Automated epidermal thickness quantification of in vitro human skin equivalents using optical coherence tomography

Tissue-engineered human skin equivalents (HSEs), also referred to as organotypic skin cultures or full-thickness models, have been used to study skin development, cytotoxicity, healing and disease [1–4]. Typically, they are analyzed post-culture through sectioning and histology. These methods of analysis conceal the three-dimensionality of the in vitro constructs and do not allow for real-time assessment of epidermal thickness during the long culture periods that HSEs require (3–10 weeks). As an alternative, optical coherence tomography (OCT) has been used before to analyze native human skin [5–8] and a few in vitro skin models noninvasively [9–13]. OCT and other noninvasive live imaging techniques [5,6,13] allow for time tracking of culture health and stability, which is especially important when using HSEs for wounding and aging studies [11]. Furthermore, as a noninvasive technique, OCT avoids the disruption of sample morphology that can occur during histology [11,12]. While OCT provides an opportunity for HSE research by enabling volumetric imaging, the three-dimensional analysis can be laborious and is often simplified to two-dimensional or one-dimensional analysis.

Often, when prior OCT studies have investigated epidermal thickness in humans and in HSE models, the common thickness calculation is acquired by manually measuring then averaging a few (3–5) predefined OCT line scans (A-lines), and it is unclear how many OCT images (B-frames) are taken into account [10,11]. Furthermore, due to this manual selection, there is potential for bias introduction from the investigator. Here we present a technique for automated analysis of epidermal thickness in tissue-engineered HSE using live, noninvasive OCT imaging, including optimized OCT imaging procedures and automated image analysis algorithms. These results were further validated against post-culture fluorescent confocal imaging and epidermal thickness quantification.

Materials & methods

Skin equivalent generation

IMR90s (human lung myofibroblast; passage 18; ATCC, VA, USA), human microvascular endothelial cells (HMEC1; passage 7; ATCC) and N/TERT-1 human keratinocytes (passage 7; a gift from Dr Jim Rheinwald [14] and Dr Ellen H van den Bogaard [15]) were routinely maintained at 5% CO₂ and 37°C. Culture media were as follows: IMR90 fibroblasts: Dulbecco’s modified Eagle medium/F12 50:50, 10% fetal bovine serum (FBS), 1% penicillin/streptomycin (P/S); HMEC1: MCDB131 base medium, 10% FBS, 1% P/S, 10 mM l-glutamine, 10 ng/ml EGF, 10 μg/ml hydrocortisone; N/TERT-1: keratinocyte serum-free medium base (Thermo Fisher Scientific, MA, USA), 1% P/S, 25 μg/ml bovine pituitary extract, 0.2 ng/ml EGF and 0.3 mM CaCl₂. Vascularized HSEs (VHSEs) were generated in culture inserts (12-well size; 0.4-μm pore; PET membrane; Corning, NY, USA) as previously described [16]. Briefly, to generate the dermal portion of the VHSE, 120 μl of acellular type I collagen (3 mg/ml) was seeded onto the insert membrane and allowed to gel for 30 min. IMR90s and HMEC1 cells were suspended in a type I collagen matrix (3 mg/ml) at 75,000 and 750,000 cells/ml, respectively. Then 250 μl of cellular collagen was seeded onto the acellular component and allowed to gel for 30 min, then submerged in growth medium (1:1 complete HMEC1 medium and complete IMR90 medium, at 10% FBS total). Next, 24 h after collagen seeding, the medium was changed to 5% FBS complete IMR90 medium and supplemented with 100 μg/ml l-ascorbic acid (L-AA). The dermal component was cultured for 7 days with the 5% FBS medium blend. To establish the epidermis, on day 7 of dermal culture, N/TERT-1s were seeded onto the dermal surface (170,000 cells/construct) and cultured to confluence for 3 days. On day 8 of culture, medium was changed to 1% FBS HSE differentiation medium [16] (3:1 Dulbecco’s modified Eagle medium/F12, 1% P/S, 0.5 μM hydrocortisone, 0.5 μM isoproterenol, 0.5 μg/ml insulin) with a supplement of 100 μg/ml L-AA. On day 9 of culture, medium was changed to serum-free HSE medium with 100 μg/ml L-AA; this medium blend was used until the culture end point as maintenance medium. Surrounding wells were filled with phosphate-buffered saline (PBS) to promote controlled humidity throughout culture and during OCT imaging. At day 10 of culture, VHSEs were brought to the air–liquid interface (ALI) and cultured with routine medium supplemented with L-AA at 100 μg/ml and maintained for 8 weeks (for every medium change, fresh L-AA supplement was used) [16]. All media and time points for media changes are as previously described [16].

