Readiness of the in vitro alveolar macrophage assay: an interlaboratory comparison using 12 blinded particles

1 Introduction

Adverse health effects caused by inhaled particles have been recognized for more than a century, particularly in occupational environments with sustained dust exposure (Zenker, 1867; Donaldson and Seaton, 2012; Donaldson et al., 2017). Persistent particles that reach the alveolar region can induce chronic inflammation and tissue damage when pulmonary clearance mechanisms are overwhelmed (Morrow, 1992; Driscoll and Borm, 2020). Particles with an aerodynamic diameter smaller than approximately 4 µm can reach the alveoli (European Committee for Standardization, 1993), where alveolar macrophages (AMs) represent the central defense system responsible for particle uptake, clearance, and the initiation of inflammatory responses (Oberdörster et al., 1992; Liu et al., 2025; Shan et al., 2026).

The macrophage response forms the mechanistic core of particle-induced lung toxicity. When macrophages efficiently clear deposited particles, inflammation remains transient and tissue homeostasis is preserved. However, when particle burdens exceed macrophage clearance capacity, clearance slows and persistent inflammation may develop. This phenomenon is conceptualized in the lung overload paradigm, which explains adverse responses to poorly soluble low-toxicity (PSLT) particles primarily as a consequence of excessive particle retention rather than intrinsic particle toxicity (Morrow, 1988; Borm et al., 2015; Driscoll and Borm, 2020; Stratmann et al., 2021; Ma-Hock et al., 2026). Within this framework, a fundamental distinction emerges between particles that cause biological effects through high retained particle burdens and those that trigger toxicity through specific physicochemical properties. Discriminating between these two mechanisms is a central challenge in inhalation toxicology and regulatory assessments.

The rapid emergence of engineered nanomaterials has intensified this challenge. Nanomaterials exhibit a wide diversity of physicochemical properties and large surface-area-to-mass ratios that can alter their biological interactions relative to larger particles (Oberdörster et al., 2005; Gottardo et al., 2017; Riediker et al., 2019). Consequently, many particulate materials now require toxicological assessment under regulatory frameworks such as the European Union’s REACH regulation. At the same time, regulatory policy increasingly emphasizes that animal studies should be used only when necessary, consistent with the principles of replacement, reduction, and refinement (3Rs) (European Commission, 2025). These developments create a strong demand for human-relevant testing strategies capable of evaluating particle toxicity while minimizing animal testing.

New approach methodologies (NAMs) are therefore becoming an increasingly important component of modern toxicology (Clippinger et al., 2018; Stucki et al., 2022; Haber et al., 2024). For particulate materials, NAMs should ideally capture the key biological processes underlying particle-induced lung toxicity. In particular, macrophage activation, oxidative stress responses, and inflammatory mediator release represent central events in the proposed adverse outcome pathways associated with particle exposure (ECETOC, 2013; Bos et al., 2019; Li et al., 2022). Mechanistically anchored in vitro assays targeting these processes therefore offer the potential to support early hazard identification and contribute to grouping and prioritization strategies for particulate materials.

One such assay is the alveolar macrophage assay (AMA), which was developed as a mechanistically informed in vitro method to evaluate the biological activity of inhaled particles. Building on earlier work by Rehn and colleagues (1999), Wiemann et al. (2016) established the AMA using the rat alveolar macrophage cell line NR8383. The assay measures four complementary endpoints: hydrogen peroxide (H₂O₂) formation, β-glucuronidase (GLU) release, lactate dehydrogenase (LDH) release, and tumor necrosis factor-α (TNFα) production. Together, these endpoints reflect key macrophage responses to particle exposure, including oxidative stress, lysosomal activation, cytotoxicity, and inflammatory signaling.

The AMA was designed to distinguish particles with intrinsic toxicity from particles that primarily act through macrophage overload. In the original study, a prediction model based on a “two-out-of-four” decision rule was introduced. According to this framework, particles are classified as biologically active if at least two endpoints show statistically significant responses below a predefined surface area-based dose threshold. It should be mentioned that most of the nanomaterials used in that study formed small agglomerates prone to gravitational settling so that cells had a chance to quantitatively phagocytose the test particles. Using this approach, the AMA successfully predicted the outcomes of standardized short-term inhalation studies (STIS) (Klein et al., 2012; Takebayashi et al., 2019) for a panel of nanomaterials with high sensitivity and specificity (Wiemann et al., 2016).

Beyond its initial development, the AMA has also been incorporated into particle grouping strategies such as DF4nanoGrouping (Arts et al., 2015; Arts et al., 2016; Janer et al., 2021a; Janer et al., 2021b). Within these frameworks, the assay contributes biological information that complements physicochemical characterization and helps differentiate biologically active particles from particles that behave as PSLT materials. Such approaches illustrate how mechanistically anchored NAMs can support regulatory decision-making by providing information relevant for grouping, read-across, and targeted testing strategies (ECETOC, 2013; Borm et al., 2015; DGUV, 2020; Driscoll and Borm, 2020; Stratmann et al., 2021; Driscoll, 2022; DGUV, 2023; Ma-Hock et al., 2026).

