A major challenge in evaluating inhalation toxicity is accurate determination of the delivered dose. In humans, breathing is a complex physical process (inhalation-pause-exhalation), and the complex anatomy of the respiratory tract makes it challenging to estimate the delivered doses to the cell surface (Alexander et al., 2008). After a substance is inhaled and deposited in the lung, particles can dissolve and absorb into the systemic/pulmonary circulation. Others are cleared from the lung by pulmonary metabolism or alveolar macrophages, and those deposited higher up in the respiratory tract are removed by mucociliary clearance (Clara et al., 2023).

The exact application of dosimetry measurements also largely depends on the exposure system being used, which should be selected based on the deposition and interaction of particles and vapors at the cell surface. Characterization of the exposure system is key to understanding the delivery of smoke/aerosol to the cell surface (Miller-Holt et al., 2023; Wieczorek et al., 2023). Ideally, dosimetry is the measure of the internal dose, or even the concentration at the molecular target (biologically effective dose) within the target cells for the chemicals of interest (Paustenbach, 2000; Escher and Hermens, 2004; Proença et al., 2021). However, directly measuring cellular dose in submerged cultures poses a significant obstacle to the application of target tissue dosimetry. For example, for nanoparticles and microparticle toxicity assessment, particularly for in vitro systems, due to nanoparticle agglomeration in liquids, which can alter the density of the nano particles has the ability to alter the particle transport and deposition, ultimately altering the dose response relationship (Hinderliter et al., 2010; Watson et al., 2016; Deloid et al., 2017; Thomas et al., 2018). As a consequence, the target tissue paradigm for dosimetry and hazard assessment for nanoparticles has largely been ignored in favor of using alternative indirect methods of potential or surrogate exposure such as μg particle/mL culture medium, particle surface area/mL, or particle number/mL for submerged cultures (Hinderliter et al., 2010).

Air-liquid interface (ALI) exposure is the most physiologically relevant approach to accurately determine both the dose deposited on the cell surface and the dose ultimately available to be absorbed by the cells. The most common methods are deposition of mass onto quartz crystal microbalance and the use of particle counters, photometers, or specialized gas analyzers. For a recent review of recommendations for conducting dosimetry studies in inhalable tobacco products please see the review by Miller-Holt and colleagues (Miller-Holt et al., 2023).

HCS enables investigative toxicity testing in vitro to provide knowledge about the affected biologic processes and functions. It generally refers to automated (high-throughput) microscopy, multi-parameter image processing, and visualization to extract quantitative data from cells growing in multi-well cell culture plates. HCS typically uses fluorescent imaging to trace the effect of chemicals on different toxicity pathways including oxidative stress, apoptotic cell death, DNA damage, and mitochondrial health and can be performed in a multiplexed fashion (Gonzalez-Suarez et al., 2016; Taylor et al., 2018; Czekala et al., 2019). In addition, HCS allows morphometric analysis for evaluating toxicant-induced effects on the morphology or size of cells and organelles. All these analyses can be performed with fixed cells or in a time-resolved mode by using live-cell imaging. HCS is also commonly used to monitor spatial (re)distribution of target molecules inside cells, which is pertinent to transcriptional activation following cell stress and inflammatory challenges (Czekala et al., 2019). Similar to flow cytometry, HCS enables information to be collected for multiple endpoints at an individual cell level or from an entire cell population, allowing analysis of dynamic ranges across treated cells in culture.

HCS methodologies are continuously expanding; they are now being applied to two-dimensional (2D) culture systems and three-dimensional (3D) organotypic tissue culture models (Booij et al., 2019). Artificial intelligence adds another level to HCS by allowing predictive in vitro toxicology analysis of new chemicals (Su et al., 2016; Lee et al., 2018).

