NPCoronaPredict: A Computational Pipeline for the Prediction of the Nanoparticle-Biomolecule Corona

Abstract

The corona of a nanoparticle immersed in a biological fluid is of key importance to its eventual fate and bioactivity in the environment or inside live tissues. It is critical to have insight into both the underlying bionano interactions and the corona composition to ensure biocompatibility of novel engineered nanomaterials. A prediction of these properties in silico requires the successful spanning of multiple orders of magnitude of both time and physical dimensions to produce results in a reasonable amount of time, necessitating the development of a multiscale modeling approach. Here, we present the NPCoronaPredict open-source software package: a suite of software tools to enable this prediction for complex multicomponent nanomaterials in essentially arbitrary biological fluids, or more generally any medium containing organic molecules. The package integrates several recent physics-based computational models and a library of both physics-based and data-driven parametrizations for nanomaterials and organic molecules. We describe the underlying theoretical background and the package functionality from the design of multicomponent NPs through to the evaluation of the corona.

Figure 1

A summary of the overall workflow for corona prediction using the suite of computational tools discussed here. Yellow parallelograms indicate data used as input/output with the expected file format indicated in brackets. Blue rectangles indicate software or methodologies used to generate or process input, with entries in italics indicating external methodologies not included in either NPCoronaPredict or PMFPredictor. A large set of precomputed Hamaker constants and Bead-surface PMFs are additionally supplied in the repository for immediate use.

Figure 2

Automatically generated bead mapping for a target molecule (DPPC) using the MolToFragment.py script included in the repository, with highlighting applied to indicate the resulting fragments and manual annotation added to indicate SMILES codes for each bead. The mapping has been constrained to use only bead types for which interaction potentials are available.

Figure 3

An example NP consisting of an anatase core decorated with carbon black beads produced using the NPDesigner software tool. The locations of beads are shown in the right-hand table, while the bottom table lists definitions of all bead types which have been added so far. A visualization of the NP is shown in the upper left corner, with the dashed blue line indicating the NP bounding radius at the nominal surface and the red dashed line indicating the limit at which adsorbates are assumed to be unbound.

Figure 4

An example heatmap plot of binding energies produced for bovine serum albumin (PDB code 3V03) to a silver NP (Ag (100), R = 27 nm, surface potential −31 mV). The location of the most favorable protein orientation is marked with a green ring at ϕ = 267.5°, θ = 102.5°.

Figure 5

An example of the protein–NP complex produced by postprocessing the results of a UA calculation for bovine serum albumin (PDB: 3V03) to a silver NP (Ag (100), R = 27 nm, surface potential −31 mV) using the VisualizeUAResults.ipynb script. The conformation shown is the energetically most favorable orientation of the protein. The inset shows the entire complex while the main figure provides a cropped region to show finer details of the protein.

Figure 6

Time-evolution of the predicted corona for a set of 20 proteins (Table 1) on a 5 nm gold (100) surface, averaging over five simulations. For clarity, only proteins still in the corona at t = 5 ms have an entry shown in the legend.

Figure 7

A plot of predicted surface-area normalized corona mass and charge for a variety of nanomaterials immersed in a medium of 20 proteins. The proteins are selected from the Daphnia magna proteome based on a cluster analysis to demonstrate the use of the pipeline to assign a simple low-dimensional representation to arbitrary nanomaterials broadly matching their chemical classes (metallic, carbonaceous, metal/semimetal oxide), with the other category including semiconductors and metals with organic ligands attached. An exponential smoothing with a time constant of 0.1 ms has been applied to reduce noise.

Figure 8

A demonstration of the NPCoronaPredict-GUI interface for performing NP–protein binding energy calculations using a simplified set of options. The first panel (top left) shows the interface for automatically downloading protein structures based on their ID. The second (bottom left) shows the setup and output of computation for a single protein–NP pair (here human serum albumin to a rutile NP). The third (right) shows the results visualized as a heatmap and schematic view of the favored orientation.

Figure 9

Results from the simulation of the corona formed for NPs in human blood plasma, presented in the style of an SDS-PAGE blot for comparison to experiment. The intensity of each band corresponds to the total mass of protein in that band, normalized within the channel while the location of the band is given by log₁₀(MW) as an approximation of where it would appear in a gel experiment.