Preclinical safety studies of human embryonic stem cell‐derived retinal pigment epithelial cells for the treatment of age‐related macular degeneration

Abstract As pluripotent stem cell (PSC)‐based reparative cell therapies are reaching the bedside, there is a growing need for the standardization of studies concerning safety of the derived products. Clinical trials using these promising strategies are in development, and treatment for age‐related macular degeneration is one of the first that has reached patients. We have previously established a xeno‐free and defined differentiation protocol to generate functional human embryonic stem cells (hESCs)‐derived retinal pigment epithelial (RPE) cells. In this study, we perform preclinical safety studies including karyotype and whole‐genome sequencing (WGS) to assess genome stability, single‐cell RNA sequencing to ensure cell purity, and biodistribution and tumorigenicity analysis to rule out potential migratory or tumorigenic properties of these cells. WGS analysis illustrates that existing germline variants load is higher than the introduced variants acquired through in vitro culture or differentiation, and enforces the importance to examine the genome integrity at a deeper level than just karyotype. Altogether, we provide a strategy for preclinical evaluation of PSC‐based therapies and the data support safety of the hESC‐RPE cells generated through our in vitro differentiation methodology.


| INTRODUCTION
Age-related macular degeneration (AMD) is the leading cause of blindness in people aged over 65 years in industrialized countries, 1 and it manifests as dry (nonexudative) and wet (exudative) forms. The dry advanced form of the disease, also known as geographic atrophy, is untreatable at present. Although its cause is known to be multifactorial, including both genetic and environmental factors, 2 a main hallmark is the degeneration of retinal pigment epithelium (RPE) cells: a monolayer of polarized hexagonal and heavily pigmented cells that constitutes the outer blood-retina-barrier and performs a number of central tasks in the eye. 3 RPE cells are crucial for the maintenance of the photoreceptor layer, and their scarcity leads to a progressive outer retina degeneration and vision loss. [4][5][6] Therefore, a potential treatment strategy would involve cell replacement of RPE cells from human pluripotent stem cells (hPSC). Several groups, including ours, have established defined, xenofree, and robust differentiation protocols to derive RPE cells from human embryonic stem cells (hESCs) with morphological, physiological, and functional features shared with native RPE cells. [7][8][9][10] Preclinical studies have shown that subretinal suspension or sheet injections of hPSC-RPE can prevent photoreceptor and vision loss, 7,[11][12][13][14] altogether supporting its translation into the first clinical studies. [15][16][17][18][19] In addition to cell therapies for AMD, differentiated RPEs have been used to treat patients with Stargardt's macular dystrophy. 20 Also, other cell types derived from PSCs are currently being tested in clinical studies for the replacement of cells in cardiac ischemia, type I diabetes (trial ClinicalTrials. gov number NCT03162926 and NCT03163511) or Parkinson's disease. [21][22][23][24] Any cell transplantation strategy should restore function but must also be thoroughly evaluated for safety, which is of particular importance when the starting material relies on PSCs that have inherent proliferative and potential tumorigenic properties. Although guidelines for assessing safety of hPSCs are still in development, some groups have paved the path, either by reporting preclinical safety studies for their ongoing clinical trials for AMD or by suggesting possible assays for such evaluation. 16,[25][26][27][28][29][30] In the present study, we have performed extensive tumorigenicity and migratory tests among other whole genome-wide studies to assess the stability and safety of our hESC-derived RPE cells, and provide novel insights for the development of hPSC-derived therapies.

