To improve the identification of chemical-induced corneal injuries, we developed a test method to analyze tissue damage over time using impedance spectroscopy. The impedance-based eye irritation test (ImAi-test) has proven successful discrimination between irreversible and reversible induced tissue damages caused by neat liquids and dilutions via TEER.

During ImAi-test development the following steps were reached: (1) The establishment of a list of 329 reference chemicals including a selection of a training set (40–70 chemicals) and a validation set (30 chemicals, marked with*) (supplementary material). The test development was made on a collection of suitable reference chemicals from the DRD of Barosso et al. (Barroso et al. 2017). (2) The development of an impedance measurement device to analyze human tissue models generated in conventional inserts in a 24 well format, allowing a wide range of applications. (3) 23 liquids of the training set were selected to demonstrate the ImAi-test method. Hereby, based on human 3D RCE models, the ImAi-test was able to discriminate liquids causing irreversible (GHS Cat. 1) and reversible effects (GHS Cat. 2) by TEER over time.

Establishment of reference list for eye-irritation testing

An extensive list of reference substances was compiled to develop the EIT protocol. Databases such as the DRD Draize eye test Reference Database published by Cosmetics Europe (Barroso et al. 2017), ECHA and OECD TGs were screened to identify suitable reference chemicals (Fig. 2).

Fig. 2

Graphical abstract outlining the procedure for creating the list of test substances to evaluate eye irritations

Chemicals were identified as suitable if reliable in vivo data were available. To cover a wide range of chemical properties and substance classes, the selection was mainly based on five criteria: (1) the selected substances cover UN GHS categories, (2) the representation of important in vivo drivers of classification based on effects on corneal opacity (CO) and conjunctival redness (CR), (3) the aggregate state of the chemicals including liquids, semi-solids and solids, (4) a wide range of functional groups and 5) different modes of action of the chemicals on the human cornea.

Solid and liquid chemicals were selected of each GHS category (Cat.) such as Cat.1 (26 liquids/29 solids), Cat. 2 (35 liquids/19 solids), No Cat. (122 liquids/72 solids). According to 2), the following drivers of classification for Cat. 1 by order of importance were taken into account: (1) CO mean ≥ 3 (days 1–3) in ≥ 60% of the animals; (2) CO persistence on day 21 in ≥ 60% of the animals (with CO mean < 3); (3) CO = 4 in ≥ 60% of the animals in the absence of persistence and with CO mean < 3 (or if unknown). For Cat. 2, by order of importance: (4) CO mean ≥ 1; (5) CR mean ≥ 2 (with CO mean < 1). Subgroups for chemicals that do not require classification (No Cat.): (6) CO > 0 (minor effects on CO observed); (7) CO = 0 (clear negative results); (8) CO = 0 ** and CO > 0 ** (only a few chemicals should be included). The test substance list is sorted according to the drivers of classification (1)—(8) (supplementary material, Table S1). To cover a broad spectrum of chemicals: 4) in principle, no chemical functional groups were excluded; alcohol, ether, carboxylic acid (ester) are most frequently mentioned; chemical functional groups such as alcohols or ketones have also been included as references, however, it is recognized that an increased false negative rate is reported in some OECD TGs (OECD 2023c, 2023d). Dye and MTT-reducing test substances have also been included in the reference list as, in contrast to MTT-based endpoint measurement, these test substances do not interfere with barrier measurement (supplementary material, Table S1, Reference list, marked withC). Furthermore, chemicals inducing cell lysis (surfactants, organic solvents, ketones, alcohols, volatile liquids, ethers, polyether’s, esters, aromatic amines), coagulation (acids, cationic surfactants, organic solvents) and ester hydrolyses (alkalis) were included in the reference list. Reactions with macromolecules (peroxides), mustards, alkyl halides, epoxides, bleaches (oxidizers) were not included due to lack of high-quality in vivo data. The developed reference list is based on 329 chemicals of all three UN GHS categories (supplementary material, Table S1) with 26 liquids and 29 solids of category 1 (Cat. 1), 35 liquids and 19 solids of category 2 (Cat. 2) as well as 122 liquids and 72 solids of no category (No Cat.). Due to the high number of suitable No Cat. chemicals of the DRD, the proportion of No Cat. chemicals also predominate in the ImAi-test reference list. Not many should be included, however, to complete the reference list, 21 liquids and five solids with CO = 0 ** and CO > 0 ** were also listed (supplementary material, Table S1). To define a test-set to validate the ImAi-test, 30 chemicals were selected from the reference list (supplementary material, Table S1, marked with*). The validation test set is based on 10 chemicals of each GHS category, including 5 liquids and 5 solids. The selection is partially based on proficiency chemicals of existing TG (OECD 2024c, 2023c, 2024b, 2023d, 2023f, 2023e). Based on the training set, the ImAi-method was developed to further optimize the prediction model to discriminate all Cat. 1 and Cat. 2 chemicals. In addition to published reference lists for eye-irritation testing (Adriaens et al. 2017; Barroso et al. 2017), the established ImAi-reference list could also support the NAM-development for ocular hazard assessment.