HCS is generally perceived as a powerful in vitro screening technology for assessing the toxicity and efficacy of chemicals. There are also several associated limitations, especially as the technology is routinely used for screening cells growing in a 2D format:

• Lack of standardization

Several groups have employed HCS to assess the biological impact of NGP aerosol fractions (Gonzalez-Suarez et al., 2016; Taylor et al., 2018; Czekala et al., 2019). All of these studies demonstrated the utility of HCS as a tool for NGP product assessment. They also provided in vitro evidence for reduced biological impacts of fractions generated from HTPs and other nicotine-containing products compared to those of cigarette smoke fractions. Moreover, complementary test methods, such as those sensitive for oxidative stress toxicity pathways (gene-expression analysis or reporter-gene assays), confirmed the trends observed by HCS. Once qualified, HCS may be applied as a standard platform for modern toxicologic analysis of NGPs.

ToxTracker™ (Toxys, Oegstgeest, the Netherlands) is a high-content assay that employs a series of mouse embryonic stem cell reporter gene cell-lines (Czekala et al., 2021). It consists of six green fluorescent protein (GFP) reporter gene lines plus a control wild-type line, providing readouts on biomarkers for oxidative stress (Srnx1 Nrf2 dependent and Blvrb Nrf2 independent), DNA damage (Bscl2 and Rtkn), cell stress (Btg2), and protein damage/misfolding (Ddit3). ToxTracker has been qualified against >450 known compounds and possesses ≥95% sensitivity and selectivity for both the Ames and in vivo micronucleus assays for mutagenicity and genotoxicity, respectively (Hendriks et al., 2016). ToxTracker could prove particularly valuable as it exhibits good/excellent concordance with classical genotoxicity testing and also offers mechanistic data on mode of action. Accordingly, ToxTracker can be a potential screening tool and/or a follow-up assay to identify mode of action in a positive in vitro response.

Several caveats exist for the application of ToxTracker:

• Although excellent concordance has been shown with classical toxicological approaches, the assay utilizes a mouse (not human) embryonic cell line

The in vitro MultiFlow^® (Litron Laboratories, Rochester, NY, USA) genotoxicity flow cytometric assay multiplexes several biomarkers that are responsive to diverse forms of DNA damage. The multiplexed biomarkers include: 1) phosphorylation of H2AX at serine 139 to detect double-strand DNA breaks, 2) phosphorylation of histone H3 at serine 10 to identify mitotic cells, 3) nuclear p53 content as an indicator of p53 activation, 4) frequency of 8n + cells to monitor polyploidization, and 5) relative nuclei counts to provide information about treatment-related cytotoxicity (Bryce et al., 2016; Bryce et al., 2017; Dertinger et al., 2019). Some multiplexing adaptations to the methodology have been described for a more integrated genotoxicological approach, such as the combination of the MultiFlow^® with a flow-based in vitro micronucleus assays (Smart et al., 2020). The MultiFlow assay has also undergone significant inter-laboratory comparisons (Bryce et al., 2017). Eighty-four chemicals split between aneugen, clastogen, and nongenotoxin groups were collectively compared and cross-analyzed to determine inter-laboratory assay variability. Compared to historical mode of actions for the three class of chemicals, the MultiFlow assay demonstrated ≥92% sensitivity, specificity, and concordance. As a result, an excellent “training” list and established data are available to cross-reference results (Bryce et al., 2017).

Some limitations of MultiFlow include:

• The variability observed in the biomarker endpoints could confound result interpretation in less experienced laboratories. Analyzing multiplexed data requires careful interpretation and consideration.

