A Network Toxicology Approach for Mechanistic Modelling of Nanomaterial Hazard and Adverse Outcomes

Abstract

Hazard assessment is the first step in evaluating the potential adverse effects of chemicals. Traditionally, toxicological assessment has focused on the exposure, overlooking the impact of the exposed system on the observed toxicity. However, systems toxicology emphasizes how system properties significantly contribute to the observed response. Hence, systems theory states that interactions store more information than individual elements, leading to the adoption of network based models to represent complex systems in many fields of life sciences. Here, they develop a network-based approach to characterize toxicological responses in the context of a biological system, inferring biological system specific networks. They directly link molecular alterations to the adverse outcome pathway (AOP) framework, establishing direct connections between omics data and toxicologically relevant phenotypic events. They apply this framework to a dataset including 31 engineered nanomaterials with different physicochemical properties in two different in vitro and one in vivo models and demonstrate how the biological system is the driving force of the observed response. This work highlights the potential of network-based methods to significantly improve their understanding of toxicological mechanisms from a systems biology perspective and provides relevant considerations and future data-driven approaches for the hazard assessment of nanomaterials and other advanced materials.

Keywords: adverse outcome pathways; complex systems; engineered nanomaterials; network theory; systems toxicology; toxicogenomics.

Figure 1

Approaches for the analysis and interpretation of the MOA of ENMs starting from toxicogenomics data. In step 0, toxicogenomics data are generated and preprocessed. In step 1, traditional methods represent molecular alterations in terms of lists of differentially expressed genes (DEGs), while other studies (including this study) model them as co‐expression networks. In step 2, previous approaches reconstruct the mechanism of action (MOA) via functional annotation or enrichment of adverse outcome pathway (AOP) associated genes, leading to an arbitrary interpretation of the retrieved functions (step 3). The proposed study allows a reconstruction of an AOP based network of alterations, requiring minimal interpretation with respect to previous approaches.

Figure 2

Workflow converting molecular profiles in networks based on the AOP framework. Network models of the exposure (co‐expression networks or preselected edgelists) are used as input of the framework. In step 1, a topological enrichment of the adverse outcome pathway (AOP) network is performed based on the input data. In step 2, possible molecular initiating events (MIEs, blue squares) and adverse outcomes (AOs, red squares) connecting to the enriched events (KEs, green squares) are prioritized based on topological properties of the AOP network and molecular alterations. In step 3, the list of MIEs, KEs and AOs is mapped on the AOP network to form an AOP‐based interpretation of the mechanism of action (MOA). The results can be visualized as a network or in tabular form, and interpreted independently or compared with other exposure mechanisms.

Figure 3

Reconstructed mechanism of response to titanium dioxide (TiO₂) exposures with varying shapes, i.e., spherical (A) versus rod‐shaped (B). Enriched key events (KEs), molecular initiating events (MIEs) and adverse outcomes (AOs) are reported in green, blue, and red, respectively. Dashed lines connect events which are not formally linked in the AOP‐Wiki database, but are related to common biological processes (not provided as output of the framework).

Figure 4

Graphical representation of the defined information score based on the possibility of reconstructing the mechanism of action (MOA) of the compound. A) The “AOP completeness score” was defined as a measure of the ability to observe a complete chain of events connecting one or more molecular initiating events (MIEs) and one or more adverse outcomes (AOs). B) The “Key event connectivity score” was defined as the ratio between the portion of connected key events (KEs) against the unconnected ones characterizing the MOA of an exposure. For each score, pie charts representing the results in the information metric values when mechanisms of response are reconstructed based on lists of differentially expressed genes (DEGs) and co‐expression networks.

Figure 5

Distribution of molecular initiating events (MIEs), key events (KEs) and adverse outcomes (AOs) across engineered nanomaterials (ENMs) toxicity classes (low, medium, high). Data are reported as a percentage of the total enriched events.

Figure 6

Biological systems influence the response to ENM exposures. A) Clustering of the 93 exposures co‐expression networks based on topological properties (edge betweenness, Jaccard, Hamming distance, SMC distance between the existing edges of the networks as well as a percentage value of shared edges between each pair of input networks). Each measure is converted into a distance when possible, and a consensus is taken as input for the clustering. Clustering results are reported as a heatmap, where the individual 93 exposures and the 5 clusters are plotted on the y axis and on the x axis, respectively. B) Similarity between biological systems in exposures of the same material. In teal, exposures where the highest similarity is between lung and THP‐1, in violet when the highest value is between BEAS‐2B and lung, in gray when the in vitro systems are not representative of the in vivo counterpart. Engineered nanomaterial (ENM) chemistry has been annotated. C–E) Top fifteen enriched pathways of the overrepresented edges for each biological system, with their respective enrichment score (ES). The enrichment was performed using a weighted Kolmogorov‐Smirnov test against the information in the KEGG pathways database. The statistical significance of the enrichment analysis was estimated by permutation analysis over 100 random shuffles of the edge sets. The p‐values were corrected for multiple comparisons using the false discovery rate (FDR) method and setting 0.05 as the significance threshold.

Figure 7

Explorable AOP network is determined by the biological system. Subgraph of the AOP network containing the events enriched by the 31 engineered nanomaterials (ENMs) in the three biological systems: mouse lungs (orange), BEAS‐2B cells (blue), and THP‐1 cells (teal). Event type has been reported in the legend.