However, for any new toxicological method, reproducibility and transferability across laboratories are prerequisites for broader acceptance. Demonstrating that an assay produces consistent results when implemented by independent laboratories is therefore a critical step in the validation of NAMs. Ring trials and interlaboratory comparisons provide the empirical evidence required to evaluate assay reliability and to identify potential sources of variability in experimental procedures or biological systems (Jacobs et al., 2024).

The AMA has previously demonstrated good repeatability within individual laboratories and promising predictive capacity for in vivo inhalation outcomes. Nevertheless, broader application of the assay requires evidence that the method can be successfully transferred to independent laboratories and that comparable results can be obtained under different experimental conditions.

The objective of the present study was therefore to evaluate the transferability and interlaboratory reproducibility of the AMA in a blinded ring trial involving three independent European laboratories. Twelve particulate materials were tested using standardized protocols and predefined acceptance criteria. By examining assay performance and consistency of particle classification across laboratories, this study provides critical evidence supporting the further validation and potential regulatory use of the AMA as a NAM for inhalation toxicology.

If macrophage-driven mechanisms represent the central determinant of particle-induced lung toxicity, assays capturing these responses must demonstrate robust reproducibility across laboratories. To this end, the present ring trial evaluates whether the AMA can meet this requirement and thereby support grouping and hazard assessment of particulate materials in regulatory toxicology.

2 Materials and methods

2.1 Study scope and quality

The present study was designed as an international interlaboratory ring trial to evaluate the transferability and reproducibility of the AMA. It was not a formal regulatory validation study under Good Laboratory Practice (GLP). Only one participating laboratory operated under a full GLP quality system, whereas the involvement of academic laboratories without GLP accreditation reflects current practice in early-stage interlaboratory studies for novel in vitro methods. Accordingly, the study was conducted under good scientific practice without formal quality assurance unit (QAU) oversight.

To ensure robustness and comparability of results, several quality elements aligned with OECD Good In Vitro Method Practices (GIVIMP) were implemented, including harmonized SOPs, use of common test materials, standardized exposure protocols, predefined acceptance criteria, and centralized data evaluation. Interlaboratory variability, including instrument- and kit-specific differences, was therefore considered at the level of qualitative assay performance and particle binary assignment rather than absolute quantitative equivalence, consistent with the exploratory and comparative nature of the ring trial.

2.2 Test particles and suspension preparations

The selection of test particles considered (1) data availability (characterization and/or in vivo study), and (2) relevance to MACRAMÉ, which focused on advanced materials. Stock dispersions for all three laboratories were prepared by the lead laboratory (Lead Lab). Sample blinding was conducted by Bio-MS (a partner in MACRAMÉ project). Ten of the 12 materials were tested in the concentration range of 22.5–180 μg/mL, apart from SiO₂ NM-203 and ZnO NM-111, which were tested in a lower dose range (2.8–22.5 μg/mL) due to known toxicity.

The selected dose range of 22.5–180 μg/mL was not arbitrary: it was originally chosen to cover the in vivo mass-based dose metrics reported in the inhalation toxicology literature, particularly the work of Rehn et al. (1999), Bruch et al. (2014), and Pauluhn (2009). The highest concentration of 180 μg/mL corresponds to 120 pg/macrophage, which was considered by Bruch et al. (2014) as the in vivo upper value. Accordingly, the AMA adopted a mass-based metric to ensure that the in vitro exposures reflect physiologically relevant particulate burdens.

For the dispersion, the test particles were weighed in 20 mL glass vials (20 mL Scint-Burk, WHEA986581; Wheaton Industries Inc.) and water was added to obtain concentrations of 2.5 mg/mL (312.5 μg/mL for SiO₂ NM-203 and ZnO NM-111). Due to different dispersibilities, the dispersion protocols differed for the materials (see Table 1): ZnO NM-111 was dispersed using a probe sonicator (Branson Sonifier 450D, 6% amplitude) for 16 min according to the final protocol for producing suitable manufactured nanomaterial exposure media within the NANOGENOTOX European project (Jensen et al., 2011). All other materials were dispersed using an ultrasonic probe (Vibra cell, Sonics and Materials, USA) for 5 min or 10 s (see Table 1). The stock dispersions were adjusted with water to concentrations of 720 μg/mL or 90 μg/mL.

TABLE 1. Characterization and dispersion of tested materials.

DLS: dynamic light scattering, D: diameter, L: length, US: ultrasonic probe, s: seconds, min: minutes.

