1. Introduction
The UN Globally Harmonized System (GHS) defines eye damage and irritation as reversible changes to the eye’s anterior surface within 21 days of chemical exposure [
1]. Traditional reliance on live animals using the Rabbit Draize eye Test (OECD TG405) is not sustainable due to ethical concerns and legislature changes [
2,
3]. While ex vivo tests, such as the Bovine Corneal Opacity Test (BCOP) and Chicken Eye Test, preserve the use of whole eye tissues, they have led to over- or under-prediction [
2], and are not very amenable for compound screening due to elaborate set-ups required to manipulate the whole organ. Based on current knowledge of the ocular anatomy and the mechanisms underpinning chemical-induced eye irritation, several different in vitro tests have been developed [
2,
4] and are currently being validated as a substitute for testing with animal ocular samples.
Typically, an in vitro test aims to simulate a specific biological mechanism by which a compound may induce eye irritation. Such tests are comprised of two primary components: (1) the experimental model, which delineates the cellular or tissue framework with which the test compounds interact, and (2) the test read-out, which quantitatively assesses the model’s response to these compounds, examples of which include cell viability, cytokine secretion, or fluorescence dye leakage [
5]. The classification of a compound as an eye irritant relies on predefined cut-off values derived from these quantitative read-outs. It has been acknowledged that no single in vitro assay can accurately predict the eye irritation potential of a chemical due to the multiple mechanisms of action. This has spurred the development of complex 3D organotypic corneal tissue models, such as the Epiocular™ assay (OECD TG492) and the Vitrigel-Eye Irritancy test (OECD TG494) [
4,
6,
7]. These models strive to more accurately mimic the physiological structure and functionality of the ocular surface, with the expectation that they will offer a more predictive response to test compounds. Nonetheless, enhancing the physiological relevance of these models introduces greater complexity, costs, and time requirements for conducting in vitro tests [
8]. This results in a trade-off regarding experimental throughput and compatibility with automated high-content imaging systems. Simpler models, such as the corneal epithelial cell monolayer utilized in the Short-Term Exposure test (OECD TG491) [
9] or the polymer matrix used in the in vitro Macromolecular test (OECD TG496) [
10] are more conducive to high-throughput screening but are limited to assessing a single mechanism which cannot predict eye irritation reliably.
We posit that the potential to enhance the accuracy of simple cellular models, such as the corneal epithelial monolayer, lies in the integration of multiple assay readouts addressing various mechanisms of eye irritation. This approach aligns with the prevailing opinion in the field that a diverse array of in vitro tests will be necessary to replace animal-based ocular surface testing [
11,
12]. However, combining multiple in vitro tests into a cohesive model for predicting eye irritation potential presents a significant challenge, particularly in merging analog read-outs that span various physical measurements, such as cytotoxicity, optical density, and permeability. This challenge can be circumvented with statistical regression analyses, such as logistical regression models and Kaplan–Meier “survival curves”, to facilitate probability-based prediction of event outcomes. The probabilities of several assay readouts can be combined mathematically. However, these regression models necessitate a substantial dataset, typically requiring more than ten observations for the least probable outcome [
13,
14]. Therefore, statistical regression has not been commonly utilized for analyzing in vitro eye irritation screening assay results, as these are typically performed on bulk cell cultures in microtiter plates or 3D-engineered tissue constructs, which offer limited throughput. To address this limitation, our study employs a cell micropatterning technique to divide a bulk human corneal epithelial cell (hCEC) culture into numerous distinct populations. This enables the measurement of responses to test compounds post-exposure. We selected cell apoptosis and the activation of the nociceptor (pain receptor) Transient Receptor Potential Vanilloid type 1 (
TRPV1) as indicators of two separate eye irritation mechanisms [
15,
16]. By miniaturizing and segregating the cell populations, we were able to generate a high volume of response readouts for both cell apoptosis and nociceptor activation. These outcomes are amenable to digitization (i.e., response = 1 or 0), facilitating the assignment of binary outcomes even for small cell populations (i.e., fewer than ten cells). The digitized response data can then be applied to a logistic regression model, enabling the estimation of a concentration-dependent response probability for various test compounds.
