Development of structure–activity relationship for metal oxide nanoparticles† Rong Liu, a Hai Yuan Zhang, a Zhao Xia Ji, a Robert Rallo, b Tian Xia, c Chong Hyun Chang, a Andre Nel c and Yoram Cohen * ad Nanomaterial structure–activity relationships (nano-SARs) for metal oxide nanoparticles (NPs) toxicity were investigated using metrics based on dose–response analysis and consensus self-organizing map clustering. The NP cellular toxicity dataset included toxicity profiles consisting of seven different assays for human bronchial epithelial (BEAS-2B) and murine myeloid (RAW 264.7) cells, over a concentration range of 0.39– 100 mg L 1 and exposure time up to 24 h, for twenty-four different metal oxide NPs. Various nano-SAR building models were evaluated, based on an initial pool of thirty NP descriptors. The conduction band energy and ionic index (often correlated with the hydration enthalpy) were identified as suitable NP descriptors that are consistent with suggested toxicity mechanisms for metal oxide NPs and metal ions. The best performing nano-SAR with the above two descriptors, built with support vector machine (SVM) model and of validated robustness, had a balanced classification accuracy of 94%. An applicability domain for the present data was established with a reasonable confidence level of 80%. Given the potential role of nano-SARs in decision making, regarding the environmental impact of NPs, the class probabilities provided by the SVM nano-SAR enabled the construction of decision boundaries with respect to toxicity classification under different acceptance levels of false negative relative to false positive predictions. 1 Introduction The generation of in vitro and in vivo toxicity characterization data is essential for risk assessment and establishment of safe- use of engineered nanomaterials (ENMs). However, this is a formidable task given the expected growth in number and diversity of ENMs. 1 Therefore, in addition to experimental toxicity studies, there is a need for in silico methods (i.e., computational approaches) that will support rapid toxicity screenings. 2,3 Accordingly, in recent years there have been increased efforts to develop data-driven structure–activity rela- tionships (SARs) 4 for ENMs (i.e., nano-SARs) 5–11 that correlate their physicochemical properties 12 with the observed bioactivity (e.g., toxicity) of the exposed target receptors. It is noted that, relative to SARs for chemicals, 4 a small number of nano-SARs have been proposed 13–17 over the last ve years, relying on modest size datasets (typically of the order of 10–100 different ENMs) with the present cumulative toxicity datasets for 200 different ENMs. The majority of published nano-SAR studies 11,13–17 have focused on metal oxide nanoparticles (NPs) 13,15,16 that have high commercial production volume. 18 Among the eight recently published nano-SAR studies, ve of the reported models clas- sied a given NP as either toxic or non-toxic, 13,14,16,17 while the other three are linear/log-linear regression models of EC 50 for bacteria cytotoxicity, 15 smooth cell apoptosis, 11 and NP uptake, 14 respectively. The above nano-SARs were developed based on in vitro toxicity assessed for different cell lines and different toxicity assays. 13–16 NP descriptors included NP primary 13,14,16 and aggregate 13 size, zeta potential, 13,14,17 concentration measure (mass concentration 13 and volume fraction 16 ), relaxiv- ities, 14,17 energies/enthalpies (atomization energy 16 and forma- tion enthalpy of a gaseous cation of the same oxidation state as in the metal oxide 15 ). Previously developed nano-SARs 11,13–16 have demonstrated that, to various levels of accuracy, NP toxicity can be correlated with their physicochemical properties. However, in order for nano-SARs to become acceptable as tools for regulatory decision making and support safe-by-design approaches to ENM development, decision boundaries and applicability domain for developed nano-SARs must be derived based on appropriate end point denitions. A recent toxicity study 19 for twenty-four metal oxide NPs has provided a rich dataset and thus an opportunity for exploring the nano-SAR development along with quantication of the a California Nanosystems Institute, University of California, Los Angeles, CA 90095, USA b Departament d'Enginyeria Informatica i Matematiques, Universitat Rovira i Virgili, Av. Paisos Catalans 26, 43007 Tarragona, Catalunya, Spain c Department of Medicine – Division of NanoMedicine, University of California, Los Angeles, CA 90095, USA d Chemical and Biomolecular Engineering Department, University of California, Los Angeles, CA 90095, USA † Electronic supplementary information (ESI) available. See DOI: 10.1039/c3nr01533e Cite this: Nanoscale, 2013, 5, 5644 Received 28th March 2013 Accepted 3rd May 2013 DOI: 10.1039/c3nr01533e www.rsc.org/nanoscale 5644 | Nanoscale, 2013, 5, 5644–5653 This journal is ª The Royal Society of Chemistry 2013 Nanoscale PAPER Published on 09 May 2013. Downloaded by University of California - Los Angeles on 26/05/2015 22:08:44. View Article Online View Journal | View Issue