Each participating laboratory prepared the working dispersions of the distributed materials before each biological replicate. First, to achieve the correct concentration of salts, the particles were diluted 1:2 in double concentrated Krebs-Ringer-Phosphate-Glucose (KRPG) buffer (129 mM NaCl, 4.83 mM KCl, 0.54 mM CaCl₂, 16.5 mM NaH₂PO₄, 5.5 mM glucose, pH 7.3–7.4) for H₂O₂ assay or serum-free Kaighn′s Modification of Ham’s F-12 medium (F-12K, Sigma-Aldrich, cat. no. N3520) for GLU, LDH and TNFα assays. Then, the dispersions were further diluted 1:2 in KRPG buffer or F-12K medium to achieve the test concentrations.

2.3 Alveolar macrophage assay (AMA)

The AMA was performed as described by Wiemann et al. (2016) with some modifications. The method will be described briefly in the following sections.

2.3.1 Cell culture

NR8383 cells (rat alveolar macrophage cell line, ATCC - CRL-2192) were cultivated in F-12K media supplemented with 15% of heat-inactivated (30 min at 56 °C) fetal bovine serum (FBS, PAN Biotech, cat. no. P30-3302) and with addition of 2 mM L-glutamine (PAN Biotech, cat. no. P04-80050), 100 U/100 μg/mL penicillin/streptomycin (PAN Biotech, cat. no. P06-07100), and 2.5 g/L sodium bicarbonate. Cells were maintained in a humidified atmosphere at 37 °C and 5% CO₂. For the subculture procedure, cells were left at room temperature (RT) for approximately 15 min to allow cells to detach. Cells were subcultured weekly at a ratio of 1:2.

For the assays, 3 × 10⁵ cells/well were seeded onto two 96-well plates previously coated with fibronectin (0.01 mg/mL in PBS, PAN Biotech, cat. no. 2705001S) and in a final concentration of 5% FBS. Cells were incubated from 18 to 24 h for stabilization.

2.3.2 H₂O₂ formation

After 18 h of stabilization, cells were exposed either to test particle suspensions (100 µL in KRPG buffer), or to KRPG buffer alone serving as the negative control (NC), or to Zymosan (final concentration 360 μg/mL, Merck, cat. no. Z4250) as the positive control (PC). To verify assay performance and the accuracy of the colorimetric measurement, control wells containing 30 µM H₂O₂ were additionally included. Next, 100 µL of reaction mix [0.1 mM Ampliflu Red (Merck, cat. no. 90101), 2 mM NaN₃, and 2 U/mL horseradish peroxidase (Merck, cat. no. P8250) in KRPG buffer] were added to the wells. Note that in this test all particle concentrations administered with the 100 µL KRPG were doubled to compensate for the dilution caused by the 100 µL reagent mix. The plate was incubated at 37 °C and 5% CO₂ for 90 min. The absorbance was read at 570 nm and 595 nm (reference wavelength). The optical density (OD) measurements were subtracted from cell-free particle controls (for each test concentration) to correct for test material adsorption and light scattering. The H₂O₂ concentration in the supernatant was calculated according to Lambert-Beer’s Law and using the molar extinction coefficient of resorufin (54,000 L × mol⁻¹ × cm⁻¹).

2.3.3 β-Glucuronidase (GLU), lactate dehydrogenase (LDH) and tumor necrosis factor (TNFα) assays

Following 24 h of stabilization, the cells were either exposed to 200 µL/well of test particles in serum-free F-12K, or to 200 µL of serum-free F-12K alone (NC), or to 1% Triton X-100 (PC for LDH and GLU) or 0.5 μg/mL LPS (PC for TNFα). The cells were incubated for 16 h at 37 °C, 5% CO₂, in a humid chamber to prevent evaporation. After treatment, the supernatant was collected and centrifuged (200 × g, 10 min) to remove particles and cell debris. To measure LDH release, 50 µL of the supernatant were added to 50 µL of LDH reaction mix (LDH Cytotoxic Detection Kit, Merck, cat. no. 11644793001) and incubated in the dark at RT for a maximum of 10 min. The reaction was stopped with 50 µL of 1 N HCl. To measure GLU activity, 50 μL of the supernatant were incubated with 50 μL of reaction mix (13.3 mM p-nitrophenyl-β-D-glucuronide, Merck, Germany), 0.1% Triton X-100 in 0.1 M sodium acetate buffer (pH 4.5) for 90 min at 37 °C. The reaction was stopped with 50 µL of 0.2 N NaOH. The absorbance was read at 492 nm and 620 nm (reference wavelength) for LDH, and at 405 nm for GLU. The OD measurements were corrected for test material adsorption and light scattering using cell-free particle controls for each concentration and normalized to the PC value (cells lysed with 1% Triton X-100) set as 100%. Finally, for the TNFα assay, 50 µL was added to wells containing 10 μL FBS, and the samples were frozen at −80 °C until analysis. Technical duplicates were pooled, and ELISA was performed using different kits in each laboratory following the manufacturer’s instructions (Lead Lab: ELISA MAX™ Deluxe Set Rat TNF-a, BioLegend, Inc., cat. no. 438204; Lab1: Rat TNF-alpha Quantikine™ ELISA, R&D Systems®, Inc., cat. no. SRTA00; Lab2: Rat TNF-alpha ELISA Kit, Invitrogen, cat. no. ERA56RBX5).