2. Methods and Materials
2.1. Cell Culture
Commercially available primary human corneal epithelial cells (hCECs) (ATCC, Manassas, VA, USA, PCS-700-010, Lot #1170, 5170 and #804003 were used in this study) were cultured in complete primary cell medium prepared by combining corneal epithelial cell growth kit (ATCC, Manassas, VA, USA, PCS-700-040) and corneal epithelial basal medium (ATCC, VA, USA, PCS-700-030) according to manufacturer’s protocol. Cells were sub-cultured after reaching ~80–90% confluence using Trypsin/EDTA solution (ATCC, Manassas, VA, USA, PCS-999-003), and washed twice with 1× Hanks’ balanced salt solution (HBSS, Hyclone, UT, USA, SH30588.01). The cells were subsequently neutralized by Trypsin neutralizing solution (TNS, ATCC, Manassas, VA, USA, PCS-999-004) and re-suspended in the complete medium after centrifugation for 5 min at 1000 rpm and later counted for experimental seeding. Excess cells were cryopreserved at a cell density of 1 million cells/mL with 80% complete medium, 10% Fetal bovine serum (FBS, Hyclone, UT, USA, SV30160.03) and 10% Dimethyl Sulfoxide (DMSO, Life Science, UK, D2660).
Immortalized hCECs (HCE-2 [50.B1], ATCC, Manassas, VA, USA, CRL-11135) were cultured in Keratinocyte-Serum free media containing 0.05 mg/mL bovine pituitary extract (BPE) and 5 ng/mL epidermal growth factor (EGF) (17005042, Gibco, Agawam, MA, USA) and supplemented with 500 ng/mL hydrocortisone and 0.005 mg/mL insulin. The cell culture surface was coated with a solution prepared using HBSS, Fibronectin (0.01 mg/mL, F1141, Sigma Aldrich, Burlington, MO, USA), Collagen (0.03 mg/mL, A1064401, Gibco, Agawam, MA, USA) and Bovine Serum Albumin (0.01 mg/mL, A3311, Sigma Aldrich, Burlington, MO, USA). Cells were sub-cultured after reaching ~80–90% confluence using TrypLE (12605010, Gibco, Agawam, MA, USA), and washed twice with 1× Hanks’ balanced salt solution (HBSS, Hyclone, Logan, UT, USA, SH30588.01). The cells were subsequently neutralized with Low Glucose DMEM with 10% FBS and re-suspended in the complete medium after centrifugation for 5 min at 1000 rpm and later counted for experimental seeding. Excess cells were cryopreserved at a cell density of 1.5 million cells/mL with 50% complete medium, 40% FBS and 10% DMSO.
HEK293-TRPV1 cell line with Red Fluorescence Protein (RFP) and Puromycin dual marker (SC027-RB, GenTarget Inc., San Diego, CA, USA) were cultured in High Glucose DMEM (11965084, Gibco, Agawam, MA, USA) supplemented with non-essential amino acids (1:100) and blasticidine 1:1000 (15205-25MG). Cells were sub-cultured after reaching ~80–90% confluence using TrypLE (12605010, Gibco, Agawam, MA, USA), post twice washing with 1× Hanks’ balanced salt solution (HBSS, Hyclone, UT, USA, SH30588.01, USA). The cells were subsequently neutralized with Low Glucose DMEM with 10% FBS and re-suspended in a complete medium after centrifugation and later counted for experimental seeding. Excess cells were cryopreserved at a cell density of 2.25 million cells/mL with 50% complete medium, 40% FBS and 10% DMSO.
All cells were maintained at 37 °C in a humidified 5% carbon dioxide (CO2) environment with medium replacement on alternative days.
2.2. Cell Micropatterning
hCECs were seeded onto commercial micropatterned glass substrates (CYTOO, Grenoble, France) which have been customized to contain an array of circular micropatterns (65 and 80 µm diameter) on which cells can adhere. The area surrounding the circular micropatterns was selectively passivated to attenuate cell and protein adhesion. Before cell seeding, the micropatterned substrates were coated with a mixture consisting of Fibronectin (Sigma-Aldrich, Burlington, MO, USA F1141 0.03 mg/mL), Bovine Collagen I (5 mg/mL Gibco, Agawam, MA, USA, A1064401, 0.01 mg/mL) and Bovine Serum Albumin (BSA, Sigma-Aldrich, St. Louis, MO, USA, A331, USA, 0.01 mg/mL) in 1× HBSS, and incubated at 37 °C for 24 h. Coated micropattern substrates were transferred to a 35 mm dish (Thermo Scientific, China, 150460), which was passivated with 10% Pluronic acid (Sigma-Aldrich, St. Louis, MO, USA, P2443) in 1× HBSS overnight. Cells were seeded onto the micropatterned substrates at a density of 300,000 cells/mL. Unattached cells were removed after 2 h by washing with 1× HBSS 5 times, followed by passivation with 0.2% Pluronic acid for 10 min, and finally replaced with 2 mL of complete medium. The cells were cultured on the micropattern substrates for 24–48 h based on the assay being performed.
