Linking Nanomaterial-Induced Mitochondrial Dysfunction to Existing Adverse Outcome Pathways for Chemicals

The adverse outcome pathway (AOP) framework plays a crucial role in the paradigm shift of toxicity testing towards the development and use of new approach methodologies. AOPs developed for chemicals are in theory applicable to nanomaterials (NMs). However, only initial efforts have been made to integrate information on NM-induced toxicity into existing AOPs. In a previous study, we identified AOPs in the AOP-Wiki associated with the molecular initiating events (MIEs) and key events (KEs) reported for NMs in scientific literature. In a next step, we analyzed these AOPs and found that mitochondrial toxicity plays a significant role in several of them at the molecular and cellular levels. In this study, we aimed to generate hypothesis-based AOPs related to NM-induced mitochondrial toxicity. This was achieved by integrating knowledge on NM-induced mitochondrial toxicity into all existing AOPs in the AOP-Wiki, which already includes mitochondrial toxicity as a MIE/KE. Several AOPs in the AOP-Wiki related to the lung, liver, cardiovascular and nervous system, with extensively defined KEs and key event relationships (KERs), could be utilized to develop AOPs that are relevant for NMs. However, the majority of the studies included in our literature review were of poor quality, particularly in reporting NM physicochemical characteristics, and NM-relevant mitochondrial MIEs were rarely reported. This study highlights the potential role of NM-induced mitochondrial toxicity in human-relevant adverse outcomes and identifies useful AOPs in the AOP-Wiki for the development of AOPs for NMs

changes, inhibition of protein complexes in the electron transport chain, failure to produce enzymes that detoxify reactive oxygen species (ROS) (e.g., manganese superoxide dismutase), disruption of mitochondrial network formation and structure, impaired clearance of dysfunctional mitochondria through mitophagy, dysregulation of cytoplasmic and mitochondrial matrix transport of Ca 2+ ions, induction of pro-inflammatory and apoptotic pathways, or damage to the mitochondrial DNA (mtDNA), including also the induction of mtDNA adducts impairing mitochondrial transcription or alteration of the activity of DNA polymerase gamma (Wallace, 2012;Meyer et al., 2013Meyer et al., , 2018;;Vuda and Kamath, 2016;Vyas et al., 2016;Fetterman et al., 2017;West, 2017;Massart et al., 2018;Daiber et al., 2020).Owing to their unique characteristics and functionality, mitochondria are prone to be affected by various chemical stressors.The negatively charged and alkaline mitochondrial matrix enables the accumulation of amphiphilic xenobiotics and metals such as lead, cadmium, mercury, or manganese (Cohen, 2010), while the high lipid content of the mitochondrial membrane facilitates the internalization of lipophilic compounds such as polycyclic aromatic hydrocarbons (Backer and Weinstein, 1980).Moreover, mtDNA is more susceptible to damage compared to nuclear DNA, probably due to its vicinity to the electron transport chain, its lack of histones, and its deficiency in certain DNA repair mechanisms (Khalifa et al., 2021).
Occupational studies have revealed that exposure to chemical stressors such as pesticides, benzene, polycyclic aromatic hydrocarbons, metal-rich particulate matter, and particle-containing welding fumes is associated with mitochondrial dysfunction and mtDNA damage (Roubicek and de Souza-Pinto, 2017).Additionally, mitochondrial dysfunction can also be induced by a broad range of other stressors, including polychlorinated biphenyls, dioxins/furans, metals/metalloids (arsenic, lead, copper, chromium, cadmium, nickel, and vanadium), air pollutants (diesel exhaust and ambient ultrafine particles, sulfur dioxide, nitrogen oxides), tobacco smoke (Fetterman et al., 2017), and algal toxins (Jayasundara, 2017).
Clinical perspectives on chemical exposure and mitochondrial dysfunction have been excellently reviewed (Zolkipli-Cunningham and Falk, 2017;Gorini et al., 2018;Tang et al., 2022).Various organs and tissues (brain, heart, liver, kidney, pancreas, muscles, arteries) can be affected by mitochondrial dysfunction and damage (Hayden, 2022).Due to its high energy demand, the heart is one of the major organs where mitochondrial disturbances have marked consequences like myocardial infarction, cardiomyopathy, and heart failure, all of which have been linked to the accumulation of aberrant mitochondria (Kirichenko et al., 2022;Li et al., 2022).Moreover, accumulation of aberrant mitochondria due to impaired mitophagy (which selectively degrades damaged mitochondria) in the myocardium has been reported for several diseases, such as obesity, impaired glucose tolerance, type 2 diabetes mellitus, insulin resistance, and metabolic syndrome (Hayden, 2022).Furthermore, impairment of mitochondrial dynamics has also been linked Additionally, AOPs play an important role in the development of alternative strategies to animal testing to inform various steps of the human health risk assessment (RA) process (OECD, 2020a).
