All experiments in this study were conducted with the approval of Nagoya University and in accordance with the Guidelines of Nagoya University.
Culture of human iPSC
The human iPSC lines (clones: 1383D2 and 1383D6) were generated from a non-HLA-homozygous donor under feeder-free conditions and provided by Drs. Masato Nakagawa and Shinya Yamanaka of the Center for iPS Cell Research, Kyoto University24. The experimental protocols dealing with human subjects were approved by the institutional review board at Ethics Committee of Graduate School and Faculty of Medicine, Kyoto University. Written informed consent was provided by each donor. The clones 1383D2 and 1383D6 were generated from peripheral blood cells by five plasmids (pCE-hOCT3/4, pCE-hSK, pCE-hUL, pCE-mp53DD and pCXB-EBNA1) for induction of pluripotency. The human iPSC line (clone: A18945) was purchased from Thermo Fischer. The clone A18945 was generated from CD34+ human cord blood cells by three episomal vectors of seven factors SOX2, OCT4, KLF4, MYC, NANOG, LIN28, and SV40L T antigen. The human iPSC were maintained on iMatrix 511 (Nippi Inc., Tokyo, Japan)-coated culture dishes in StemFit AK02N (Ajinomoto, Tokyo, Japan)24. For passaging, when hiPSC grew almost to confluency, hiPSC were treated with accutase (Nacalai Tesque, Kyoto, Japan), dissociated into single cells, and replated at 5.0 × 103 cells/cm2 onto iMatrix 511-coated dishes in AK02N in the presence of 10 μM Y-27632 (Wako, Osaka, Japan) for 24 h. The medium was changed to a fresh one without Y-27632 on the next day of plating and thereafter every 2 days. hiPSC were maintained in a humidified atmosphere of 5% CO2 and 95% air at 37 °C. hiPSC at passage 6 to 13 were used for the present study to guarantee pluripotency. Data from the clone 1383D6 were presented in “Results” and Figures, unless otherwise indicated.
RPE differentiation and RPE sheet production
For differentiation, undifferentiated hiPSC were treated with accutase for 5 min, dissociated into single cells, and plated onto iMatrix 511-coated 6-well culture plates (Thermo Fisher Scientific, Waltham, MA) at 3.0 × 104 cells/well. For RPE induction, cells were treated with 100 nM LDN193189 (Sigma, St. Louis, MO), 500 nM A-83-01 (Wako), 1 μM IWR-1-endo (Wako), and 10 μM Y-27632 were added to IMDM/Ham’s F12 (1:1, both from Sigma) supplemented with 10% KnockOut Serum Replacement (Thermo Fisher Scientific), 0.5 mM Monothioglycerol Solution (Wako), 1% Chemically Defined Lipid Concentrate (Wako), and 2 mM l-glutamine (Wako) for the initial 6 days, and then with 3 μM CHIR99021 (Wako), 2 μM SU5402 (Wako), and 10 μM Y-27632 in IMDM/F12 for another 12 days. From day 18, the medium was changed to DMEM/F12 (Sigma) supplemented with 10% KnockOut Serum Replacement, 1% N2 Supplement (Wako), and 2 mM l-glutamine. In some experiments, 10 mM nicotinamide (Wako) was added from day 12 to day 24. For further maturation, hiPSC-RPE were cultured in RPE maintenance medium (67% high glucose DMEM (Wako), 29% Ham’s F12, 2% B27 supplement minus vitamin A (Thermo Fisher Scientific), 2 mM l-glutamate, 100 U/mL Penicillin and 100 μg/mL Streptomycin. The culture medium was changed with a fresh one every day.
For RPE sheet generation, the hiPSC-RPE were treated with 0.25% Trypsin–EDTA (Wako) for 10 min and dissociated into single cells by pipetting. The cells were filtered by passing through a 35-μm cell strainer (Corning Inc., Corning, NY) and plated onto iMatrix 511-coated 12-well transwell insert (Corning) for functional analysis or iMatrix 511-coated 6-well culture plates for further proliferation. For the functional analysis, hiPSC-RPE were cultured in RPE growing medium containing Ham’s F12 with 10% fetal bovine serum (Thermo Fisher Scientific), 2 mM l-glutamate, 100 U/mL Penicillin and 100 μg/mL Streptomycin. After 14 days, the hiPSC-RPE growing medium was replaced with RPE maintenance medium containing 10 ng/mL basic fibroblast growth factor (Wako), and 0.5 μM SB431542 (Wako)13. All the cells were used until passage 3.