| hESC-RPE in vitro differentiation
hESC line HS980 was characterized and cultured under xeno-free and defined conditions according to the previously described method 31 (with ethical permit from the Swedish Ethical Review Authority, EPN 2011/745:31/3). Donors gave their informed consent for the derivation and subsequent use of the hESC lines. hESCs were cultured to confluence on 10 μg/mL recombinant human laminin (rhLN)-521 (Biolamina) and manually scraped to produce embryoid bodies (EBs) using a 1000-μL pipette tip as described previously in our group. 7 EBs were cultured in suspension in low attachment plates (Corning) at a density of 5 to 7 × 10 4 cells/cm 2 . Differentiation was performed in custommade NutriStem hESC XF medium without bFGF and TGFβ (Biological Industries), and 10-μM Rho-kinase inhibitor (Millipore, Y-27632) was added to the suspension cultures only during the first 24 hours. Media was changed twice per week. Following 5-weeks of differentiation, optical vesicles (OVs) were mechanically dissected out of the EBs using two needles. Cells were dissociated using TrypLE Select (Ther-moFisher Scientific), followed by flushing through a 20G needle. Cells were seeded through a cell strainer (ø 40 μm, BD Bioscience) onto 20 μg/mL LN-coated dishes at a cell density of 0.6 to 1.2 × 10 4 cells/ cm 2 and fed twice a week with NutriStem hESC XF medium without bFGF and TGFβ (Biological Industries), resulting in a pure and homogenous hESC-RPE culture, as described previously. 7

| Quantitative real-time PCR
Total RNA was isolated using the RNeasy Plus Mini Kit and treated with RNase-free DNase (both from Qiagen). cDNA was synthesized using 1 μg of total RNA in 20 μL reaction mixture, containing random hexamers and SuperScript III reverse transcriptase (ThermoFisher Scientific), according to the manufacturer's instructions.

Significance statement
This study evaluated the preclinical safety of in vitro differentiated retinal pigment epithelial cells from embryonic stem cells by (a) examining karyotype and performing whole genome sequencing to assess genome stability; (b) performing single-cell RNA sequencing to ensure purity and absence of undifferentiated cells; and (c) examining biodistribution and tumorigenicity of transplanted cells to rule out malignant growth and migratory properties. The derived cells proved to be safe, and this study altogether provided a strategy for preclinical evaluation of PSC-based therapies. Also, the whole genome sequencing analysis illustrates that the preexisting load of germline variants is significantly higher than the introduced variants acquired through vitro culture or differentiation, which enforces the importance to evaluate the genome integrity at a deeper level than just karyotype.   2.10 | Whole-genome sequencing analysis gDNA was isolated as above and 250 ng were used for sequencing with Ilumina HiSeq X, 30X coverage. Whole-genome paired-end DNA sequencing reads of HS980 (p22), HS980 (p38), and hESC-RPE cells in biological triplicate experiments were aligned to the human reference genome (NCBI reference genome GRCh37 based "human_ g1k_v37_decoy") using the Burrows-Wheeler Aligner (BWA-0.7.8). 33 Sequencing quality and coverage was analyzed using QualiMap (v2.2.1). 34 Reference genome aligned BAM files was sorted and marked duplicated using Picard (v2.0.1). "GATK Best Practice" guidelines was followed to generate analysis-ready BAM files which includes local realignments and base quality recalibration using GATK bundle "b37" files that include data sets from HapMap, Omni, Mills Indels, and 1000 Genome Indels databases. Additionally, SNPs from NCBI-dbSNP (dbSNP-150) were included in the analysis. Data are available at Stockholm Medical Biobank upon request through the authors.
2.11 | Germline single nucleotide variant discovery, filtration and annotation Analysis-ready BAM files of HS980 (p22) (replicates 1, 2, and 3) were processed individually using GATK 3.7 HaplotypeCaller walker in genomic variant call format (gVCF) mode with default parameters. Output gVCF files of individual HS980 (p22) replicate was then used for raw germline single nucleotide variant (SNV) identification using Genotype-GVCFs walker. Furthermore, variant quality score recalibration was performed using VariantRecalibrator walker with default parameters followed by ApplyRecaliberation walker (truth sensitivity level 99.5 for SNP and 99.0 for indels) to select filter "PASS" variants separately for individual replicates. Finally, BCFtools "isec" utility was used to identify SNVs commonly present in all three replicates for further downstream analysis.
As an additional control set for analysis, publicly available preprocessed germline SNVs from 11 participants from Personal Genome Project UK were downloaded (https://www. personalgenomes.org.uk/data/) and annotated for clinical significance.