Common in vitro models for studying the biological impact of inhaled toxicants on human airway epithelial cells are based on 2D cell culture systems. Tumor cell lines and immortalized primary epithelial cells (Calu-3, BEAS-2B, 16HBE14o-, NCI-H292, NCI-H441, RERF-LC-AI, and A549) grown in submerged (2D) culture conditions have been frequently used; however, they do not accurately recapitulate the native airway epithelia (Gordon et al., 2015). Although most of them can also be grown at the ALI (Haswell et al., 2010), drawbacks of using these epithelial cell lines for in vitro toxicity studies include their limited metabolic competency that might affect their responsiveness to toxic stimuli (Garcia-Canton et al., 2013), the absence or reduced formation of tight junctions, and their refractoriness to differentiation (Stewart et al., 2012). Non-differentiated primary epithelial cells grown in submerged conditions do not fully reflect native airway epithelia because they either lack the polarized morphology and expression of markers (ion channels) normally found in airway epithelia (Kunzelmann et al., 1996) or show altered responsiveness to solid-particle exposure relative to differentiated primary epithelial cells (Ghio et al., 2013). Due to technical convenience (propagation and possibility for higher throughput), 2D culture systems based on cell lines are most suited for screening purposes to investigate the toxicity of a greater number of test items. Transformation techniques based on hTERT/Cdk4 allowed the generation of immortalized cell lines from primary lung epithelial cells that can differentiate into mucin-producing and ciliated airway models that more closely resemble the lung epithelium (Vaughan et al., 2006).

Reconstituted 3D organotypic culture systems from primary epithelial cells grown at the ALI (Nichols et al., 2013) also replicate the cellular complexity of pseudostratified epithelium found in human airways, containing ciliated/non-ciliated epithelial cells as well as basal (progenitor) cells (Gray et al., 1996; Prytherch et al., 2011). Furthermore, they closely mimic epithelial functionality by means of cilia beating, apical mucus secretion, ion channel expression, and the presence of tight junctions (BeruBe et al., 2009). These models also match the metabolic competency of the native tissue (Iskandar et al., 2013) and are amenable to whole aerosol exposure. The last characteristic can make an important difference regarding the potential underestimation of the biological impact of complex mixtures such as cigarette smoke, because frequently used smoke extracts such as TPM or GVP do not contain the totality of cigarette smoke constituents.

Three-dimensional epithelial cells from different zones of the human aerodigestive tract are commercially available as ready-to-use ALI cultures under brand names such as EpiAirway™ (MatTek Corporation, Ashland, MA, USA) MucilAir™, and SmallAir™ (both from Epithelix Sàrl, Geneva, Switzerland). They can be cultured for weeks to months and may be used for repeated (chronic) exposure (Czekala et al., 2021). The primary alveolar test model systems is increasing in use and commercial availability, with models currently available from suppliers such as Epithelix, ImmuONE (Hatfield, UK), and Invitorlize (Belvaux, Luxembourg). The use of these models is relatively new and beyond the scope of this paper.

Even though 3D organotypic cell cultures closely resemble the native tissue, they usually lack the cellular components of the immune system. More complex model systems have been evaluated as potential candidates for routine aerosol testing (Lehmann et al., 2011; Klein et al., 2013; Marescotti et al., 2019). They use ALI conditions and combine immune cells from the innate immune system or fibroblasts with epithelial cell layers. Attempts to produce functional airway epithelial model systems originating from pluripotent stem cells are also promising and might provide additional mechanistic understanding of aerosol-induced toxicity while also allowing the generation of in vitro models from specific patient populations (Wong et al., 2015). No single in vitro model can answer every study question, so a pragmatic approach should be tailored depending upon the exact hypotheses.

The use of precision-cut lung slices (Behrsing et al., 2013; Sewald and Braun, 2013) and lung-on-chip model systems (Huh et al., 2010; Stucki et al., 2015) resembling additional aspects of the human respiratory tract provide opportunities for characterizing aerosol-induced toxicity in vitro, with an even greater relevance to native tissue and the possibility to emulate pharmacokinetics. Other approaches include using co-cultures of various organ cell types as a cell-based disease screening model (BioMap^®) to screen drugs and NGPs for potential adverse effects (Simms et al., 2021).