Development of an impedance spectroscopy measuring device

As part of the ImAi-test development, an impedance spectroscopy measurement device was specifically designed and adapted to the used in vitro models. The CSM 2100 impedance spectrometer device is capable of non-invasively measuring 24 in vitro models in parallel within a standard 24-well culture plate (Fig. 3). The CSM 2100 spectrometer consists of a main body containing the electronic and provides the connection to the electrodes (Fig. 3a). The electrodes are coated by TiN, which allow sensitive full spectrum measurements of tissue integrity and barrier properties also in low frequencies (Fig. 4a) (Schmitz et al. 2018). The TiN-coated electrodes, are mounted in a glass plate which is arranged by the plate holder. The impedance spectrum of a single tissue model is measured using two electrodes, placed inside of the insert and outside of the insert. With the fixed position of the electrodes, the inserts are always placed in the same order, allowing non-contact measurements of the tissue models in the same position. The measurement setups and experimental study designs are defined in the CSM 2100 software (Fig. 3b). To provide information on the tissue barrier properties, measurements can be taken at single frequencies of 12.5 and 1000 Hertz (Hz) or in full spectrum mode ranging from 1 to 200,000 Hz. Blank values including only medium, and inserts showed reproducible values within all 24 electrodes 73.2 ± 2.1 Ohm*cm2 (Fig. 4b). Raw measurements were saved into a database and exported as excel-format (.xlsx). Raw data includes data tables and graphs for further evaluation.

Fig. 3

Impedance spectroscopy measuring device and software to analyze tissue properties over time. a overview of the construction of the CellSpectrometer CSM 2100. b software to analyze and export impedance spectroscopic measurements

Fig. 4

Reproducible impedance spectroscopic measurements of the TIN electrodes. 24 inserts filled with 500-µl impedance measuring media were placed in a 24-well plate containing 1.8-ml measurement media per well. Measurements were performed without cells to ensure comparable and robust resistance of electrodes and media. a impedance spectroscopic measurements from 1 to 2 00.000 Hertz of 24 inserts at each well position. b TEER1000Hz values of a 24-well plate showing a mean of 73.2 ± 2.1 Ohm*cm2 of n = 10 measurements

Discrimination between category 1- and category 2-induced tissue damages following exposure to liquid substances

Tissue damages from both category 1 (Cat. 1) and category 2 (Cat. 2) liquids were assessed by their membrane barrier functionality. Therefore, the tissue integrity of the RCE models were measured directly before the treatment (TEERpre), 120 min after post incubation (TEERpost) and in the following on day 7 (TEER7d) and 14 (TEER14d). The EIT was carried out with 23 liquids of Cat. 1 and 2 of the training set (Table 2). Focusing on reversible damage, 12 Cat. 2 (10 Cat. 2A and 3 Cat. 2B liquids) and 11 Cat. 1 liquids were selected, covering the main drivers of classification: CO mean ≥ 3, CO mean < 3, CO = 4, CO mean ≥ 1, CO mean < 1 (Barroso et al. 2017).