| Somatic SNV calling and annotation
Somatic SNV calling was performed using GATK 3.7-MuTect2 in a pairwise manner with default parameters. Comparisons were made between HS980 at passage 22 (HS980 p22) and hESC-RPE (differentiated from passage 22), followed by HS980 p22 compared with passage 38 (HS980 p38) to find somatic SNV mutations. All analysis was performed in three independent replicates. dbSNP150 and COSMIC-v83 VCF files were considered as an argument for -dbsnp and -cosmic, respectively. In addition, filter "PASS" somatic SNVs identified as a final outcome of MuTect2 pairwise analysis were merged to create a nonredundant set of somatic SNVs for HS980-p22 vs hESC-RPE and HS980-p22 vs HS980-p38.
Affymetrix CEL files were imported to the PartekGenomic Suite 6.6 (Partek Inc, St. Louis, Missouri) to perform pairwise CNV analysis.  For sequencing analysis, single-cell transcriptome sequencing reads were mapped to the human genome (hg19) including ERCC sequence using STAR aligner with default settings 40 and uniquely aligned reads were retained. The number of reads for each RefSeq and Ensemble annotated genes were calculated using featureCounts. 41 Cells were quality-filtered based on the exclusion criterium: had total aligned reads (within transcriptomic boundaries) lesser than 10 3 and showed expression of fewer than 2000 unique genes. Read count matrix from qualityfiltered cells was processed using R package Seurat (version 2.2.0). 42 Gene expression measurement was performed using NormalizeData function in Seurat with scale factor 10 000 followed by log-transformation. RunPCA, JackStraw, FindClusters, and RunTSNE functions were used to further process the data and obtain t-SNE cluster of cells. Two hundred microliters of the Matrigel cell suspension were injected subcutaneously in the mouse necks using a 27G needle.

| Biodistribution analysis
For rabbits, native RPE would most likely be removed by the mechanical pressure of the injection, but not a priori with any mechanical/chemical treatment as demonstrated previously. 7,14 In any case, if integration was successful, it implies that native RPE was removed and the retinal barrier was kept intact thus avoiding immune cell infiltration. At, 1, 4,

| Efficient differentiation of functional RPE from hESC
Cells were differentiated in accordance with our previously published xeno-free and defined methodology. 7 In brief, hESCs were aggregated as EBs in NutriStem hESC XF medium without exogenous growth factors. After 5 weeks, the pigmented structures were manually dissected and plated following single cell dissociation on (rhLN)-521 for an additional 5 weeks to generate mature pigmented hexagonal hESC-RPE ( Figure 1A). hESC-RPE were uniformly positive for protein expression of specific RPE markers such as BEST-1 and CRALBP ( Figure 1B). Transcriptional analysis showed robust expression of neuroectoderm transcripts SOX9 and PAX6, and RPE differentiation was apparent by the expression of genes such as BEST-1, RPE65, MITF, PMEL, and TYR, whereas pluripotency-related gene expressions were low (Figure 1C).
At a functional level, cultures showed active phagocytosis of isolated FITC-labelled bovine POS at 37 C when compared to the 4 C condition ( Figure 1D). Finally, hESC-RPE showed a normal karyotype with no clonal aberrations ( Figure 1E).