In vitro organ models have received considerable attention in the attempt to reduce animal experiments and decrease the rate of preclinical failure associated with drug development. Two major advances in the in vitro research field are helping address the weaknesses of the simple models used in the past, rendering them more physiologically relevant and predictive for the effects of compounds in humans: 1) simple 2D cell culture systems are increasingly replaced by organotypic culture systems with 3D architecture, and 2) static culture systems have been switched to dynamic exposure models. The merging of these two developments has culminated in OoC technology, which combines 3D organotypic culture systems (organs) with microfluidic devices (Alepee et al., 2014; Bovard et al., 2017).

Microfluidics enable interconnection of different organ models on the same chip platform by circulating culture medium from a common reservoir. These multi-organ-on-a-chip (MOC) devices permit organ-to-organ crosstalk, which allows compounds to be tested in a more in vivo-like environment (Rogal et al., 2017). To date, long-term, MOC-based co-cultures of different organ combinations have been established for liver spheroids with human 3D lungs (Bovard et al., 2018) or intestine (Maschmeyer et al., 2015a), neuronal (Materne et al., 2015) and pancreatic islet (Bauer et al., 2017) tissue models, as well as a skin–intestine–liver kidney chip (Maschmeyer et al., 2015b). At least two reported OoC systems have been designed for whole aerosol exposure of 3D lung models (Benam et al., 2016; Stucki et al., 2018). However, these platforms are not compatible with smoking machines or the in vitro exposure instruments commonly used in the tobacco industry.

Current OoC or MOC models have potentially improved the human relevance of in vitro assessment of chemicals (see Table 3 for systems strengths and challenges). While there are demonstrated examples in drug candidate profiling and investigative toxicology, their application for NGPs is currently in its infancy.

In summary, there are a number of established alternative in vitro models to animal testing that are more sophisticated than simple cell lines and suitable for investigating the biological impact of inhaled toxicants in an environment more relevant to human exposure. However, model validation remains important for ensuring acceptance by regulatory bodies. Additionally, inter-donor differences are difficult to capture in in vitro aerosol exposure studies, particularly with 3D organotypic ALI exposure, which is usually not sufficiently scalable to test a variety of donors (Mori et al., 2022). The development of a relevant human lung tissue model is of ultimate importance for method standardization; the in vitro model must have high reproducibility and clinical relevance for inhalation toxicity assessment. However, constructing an in vitro lung tissue model derived from human cells currently remains challenging due to the inherent complexity of the human lung and the number of different cell types involved (Petpiroon et al., 2023).

An AOP is defined as “an analytical construct that describes a sequential chain of causally linked events at different levels of biological organization that leads to an adverse health or ecotoxicological effect” as defined by the OECD (OECD, 2023). AOPs are used to help to organize the available mechanistic information relating to an adverse outcome into the key events (KEs) that are required for the adverse outcome, spanning all organizational levels of a biological system(s) (Lowe et al., 2017; Luettich et al., 2017). Key event relationships (KERs) define the relationship between a pair of KEs by showing which is up-/downstream and are supported by both biological plausibility and empirical evidence (Villeneuve et al., 2014). The use of AOPs to organize information into AOP networks has the potential to improve the use of mechanistic data and lead to improved regulatory decision making (Villeneuve et al., 2014). In this way, the use of AOPs can greatly improve the biological understanding of a particular disease process through a simplified series of events. The critical element is causality as the AOP moves from 1 KE to another (i.e., KE 1 always occurs before KE 2, etc.). AOPs can also help link biological exposures to the eventual toxic effects at the population level.

In terms of regulatory context, knowledge of disease mechanisms can guide the design of testing strategies using in vitro methods that can measure or predict KEs relevant to the biological effect of interest. AOPs are not chemical specific, but they link the molecular initiating event (MIE) for a chemical to the apical end point, which is the observable outcome in the whole organism and typically a clinical sign or pathological state (Burden et al., 2015). The well-considered use of AOPs could drive positive changes in toxicology testing, moving toward less reliance on making predictions based on animal models and focusing on the measurement of apical toxicity endpoints (Burden et al., 2015).