2.4 Acceptance criteria

A test run was considered valid if all acceptance criteria described in Table 2 were fulfilled. These criteria were originally established by the lead laboratory and were slightly adjusted during the transferability phase of the project to accommodate differences among the three participating laboratories. In practice, the labs used not only different measurement devices (e.g., photometers) but also occasionally different assay kits, which resulted in higher variability compared to the lead/developer laboratory.

TABLE 2. Acceptance criteria for considering a run valid in the Alveolar Macrophage Assay.

2.5 Data analysis and prediction model

All analyses were performed in R version 4.5.0 (R Core Team, 2025). TNFα standard curves were fitted using a four-parameter log-logistic model. The best-fitting parameterization was selected by Akaike Information Criterion (Akaike, 1974) from four-parameter candidate sigmoidal functions (LL.4, LN.4, L.4, LL2.4) implemented in the drc package (Ritz et al., 2016) version 3.2.0. The 4 PL model is defined by Equation 1:where b denotes the Hill slope, c and d the lower and upper asymptotes, respectively, and e the inflection point. TNFα concentrations were interpolated from the fitted standard curve.

f x = c + d − c 1 + exp b ln ⁡ x − ln ⁡ e (1)

For LDH, GLU, H₂O₂ and TNFα, model fit was assessed via quantile residual analysis (Dunn and Smyth, 1996) using the DHARMa package (Hartig, 2024) version 0.4.7 accessed through the performance package (Lüdecke et al., 2021) version 0.15.1, which indicated that a gamma distribution was most appropriate. Data were therefore analyzed using a generalized linear model (GLM) with a gamma error distribution and a log link, as defined in Equation 2:where μi = E(Yi) is the expected response, β the vector of fixed-effect coefficients, xi the vector of dose groups (as factor) per experiment and ϕ the dispersion parameter. Models were fitted with the glm() function from the stats package. Dunnett’s post hoc multiple-comparison test was applied to assess statistical significance of each dose group against the control using the emmeans package (Lenth and Piaskowski, 2025) version 2.0.1 with multiplicity adjustment applying the multivariate-t distribution.

ln µ i = x i ⊤ β , ∼ G a m m a µ i , ϕ (2)

The prediction method was based on Wiemann et al. (2016), with slight modifications. In this method, the mass-based lowest effect concentration (LOEC, defined as the lowest particle concentration with a statistically significant increase (p < 0.05) relative to the NC) in each assay for each particle is converted to surface area-based concentrations using the respective test material’s surface area (m²/g) determined by the method of Brunauer Teller and Emmett metrics (BET). The use of surface area metrics is based on the large body of scientific data demonstrating the correlation of surface area with the effects of nanomaterials (Oberdörster, 1996; Sager et al., 2008; Braakhuis et al., 2014; Schmid and Stoeger, 2016; Cosnier et al., 2021). The threshold of 6,000 mm²/mL was empirically defined by Wiemann et al. (2016), based on in vivo data.

Accordingly, the criteria to assign particles as active (i.e., particles with intrinsic toxicity) were:

• Two or more out of the four assays included in AMA were statistically significant below concentrations equivalent to 6,000 mm²/mL (particle surface area), and
• If there was a significant increase in LDH and GLU, a monotonic dose-response relationship should be present, because it is the only biologically plausible outcome. For TNFα and H₂O₂ assay, other types of dose-response patterns were considered acceptable, since pronounced cytotoxicity may affect both the release of TNFα and formation of H₂O₂.

Otherwise, particles were assigned as passive (i.e., effects due to test material-unspecific cellular overload).

3 Results

3.1 Phase I: transferability

In Phase I, the Lead Lab provided the standard operating protocols (SOPs) and on-site training. The transferability of the method was assessed by the ability to fulfill the acceptance criteria and tests with the particle controls (i.e., corundum and quartz DQ12). During this phase, basal cytotoxicity levels (LDH) were higher than the acceptable values. Through extensive troubleshooting, the lot of FBS used for routine culturing was identified as the cause. Among four different batches of FBS tested, only cells cultivated in one of them resulted in values within acceptance criteria. After adopting the same FBS lot as used in Lead Lab, LDH levels fell below 25% in the NC (F-12K), and corundum and DQ12 showed the expected effects of passive and active particles, respectively, in both Lab1 and Lab2 (data not shown).