| Germline variations in the starting hESC
Genetic stability is a very relevant concern to translate PSC aligned to the NCBI reference genome GRCh37 based "human_ g1k_v37_decoy" using BWA-MEM (Table S2). The mapped reads were investigated further to identify putative germline and somatic SNVs, CNVs and large structural variants (SVs) following the Genome Analysis Toolkit (GATK) best practice guidelines (Figure 2A).
Next, mutational subtype analysis of these SNPs revealed that transition (Ts) was more common than transversion (Tv) (Figure S1), and heterozygous variants contribution was more than homozygous variants (62% and 38%, respectively). Furthermore, 67.99% SNVs SNVs were reported as "pathogenic" in ClinVar (Table S3), whereas 35, 20, and 18 germline SNVs with pathogenic FATHMM score 44 were annotated within COSMIC cancer gene census 36,45 (Table S4A)  Importantly, none of the identified mosaic variants were commonly present within the samples neither within TP53, as reported earlier. 26 3.4 | Acquired SNV changes during hESC culture and hESC-RPE differentiation

| Nontumorigenic growth after hESC-RPE injection
Although genomic and transcriptional analyses are informative, they still should be complimented by functional studies. Tumorigenicity studies were therefore performed to evaluate the risk of tumorigenic growth capacity of the hESC-RPE cells, especially of undifferentiated pluripotent cells. A well-established methodology, also recommended by the International Stem Cell Initiative, 48,49 is to screen for tumor formation following subcutaneous injection in immunocompromised NOG mice. 50 A benefit with subcutaneous injection is that the cell number is not as limited as it is in the subretinal space (the planned clinical site of delivery). Six groups of NOG mice (10 mice per group) were subcutaneously injected with increasing numbers of hESC (ranging from 1 to 10 million cells) to establish the number of cells that could potentially generate a tumorigenic growth. In parallel, 10 million cells from three time points along the differentiation protocol were injected in three groups (Table S1). 10/10, 9/10, and 8/10 mice injected with 1 million; 100 000 and 10 000 hESCs, respectively, developed detectable cell masses, which were not observed in any of the mice injected with lower amounts of hESCs. All teratomas analyzed showed contribution to all three germ layers ( Figure 4B). A subset showed formation of yolk sac tissue that has been suggested to be a malignant indicator associated with teratocarcinoma ( Figure S4A). Furthermore, 9/10 and 1/10 mice injected with 10 million cells obtained from partially differentiated EBs that were in culture for 3 and 5 weeks respectively showed cell mass formation ( Figure 4A). Importantly, and in contrast to the hESC-derived teratomas, these masses were only com-  Figure S4B). Several studies have also examined the tumorigenic risk at the clinical site (Table 1). 10

| Biodistribution studies only detect cells at site of transplantation
To analyze the migratory capacity of the injected cells, we first ensured that none of the genes with nonsynonymous acquired SNVs were associated with cell migration gene ontology classifications.
Secondly, we evaluated the presence of human cDNA by qPCR analysis of several organs/body parts (lung, spleen, liver, heart, kidney,    Lung Lung Abbreviations: hESC, human embryonic stem cell; RPE, retinal pigment epithelial. and biodistribution studies would overcome some of the limitations observed in previously described methods, assessing any tumorigenic and/or migratory potential of the final product, and providing more relevant information of the possible impact caused by the variations. [27][28][29][30]48 Till date, several groups leading the first clinical trials with hPSC-RPE cells have proven the safety of their products through preclinical studies, 15,16,27,50,54 which have been summarized in Table 1. In all these studies, they use rodent, pigs, or minipigs as preclinical models for teratoma studies with differences regarding injection site, duration, and number of animals and injected cells but overall showing that the differentiated cells fail to form any tumorigenic growths. Only da Cruz et al 15

| CONCLUSION
In the present study, we show that our differentiation protocol 7 generates pure, safe, and stable hESC-RPE cells without any abnormal chromosomal organization nor carcinogenic mutation load at a SNV or CNV level. In addition, we demonstrate that our hESC-RPE cells do not form any tumorigenic structures after 7 months when injected subcutaneously in immunodeficient mice, neither migrate to other organs in mouse or rabbit models. Even though functional assays like tumorigenicity and biodistribution studies are considered gold standard studies to prove the safety of these therapies, we would like to argue that comparative genome-wide genomic analysis together with single-cell characterization using flow cytometry and transcriptional analysis may be equally or even more informative when developing and testing new hPSC-derived therapies.