2.3. Test Compound Treatment for Micropatterned Cells
Cells seeded on the micropatterned glass substrates were assembled into a specialized CYTOO chamber (CYTOO, Cambridge, MA, USA, 30-011) to partition a single micropatterned glass substrate into 4 wells. This allowed for 4 different compound treatments on the same micropatterned substrate. The chamber consisted of a bottom plate to fit the micropatterned substrate. The bottom plate was then magnetically assembled into the main body, which was locked using a silicon gasket. The silicon gasket and the main body partitioned the micropatterned glass substrate (i.e., CYTOO chip) into 4 wells. A glass lid was placed at the top to cover the whole setup and keep it sterile (
Figure S1). For each compound treatment experiment, every CYTOO chip was administered with 3 different compound concentrations and 1 vehicle control as an internal control to minimize inter-experimental variation.
2.4. Bulk Apoptosis Assay
Primary hCECs were plated in 96 well plates at a density of 20,000 cells/well. After 24 h of culture, cells were treated with 100 µL of each of the concentrations of staurosporine (positive apoptosis inducer), dissolved in DMSO to obtain the test solutions of 0.1–104 nM with serial dilutions prepared in a complete medium. As a vehicle control, 0.1% DMSO in culture medium was used. Sodium lauryl sulphate (SLS), an OECD category 1 compound, was dissolved 1× HBSS to obtain test solutions of 103–108 nM. As a vehicle control, HBSS solution in the culture medium was used. The untreated category of cells was exposed to only media. Post drug treatment, the cells were labelled using a staining solution consisting of Hoechst 33342 (cell nucleus stain, Thermo Fisher–Scientific, Waltham, MA, USA, 62249, 1:500), CellEventTM Caspase-3/7 green detection reagent (Thermo Fisher–Scientific, Waltham, MA, USA, C10423, 1:1000) and Ethidium homo dimer (necrotic cell stain, Sigma-Aldrich, St. Louis, MO, USA, cat. 46043, 1:500) dissolved in 1× HBSS. Cells were incubated with the staining solution for 40 min at 37 °C. The cells were subsequently fixed post-washing with 1× HBSS containing 4% Paraformaldehyde (PFA) for 10 min at room temperature. The wells were subsequently imaged using a Nikon Fluorescence microscope. The percentage of apoptotic cells (caspase stained) and dead cells (Ethidium homo dimer stained) were calculated.
2.5. Apoptosis Assay on Micropattern Substrates
The apoptosis assay was performed 24 h post-seeding. Before drug treatment, the micropatterned substrates were removed from 35 mm dish before being assembled and partitioned into 4 replicate wells using CYTOO chambers as described in
Section 2.3. The micropatterned hCECs were treated with staurosporine (positive apoptosis inducer) and dissolved in DMSO to obtain the test solutions of 0.1–10
6 nM with serial dilutions. As the vehicle control, 0.1% DMSO in the culture medium was used. Sodium lauryl sulphate (SLS), OECD category 1 compound, dissolved in 1× HBSS used to obtain test solutions of 100–10
7 nM. As the vehicle control, HBSS solution in the culture medium was used. Ethyl methyl acetoacetate (EMA), OECD Category 2B compound dissolved in DMSO to obtain the test solutions of 500–5 × 10
6 nM with serial dilutions. As the vehicle control, 0.1% DMSO in the culture medium was used. Drug treatment was conducted for 24 h at 37 °C. A vehicle control was included in one of the four CYTOO chambers for each micropattern substrate.
After treatment with test compounds, the cells were stained using a staining solution consisting of Hoechst 33342 solution (cell nucleus stain, Thermo Fisher–Scientific, Waltham, MA, USA, 62249, 1:500), cell event caspase-3/7 green detection reagent (Thermo Fisher–Scientific, Waltham, MA, USA, C10423, 1:1000) and Ethidium homo dimer (dead cell stain, Sigma-Aldrich, St. Louis, MO, USA, cat. 46043, 1:500) dissolved in 10% FBS with 1× HBSS. Then, cells were incubated with the staining solution for 40 min at 37 °C. The cells were subsequently fixed post washing with 1× HBSS with 4% Paraformaldehyde (PFA) for 10 min at room temperature. The culture chambers were removed and disassembled to retrieve the micropatterned substrates, which were then mounted on glass coverslips for imaging. The cells were imaged at 10× magnification using a Nikon Eclipse Ti Fluorescence microscope, NY, USA.
