Mechanism of RGD-conjugated nanodevice binding to its target protein integrin αVβ3 by atomistic molecular dynamics and machine learning.

Active targeting strategies have been proposed to enhance the selective uptake of nanoparticles (NPs) by diseased cells, and recent experimental findings have proven the effectiveness of this approach. However, no mechanistic studies have yet revealed the atomistic details of the interactions between ligand-activated NPs and integrins. As a case study, here we investigate, by means of advanced molecular dynamics simulations (MD) and machine learning methods (namely equilibrium MD, binding free energy calculations and training of self-organized maps), the interaction of a cyclic-RGD-conjugated PEGylated TiO2 NP (the nanodevice) with the extracellular segment of integrin αVβ3 (the target), the latter experimentally well-known to be over-expressed in several solid tumors. Firstly, we proved that the cyclic-RGD ligand binding to the integrin pocket is established and kept stable even in the presence of the cumbersome realistic model of the nanodevice. In this respect, the unsupervised machine learning analysis allowed a detailed comparison of the ligand/integrin binding in the presence and in the absence of the nanodevice, which unveiled differences in the chemical features. Then, we discovered that unbound cyclic RGDs conjugated to the NP largely contribute to the interactions between the nanodevice and the integrin. Finally, by increasing the density of cyclic RGDs on the PEGylated TiO2 NP, we observed a proportional enhancement of the nanodevice/target binding. All these findings can be exploited to achieve an improved targeting selectivity and cellular uptake, and thus a more successful clinical outcome.