RA related to nanomaterials (NMs) is challenging due to their large number and the extensive variability in their physicochemical properties and associated nano-specific effects.AOPs provide a framework for understanding the molecular and cellular events that lead to adverse outcomes (AOs) and can help to identify the mode of action specific to critical properties of NMs (Vietti et al., 2016;Rolo et al., 2022).By utilizing AOPs, researchers and regulators can overcome the challenges posed by NMs, which will lead to more accurate and comprehensive risk assessments (Rolo et al., 2022).
AOPs are stressor-agnostic, and consequently, AOPs developed for chemicals are also in principle applicable to NMs (Halappanavar et al., 2020b).In recent years, efforts have been made to develop nano-relevant AOPs by adapting existing AOPs to NMs (Nymark et al., 2018;Brand et al., 2020;Halappanavar et al., 2020a;Murugadoss, 2021).Based on a nanotoxicological systematic literature review, an AOP network relevant to NM was proposed by combining linear AOPs for lung fibrosis (AOP 173), lung emphysema (AOP 1.25), acute lung toxicity (AOP 302), lung cancer (AOP 303), and atherosclerotic plaque formation (AOP 237) as an outcome of the OECD Working Party for Manufactured Nanomaterials (WPMN) project on advancing the development of nano-relevant AOPs (Halappanavar et al., 2020a).By leveraging AOP 144 and AOP 34, AOPs associated with steatosis, edema, and fibrosis of the liver based on existing information on TiO 2 , including its nanoform, have been postulated (Brand et al., 2020).AOP 144 has also been used to propose a putative liver fibrosis AOP based on available information for metal oxide NMs (Gerloff et al., 2017).Several ongoing EU and international projects, including RiskGONE, seek to further the development and utilization of AOPs for NMs (Ede et al., 2020).
Recently, we presented a simple and testable strategy to develop AOPs for NMs based on existing AOPs (Murugadoss et al., 2021b).This work was focused on searching molecular initiating events (MIEs) and key events (KEs) reported for NMs in scientific literature and identifying AOPs associated with these MIEs and KEs in the AOP-Wiki.In a next step, we analyzed these AOPs and found that mitochondrial toxicity plays a significant role in several of them at the molecular and cellular levels.Although there is a growing body of studies on NM-induced mitochondrial toxicity (Wu et al., 2020), the extent to which it contributes to adverse effects is not yet fully understood.
Stressor-induced mitochondrial toxicity primarily results in mitochondrial dysfunction.The term "mitochondrial dysfunction" encompasses a wide variety of changes in the structure and functioning of the mitochondria.The most reported dysfunction is disturbance in the production of adenosine triphosphate (ATP) via oxidative phosphorylation.Other dysfunctions encompass the loss of mitochondrial membrane potential (MMP) and pore permeability tion of keywords used in the Web of Science database to search for relevant studies included [(nanoparticles or nanomaterials) AND (mitochondria) AND (in vitro)].The search resulted in 1473 papers in total for studies published before 25/01/2023.After screening and applying different exclusion and inclusion criteria (see Fig. 2), 78 studies covering in vitro experiments on human cells and pristine NMs were selected for further analysis.
These 78 studies were then evaluated by a data quality scoring approach based on the GUIDEnano system (Fernández-Cruz et al., 2018).This approach follows the principles of the Klimisch score, related to test design and reporting considerations, and it is complemented by the GUIDEnano S score, which considers the reported physicochemical properties of the NMs, including those characterized in the exposure medium.The Klimisch score and the S score are combined to obtain an overall quality score (Q score), which is numerically classified as follows: Q = 1 (very high quality), Q = 0.8 (high quality), Q = 0.5 (medium quality), and Q = 0 (unacceptable quality).