Quantitative real-time polymerase chain reaction
Total RNAs of hiPSC and differentiated cells were extracted using Tissue Total RNA Mini Kit (Favorgen Biotech Corp., Taiwan), and then reverse-transcribed with PrimeScript RT Master Mix (TaKaRa, Shiga, Japan) according to the instructions of the manufacturer. Quantitative real-time PCR was performed with TB Green Fast qPCR Mix (TaKaRa) on LightCycler 96 system (Roche, Indianapolis, IN). Expression levels were normalized to those of GAPDH. The primers used are listed in Table S1.
Cells were immunolabeled as described previously6. Cells were washed with PBS and then fixed with 4% paraformaldehyde in phosphate-buffered saline (PBS) for 15 min at room temperature. After rinsing with 0.1% Triton three times, Cells were blocked with blocking one (Nacalai Tesque) for 1 h, then incubated with primary antibodies overnight at 4 °C. After washing three times by PBS, cells were incubated with appropriate second antibodies and DAPI (Wako) for 2 h at room temperature. The primary antibodies and their working dilutions were as follows: mouse anti-Oct3/4 (1:1,600, BD Pharmingen, San Diego, CA), mouse anti-Mitf (1:100, Abcam, Cambridge, MA), rabbit anti-Pax6 (1:600, Covance, Munich, Germany), rabbit anti-Pax6 (1:500, Abcam), sheep anti-Chx10 (1:200, Exalpha Biologicals, Shirley, MA, USA), rabbit anti-ZO-1 (1:300, Thermo Fisher Scientific), mouse anti-Bestrophin (1:500, Abcam), and mouse anti-N-cadherin (1:200, Cell Signaling Technology, Danvers, MA, USA). The secondary antibodies used were as follows: anti-mouse IgG, anti-rabbit IgG, and anti-sheep IgG conjugated with Alexa488 or Alexa596 (1:1,000, Jackson Immunoresearch Laboratories Inc., West Grove, PA). For labeling of F-actin, fixed cells were treated with blocking one and then treated with Rhodamine-X-conjugated phalloidin (Wako) and DAPI for 2 h at room temperature. For lectin labeling for hiPSC, fixed cells were blocked with blocking one and then treated with FITC-labeled recombinant BC2L-C N-terminal domain (rBC2LCN-FITC, Wako) and DAPI for 2 h at room temperature. Labeled cells were imaged with a confocal laser-scanning microscope with GaAsP detectors (LSM800, Zeiss, Jena, Germany) using a 20× objective lens (NA 0.75, Zeiss) or 40× objective lens (NA 1.2, Zeiss).
To evaluate the purity of hiPSC-RPE, cells were treated with 0.25% Trypsin/EDTA for 15 min, dissociated into single cells by gentle pipetting, passed through a 35-μm cell strainer, and seeded on iMatrix 511-coated 3.5 cm culture dishes at a density of 1.0 × 105 cells/cm2. Cells were cultured for 7–8 days in RPE growing medium until confluency. Then, the cells were fixed with 4% paraformaldehyde in PBS for 15 min at room temperature and subjected to immunostaining with anti-Chx10 and anti-Mitf antibodies. Labeled cells were imaged with a confocal laser-scanning microscope with GaAsP detectors. At least three visual fields were randomly selected. The number of PAX6-positive cells and MITF-positive cells were counted. Over 800 cells were counted in one sample. The cell purity was determined as the percentage of the number of positive cells to the number of total cells.
Transepithelial electrical resistance (TER) measurement
hiPSC-RPE were seeded on the transwell insert for TER measurement using the electrical resistance system (Millicell, Millipore, Billerica, MA). A test electrode of the electrical resistance system was connected to the input port on the meter and confirmed at the value of 1,000 Ω. The electrode was rinsed with 70% ethanol for 15 min, and then with RPE maintenance medium for 15 min. Measurement was performed within 10 min after removal from the incubator to avoid reducing temperature. Net TER (Ω cm2) was calculated by subtracting the value of a blank insert from the experimental value and multiplying by the area of the insert membrane.
hiPSC-RPE sheets were incubated in the RPE maintenance medium containing 7 μg of pH-Rhodo-labeled bioparticles (Thermo Fisher Scientific) at 37 °C in incubators for 4 h. As a negative control, hiPSC-RPE sheets were cultured in the presence of pH-Rhodo-labeled bioparticles at 4 °C. The cells were fixed with 4% paraformaldehyde in PBS, washed with PBS three times, and stained with phalloidin and DAPI. Fluorescent signals were imaged with a confocal laser-scanning microscope with GaAsP detectors using a 20× or 40× objective lens.