Imaging

Imaging occurred at weeks 2, 3, 4, 5, 7 and 8 of the ALI culture period. To ensure sterility during OCT imaging and prevent optical artifacts caused by the plastic well plate lid, condensation and residual medium at the ALI, specific procedures were followed. Culture plates were equilibrated at room temperature with the lid off in a sterile biosafety cabinet for 20 min after being removed from the incubator, reducing condensation. Immediately prior to imaging, the constructs were checked for any residual medium at the air interface of the VHSE, and this was removed with aspiration if necessary while maintaining the medium below the culture insert. Even with the removal of visible medium from the epidermal surface, the sample remained moist and reflective. To minimize the reflective effect of the epidermal surface and well plate lid, the sample arm of the OCT system was tilted at 15 degrees, reducing reflection while maintaining an adequate sample illumination for structural imaging. During OCT imaging, constructs were held at room temperature for a maximum of 2 h; no loss of sample viability was observed.

OCT imaging was conducted with a custom-built fiber-based spectral domain OCT system centered at 1310 nm, as described previously [17]. In this system the reference arm remains stationary, and the interference pattern is detected in a spectrally resolved manner on a line scan camera. Light is sent into a circulator and then to a 90/10 fiber coupler that splits the light between the sample and reference arms, respectively. The reflected light from each arm travels back through the splitter and is redirected by the circulator to a custom-built spectrometer in the detection arm. Raster-scanning across the sample is controlled with two-dimensional galvanometer-mounted mirrors in the sample arm. Technical settings for data acquisition were as follows: scanning area 4 × 4 mm (-1 × 1 V); 4096 A-lines per B-frame, 400 B-frames per volume, 10 volumes averaged per construct. Resulting image size was 4096 × 512 × 400 voxel (4 × 2 × 4 mm).

Fixation, staining & confocal imaging

After 8 weeks of ALI culture, VHSEs were processed as previously described [16]. Briefly, samples were fixed and permeabilized at culture end point: 5 min pre-fix with 4% paraformaldehyde in PBS, then 1 h fix with 0.5% Triton™ X-100 and 4% paraformaldehyde in PBS at room temperature. For staining, primary antibodies used were filaggrin (AKH1) mouse monoclonal IgG (sc-66192, 1:250 dilution, Santa Cruz Biotechnology, CA, USA), involucrin rabbit polyclonal IgG (55328-1-AP, 1:250 dilution, Proteintech Group, IL, USA), Cytokeratin 10 (DE-K10) mouse IgG supernatant (sc-52318, 1:350 dilution, Santa Cruz Biotechnology), then DRAQ7 was used as a nuclear counterstain (7406, 12 μM final concentration, Cell Signaling, MA, USA). All sample staining was completed in blocking buffer [16] at 4°C for 48 h. Following staining, samples were imaged via confocal microscopy (TCS SPEII, Leica Microsystems, IL, USA).