To ensure the correct identification of mitochondrial MIEs and KEs and their reliable connection to AOPs in the AOP-Wiki, we analyzed the descriptions of mitochondrial KEs in Table 1 and summarized mitochondrial toxicity endpoints and related methods/assays to measure them (Tab.2).This includes endpoints such as mitochondrial complex inhibition, decrease in oxidative phosphorylation, MMP, ATP production as well as increase in mitochondrial ROS production and mitochondrial damage/dysfunction/disruption.These endpoints were then used to design a template, which was used to extract and consolidate data from the selected studies.
We extracted data from the selected studies and analyzed it in two steps: 1) Analysis and evaluation of the evidence for NMinduced mitochondrial toxicity.This step summarized the studies reporting mitochondrial and cellular toxicity endpoints induced by to vascular endothelial dysfunction (Qu et al., 2022).Mitochondria have been shown to play a crucial role not only in cardiovascular disturbances but also in several neurological disorders, such as Alzheimer's disease, Parkinson's disease, Huntington's disease, ischemic stroke, traumatic brain injury, and epilepsy (Cabral-Costa and Kowaltowski, 2020; Norat et al., 2020;Delp et al., 2021).
In light of the potential adverse effects induced by NMs, it is crucial to identify the role of NM-induced mitochondrial toxicity.This can be achieved by utilizing the AOP framework.The objective of this study was to generate hypothesis-based AOPs related to NM-induced mitochondrial toxicity by integrating knowledge on NM-induced mitochondrial toxicity into all existing relevant AOPs in the AOP-Wiki, which already includes mitochondrial toxicity as a MIE/KE.

Methods
The approach to generate hypothesis-based AOPs related to nanomaterial (NM)-induced mitochondrial toxicity is schematically depicted in Figure 1.Firstly, we analyzed AOPs identified in Murugadoss (2021) that are associated with MIEs and KEs reported for NMs and found that mitochondrial toxicity plays a key role in several of them at the molecular and cellular levels (Tab.1).Furthermore, we performed a keyword search for "mitochondria" in the AOP-Wiki KEs, which revealed that mitochondrial toxicity is involved in several other AOPs not included in Table 1.Here, we aimed to establish a plausible link between NM-induced mitochondrial toxicity and existing AOPs in the AOP-Wiki and to propose a conceptual AOP (network) relevant to human health effects.
To identify relevant studies, we conducted a literature review on mitochondrial toxicity induced by NMs (Fig. 2).The combina-

Endpoints related to KEs provided in Table 1
Methods / assays Mitochondrial complex inhibition Complex inhibition assays specifically for complex III, IV, V, MitoTox (inhibition of complex I), enzyme activity assays, mitochondrial membrane potential (MMP) measurement using fluorescent dyes Decrease in oxidative phosphorylation MMP measurement using mitochondrial dyes (JC-1, rhodamine 123, DiOC6, TMRE), extracellular lactate reflecting an increase in glycolytic rate (colorimetric assay), oxygen consumption measurement using the Seahorse assay Decrease in MMP MMP measurement using mitochondrial dyes (JC-1, rhodamine 123, colorimetric DiOC6, TMRE) Decrease in adenosine triphosphate (ATP) synthesis ATP bioluminescent assay, ATP synthase assay, Complex V activity assay Increase in mitochondrial reactive oxygen species (ROS) MitoROS (targeting mitochondrial ROS) Increase in mitochondrial injury/damage/disruption Cellular oxygen consumption, MMP, enzymatic activity of the electron transport system, ATP content also included.A summary of the findings with positive outcome is presented for each target organ/tissue/system in Table 3.The analysis revealed that multiple studies have investigated the mitochondrial and cellular toxicity induced by various types of NMs across different human cell types that are related to the lung, liver, skin, cardiovascular, immune or nervous systems.Based on these findings, AOPs related to these target organs/tissues/systems in the AOP-Wiki were considered for further investigation.The low number of studies for other target organs/tissues, such as the kidney or eye, may be due to the scarcity of available cell models, or at least their infrequent utilization in toxicology testing.

