral resolving agents, has emerged as a promising approach for enantiomer differentiation, although its effectiveness is often limited by subtle resolution and an incomplete molecular-level understanding.2 In this study, we employ a chiral macrocycle (Chirabite-AR, MAC) as a resolving matrix to enhance enantiodifferentiation and enable the assignment of the absolute configuration of mandelic acid (MA) enantiomers via diffusion-NMR. 1H-DOSY and selective 1H-ROESY experiments were performed in CDCl3 at 500 and 600 MHz. The results reveal the stereopreference of the macrocycle for (S)-MA, evidenced by a lower diffusion coefficient and greater shielding relative to (R)-MA. Computational investigations, including classical molecular dynamics, metadynamics, and density functional theory, show that this enantioselectivity arises when the macrocycle adopts a closed conformation rather than the expected open structure. Temperature-dependent NMR experiments further support a dynamic equilibrium between open and closed MAC forms, consistent with simulations. To enhance conformational sampling and capture the complex free-energy landscape, machine learning interatomic potentials were employed, enabling efficient exploration of host–guest configurations and providing deeper insight into the stability and dynamics of the complexes. Overall, the combined results demonstrate that only the closed MAC–MA complex accounts for both diffusion and chemical shift behavior, identifying it as the key structure responsible for enantiodifferentiation. This study highlights the critical role of conformational flexibility in chiral recognition and underscores the value of combining diffusion-NMR with advanced simulation techniques, including machine learning, to elucidate enantioselective mechanisms in solution.