In this work, we critically assess the ability of the all-atom enhanced sampling method accelerated molecular dynamics (aMD) to investigate conformational changes in proteins that typically occur on the millisecond time scale. We combine aMD with the inherent power of graphics processor units (GPUs) and apply the implementation to the bovine pancreatic trypsin inhibitor (BPTI). A 500 ns aMD simulation is compared to a previous millisecond unbiased brute force MD simulation carried out on BPTI, showing that the same conformational space is sampled by both approaches. To our knowledge, this represents the first implementation of aMD on GPUs and also the longest aMD simulation of a biomolecule run to date. Our implementation is available to the community in the latest release of the Amber software suite (v12), providing routine access to millisecond events sampled from dynamics simulations using off the shelf hardware.

Idiosyncratic adverse drug reactions are unpredictable, dose-independent and potentially life threatening; this makes them a major factor contributing to the cost and uncertainty of drug development. Clinical data suggest that many such reactions involve immune mechanisms, and genetic association studies have identified strong linkages between drug hypersensitivity reactions to several drugs and specific HLA alleles. One of the strongest such genetic associations found has been for the antiviral drug abacavir, which causes severe adverse reactions exclusively in patients expressing the HLA molecular variant B*57:01. Abacavir adverse reactions were recently shown to be driven by drug-specific activation of cytokine-producing, cytotoxic CD8 + T cells that required HLA-B*57:01 molecules for their function; however, the mechanism by which abacavir induces this pathologic T-cell response remains unclear. Here we show that abacavir can bind within the F pocket of the peptide-binding groove of HLA-B*57:01, thereby altering its specificity. This provides an explanation for HLA-linked idiosyncratic adverse drug reactions, namely that drugs can alter the repertoire of self-peptides presented to T cells, thus causing the equivalent of an alloreactive T-cell response. Indeed, we identified specific self-peptides that are presented only in the presence of abacavir and that were recognized by T cells of hypersensitive patients. The assays that we have established can be applied to test additional compounds with suspected HLAlinked hypersensitivities in vitro. Where successful, these assays could speed up the discovery and mechanistic understanding of HLA-linked hypersensitivities, and guide the development of safer drugs.3D structure | small molecule | binding site A bacavir is a nucleoside analog that suppresses HIV replication. In approximately 8% of recipients, abacavir is associated with significant immune-mediated drug hypersensitivity, which is strongly associated with the presence of the HLA-B*57:01 allele (1, 2). Three complementary models for the mechanism of immune-mediated severe adverse drug reactions have traditionally been discussed (3, 4). The hapten (or prohapten) model states that drugs and their metabolites are too small to be immunogenic on their own, but rather act like haptens and modify certain self-proteins in the host that lead to immune recognition of the resulting hapten-self-peptide complexes as de novo antigens (5-7). The pharmacologic interaction with immune receptors (p-i) model states that drugs can induce the formation of HLA-drug complexes that can activate T-cell immune responses directly without requiring a specific peptide ligand (8). The danger model, which is in principle compatible with other models, states that danger signals other than the drug itself (e.g., chemical, physical, or viral stress) are required to overcome immune tolerance barriers that otherwise suppress drug hypersensitivity reactions (7).None of these existing models provides a convincing mechanism explaining how abacav...

Slow diffusive conformational transitions play key functional roles in biomolecular systems. Our ability to sample these motions with molecular dynamics simulation in explicit solvent is limited by the slow diffusion of the solvent molecules around the biomolecules. Previously, we proposed an accelerated molecular dynamics method that has been shown to efficiently sample the torsional degrees of freedom of biomolecules beyond the millisecond timescale. However, in our previous approach, large-amplitude displacements of biomolecules are still slowed by the diffusion of the solvent. Here we present a unified approach of efficiently sampling both the torsional degrees of freedom and the diffusive motions concurrently. We show that this approach samples the configuration space more efficiently than normal molecular dynamics and that ensemble averages converge faster to the correct values.