Plausible linking of NM-induced mitochondrial toxicity to existing AOPs in the AOP-Wiki
The summary of cellular toxicity endpoints in the studies reporting mitochondrial MIEs and KEs (Tab.3) suggests that mitochondrial toxicity induced by NMs (upstream KEs) can be potentially linked to cellular toxicity endpoints such as cytotoxicity, oxidative stress, inflammation, and DNA damage (downstream KEs).The biological plausibility of mitochondrial toxicity leading to oxidative stress and/or cytotoxicity, and their subsequent cellular toxic responses, such as pro-inflammatory responses and DNA damage, are well established in the literature as well as through the KERs described in the AOP-Wiki (Tab.4).We have also indicated the status of the AOPs in which these KERs are included.
To make a biologically plausible and causal link of NM-induced mitochondrial and cellular toxicity to existing AOPs, we searched for AOPs in the AOP-Wiki related to lung, liver, skin, cardiovascular, immune, and nervous systems using the keywords indicated in Table 5.Then, it was assessed whether a given AOP contains NMs (such as titanium dioxide (TiO 2 ), amorphous silica (SiO 2 ), silver (Ag), etc.) in human cell types that are representative of a potential target organ/tissue (such as lung, liver, etc.) or system (such as nervous).2) Establishment of a biologically plausible link between NM-induced mitochondrial toxicity and existing AOPs in the AOP-Wiki and proposal of a conceptual AOP network relevant for NMs.This step focused on finding AOPs related to each target organ/tissue/system in the AOP-Wiki that had mitochondrial endpoints identified in step 1 and linking NM-induced mitochondrial toxicity to these AOPs via other cellular toxicity endpoints.

Evidence analysis of mitochondrial toxicity induced by different NMs
Supplementary file 11 includes the list of selected studies, extracted raw data, and evaluation of these studies using a data quality scoring approach based on the GUIDEnano system.All studies are also marked by quality Q score.However, in order to address the gaps in the available data, we did not exclude studies based on their data quality but included all 78 studies in our data analyses.Indeed, only 25 out of the 78 studies had an acceptable Q value with very high, high or medium quality.
In supplementary file 22 , we organized the data from individual studies related to mitochondrial and cellular toxicity endpoints, cell types used, and their corresponding organs/tissues/system.Each study was assigned to one row in the file, and information on whether a positive or negative outcome was observed for each endpoint as well as the Q score associated with each study was  MMP, ATP production, and oxygen consumption, as well as an increase in mitochondrial ROS production and physical damage to mitochondria are widely reported as a sign of mitochondrial dysfunction at the organelle or cellular level.Thus, we centralized mitochondrial dysfunction as the subsequent KE hub and linked it to oxidative stress, cytotoxicity, inflammatory responses, and DNA damage, including interconnections.Figure 3 presents this conceptual AOP network connecting different AOPs related to different target organs/tissues/systems.

Discussion
In this study, we applied a systematic approach and established a biologically plausible link between mitochondrial toxicity induced by NMs and already existing AOPs in the AOP-Wiki.This strategy can both inform on the potential role of NM-induced mitochondrial toxicity in several human-relevant AOs and identify potential AOs and AOPs in the AOP-Wiki that can be prioritized in the further development of nano-relevant AOPs.AOPs are, in principle, designed to be modular with re-usable elements and stressor-agnostic.Our results demonstrate how several components from the AOP-Wiki that were extensively defined (such as KEs and KERs) can be utilized to develop NM-relevant AOPs, rather than investing considerable resources in developing new nano-related AOPs.
KEs related to mitochondrial and/or cellular toxicity endpoints (cytotoxicity, oxidative stress, pro-inflammatory responses, and DNA damage).This resulted in 28 AOPs that include the selected target organs/tissues/systems.Among them, several AOPs that can be potentially linked to mitochondrial toxicity induced by NMs were identified for cardiovascular (n = 9), lung, (n = 7), liver (n = 6), and nervous systems (n = 6).No potential AOPs related to the mitochondrial toxicity for the immune system and skin were found in the AOP-Wiki (Tab.5).A summary of AOPs with mitochondrial and/or cellular KEs corresponding to the KERs (indicated in Tab. 4) is given in Table 6.The following AOs were identified for each target organ/tissue/ system: Lung: decreased lung function, lung cancer, lung fibrosis, and mesothelioma; liver: liver steatosis, liver injury, liver fibrosis, immune-mediated hepatitis, cholestasis, and liver cancer; nervous system: neurodegeneration, parkinsonian motor deficits, and impairment of learning and memory; cardiovascular system: heart failure, decreased cardiovascular growth, increased cardiovascular morbidity, and mortality of cardiovascular diseases.
