Electronic structure governs biological fate: Multi-scale computational nanotoxicology of TiO₂ nanoparticles across...

Titanium dioxide nanoparticles (TiO₂ NPs) are among the most widely produced engineered nanomaterials, yet a mechanistic understanding of how their physicochemical properties govern biological interactions remains incomplete. Here we present an integrated, multi-scale computational workflow for the mechanistic in silico characterization of TiO₂ NPs spanning two crystallographic polymorphs (anatase and rutile) and four morphology classes (bipyramidal, cuboctahedral, spherical and amorphous), across a systematic size series of approximately 1-3 nm. At the quantum-mechanical level, density functional theory (DFT) calculations characterize the electronic structure, surface energetics and reactive-site distribution of each nanoparticle. These intrinsic descriptors are linked to mechanistic bio-interaction endpoints through three complementary approaches: (i) molecular docking followed by molecular dynamics (MD) simulations against a curated panel of 98 proteins (97 of them human), mapping the NP-protein binding landscape for individual proteins in isolation, which informs but does not by itself determine protein-corona composition; (ii) COSMO-RS-based thermodynamic modelling of nanoparticle-lipid bilayer interactions to estimate passive membrane permeability; and (iii) explicit all-atom MD simulations of NP interactions with two model cell-membrane systems, representing human keratinocytes and the rainbow-trout gill cell line RTgill-W1. Quantitative cross-tier analysis shows that these endpoints are not redundant: docking affinity and predicted membrane permeability are statistically independent across the nanoparticles common to both tiers (Spearman ρ = +0.03, n = 6), whereas particle size correlates strongly with both docking affinity (ρ = -0.77, n = 10, p = 0.009) and permeability (ρ = -0.83, n = 6, p = 0.042). The most robust structure-activity signal is morphological: faceted particles (bipyramidal, cuboctahedral, rutile) and rounded particles (spherical, amorphous) are completely separated in predicted permeability (mean log P - 5.8 versus +1.3; Mann-Whitney U = 0, p = 0.036). A complementary nano-SAR classification model built on 61 curated literature cytotoxicity records retains predictive performance under publication- and material-grouped cross-validation (AUC-ROC 0.83 versus 0.87 for random splits) and passes y-randomization, but its applicability domain is narrow and is defined explicitly here. We present this workflow as a tool for mechanistic interpretation and early-stage hazard prioritization rather than as a validated, regulatory-grade hazard-assessment platform: protein adsorption, membrane association and changes in membrane properties are mechanistic descriptors and are not equivalent to toxicity, and the framework has not yet been externally validated against experimental data. Used with these limitations in mind, it can support the hazard-screening component of Safe-and-Sustainable-by-Design workflows, which additionally require exposure, functional performance, environmental and life-cycle assessment that are outside the scope of this study.