Image processing & quantification

Confocal and OCT volumes were automatically segmented, and epidermal thickness was quantified using a custom algorithm implemented in MATLAB 2020b (MathWorks, MA, USA). Example MATLAB code is included as a Supplementary File. Briefly, the OCT algorithm was as follows. First, depth profiles with excess specular reflection were removed by performing intensity thresholding. Ten volumetric scans were acquired from each HSE and averaged in order to reduce speckle, resulting in a single 16-bit 4096 × 512 × 400 voxel (3.9 × 2 × 3.9 mm) volume of corrected intensity values. From the single volume per sample, intensity thresholding was performed to detect the upper surface of the skin epidermis. To smooth small gaps in the image, a morphological closing with a two-dimensional disk (40-voxel radius) followed by a Gaussian filter (standard deviation 1.5 × 8 × 1.5 voxels) was applied to the image volume. To segment the epidermis from the bulk, hysteresis thresholding was applied with a high intensity threshold of 16,000 (corresponding to ∼47.8 dB above the OCT system noise floor) and a low threshold of 8000 or 4200 (corresponding to ∼40.9 and ∼34.4 dB, respectively). Thresholds were empirically determined for this 16-bit dataset. The resulting binary volume was cleaned via area opening to remove small artifacts and a morphological closing with a 40-voxel radius disk. The bottom and top surfaces of the binary object were detected. Streaking artifacts, common in OCT imaging, presented as sharp peaks on the surfaces and were removed via automatic detection of high gradients and local averaging. Volumetric thickness was calculated from the voxel difference between the top and bottom surfaces scaled by the acquisition size (4 × 2 × 4 mm). For each volume, 30 A-lines were removed from left and right sides (1.4%) for each frame due to edge effects. Thickness average was calculated from imaging volumes of 3.9 × 2 × 3.9 mm. An example MATLAB script file is attached as Supplementary File 1.

For confocal volumes, epidermal thickness was calculated from imaging volumes of ∼3.3 mm × 370 μm × 250–400 μm (imaging depths were adjusted per sample, but a consistent voxel size of 0.7 × 0.7 × 3 μm was used). A similar custom algorithm was applied to calculate epidermal thickness from confocal volumes. Briefly, noise was removed via a median filter (5 × 5 pixel kernel for epidermal-specific markers and 3 × 3 kernel for nuclear markers) applied to each XY-plane, and intensities were scaled by linear image adjustment. Background autofluorescence was removed using a 20-voxel radius rolling ball filter on each XY-plane, and the epidermis was segmented using hysteresis thresholding with values empirically determined for each fluorescent channel (for epidermal markers, 3000 and 65,000 were used, while nuclei thresholds were set at 1150 and 64,000; all images were 16-bit and thresholds were empirically determined for this dataset). Small artifacts and gaps in the epidermal binary volume were removed through morphological closing and opening with a disk structuring element (pixel radius 10) for epidermal stains and a sphere structuring element (pixel radius 8) for nuclear stain. Epidermal thickness was calculated from the binary epidermal volume scaled by the voxel size (0.7 × 0.7 × 3 μm).

Data analysis

OCT imaging was performed at ALI culture weeks 2, 3, 4, 5, 7 and 8 (n = 4 culture replicates for every time point). Confocal imaging was completed on three of the same samples that were OCT imaged, after fixation and staining (n = 3). Analysis of variance followed by Tukey’s honestly significant difference post hoc test was used to test for statistically significant differences between time points; pairwise comparisons of 8-week confocal and OCT data were completed through two-tailed t-test. Significant differences (p < 0.05) are represented by a single asterisk.

Results & discussion

The epidermis of both native and in vitro skin is highly reflective, producing a strong signal in OCT [5]. This allows for effective automatic segmentation of the epidermis, as demonstrated with our algorithm (Figure 1A). Here, the higher threshold (8000; ∼40.9 dB) limits the automated segmentation to the highly reflective epidermis. As validation, after culture the HSEs were fixed, stained and confocally imaged for epidermal markers (Figure 1B). Median thickness values for OCT from weeks 2 to 8 and post-fixation confocal images are shown in Figure 1C. No statistically significant difference between the OCT time points and confocal thickness quantifications was found using the high threshold value, indicating stable culture morphology and agreement between the OCT and confocal modalities. As a further test, an additional sample was fixed, stained and confocally imaged at the week 4 time point. The fixed week 4 epidermis was qualitatively similar to that of the fixed week 8 cultures (Supplementary Figure 1), although quantification was not performed due to the limited sample size. The values obtained using the high threshold value (Figure 1A & C) are similar to in vivo epidermal thickness determined by OCT found in young and aged humans at the forearm (mean ± standard deviation: young, 71.8 ± 10 μm; aged, 60.8 ± 7.4 μm) [11]. The confocal thickness values at week 8 are slightly low compared with the human epidermal thickness values found in literature [10,11] as well as the OCT measurements. This may be due to shrinkage of the epidermis during fixation, as formaldehyde-based fixations have been shown to reduce cell and tissue size [18]. Furthermore, it is important to note that the in vivo thickness values are often obtained manually, for example via measurement of five predefined positions in the OCT image and then the average of five measurements as calculated by a single user [11]. The automated method presented here eliminates the necessity of an investigator to manually select locations and determine thickness by enabling automatic detection of every XY position in a volume.

