Yoram Cohen: A nanoinformatics platform for environmental impact assessment of manufactured nanomaterials

Yoram CohenYoram Cohen of the University of California Center for Environmental Implications of Nanotechnology gave a talk co-authored by colleagues at the University of California. Nanoinfo.org is a nanoinformatics platform that supports the environmental impact assessment of engineered nanomaterials (ENMs) with a central database of ENM safety data and a toolkit for various exploration and analysis methods.21 These methods include the estimation of environmental exposure levels of ENMs (MendNano), evaluation of environmental releases of ENMs (LearNano), analysis of high throughput toxicity data of ENMs (ToxNano), and predictive toxicity models, and analysis of the environmental impact of ENMs via Bayesian inference (NanoEIA).

NanoDatabank is a data repository of ENM properties, and experimental and simulation datasets of ENM toxicity and environmental fate and transport (F&T). It contains databases that include physicochemical properties; toxicological properties; experimental datasets of ENM toxicity and F&T; and results of model simulations and estimation of ENM toxicity and F&T behavior, and physicochemical properties. It includes data for over 300 nanomaterials, and toxicity data for various cell lines, zebrafish and bacterial strains, from 325 publications. ToxNano is a high-content data analysis tool (HDAT)22,23 offering QSARs using random forest and Bayesian network toxicity models; analysis of knowledge evidence, and data visualization. MendNano (multimedia environmental distribution of nanomaterials) is a Web-based modeling platform.24,25 Nanoinf.org has 400 users from more than 50 countries.

As an example of work on the toxicity of nanomaterials, Yoram presented unpublished results on evaluating the body of evidence on quantum dots (QDs) via meta-analysis. QDs are very small semiconductor particles, only several nanometers in size, so small that their optical and electronic properties differ from those of larger particles. Many types of quantum dot will emit light of specific frequencies if electricity or light is applied to them, and these frequencies can be precisely tuned by changing the dots’ size, shape, and material.

QD data were collected from 448 publications, reporting 2,703 samples, with 7 core types, 12 shell types, 13 surface modifications, 14 surface ligands, and 20 assay types. In the predictive toxicity model R2 was about 0.81 for cell viability, and about 0.83 for IC50. Yoram and his colleagues studied cause-effect relationships between cellular bioactivity and QD attributes. Median IC50 was ≤ 10 mg/L, for the surface ligands of type amphiphilic polymer, lipid, other hydrophobic, aminothiol, and other amphiphilic. It was uniformly distributed for silica. There was no correlation between surface charge and IC50. The sensitivity distribution of IC50 for cell anatomical type suggests that more differentiated cells are more adversely affected by exposure to QDs. Toxicity is not governed by QD size alone: there is a wide range of IC50 for a given size, and toxicity can be high or low irrespective of the size. Core type affects toxicity, but the wide range of IC50 for a given core type suggests that there are other important attributes.

Bayesian network models can be useful for handling uncertainties, mixed attributes, and hidden conditional relationships since they provide rigorous and simple mathematical means of handling data uncertainty; they integrate graphical representation of the problem with probabilistic evaluation of variable relationships; they can incorporate prior knowledge based on data as well as expert opinion in a convenient representation of probability distributions; and they calculate the likelihood of specific scenarios based on prior knowledge.

Bayesian network model sensitivity analysis showed that QD toxicity is correlated with the most relevant (or significant) attributes tabulated below. The QD attributes identified in this study were consistent with previous analysis via random forest.26

Bayesian network for IC50 Random forest for IC50
Surface ligand QD diameter
Shell Surface ligand
QD diameter Shell
Assay type Assay type
Exposure time Exposure time
Surface modification Surface modification
Surface charge Surface charge
Bayesian network for cell viability Random forest for cell viability
Surface ligand QD diameter
QD diameter QD concentration
QD concentration Surface ligand
Exposure time Exposure time
Shell Surface modification
Assay type Assay type
Surface modification Surface charge
Surface charge  

Bayesian networks for new explorations of association rules among various biological responses as a result of exposure to manufactured nanomaterials have also been demonstrated in zebrafish toxicity studies. Yoram and his co-workers used a nanomaterial biological interaction knowledge base of zebrafish phenotype data with 1,147 samples, and 11 biological responses (including mortality). The data included exposure to seven material types (carbon, cellulose, dendrimer, metal, (metal) oxide, polymeric, and semiconductor) of 0.8–250 nm average primary size; concentration; number of embryos per experiment; and responses recorded for each exposure scenario.

The Bayesian network model for zebrafish mortality (percentage of dead embryos) had an R2 of about 0.79. Sensitivity analysis of the key material properties and exposure conditions that correlate with zebrafish mortality was carried out, and cause-effect relationships between zebrafish phenotypes and material properties and exposure conditions were investigated. Attribute significance was determined by exhaustive search of 13 attributes using bootstrapping. Mortality at 120 hours post-fertilization correlated with concentration used, core atomic composition, outermost surface, average particle size, surface charge, shell composition and purity. The significant attributes at 24 hours post-fertilization were the same but the ranking of the top four differed slightly.

The responsible development of beneficial manufactured nanomaterials requires a thorough understanding of their potential adverse environmental and human health impacts. This requires predicting the biological response of various receptors when exposed to these materials, along with an understanding of their fate and transport, and their range of likely exposure concentrations. Yoram’s work helps to rank various nanomaterials with respect to their potential environmental impact.

References

  1. Cohen, Y.; Rallo, R.; Liu, R.; Liu, H. H. In Silico Analysis of Nanomaterials Hazard and Risk. Acc. Chem. Res. 2013, 46 (3), 802-812.
  2. Liu, R.; Jiang, W.; Walkey, C. D.; Chan, W. C. W.; Cohen, Y. Prediction of nanoparticles-cell association based on corona proteins and physicochemical properties. Nanoscale 2015, 7 (21), 9664-9675.
  3. Liu, R.; Rallo, R.; Bilal, M.; Cohen, Y. Quantitative Structure-Activity Relationships for Cellular Uptake of Surface-Modified Nanoparticles. Comb. Chem. High Throughput Screening 2015, 18 (4), 365-375.
  4. Liu, H. H.; Bilal, M.; Lazareva, A.; Keller, A.; Cohen, Y. Simulation tool for assessing the release and environmental distribution of nanomaterials. Beilstein J. Nanotechnol. 2015, 6, 938-951.
  5. Liu, H. H.; Cohen, Y. Multimedia Environmental Distribution of Engineered Nanomaterials. Environ. Sci. Technol. 2014, 48 (6), 3281-3292.
  6. Oh, E.; Liu, R.; Nel, A.; Gemill, K. B.; Bilal, M.; Cohen, Y.; Medintz, I. L. Meta-analysis of cellular toxicity for cadmium-containing quantum dots. Nat. Nanotechnol. 2016, 11 (5), 479-486.