Xinqidi,Your biological research partner
Xinqidi Biotech Co.,Ltd,Wuhan,China 2008-2021
R&D 13th year

Retinoblastoma from human stem cell-derived retinal organoids

Issuing time:2021-07-30 14:32


Retinoblastoma is a childhood cancer of the developing retina that initiates with biallelic inactivation of the RB1 gene. Children with germline mutations in RB1 have a high likelihood of developing retinoblastoma and other malignancies later in life. Genetically engineered mouse models of retinoblastoma share some similarities with human retinoblastoma but there are differences in their cellular differentiation. To develop a laboratory model of human retinoblastoma formation, we make induced pluripotent stem cells (iPSCs) from 15 participants with germline RB1 mutations. Each of the stem cell lines is validated, characterized and then differentiated into retina using a 3-dimensional organoid culture system. After 45 days in culture, the retinal organoids are dissociated and injected into the vitreous of eyes of immunocompromised mice to support retinoblastoma tumor growth. Retinoblastomas formed from retinal organoids made from patient-derived iPSCs have molecular, cellular and genomic features indistinguishable from human retinoblastomas. This model of human cancer based on patient-derived iPSCs with germline cancer predisposing mutations provides valuable insights into the cellular origins of this debilitating childhood disease as well as the mechanism of tumorigenesis following RB1gene inactivation.


Retinoblastoma is a rare pediatric cancer of the developing retina that initiates in utero and is diagnosed within the first few years of life1. The vast majority of retinoblastomas initiate with biallelic inactivation of RB1 and a small subset (1–2%) initiate with MYCN amplification in the absence of RB1 inactivation2,3. Approximately half of all retinoblastoma cases involve a germline mutation in RB1 and 25% of germline retinoblastoma patients inherited the mutant allele from a parent4. Patients with germline RB1 mutations have an earlier age of onset because they only require inactivation of the remaining RB1 allele for tumor initiation1,5. Genomic studies have indicated that RB1 inactivation is sufficient for tumorigenesis6. Therefore, retinoblastoma is an important model of tumor initiation as a result of biallelic inactivation of a single tumor suppressor gene.

Early attempts to model retinoblastoma in mice by mutating the Rb1 gene failed to produce retinal tumors in Rb+/– mice7,8,9. Subsequent conditional inactivation of both copies of Rb1in the developing murine retina also failed to produce retinoblastoma10,11,12,13. Further molecular and genetic studies demonstrated species-specific intrinsic genetic redundancies and compensation among Rb family members prevent retinoblastoma in mice14. Conditional inactivation of Rb1 and p107 or Rb1 and p130 can lead to retinoblastoma10,11,12,13,14 in mice and more recently, a murine model of the MYCN amplified form of retinoblastoma was developed15.

While these genetically engineered mouse models (GEMMs) have provided important insights into retinoblastoma biology, there are some differences in the molecular and cellular features of retinoblastoma across species11,14,16,17. For example, in a recent epigenetic study of the developing retina and retinoblastoma from mice and humans, the human tumor epigenome mapped to a later developmental stage than that of the mouse tumors18. Those differences may be related to the initiation of tumorigenesis during retinal development, the cellular origin of the tumors, or both19. Another important difference between human and murine retinoblastoma is their drug sensitivity. In side-by-side preclinical studies of murine retinoblastoma and orthotopic patient-derived xenografts (O-PDXs) of retinoblastoma, there were differences in the response to the conventional systemic chemotherapy regimen currently used to treat retinoblastoma patients20. Virtually, all of the GEMMs had a complete and durable response while few of the O-PDXs tumor-bearing mice had any response20. Therefore, a laboratory model of human retinoblastoma could contribute to our understanding of the retinoblastoma cell of origin and provide patient-specific tumors for preclinical testing.