3.2 Phase II: inter-laboratory reproducibility

After the transferability phase, tests with the blinded particles were performed. The particle uptake was confirmed visually, and micrographs after 16 h of exposure in Lead Lab and Lab1 are presented in the Supplementary Material. Model estimates derived from the generalized linear model (see Section 2.5) and 95% confidence intervals are presented in Supplementary Table S2.

To evaluate if the values obtained in the ring trial were within expected ranges, mean control values were compared with the historical control data (HCD) from the last 5 years (2020–2025) in the Lead Lab (Supplementary Table S1). All mean control values between biological replicates were within the minimum and maximum values of the HCD, except for GLU-NC in Lab2, which had a mean value of 5.13%, slightly above the maximum value of 4.92% in the Lead Lab.

Regarding individual control values (from each biological replicate), Lab2 had 4 out of 9 values above the HCD maximum for GLU (9.32, 8.09, 7.51% and 5.84%), and two LDH values above the HCD (31.22% and 39.86%). Although these values were not within the acceptance criteria, they were retained in the overall assessment to enable a meaningful interlaboratory comparison and were discussed on a case-by-case basis below. For TNFα, Lab2 used a kit with the lowest standard calibration point as 82.3 pg/mL, which is below range produced by NR8383 cells for most tested particles (Wiemann et al., 2016). With most measured OD values falling below the standard calibration range, TNFα data of Lab2 was excluded from further analyses.

In Lab1, five TNF-PC (LPS) values exceeded the maximum absorbance range of the microplate reader (OD > 4). Based on extrapolation from the assay-specific standard curves, these values then exceeded the HCD maximum (1,277 pg/mL). However, the purpose of the TNF-PC is to confirm that the NR8383 cells respond to LPS stimulation. Therefore, exceeding the HCD maximum does not affect the validity of the TNFα measurements in Lab1.

When comparing results from all laboratories, similar dose-response patterns were observed for most particles in LDH and GLU assays (Figure 1). For the H₂O₂ assay, however, Lab2 obtained negative values after subtraction of the reference wavelength (595 nm), which prevented the analysis for a few concentrations of some particles (Supplementary Table S2).

In Table 3, the LOECs are defined as the lowest particle concentration with a statistically significant increase (p < 0.05) relative to the NC. The LOECs were converted to surface area-based doses using the BET surface area of each particle. Considering the criteria mentioned above and applying the prediction model determined by Wiemann et al. (2016), tested particles were assigned as passive or active. Due to the absence of reliable TNFα data for Lab2, particles were tentatively assigned as passive when no assay was significant below the threshold, and active when 2 out of 3 assays showed significant effects below the threshold. Particles with only one positive assay below the threshold were considered as uncertain. Seven out of the 12 particles were consistently assigned as either active (Mn₂O₃, tungsten carbide-cobalt (WC-Co), SiC nanowires, quartz DQ12, and SiO₂ NM-203) or passive (corundum and tungsten carbide (WC)) by all three laboratories (Table 3).

TABLE 3. Lowest observed effect concentrations (LOEC) in mass- and surface-area based doses for the three laboratories and the resulting assignment of tested particles.

n.s. = not statistically significantly increased.

n. a. = not assessed.

Bold indicates LOEC below 6,000 mm²/mL.

statistically significant, but not concentration-related, considered not substance-related.

TNFα, levels in the Lead Lab’s non-exposed controls were consistently low: at least half of those in the lowest particle doses. In this lab, all controls were run on one ELISA plate with the standard curve, while treated samples were measured on separate plates without a standard curve. This likely contributed to the statistically significant increases observed at all doses for all particles, even when no monotonic dose-response was present (e.g., corundum, WC).

Of note, corundum showed an increase in H₂O₂ formation at Lead Lab and Lab1. Due to its very low surface area (1.7 m²/g), the tested doses do not reach the defined threshold (i.e., 6,000 mm²/mL). However, corundum is historically used as a negative benchmark material (Bruch et al., 1993a; Bruch et al., 1993b; Bruch et al., 2014) due to its known insolubility (both in water and biological fluids) and low toxicity, with effects occurring only at high concentrations inducing lung overload (Hartwig and MAK Commission, 2020). For such particles, other metrics other than surface area may be more relevant (see Section 4.4 for detailed discussion). Thus, corundum was maintained as passive particle control for AMA, due to its numerous historical control data.

For most materials, Wiemann’s prediction model was applicable; however, this model cannot be used for fiber-like particles such as SiC nanowires. For fibers, number, length and rigidity of fibers are more appropriate dose metrics than surface area (Padmore et al., 2017). Nevertheless, SiC nanowires were provisionally assigned as active because they induced concentration-dependent increases in GLU, LDH, and TNFα at concentrations as low as 22.5 μg/mL.