Inhibition of mitochondrial complexes, uncoupling of oxidative phosphorylation, and mtDNA damage are identified as potential MIEs leading to mitochondrial dysfunction (Dreier et al., 2019), and these were also observed in some studies concerning NMs (in Tab. 3).Thus, we propose these endpoints as MIEs for the conceptual AOP network on NM-induced mitochondrial toxicity.Table 3 shows that other endpoints, such as decrease in mitochondrial  ence could be a confounder in broader contexts such as saferby-design approaches, which require linking of material-specific properties to the toxicological outcome.As previously indicated (Barbir et al., 2021;Murugadoss et al., 2021a;Cheimarios et al., 2022), to reliably link physicochemical properties to toxicological effects, one should start with the standardization of NM dispersion protocols and experimental conditions.In this way, one can utilize the data from the literature to perform a meta-analysis with minimal bias introduced by experimental conditions and establish reliable linking.Several dispersion protocols have been established, such as NANOGENOTOX and ENPRA (Hartmann et al., 2015;Deloid et al., 2017) or Nanodefine (Mech et al., 2020), which can be applied to many types of NMs, but ideally, there is also a regulatory ambition to move towards a one substance-one assessment approach.Secondly, systematic and case studies should be designed to establish an in-depth understanding of the influence of material-specific properties, such as primary size, shape, dissolution, surface properties, surface reactivity, crystal phase, surface functionalization, and exposure medium-specific properties such as agglomeration in different experimental conditions.Alternatively, the same NM should be tested under different experimental conditions such as with and without serum, and NM characteristics such as catalytic property, the release of ions, etc., should be thoroughly characterized in each condition.In this way, we can establish a scientific understanding of the link between material-specific properties and agglomeration, and therefore, a better understanding of the association of materialspecific properties with the observed toxicity.
Establishing a dose-effect relationship for NMs is vital for hazard characterization, hazard potency ranking, and hazard testing according to the safer-by-design principles.However, establishing the dose-effect dependency for NMs based on existing literature is challenging for several reasons.One of the main difficulties is the absence of a common dosimetry approach in reporting such data, which may cause unreliable and biased conclusions about NMs' hazard properties.At the scientific and regulatory level, there is still no consensus on the best approach and which units should be used for reporting dose-response results following exposure and treatments with NMs (mg/L or number of particles/L or specific surface area/L).Furthermore, the administered dose can differ greatly from the delivered dose reaching the cells.Under in vitro submerged conditions, NMs are introduced into the cell culture medium, and the subsequent settling of NMs depends on their density, size, and the properties of the medium (Pyrgiotakis et al., 2013).Moreover, in a submerged environment, NMs often experience dynamic agglomeration in the medium, which significantly impacts the dose reaching the cells.Approaches are now available to model delivered dose including the distorted grid (Deloid et al., 2017) or ISD3 model (Thomas et al., 2018).These models require the effective density of NMs as input, among others.Effective density is the density of NM in the dispersion medium, and in the case of agglomerates, this includes the density of medium trapped inside the agglomerates.The effective density can be measured using the volume centrifugation method or analytical ultracentrifugation.
In order to identify and address gaps in the available data, the quality of the studies was analyzed using the GUIDEnano approach (Fernández-Cruz et al., 2018).The results of this analysis revealed that only 25 out of 78 studies had an acceptable Q score, which is a result of a combination of Klimisch and S scores (supplementary file 1 1 ).Further analysis of the scores showed that the S score, which is based on the reported physicochemical properties of the NMs, had a greater impact on the overall Q score.A closer examination of these physicochemical data revealed that most studies failed to report relevant information such as endotoxin content, impurities, NM concentration, and NM stability in the exposure medium.In addition, the surface area and hydrodynamic diameter of the NM at the beginning and/or end of the exposure period were rarely reported.To improve the overall quality, reusability, and reliability of the data for specific purposes, such as the development of nano-relevant AOPs with less overall risk of bias and/or meta-analyses, it is important to address these characteristics in future in vitro nanotoxicity studies.