To produce a laboratory model of human retinoblastoma, we created iPSCs from 15 participants with germline RB1 mutations or deletion. Multiple clones from each iPSC line were validated for retention of the germline mutation, subjected to molecular profiling, including whole-genome sequencing, and characterized for embryoid body formation, neural rosette formation, and retinal differentiation using an improved human retinal organoid procedure. Representative clones from each participant were then differentiated into retinal organoids, dissociated, and injected into the eyes of immunocompromised mice to monitor tumor formation. Parallel experiments were performed with targeted RB1 gene inactivation using CRISPR-Cas9. Individual tumors derived from the organoids were passaged by intraocular injection and cryopreserved, as done previously21. Complete molecular, cellular, and genetic profiling were completed and the iPSC lines and tumors have been made available to the biomedical research community free of charge with no obligation to collaborate through the Childhood Solid Tumor Network21.


Isolation and characterization of patient-derived iPSC lines

In total, 11 patients and 4 family members were enrolled on the RETCELL (NCT02193724) protocol at St. Jude Children’s Research Hospital (Supplementary Data 1 and Fig. 1A, B). Four samples for reprogramming were obtained by skin biopsy at the time of anesthesia for examination of the eyes as part of routine care, and eleven samples were obtained with the peripheral blood draw. Germline DNA samples were also obtained from 14 of the 15 participants. Participants were selected based on their clinical presentation to represent a broad spectrum of penetrance in heritable retinoblastoma disease. Four participants had no family history and were diagnosed at less than 12 months of age (bilateral, n = 3; asynchronous bilateral, n = 1) and two participants had 13q deletions (manifested as a unilateral and bilateral disease). Four family cohorts (n = 9) consented to the protocol, including two families where the parent was previously unknown to carry the RB1 mutation until the children were diagnosed with bilateral or trilateral retinoblastoma (Supplementary Data 1 and Fig. 1A, B). Offspring of the family cohorts presented with bilateral retinoblastoma (n = 4) and trilateral disease (n = 1) at diagnosis, and the parents had no evidence of tumors by ophthalmologic screening.

Fig. 1: Generation of iPSC lines from patients with germline RB1 mutations.

A Map of the genomic locus showing each of the 27 exons of the RB1 gene and the color-coded protein-coding domains. The germline mutations in each patient/family are shown below the full-length protein. B Representative pedigree for an unaffected carrier (SJRB-iPSC-10) and two affected children (SJ-iPSC-8 and SJ-iPSC-9). C Representative micrographs of SJRB-iPSC-3 colony immunostained for POU5F1 (green) and SOX2 (red) with DAPI (blue) nuclear stain. Staining was repeated on another iPSC line with the same results. D Sanger sequencing chromatogram showing the heterozygous nonsense mutation in the RB1 gene (GAA→UAA). E Representative two-color fluorescence in situ hybridization (FISH) of SJRB-iPSC-13 showing the 13q deletion (200 cells were analyzed). F Drawing of the neural rosette differentiation experiment. G, H Differential interference contrast (DIC) and DAPI-stained colony from the neural rosette induction procedure. Arrows and the enlarged box indicate neural rosettes. Three clones of each line were differentiated and each line was able to produce rosettes. I Boxplot of the normalized relative fold of neurogenic genes for each iPSC line (n = 15) and H9 ESCs from qRT-PCR of the neural induction assay. J Micrograph of representative retinal organoid from H9 ESCs and SJRB-iPSC-15 indicating retina (arrow). K GAPDH normalized relative fold gene expression for 3D retinal organoids from representative iPSC lines using H9 ESCs as controls. Each dot represents the mean of two technical replicates for an individual organoid. L Micrograph of cryosection of day 45 retinal organoid that was labeled with EdU for 1 h prior to harvest showing recoverin expressing photoreceptors and EdU+ retinal progenitor cells (red) with DAPI counterstain. Each of the 15 lines was sectioned showing similar results in neural retina regions. Box and whisker plots include center line as median, box as Q1 and Q3, and whiskers as 1.5× interquartile range. Scale bars: C, 50 μm; E, L, 10 μm.

