(406b) Development of Supervised Learning Models for Protein Adsorption to Engineered Nanoparticles

Engineered nanoparticles are an advantageous platform for developing sensors and delivery vehicles for numerous biotechnology applications^{1, 2}. Most such technologies are developed and validated in vitro but intended for use in complex biological environments in vivo. To-date, testing the compatibility of nanotechnologies in biological systems requires a heuristic approach, whereby unpredictable biofouling and off-target effects often prevent implementation of nanotechnologies. Such biofouling effects are frequently the result of spontaneous adsorption of proteins to the nanoparticle surface, forming the ‘protein corona’ and altering the physiochemical properties, and thus intended function, of the nanotechnology. To implement engineered nanoparticles in biological applications, we require predictive power over which proteins and to what extent they preferentially adsorb to the nanoparticle surface, in different biofluids, to form the protein corona.

In this work, we develop a random forest classifier (RFC) trained using mass-spectrometry data to model which proteins populate the corona of a candidate optical nanosensor composed of single-walled carbon nanotubes, SWNTs, functionalized using single stranded DNA molecules, (GT)₁₅. (GT)₁₅–SWNTs were incubated in biofluids relevant to their in vivo applications – blood plasma and cerebrospinal fluid – to populate the nanoparticle’s protein corona. Proteins adsorbed in the corona phase were purified then identified and quantified via proteomic mass spectrometry. Using publicly available protein databases, the RFC was trained to predict the propensity for specific proteins to populate the corona phase from each biofluid. We expand upon this work by investigating a panel of other relevant nanoparticles, including multiple functionalizations of SWNTs, polystyrene nanoparticles, and gold nanoparticles, evaluating their protein corona compositions in blood plasma, cerebrospinal fluid, and plant lysate. Moreover, we include additional classification techniques to understand protein physicochemical and biological factors important in predicting selectively adsorbed proteins. Lastly, we apply the results of this model to rapidly identify candidate proteins with high binding affinity to different nanoparticles, and experimentally validate adsorption of these proteins to DNA-SWNTs with a fluorescence-based corona exchange assay to evaluate the predictive power of the model³. The classifier presented here provides an important tool for predicting protein-nanoparticle interactions, which is needed for effective translation of bionanotechnologies from in vitro synthesis to in vivo use.

References

(1) Beyene, A. G.; Delevich, K.; Del Bonis-O’Donnell, J. T.; Piekarski, D. J.; Lin, W. C.; Wren Thomas, A.; Yang, S. J.; Kosillo, P.; Yang, D.; Prounis, G. S.; Wilbrecht, L.; Landry, M. P. Imaging Striatal Dopamine Release Using a Nongenetically Encoded near Infrared Fluorescent Catecholamine Nanosensor. Sci. Adv. 2019, 5 (7), eaaw3108.

(2) Demirer, G. S.; Zhang, H.; Matos, J. L.; Goh, N. S.; Cunningham, F. J.; Sung, Y.; Chang, R.; Aditham, A. J.; Chio, L.; Cho, M. J.; Staskawicz, B.; Landry, M. P. High Aspect Ratio Nanomaterials Enable Delivery of Functional Genetic Material without DNA Integration in Mature Plants. Nat. Nanotechnol. 2019, 14 (5), 456–464.

(3) Pinals, R. L.; Yang, D.; Lui, A.; Cao, W.; Landry, M. P. Corona Exchange Dynamics on Carbon Nanotubes by Multiplexed Fluorescence Monitoring. J. Am. Chem. Soc. 2020, 142 (3), 1254–1264.