Additionally, we observed a prominent reflective region immediately basal to the strongly reflective epidermis (Figure 2A). This region can be efficiently segmented from the bulk of the dermis using a lower threshold value (4200; ∼34.4 dB) and quantified (Figure 2B). Using the lower threshold value, a significant difference in thickness was found between the 8-week time points of OCT and confocal measurements (p < 0.05). No significant variation was found among the OCT measurement time points. The thickness values obtained using the lower threshold are similar to prior OCT studies in half-thickness skin equivalents at 22 days (100–200 μm, as reported by the authors) [12] These higher thicknesses may represent inclusion of the basement membrane and reticular dermis [7], differences in epidermal morphology, or differences in quantitation methodology.

All data presented here were obtained for whole sample volumes. To demonstrate the increased dimensionality of the dataset, representative-thickness surfaces of single samples analyzed with high and low thresholds are shown in Figure 3A. To determine the volumetric stability of samples over time, interquartile range (IQR) of the thickness for each sample is presented for both thresholds (Figure 3B). IQRs for all samples were similar from weeks 2–8 with no significant changes (p < 0.05), again demonstrating stability of the culture model. The small spread of IQRs shows the spatial uniformity of the HSE epidermal volumes. These data, combined with the thickness values, demonstrate both volumetric and temporal stability of HSE epidermal thickness. Importantly, tracking coverage and stability of the epidermis could be used to optimize HSE protocol steps including epidermal seeding and time points of ALI establishment, which can be technically challenging [16]. Long-term culture stability of HSEs has previously been indicated by epidermal thickness and the presence of viable epidermal cell layers [19]; the automated thickness measurement presented here enables the tracking of culture stability in real time as indicated by epidermal thickness. This method can additionally be used to investigate disease and aging models over time, and for wound tracking in vitro, as previously shown [9,10].

Conclusion

This work establishes an automated epidermal thickness tracking technique for tissue-engineered skin constructs using image analysis of OCT images. Detailed OCT scanning parameters and analysis algorithms are provided. To demonstrate the method, HSEs were generated and their epidermal thickness was tracked throughout the ALI culture period of 8 weeks. Quantification of epidermal thickness was completed through custom MATLAB algorithms using two different threshold values (high and low). Thickness values (Figures 1C & 2B) were similar to those found from epidermal thickness calculated from in vivo human skin OCT studies and prior manual OCT studies of half-thickness skin models [10,11]. Values obtained with the high-threshold analysis were similar to our validation with post-fixation confocal imaging (Figure 1C). We conclude that the method of automated epidermal thickness detection using the noninvasive OCT system is a simple and straightforward method of assessing HSE morphology throughout culture.

Future perspective

Over the next several years, tissue engineering will continue to grow as a field, and tissue-engineered skin equivalents will become a more routine tool in research labs. While useful, these models will present new challenges in analysis, especially in regard to mapping of three-dimensional morphology. Consistent with trends in other fields, we expect dramatic expansion in the amount of volumetric data produced in skin biology; automated analysis will be needed to reduce bias and enable analysis of larger datasets. The minimally invasive method presented here contributes toward the goal of ensuring accessible automated tracking of these complex cultures.