Fibroblasts or blood were reprogrammed with the CytoTune Sendai reprogramming kit (Supplemental Information). In total, 71 iPSC clones were isolated for 15 participants and the germline DNA mutations in RB1 were verified by Sanger sequencing for most lines, and fluorescent in situ hybridization for SJRB-iPSC-13 which had a RB1 locus deletion (Supplementary Data 2, Fig. 1A–E and Supplemental Information). Only fully reprogrammed clones with trilineage differentiation potential and a normal karyotype were used for subsequent studies, hereafter referred to as SJRB-iPSC-1-15 (Supplementary Fig. 1A–H and Supplementary Data 2). Whole-genome sequencing was performed on the iPSC clones, and the patient germline DNA to ensure additional deleterious gene mutations were not accumulated during the process of iPSC production (Supplementary Data 3).

To test the propensity of individual clones to form neurons, we produced embryoid bodies (EBs) in neural induction medium for 2 days and then transferred them to Matrigel-coated plates to form neural rosettes (Fig. 1F–H). We performed quantitative PCR (qRT-PCR) with a panel of primers for early neurogenic genes (Supplementary Data 2 and Fig. 1I). Using these morphologic (rosette formation) and molecular (qRT-PCR) criteria, we identified multiple clones for each patient that had neurogenic potential (Supplementary Data 2). To determine if the iPSCs could make retina in 3D organoid cultures, we used the protocol developed by Sasai22 and compared retinal formation for each iPSC clone to that of the H9 ESC line using qRT-PCR and immunostaining (Fig. 1J–L and Supplementary Data 2). Overall, we were able to isolate multipotent iPSC clones from each donor with normal karyotype and neurogenic/retinal differentiation competence that retained their germline RB1 alterations.

Optimization of retinal specification in 3D organoid cultures

The original Sasai method for producing 3D retinal organoids was developed and optimized using a highly neurogenic H9 human ESC line23 and has not been systematically tested for patient-derived iPSCs. While each of our iPSC lines was able to produce retinal organoids using the Sasai method, the efficiency varied from 4% (4/96 for SJRB-iPSC-12) to 27% (26/96 for SJRB-iPSC-7) across lines (Fig. 2A). In order to improve the retinal specification, we carried out several rounds of optimization and made six changes to the procedure. In our modified 3D-RET protocol (Fig. 2A), we added the BMP signaling inhibitor (dorsomorphin) and ROCK inhibitor to promote the neuroectodermal lineage24,25,26. We increased Matrigel to 2% to increase eye-field specification23. We change the medium every 2 days rather than every 5 days during the early stages of the eye-field specification to prevent depletion of growth factors. We moved the addition of smoothened agonist (SAG) earlier to day 12 based on a previous publication27. We replaced retinoic acid with the more stable synthetic agonist EC23 (Stemgent Inc). Our modified protocol called 3D-RET achieved 22–30% retinal organoid formation across all of our iPSC lines, and the individual organoids from each line were more homogenous (Fig. 2B, C and Supplementary Fig. 2). Using the H9 ESC line and the iPSC lines produced from participants with germline RB1 alterations, we showed that molecular, cellular, functional, and anatomic features of retinal organoids produced by the Sasai method and the 3D-RET method were indistinguishable in side-by-side comparisons (Fig. 2D and Supplementary Fig. 2). While the quality of retinal tissue was similar in each method, the increased efficiency of retinal organoid formation for iPSC lines and the more uniform and consistent size and shape of the organoids, led us to use the 3D-RET protocol for all subsequent experiments.

Fig. 2: The 3D-RET protocol for retinal organoid formation.

A Drawing of the steps in the Sasai and 3D-RET retinal organoid protocol. The number of retinal organoids produced from a 96-well dish is indicated for representative lines for each protocol. Red arrows indicate media changes. B, C Micrograph of representative retinal organoids using the Sasai and 3D-RET protocols in a side-by-side comparison for SJRB-iPSC-4. All organoids from a 96-well dish were analyzed. Arrows indicate retina organoids and (*) indicates cystic structures that are common in the Sasai protocol. D Micrographs of dissociated cell immunofluorescence of retinal organoids and a dot plot showing the percentage of recoverin immunopositive cells from individual retinal organoids (n = 8). MG Matrigel, SAG smoothened agonist, RMM retinal maturation medium, FBS fetal bovine serum, CHIR GSK3 inhibitor, ec23 retinoic acid analog. Scale bars: B, C, 100 μm. D, 5 μm.

