Understanding the Variable Drivers of Toxicity for the Broad Class of Carbon Nanotubes and Nanofibers from U.S. Facilities

Pulmonary exposure to carbon nanotubes or nanofibers (CNT/F) is known to induce inflammation, toxicity, or tumorigenesis, and is a concern in the occupational setting. U. S. facility employees are at risk of inhalation exposure of multi-walled carbon nanotubes and carbon nanofibers during primary and secondary manufacturing. To date, only one MWCNT, Mitsui-7 has been classified as possibly carcinogenic to humans (Group 2B), while all other materials were subsequently categorized as unclassifiable (Group 3). This class of material has recently been listed as a high priority to the International Agency for Research on Cancer due to this significant knowledge gap. Furthermore, expressed desire to better understand the toxicity profiles of these materials has emerged from the National Institute for Occupational Safety and Health. While human research to date is limited, the use of in in vivo and in vitro model systems can be implemented for the assessment of toxicity outcomes following respiratory exposure to CNT/F. The goal of this study was to generate an accurate an effective safety profile of MWCNT and CNFs from U. S. facilities, and to adapt a multi-disciplinary approach using machine learning to identify pertinent physicochemical characteristics that act as drivers of these toxicity outcomes. This study established toxicity profiles from male C57BL6/J mice aged 8-10 weeks exposed to either 4 or 40 µg of one of nine different CNT/F via oropharyngeal aspiration as well as human epithelial BEAS-2B cells (0-24 µg/ml), differentiated THP-1 cells (0-120 µg/ml), and human fibroblasts (0-2 µg/ml) for four primary outcomes of genotoxicity, inflammation, pathology, and translocation. The nine materials used in this study had a wide range of characteristics including diameter (6-397 nm), length (0.1-50 µm), surface area (18-238 m2/g), aspect ratio (2-1396), residual metal catalyst (0.3-6.2 %), density (0.007-0.220 g/cm3), etc., to consider. Toxicity profiles were generated regarding these four primary toxicity outcomes, and both supervised and unsupervised machine learning was used to identify the key drivers of these adverse health effects. While some physicochemical characteristics were determined to be key drivers of specific toxicity outcomes, different characteristics were essential when considering other toxicity endpoints. No single characteristic could be used as a toxicity predictor, therefore, multifactorial processes, or combination of characteristics, were necessary for an accurate and effective prediction model for responses. The study identified physicochemical drivers of CNT/F toxicity using an integrated approach, combining experimental evidence with computational modeling, with potential for broad application. This study provides necessary information for the consideration of the potential human health effects that can result from CNT/F exposure. The safety profiles and identified drivers of toxicity may be useful for future predictive risk assessment studies and translational studies as well as contributing to safety-by-design for future material designs