TiO₂ NM-101, TiO₂ NM-105, CeO₂ NM-212, BaSO₂ NM-220, and ZnO NM-111 showed inconsistent assignments across the three laboratories. TiO₂ NM-101 was assigned as passive and TiO₂ NM-105 as active by both the Lead Lab and Lab 1; however, Lab 2 rated both materials as uncertain based solely on a monotonic LDH dose-response without corresponding changes in other endpoints. CeO₂ NM-212 was assigned as active by the Lead Lab and Lab1, but passive by Lab2. BaSO₄ NM-220 and ZnO NM-111 were assigned as passive in Lab1 and Lab2, whereas the Lead Lab assigned both as active.

4 Discussion

The ring trial was conducted to evaluate the transferability and reproducibility of the AMA across three independent laboratories. The study also provided an opportunity to refine SOPs and identify additional factors influencing assay outcomes. With meticulous standardization and strict quality criteria reestablished throughout the study, the reproducibility of the AMA could be demonstrated. The factors affecting the results and the criteria needed for achieving validity of the assay as well as the remaining gaps are thoroughly discussed below.

4.1 Main factor influencing assay performance: fetal bovine serum

During the transfer phase, FBS batch was identified as the cause of elevated basal LDH levels in Lab1 and Lab2. Although the assays are performed under serum-free conditions, the routine maintenance of NR8383 cells is dependent on FBS, which is a known source of variability in in vitro assays due to undefined composition and batch-to-batch variation (van der Valk et al., 2018; Weber et al., 2025). Along the same line, serum albumin, a primary component of FBS, has been shown to affect cell adhesion (Carré and Lacarrière, 2010; Verdanova et al., 2017). As NR8383 cells are a semi-adherent growing cell line, the FBS could potentially favor one subpopulation in spite of the other. The replacement of FBS by chemically defined animal-free ingredients could improve method reliability and reproducibility. Until such replacement is implemented, testing each FBS batch in the full assay has to be conducted.

4.2 Overall reproducibility across endpoints

After the successful transfer, each of the three laboratories tested the 12 blinded particles as suspensions. Overall, the results showed a similar dose-response pattern across the three institutes for all assays, with the exception of TNFα for Lab2. When comparing treatments with p < 0.05, the LDH assay showed the highest concordance across the three institutes. One likely explanation is the use of a well characterized commercial kit, while for GLU and H₂O₂ other sources of variability are introduced with the in-house preparation of reagents.

For TNFα, several factors may explain the variability observed between the institutes. While commercially available kits were used, they differed in manufacturer, detection range and sensitivity. Further variability could also arise from plate design and placement of standard curves. Taken together, these factors indicate that, because of the low TNFα production range in NR8383 cells upon particle exposure, a higher standardization would be required to minimize variability and improve reproducibility. This could include specifying assay range and standards, plate layout, and pipetting technique (e.g., forward versus reverse pipetting). These details should be in the future SOP.

4.3 Causes for differences across laboratories in the assignment of particles to the groups of passive or active particles

As mentioned before, the negative control values of GLU and LDH of Lab2 exceeded 6% and 30%, respectively, in at least one biological replicate of each particle. We nevertheless decided to retain these data in the analysis to ensure full transparency and to report all values generated within the ring trial. Aware that these elevated controls reduced the dynamic range and introduced additional variability, such values were important contributors to the inconsistent particle assignment. Therefore, the importance of adherence to acceptance criteria should be emphasized. Each mismatch case is discussed below.

TiO₂ NM-101 and TiO₂ NM-105 were assigned as uncertain in Lab2, whereas the lead laboratory and Lab1 produced consistent assignments (NM-101 as passive; NM-105 as active). For NM-101, only LDH showed a monotonic, concentration-dependent increase in Lab2; without TNFα, the required second confirming endpoint was missing. For NM-105, Lab2 observed increases in GLU and LDH, but GLU failed to reach statistical significance due to high variability between replicates and inability to meet acceptance criteria for the NC. Similarly, CeO₂ NM-212 was assigned active by the Lead Lab and Lab1, but passive by Lab2, since LDH values across concentrations were too similar to meet the monotonic dose-response criterion. Altogether, these discrepancies show that strict standardization of measurement ranges, assay setup, and having runs within the acceptance criteria are essential to ensure consistent assignment across laboratories.

Moreover, ZnO NM-111 also produced divergent results: it was active in the Lead Lab based on clear LDH and TNFα increases, but passive in Lab1 and uncertain in Lab2. Likewise, BaSO₄ NM-220 was only assigned as active in the Lead Lab, where slight but concentration-dependent increases in TNFα and LDH were observed. Although this meets the “two out of four” criteria, the magnitude of the responses was minimal and of questionable biological relevance. In contrast, both Lab1 and Lab2 assigned BaSO₄ as passive.