Table 2 Category 1 and 2 liquids of the training set tested in the ImAi-test. Chemicals are listed based on the DRD. According to the Globally Harmonized System of Classification and Labelling of Chemicals (GHS), GHS 1 indicates irreversible tissue effects or severe damage to the eye. GHS 2A/2B causes eye irritation and fully reversible effects within 21/7 days

Results of TEER-measurements of Cat. 1 and Cat. 2 liquids were represented as mean of raw TEER-values (Ohm*cm2, Figs. 5, 6) and as mean of normalized TEER-values (%) (Table 3).

Fig. 5

Raw TEER1000Hz (Ohm*cm2) measurements of treated in vitro RCE models indicated strong tissue damages after exposure to 11 category 1 liquids. TEER measurements were taken before (pre), after (post) the test on day 0 and on days 7 and 14 after the test (7d, 14d). Statistical comparison was made of pre vs. post and of post vs. 14d values. Measurements were performed using N = 3; n = 3. Statistics are based on Tukey’s multiple comparisons test using alpha ≤ 0.05 for significance

Fig. 6

Tissue damage and recovery were observed in treated RCE models exposed to 12 category 2 liquids indicated by raw TEER1000Hz (Ohm*cm2) values. TEER measurements were taken on day 0 before (pre) and after (post) the test and on days 7 and 14 after the test (7d, 14d). Values of pre vs. post and post vs. 14d were statistically compared. EIT was performed using technical triplicates of three different biologic donors. Results are based on Tukey’s multiple comparisons test (alpha ≤ 0.05 for significance)

Table 3 The hazard potential of category 1 and 2 liquids was assessed using normalized TEER values at day 0 after (0d post) and 14 days after the test (14d post). Liquids were classified into ImAi-GHS categories using the prediction model of Lotz et al. (Lotz et al. 2018) and were compared to the UN GHS classification (OECD 2023b)

Raw TEER-values were plotted as data bars for each time point from pre, post, 7–14 days (Figs. 5, 6). The data bars represent the average of TEER-measurements obtained from three RCE models per condition at each time point (Figs. 5, 6). Thus, the TEER-mean of each liquid and control per time point contains 9 TEER-values in Ohm*cm2. In total, three independent experimental runs were performed using three different biologic donors.

Beside the representation of raw TEER, the chemical classification is based on the membrane barrier functionality (%) of RCE triplicates treated with the same chemical. The barrier functionality (%) is calculated within each condition by normalizing post-exposure TEER to the respective TEER pre values. This means, a single model within a condition that is measured on day 0 post, 7 days and 14 days was normalized to its pre-TEER-value, allowing individual models to be accurately assessed over time (Table 3).

Discrimination of Cat. 1 and Cat. 2 liquids were shown by raw TEER-values (Ohm*cm2, Figs. 5, 6). To predict the classification of chemicals, tissue induced damages of Cat. 1 liquids were non-invasively measured over 14 days and compared with effects of Cat. 2 liquids (Fig. 5). 11 out of 11 Cat. 1 liquids demonstrated a significantly reduced TEER from pre to post (p < 0.0001, Fig. 5) in the RCE models. To induce irreversible damage by Cat. 1 liquids on the RCE models, TEER was not expected to increase over time. Indeed, no significant (ns) increase of TEER values from TEERpost to TEER14d was shown for all Cat. 1 liquids. RCE models treated with Hydroxyethyl acrylate, No. 34 indicated increased TEER from post to day 7 and 14 without significance (Fig. 5). However, except for chemical Hydroxyethyl acrylate, No. 34, 10 out of 11 Cat. 1 liquids induced significant damage on RCE models until day 14 after exposure. In summary, when TEERpost was compared to TEER14d values, no significant increase was observed in RCE models for all 11 chemicals.