2.6. Quantification and Digitization of Apoptosis Assay Readout
Images from the 2 fluorescent channels (green: Caspase3/7 and blue: DAPI) were quantified and batch analyzed using an ImageJ (Version 1.51) macro [
17,
18]. The macro first loads a predefined set of regions of interest (ROIs) that serve as initial locations of the micropatterns. To account for sight variation in the plate placement, iterations of random small displacement were applied to this set of ROIs to optimize for the maximum area of nuclei. Next, each ROI was used to count the nuclei per micropattern from the DAPI channel image that had binarized by auto-threshold. The SetOption (“BlackBackground”, true) function from Image J was used to remove the background signal in the image before the auto-threshold and binarization functions were performed. Next, each ROI was enlarged by 2 pixels to measure the fluorescence intensity of the green caspase3/7 channel. Eventually, the fluorescence intensity of the caspase staining (apoptosis indicator) was determined for individual cells within every micropattern. The caspase intensity was obtained for vehicle control as well as each of the drug treatment groups. The caspase intensity of every compound-treated sample was subtracted from the average caspase intensity of its corresponding vehicle control. The normalized fluorescence intensity was used for data binarization, wherein intensities above 20% of the averaged vehicle control were defined as positive apoptosis (“1”) and values below 20% of the averaged vehicle control were defined as negative apoptosis (“0”). Each micropattern was defined as apoptotic (“1”) if ≥50% of cells in the micropattern were apoptotic (“1”) and vice versa for non-apoptotic micropatterns.
2.7. Calcium Influx TRPV1 Channel Activation Assay
After 24 h of seeding on CYTOO chips, primary hCECs were labelled with Fluo-4 AM (F14201, Invitrogen, OR, USA, 2 µM concentration) and Hoechst (1:500) in phenol red-free DMEM (11054001, Gibco, Agawam, MA, USA) supplemented with 10% FBS and incubated at 37 °C for 45 min. Labelled cells were then recovered in phenol red-free DMEM for 30–40 min at 37 °C before replacing with fresh medium for imaging. Live-cell imaging was performed using a laser scanning confocal microscope (LSM800 Airy Scan, Carl Zeiss, Jena, Germany) at 10× magnification for a period of 5 min at 5 frames per second. Test compounds at designated concentrations were administered approximately after 1 min to obtain the baseline signals. The cells were treated with a positive TRPV1 activator, capsaicin (10 µM), a category 1 compound SLS (100–106 nM) and a category 2B compound EMA (500–5 × 106 nM) with their respective vehicle controls, i.e., DMSO, HBSS and DMSO, respectively. A vehicle control was included in one of the four CYTOO chambers for each micropattern substrate.
2.8. Quantification and Digitization of TRPV1 Activation Assay Readout
Images from the two fluorescent channels (green: Fluo-4 and blue: DAPI) were quantified and analyzed using an ImageJ macro similar to the apoptosis assay. The micropatterns were identified using the pre-defined ROI and cell nuclei per micropattern were determined using the “binarization” function on the DAPI channel image. Each nuclei mask was enlarged by 2 pixels to obtain the mask for the Fluo-4 image. The Fluo-4 fluorescence intensities of individual cells in every micropattern were quantified at each time frame. The Fluo-4 intensity was obtained for vehicle control as well as each of the drug treatment groups.
The Fluo-4 fluorescence intensity of each cell was plotted against the time frames to obtain the calcium influx response of each cell. Since the baseline Fluo-4 intensity of each cell is highly heterogeneous due to variability in the intracellular Ca
2+ concentration, we normalized the intensity of each image frame by subtracting the baseline intensity at the initial frame (at
). The change in intracellular Ca
2+ concentration induced by a test compound was determined by calculating the integrated peak area between the time of induction (
) and 20 frames after induction (
) (~4 s post induction) as follows:
A positive integrated peak area was digitized to “1” binary outcome, while a negative integrated peak area was binarized to “0”. Each micropattern was defined as a positive TRPV1 activator (“1”) if ≥50% cells in the micropattern had a positive outcome and vice versa for non-activator of TRPV1.
2.9. Statistical Analysis and Logistical Regression Modelling
Each experiment was carried out in 3 biological repeats unless otherwise mentioned. The digitized responses (i.e., cell apoptosis or Ca
2+ influx) were fitted into a Logistic Regression Model (Prism version 9, GraphPad, Boston, MA, USA) to determine the response probability as a function of test compound concentration. The collective probability of a test compound causing eye irritation
by either inducing cell apoptosis (
) or activating the nociceptor (
) at a given concentration was determined using the following formula, assuming that the events are non-mutually exclusive (i.e., apoptosis and
TRPV1 action can occur at the same time):