Image processing and prediction model analysis
F-actin labeled microscopic images (total 54 images, 20× magnification) were processed by CL-Quant (Nikon corp., Tokyo, Japan) by designing filter sets according to the manufacture’s protocol. The labeled image processing and morphological feature measurement procedures are illustrated in Fig. S6. After cell recognition (Fig. S6A), cells with size < 500 pixels (305 μm2) were deleted from the recognized cell data because they tended to include mis-recognized objects or noise. After the data cleansing, 8 morphological features were measured by CL-Quant: (1) area, (2) compactness, (3) inner radius, (4) length, (5) length–width ratio, (6) outer radius, (7) perimeter, and (8) width (Table S2). The morphological features were carefully selected to eliminate multi-collinearity issues. After measuring cellular objects from each image, three replicate datasets consisting of 100–200 cells randomly picked from total cells by bootstrap in one image were produced from the original data. For 162 samples (= 54 images × 3), each average (AVE) and standard deviation (SD) of 100–200 cells were summarized and listed as explanatory variable sets (16 parameters) for each sample (Fig. S6B). The experimentally determined TER values (ranging from 70 to 390 Ω cm2) per each cellular sample were tagged as objective variables.
Before the construction of a prediction model, hierarchical clustering (average linkage) was applied to analyze the data distribution (Fig. S7A). The clustering indicated that the total data could be clustered in two groups: one cluster of which the majority correlated with high TER (group A, the right cluster) and the other cluster of which the majority correlated with low TER (group B, the left cluster). Although there was a general tendency of correlation between the morphological pattern and its functional TER value, there also existed a sub-population of samples that did not follow this pattern. In this analysis, we modeled the “primary morphological correlation”. Samples (> 250 TER) from group A and samples (< 150 TER) from group B were selected as “primary cells that represent the TER value”. The TER prediction model and TER-positive (> 250 TER) or negative (< 150 TER) discrimination model were constructed using the Least Absolute Shrinkage and Selection Operator regression as previously described21. The selected samples (total n = 98) were used for training the models. For checking their prediction performance of the models, the correlation of experimentally determined values (TER values, x-axis) and prediction values (predicted TER values only from the morphological features, y-axis) was plotted. The performance was also evaluated by root mean square error (RMSE) value, which indicates the error between the experimentally determined value and the predicted value. The accuracy, sensitivity, specificity, and its prediction probabilities were calculated for assessment of the discrimination performance. The performance was validated through leave-one-out cross-validation.
To adopt the constructed morphological TER prediction model to the non-labeled phase-contrast microscopic images, the 2 lots (6 samples), which showed high TER (mean TER value 342 ± 26 Ω cm2) and low TER (mean TER value 150 ± 12 Ω cm2) were selected for image processing and prediction (Figs. 5A and S8). From the non-labeled images, randomly chosen individual cells (~ 500 cells) were traced using the Cell Magic Wand tool (https://github.com/fitzlab/CellMagicWand) in ImageJ/Fiji with minimal manual assistance. By the traced clearer border of cells, the phase-contrast images were processed with the original filter set by CL-Quant as illustrated in Fig. S8A. From the recognized image, the morphologies of individual cells were measured by CL-Quant. From the measured group of cellular objects, their average and SD of 8 morphological features were measured to list 16 explanatory variables. These variables were applied to the constructed TER prediction model and TER-positive or -negative discrimination model, and their prediction performances were evaluated in the same manner.
To identify whether our model could be used to predict TER values of RPE sheets across cell lines, six different hiPSC-RPE sheets derived from three different cell lines were generated in a clinical manufacturing facility. Three RPE sheets were produced from the iRTA-01 cell line on different dates. Two RPE sheets were derived from the 253G1 hiPSC line, and one sheet was derived from the Ff-I01 hiPSC line. Phase-contrast images of the six hiPSC-RPE sheets were processed for morphological analysis as described above. The morphologies of individual cells were applied to the constructed TER prediction model, and their prediction performances were evaluated as described above.
Data processing and model analyses were coded with R 3.4.4. (https://cran.r-project.org). The codes used for the model construction are available at https://mega.nz/folder/zhA0SYAQ#g3uRzUiGKYbdDJmM6t3Lvw.
Data in the experimental sections are expressed as means ± SEM. All sets of experiments were performed at least three times. The statistical significance of the differences between groups was determined using unpaired t-tests with GraphPad Prism 7 (GraphPad Software, Inc., San Diego, CA). Probability values less than 5% were considered statistically significant.