Initial loss of TEER could also be observed in RCE models after exposure to Cat. 2 liquids (Fig. 6). A significant decrease in TEER1000Hz from pre to post measurements was observed for 10 out of 12 Cat. 2 liquids with p values for 2-Ethyl-1-hexanol, No. 167 (p < 0.0001), Cyclopentanol, No. 170 (p < 0.0001), Methyl acetate, No. 176 (p = 0.0097), N-Hexanol, No. 180 (p < 0.0001), Propasol Solvent P, No. 183 (p = 0.0435), Lauryl sulphobetaine (10%), No. 187 (p = 0.0206), Furfural, No. 202 (p < 0.0001), N-Lauroyl sarcosine Na salt (10%), No. 209 (p = 0.0003), 2-Methyl-1-pentanol, No. 218 (p = 0.0003), 3-Chloropropionitrile, No. 219 (p < 0.0001) (Fig. 6). Two chemicals, Gamma-Butyrolacetone, No. 173 and Methyl cyanoacetate, No. 204 indicated less severe tissue damage by decreased TEERpre to TEERpost values: with 1597 ± 574.75 to 1179 ± 628.44 Ohm*cm2 for Gamma-Butyrolacetone (p = 0.4779) and values ranging from 1742 ± 405.67 to 1339 ± 340.9 Ohm*cm2 for Methyl cyanoacetate (p = 0.5101).

For GHS Cat. 2, 10 out of 12 liquids indicated increased TEER-mean of treated RCE models on day 7 and 14 after exposure when compared to TEERpost (Fig. 6). Significantly increased TEERpost to TEER14d values were measured for 7 Cat. 2 liquids: from 1179 ± 806,23 to 2093 ± 737,03 Ohm*cm2 for Gamma-Butyrolactone, No. 173 (p < 0,0001), from 509 ± 181.39 to 1384 ± 463.63 Ohm*cm2 for Methyl acetate, No. 176 (p = 0.0147), from 426 ± 131.41 to Ohm*cm2 for N-Hexanol, No. 180 (p = 0.0099), from 413 ± 61.49 to 1241 ± 652.56 Ohm*cm2 for Propasol Solvent P, No. 183 (p = 0.0238), from 612 ± 739.29 to 1558 ± 353.91 Ohm*cm2 for Lauryl sulphobetaine (10%), No. 187 (p = 0.0067), from 209 ± 79.31 to 1590 ± 1091.44 Ohm*cm2 for N-Lauroyl sarcosine Na salt (10%), No. 209 (p < 0.0001) and from 549 ± 209.36 to 1576 ± 947.74 Ohm*cm2 for 2-Methyl-1-pentanol, No. 218 (p = 0.0026). As exception, chemical 2-Ethyl-1-hexanol, No. 167 showed irreversible effects on tissue models indicated by low TEER-mean from post to 14 days.

The membrane barrier functionality (%) was evaluated by the normalized TEER-values to TEERpre of each condition (Table 3). As classified into GHS Cat. 1, 10 out of 11 Cat. 1 liquids significantly decreased membrane barrier functionality of RCE models below 30% after chemical exposure. For 6 out of 11 Cat. 1 liquids, TEERpost indicated below 10% of the barrier functionality when normalized to TEERpre of intact RCE models.

Compared to the UN GHS classification, ten out of 12 Cat. 2 liquids were correctly classified by the ImAi-test as Cat. 2 chemicals. In the case of tissue recovery, the mean of TEER14d was 104.5% ± 21.2. TEERpost values below 10% of the membrane barrier functionality showed no increase in TEER until day 14 which was observed for chemical 2-Ethyl-1-hexanol, No. 167 (Table 3). Chemical Methyl cyanoacetate, No. 204, indicated no tissue recovery with TEER values ranging from 80% ± 30 to 68% ± 41 at post and 14 days.