The vast majority of in vitro toxicity testing is restricted to one-time (acute) treatment of cells with a test item for a relatively short exposure duration (hours to a few days) before endpoint measurement. Such tests include escalation of the dose to a point where toxic (adverse) effects are either visible at the morphological level or quantifiable by standard cytotoxicity (viability) tests. While dose–response analyses in cell-based assays to determine acute effects on cell cultures are commonly used to rank the toxicity of different chemicals, they have several limitations: 1) the doses for inducing adverse effects are often selected based on potential toxicity hazard that are not relevant for human exposure levels (environmental or use levels); 2) therefore, the toxicity mechanisms observed in vitro, especially after high dosing for eliciting acute toxicity in cell-based assays also may not reflect relevant mechanisms in humans; and 3) they cannot fully predict the effects of subtoxic concentrations of a chemical when administered repeatedly (Ito et al., 2020; Czekala et al., 2021; Chapman et al., 2023). The latter is pertinent, as smoking-related diseases such as chronic obstructive pulmonary disease or cardiovascular diseases only manifest after chronic exposure to cigarette smoke (Yoshida and Tuder, 2007; Luettich et al., 2021; Farcher et al., 2023). Chronic exposure may also result in distinct adverse effects not seen after acute exposure or cause adaptive responses that are not detectable following a single treatment. Adaptations often invoke compensatory repair mechanisms elicited upon chronic stress. These considerations are also relevant for testing inhalable toxicants in vitro and are reflected in a number of published studies on assessing NGPs through repeated exposure of cells, as discussed below.

While in vitro experiments have many advantages over in vivo testing in terms of increased human relevance as presented above, using in vitro data for the purpose of risk assessment (e.g., quantifying the margin of safety) has many challenges, including uncertainty around how to interpret and link exposure or dose. Quantitative in vitro-to-in vivo extrapolation (IVIVE) is an emerging NAM-based computational tool to facilitate this process, generally referred as an extrapolation of (human tissue-derived) cellular (in vitro) data to predict the exposures or outcomes in humans (in vivo). IVIVE implies certain quantitative extrapolations of “dose (or exposure)” and “response” to enable in vitro-based toxicological risk assessment (Zhang et al., 2021; Hines et al., 2022; Zhang and Wright, 2022). Dose extrapolation is necessary to estimate the human exposure or intake level (e.g., daily human equivalent dose) of individual toxicants, commonly by computational (e.g., physiologically-based pharmacokinetic [PBPK]) models and in silico and/or human-based in vitro data. The human equivalent dose can be extrapolated back to the corresponding exposure levels (commonly called “reverse” dosimetry) and compared to measured (environmental) exposures to evaluate the safety margin or guide regulatory safety decision-making. Therefore, one may first define and characterize the in silico or in vitro-based dose–response relationship, and the relevance of the MIEs or KEs can be selected based on the AOPs of interest. Then, using the in vitro dosimetry and disease-relevant response data and with the help of computational modeling, IVIVE enables “human-relevant” safety assessment, moving away from a need for conventional animal toxicity data.

While the concept of IVIVE is gaining interest in scientific communities (Bell et al., 2018; Chang et al., 2021; Hines et al., 2022), it is still an emerging field with limited application to regulatory risk assessment, especially for tobacco and nicotine products. The early successes—mostly from pharmaceutical and environmental safety assessment—typically involve single chemicals, although concerted effort is being made to apply IVIVE to mixtures (Chang et al., 2021; Zhang et al., 2021; Zhang and Wright, 2022). Many of the challenges in using IVIVE for tobacco or nicotine products are similar, including uncertainty in dose selection and quantification (which individual toxicants should be modeled as is or as mixtures in the complex and dynamically unstable aerosol mixtures) and the target test systems (which cellular/tissue systems should be used to reflect what clinically relevant outcomes at cellular levels). Additionally, there is a lack of chemical or mixture-related in vitro data that can be used for IVIVE model development and the independent verification of modeling outcomes.