Linking physicochemical characteristics of the test material with the toxicological responses is crucial in NM hazard assessment and for safer-by-design approaches to develop and promote the use of safer NMs.However, such linking is not yet possible in our study due to the lack of experimental data related to the characterization of NMs before and after dispersing in the exposure medium.There is some consensus on the minimal set of materialspecific properties (e.g., size, shape, surface area, chemical composition, surface charge, surface reactivity, agglomeration/aggregation, and solubility) that are essential to be evaluated in NM toxicological assessment (ISO, 2012).When assessing the studies in our review, we found that the majority of the studies did not report this minimal set of characteristics, which makes it difficult to even predict the toxicological behavior of NMs with the same composition.
Material-specific properties can be largely influenced by experimental conditions such as the use of serum.The reactive surfaces of NMs can interact with their environment, and this may lead to the formation of a corona (e.g., of proteins) on the particle surface, which can further modify their chemical behavior (Barbir et al., 2021).NMs also tend to agglomerate in cell culture media, which can affect their cellular uptake behavior.On the other hand, sonication, a widely used technique to disperse NMs, can introduce changes to size and shape and destroy the surface properties of the NMs under consideration.We have shown (Murugadoss et al., 2021a) that the agglomerate size in exposure medium is an important nanodescriptor to assess the toxicological effects of TiO 2 NMs in in vitro submerged conditions and emphasized that the agglomeration state of NMs can be potentially influenced by in vitro exposure conditions.Upon examining the data in supplementary file 1 1 , it becomes clear that the amount of serum and the applied sonication protocols used varied greatly across different studies.Moreover, as indicated previously, in addition to other characteristics, the agglomeration of the NMs at the start and/or end of the exposure period was also rarely reported in the studies reviewed and analyzed here.Determination of the influence of exposure conditions on physicochemical characteristics is crucial because such influ-coupling, redox cycling, or inhibition of protein complexes (Dreier et al., 2019).Future nanotoxicity studies should characterize nano-relevant mitochondrial MIEs as well as subsequent KEs.
AOPs can serve multiple purposes in the RA framework, particularly to inform integrated approaches to testing and assessment (IATA) and as an integral part of NGRA workflows (Bajard et al., 2023).AOP-based IATAs integrate NAMs and mechanistic knowledge for hazard characterization within a specific regulatory context, potentially eliminating the need for animal testing and fully supporting the 3Rs concept (Russell and Burch, 1959).The AOP framework can be particularly useful in identifying the most suitable assays for measuring MIE or KEs that can assess the likelihood of an AO (van der Zalm et al., 2022).Several IATA OECD case studies are already based on AOPs describing pathways for non-genotoxic carcinogens, skin sensitization, chemicalinduced liver steatosis, and neural development to assess the applicability of in vitro testing batteries for hazard identification and characterization (OECD, 2017;Jacobs et al., 2020;Bajard et al., 2023;Kubickova and Jacobs, 2023).These case studies illustrate that AOPs can help increase confidence in the predictive capabilities of NAMs and further promote their regulatory acceptance.For example, the endorsed AOP 3, which includes mitochondrial dysfunction as a KE, has informed an OECD IATA case study on the identification and characterization of the parkinsonian hazard liability of rotenone and deguelin, two structurally similar mitochondrial complex I inhibitors (Alimohammadi et al., 2023).
As shown here, NM-induced mitochondrial dysfunction has also been associated with cardiovascular disease (Tab.6).The heart is particularly vulnerable to changes in energy production, as it is the organ with the highest energy demand per kilogram (Wang et al., 2010), and proper cardiac contractile function requires a constant supply of ATP (Werbner et al., 2023).Therefore, mitochondrial dysfunction can cause cardiomyocyte death, resulting in increased cardiac remodeling and, consequently, an elevated risk of heart failure (Werbner et al., 2023).However, the current RA of chemicals, including NMs, does not adequately cover cardiotoxicity (Schaffert et al., 2023).Developing AOPs that specifically address cardiovascular AOs through mitochondrial dysfunction may be particularly useful to improve the regulatory safety assessment of cardiotoxicity.There are ongoing efforts to address this need, such as the EU Horizon 2020 project ALTERNATIVE 3 , which aims to develop AOPs based on mitochondrial dysfunction leading to heart failure via oxidative stress or ATP production decrease (AOP-Wiki AOPs 479 and 480, respectively).The mechanistic knowledge of these AOPs will be used for drafting of an IATA addressing cardiotoxicity assessment.