A potential explanation for the mismatch in the assignment of BaSO₄ NM-220 and ZnO NM-111 might be related to the dynamic dissolution of these particles. BaSO₄ NM-220 and ZnO NM-111, although poorly soluble in cell-free pH neutral environments, are cleared from the lungs much faster than other poorly soluble particles (Molina et al., 2019; Thoma et al., 2024). In phagolysosomal simulant fluid (PSF, pH 4.5), the calculated clearance half time for BaSO₄ NM-220 was only about 7 days and that of ZnO NM-111 was 1.2 days in dynamic dissolution assay. For comparison, the clearance half time of CeO₂ NM-212 and TiO₂ NM-105 were > 146 and >4,356 days, respectively (Koltermann-Jülly et al., 2018). Thus, for BaSO₄ NM-220 and ZnO NM-111, toxicity may therefore be driven by the released ions.

There were also further possibilities that may cause this mismatch. For example, the low reactivity of BaSO₄ makes it a borderline case, consistent with earlier in vivo inhalation studies in rats (Konduru et al., 2014; Landsiedel et al., 2014). This indicates a potential disadvantage in the binary classification system and supports the implementation of a borderline or ambiguous category to account for particles inducing minimal but statistically significant responses, as well as particles with responses near the defined threshold. In addition, if BaSO₄ is kept long in pH-neutral media, it may undergo transformation such as Ostwald ripening, a process in which smaller particles dissolve and recrystallize onto larger particles, leading to an increase in particle size, a reduction in total surface area, and altered toxicological behavior (Keller et al., 2020). Since the Lead Lab performed the test earlier while the other laboratories required more time for transferring and implementing the method, this delay may have affected the particle size in suspensions, potentially altering their properties.

Moreover, ZnO NM-111 had to be tested at lower concentrations based on previous data obtained at the lead laboratory. However, the tested concentration range may have been insufficient to detect effects below the 6,000 mm²/mL threshold in Lab1 and Lab2. In these less experienced laboratories, high basal levels and variability of LDH and TNFα measurements limited assay sensitivity. Consequently, effects became apparent only in the lead laboratory, which generated highly reproducible data with low standard deviation.

4.4 Remaining gaps

FBS was identified as an important source of variability during the ring trial. Beyond the scientific concerns related to this variability, the use of FBS also raises ethical issues (Rosolowski et al., 2025; Weber et al., 2025). As such, full replacement of animal use will only be achieved when in vitro methods are entirely based on defined, animal-component-free media. Therefore, replacing FBS should be pursued in the future to make the AMA a truly animal-free method.

The AMA is performed using a rat alveolar macrophage cell line, NR8383, which is convenient due to its wide availability and unlimited lifespan. In addition, it allowed direct comparison of its results to those of rat experiments. However, human AMs are about four times larger by volume than rat macrophages (about 5,000 and 1,200 μm³, respectively (Krombach et al., 1997)). Furthermore, rats are known to be more susceptible than other species, including humans, to particle-induced lung overload and its downstream consequences (Driscoll and Borm, 2020). Therefore, using these cells may overestimate human sensitivity (Scott et al., 2025).

To improve human relevance, the assay needs to be transferred to human macrophages. At present, this remains challenging since there is no human alveolar macrophage cell line available. Human primary AMs can be difficult to obtain and are terminally differentiated and non-proliferative (David et al., 2025). This is a challenge for AMA, as it requires a large number of cells, which may not be broadly available with reproducible quality. One opportunity lies in induced pluripotent stem cells (iPSCs), which could be expanded and differentiated into AMs for the assay. However, the differentiation of these cells to the AM phenotype at scalable and reproducible yields is still under development (Schinke et al., 2025). Thus, the search for the ideal human cell for application in this assay is ongoing (Kelsall et al., 2025; Scott et al., 2025).

The relation of in vitro particle concentrations to in vivo lung burdens (which in turn are related to the inhaled aerosol concentrations) and its relevance to the in vitro assays has been described (Ma-Hock et al., 2021; Azizah et al., 2024; Landsiedel et al., 2024). Yet some open questions remain: the most appropriate dose metric for the assay remains to be determined. There has been debate on which metric is the most relevant for inhalation toxicity of particles (Schmid et al., 2009; Pauluhn, 2011). Surface area appears to be the most biologically relevant dose metric for NMs with surface reactivity, since they have extremely high specific surface area per unit mass, and their effects are often driven not by the absolute particle mass but by the total surface area. On the other hand, volume correlates better with effects caused by chronic exposure to high particle doses (overload effects) (Pauluhn, 2014; Schmid and Stoeger, 2016), such as for corundum.

Rehn et al. (1999) set their in vitro particle doses to reflect the mass range of particulate burdens that AMs are likely to encounter in vivo in rats. Similarly, Wiemann et al. (2016) applied the same concept with the addition of retrospective conversion to surface area doses, based on the large body of work demonstrating the correlation of this dose metric with NM effects (Oberdörster, 1996; Sager et al., 2008; Braakhuis et al., 2014; Schmid and Stoeger, 2016; Cosnier et al., 2021). To maintain methodological continuity with the original assay development and to avoid the introduction of uncertainties in assay performance, the authors deliberately retained the previously published four mass-based dose levels.