The implementation of the AOP framework in RA can bridge the gap between mechanistic toxicological data and regulatory safety assessment for NMs.Ideally, AOPs suitable for RA should undergo a thorough weight-of-evidence evaluation process and be reviewed and endorsed/approved by experts.However, the number of such AOPs is still limited.Among the AOPs identified here that are NM-relevant and cover mitochondrial KEs, the major-Underlining this significance, utilizing the volume centrifugation method and distorted grid model, we previously showed that about 56-58% of the applied doses of nano-TiO 2 were delivered in a 24-hour exposure, whereas only 7-9% of nano-SiO 2 was delivered at the same time under identical experimental conditions (Murugadoss et al., 2020a,b).This difference might be attributed to the density and effective density of the nano-SiO 2 being similar to that of the density of the cell culture medium.Using a variety of NMs, Pal et al. (2015) demonstrated that the delivered dose can differ significantly from the administered dose.Correcting for the delivered dose led to a substantial shift in the hazard ranking of several NMs, aligning more closely with in vivo inflammation data.In the studies underlying our review, NM-induced in vitro effects were typically observed at unrealistically high doses.However, most studies were conducted under in vitro submerged conditions and presented results in terms of administered doses/ concentrations without evaluating the delivered doses/concentrations.This omission currently obstructs meaningful hazard potency evaluations, emphasizing that the estimation of delivered dose under in vitro submerged conditions is essential for meaningful future studies.An alternative solution involves using an air-liquid interface (ALI) for inhalation or ingestion exposures.
Here, NMs can be directly administered to the cells cultured at ALI, and a quartz crystal microbalance can offer precise measurements of the deposited doses.In line, it has been shown that NMs delivered via the ALI can cause toxic effects at significantly lower deposited doses that when administered in submerged conditions (Diabaté et al., 2020;Bessa et al., 2021), indicating that toxicity observed in submerged conditions could underestimate NM-induced effects due to discrepancies between administered and delivered doses.Another option to address this issue is a web application for cellular dosimetry based on the distorted grid model that enables the prediction of the NM concentration reaching the cell surface (Cheimarios et al., 2022).The use of this open access web tool allows correlation of the real exposure concentration with the observed toxicity, which may greatly increase the reliability of toxicity data.
The identification NM-relevant MIEs is also crucial in the development of predictive models for MIE activation, including quantitative structure-activity relationship (QSAR) models, which assume that different compounds showing similar structural features have similar mechanisms of action and induce similar toxicological effects (Singh and Gupta, 2014).These models could be used to predict whether a given NM with certain characteristics would trigger the MIE and thus could be useful for initial screening or prioritization of NMs for hazard assessment.The development of predictive models defining MIEs and early KEs is one of the long-term perspectives in the next generation risk assessment (NGRA), according to the OECD (2020b).Our analysis showed that most studies investigated mitochondrial toxicity as a KE but not as a MIE.This means that insufficient attention has been paid so far to the direct interaction of NMs with mitochondria, or potential mitochondria-related MIEs, such as mtDNA damage, un-

Fig
Fig. 1: Schematic workflow of the approach to generate hypothesis-based adverse outcome pathways (AOPs) related to nanomaterial (NM)induced mitochondrial toxicity MIEs, molecular initiating events; KEs, key events; KERs, key event relationships; AOs, adverse outcomes

Fig
Fig. 2: Systematic selection process to identify relevant studies

Target Cell line or type NM No. of Type of mitochondrial toxicity endpoint Cellular toxicity endpoints organ/system studies observed (positive outcome) (positive outcome)
BECs, bronchial epithelial primary cells; HPAEpiC, human pulmonary alveolar epithelial cells; HUVECs, human umbilical vein endothelial cells; LUHMES, Lund human mesencephalic; HCASMC, primary human coronary artery smooth muscle cells; HPAEC, primary human pulmonary artery endothelial cells; VSMC, vascular smooth muscle cells; PBMCs, peripheral blood mononuclear cells; HEK, human embryonic kidney; hCECs, human corneal epithelial cells; hCjECs, human conjunctival epithelial cells; MM, multiple myeloma 3: Proposed conceptual adverse outcome pathway (AOP) network linking nanomaterial (NM)-induced mitochondrial and cellular toxicity to different AOPs related to different organs/tissues/systems The dotted lines indicate that additional key events (KEs) are included in the referenced AOP but they are not included in this figure.Tab.