The volume-based metric is consistent with classical overload concepts, such as Morrow’s proposed threshold of ∼60 μm³ particle volume per rat AM, which describes the point at which macrophage activity begins to be impaired (Morrow, 1988; Oberdörster, 1988). Accordingly, volume-based dose metrics could provide a more mechanistically relevant basis for predicting lung overload and distinguishing intrinsic toxicity from effects driven purely by particle burden. Characterization data on the particle agglomerate density allowed retrospective conversion to volume-equivalent metrics (Supplementary Table S3). However, for the majority of particles examined, even the highest concentration used in the AMA (180 μg/mL) did not correspond to the particle volume of ∼60 μm³ per cell. As a result, the study design did not reach particle burdens high enough to meaningfully test volumetric overload conditions. Consequently, although volume-based dosimetry provides valuable mechanistic insight, the applicability of overload theory could not be evaluated within the current experimental confines and should be specifically addressed in future studies through adjusted dosing strategies or extended dose ranges. With this, the prediction model could be further refined to also account for particles with low surface area, such as corundum.

4.5 Readiness level

Technology Readiness Levels (TRLs) were originally developed by NASA to describe the maturity of a technology from early concept to routine operational use (NASA, 2017). For toxicological methods and NAMs, the relevant “technology” is not the cell model alone, but the complete method: biological test system, exposure procedure, SOPs, controls, acceptance criteria, analytical endpoints, data analysis, prediction model, applicability domain and context of use (Worth and Balls, 2001; European Commission, 2014; Cöllen et al., 2024).

The AMA has progressed beyond typical TRL 6–7 characteristics. It is no longer an exploratory assay or a prototype demonstrated only under developer-laboratory conditions. The method has a defined mechanistic rationale, standardized procedures, predefined acceptance criteria, benchmark materials, a structured data interpretation procedure and prior evidence of relevance for short-term inhalation outcomes (Wiemann et al., 2016). The present interlaboratory comparison adds the critical next step: it shows that the assay can be transferred to independent laboratories and can generate reproducible biological response patterns when strict standardization and quality criteria are applied, consistent with GIVIMP principles (Bal-Price et al., 2018; OECD, 2018). Importantly, this interlaboratory demonstration extends the method beyond proof-of-concept in a single laboratory and supports operation under relevant conditions across independent sites. The available evidence therefore supports positioning the AMA within the TRL 7–8 range within its defined context of use. This context is the assessment and grouping of respirable particulate materials based on macrophage-mediated biological activity. Within this domain, the assay represents a qualified NAM candidate rather than a research model. It provides a standardized, mechanistically anchored and decision-oriented method that can contribute to integrated particle safety assessment and grouping frameworks, including DF4nanoGrouping and the ECETOC NanoApp (Arts et al., 2015; Arts et al., 2016; Janer et al., 2021a; Janer et al., 2021b).

The present study also indicates that the prediction model could be revised, particularly with respect to dose metrics, partially soluble materials, low-surface-area particles, borderline responses and fiber-like materials. This does not question the reproducibility or technical transferability of the assay. It concerns the interpretation of how biological responses are converted into particle classifications across different material classes. Refining the prediction model is therefore part of the maturation of the AMA, not evidence of a technical deficiency.

The AMA has not yet reached TRL 9. It has not been adopted as an OECD Test Guideline and has not yet been established as a routinely accepted regulatory method across its full intended operational scope. Taken together, the current evidence places the assay within the TRL 7–8 transition range, reflecting demonstrated transferability and partial reproducibility across laboratories, while indicating that further refinement is required to meet all criteria associated with TRL 8. Further work should focus on revising the prediction model, exploring a wider applicability domain, harmonizing remaining assay components and building regulatory confidence.

5 Conclusion

Overall, this ring trial demonstrates that AMA yields reproducible results when strict standardization and adherence to quality measures are ensured. Several factors, including FBS batch variability and TNFα assay sensitivity, were identified as causes for variability and require further optimization. Gaps that remained to be filled also include the identification of appropriate dose metrics and the transition to a more human relevant and truly animal-free assay through the use of human cells and animal-component-free media. Therefore, our work paves the way for future research focusing on establishing serum-free or chemically defined culture media, fully harmonizing TNFα measurements, integrating surface area or volume metrics into dose selection and transferring the assay to human cells. By establishing robust interlaboratory transferability and reproducibility, this study provides key evidence needed for the formal validation and regulatory uptake of the AMA as a NAM. It positions the AMA for integration into inhalation safety assessment frameworks for regulatory decision-making on particle-induced lung toxicity.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/ftox.2026.1844090/full#supplementary-material

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Footnotes

Edited by: Stina Oredsson, Lund University, Sweden

Reviewed by: Il Je Yu, HCT, United States

Fiona Murphy, University of Strathclyde, United Kingdom