Retinoblastoma from human retinal organoids

Four of the 15 participants in our cohort had surgical enucleation as part of their treatment. We were able to produce O-PDXs from two of those tumors (SJRB-158 and SJRB-124) (Supplementary Data 1). These four patient samples and two O-PDXs will complement our large collection of reference retinoblastoma samples ( To determine if retinoblastoma can form from 3D retinal organoid cultures of the patient-derived stem cells, we grew retinal organoids from iPSCs from each patient to day 45 when the tissue was dissociated and injected into the vitreous of immunocompromised mice (Fig. 3A). As a negative control, we used RB1 wild-type stem cell (H9 ESCs and GM23710 iPSCs)-derived retinal organoids and as a positive control, we induced RB1 mutations in exon 4 with CRISPR-Cas9 in all 15 participant derived iPSC lines and H9 cells, hereafter referred to as SJRB-iPSC-1CR - 15CR (Fig. 3B). The CRISPR-Cas9 RB1 inactivation was intentionally left mosaic (<10% of cells) in the starting stem cell populations to more closely mimic the clonal heterogeneity in human retinoblastoma (Fig. 3C and Supplemental Information). We reverted the germline RB1 mutation in two of the lines (SJRB-iPSC-4-REV and SJRB-iPSC-6-REV) using CRISPR-Cas9 and performed the same tumor formation experiments in parallel with those lines (Supplementary Data 4 and Supplemental Information). We also injected undifferentiated H9 ESCs to form teratomas in the eye as an additional negative control. Each iPSC line and controls were tested with and without the RB1 CRISPR-Cas9 by two different technicians with three replicate injections per line for a total of more than 500 individual eyes injected (Supplemental Information). After 1 year, we identified 13 independent tumors from iPSC lines SJRB-iPSC-1,2,4,5,6,8 with CRISPR-Cas9 inactivation of RB1 and 7 independent tumors from SJRB-iPSC-3,8,15 lines without CRISPR-Cas9 inactivation of RB1 (Fig. 3D and Supplementary Data 4). For H9 ESC retinal organoids, four independent tumors were formed with CRISPR-Cas9 inactivation of RB1 and none was formed without RB1 inactivation (Supplementary Data 4). The two lines that had the germline RB1 mutation reverted did not form tumors after more than a year. We developed a custom Taqman qRT-PCR microfluidic card to rapidly distinguish between teratomas, retinoblastoma, and other malignancies (Fig. 3E and Supplemental Information). None of the tumors that formed from intravitreal injections of day 45 retinal organoids were teratomas based on qRT-PCR (Supplementary Data 4). Two tumors (SJ-iPSC-15-T-A/B) that arose rapidly (37 and 100 days) had a pan-neuronal gene expression pattern but were not retinoblastoma based on gene expression (Supplementary Data 4). All other tumors from the patient-derived iPSC retinal organoids (iPSC-RBs) were retinoblastomas that were indistinguishable from patient retinoblastomas or orthotopic patient-derived xenografts (O-PDXs) (Fig. 3E and Supplementary Data 4).

Fig. 3: Retinoblastoma from 3D retinal organoids.

A Schematic drawing of the retinoblastoma workflow. After 45 days of differentiation, retinal organoids are dissociated and injected into the eyes of immunocompromised mice and they are held for 1 year to wait for tumor formation. B Drawing of the RB1 genomic locus with the location of the gRNA targeting exon 4 and the corresponding sequence. C Plot of mutation frequency in a representative iPSC line for gRNA-3 with insertions (red) and deletion (blue) flanking the cut site (*). D Photograph of a mouse with retinoblastoma from a retinal organoid. E Barplot of qRT-PCR for genes found in the retinoblastoma (SYK, SIX3, HMX1), human pluripotent stem cells (MYC, SOX2, FGFR2), and teratomas (WLS, BMP2, COLA1) from one teratoma, one patient tumor (RB-169), two PDX tumors (RB116 and RB121), two spontaneous retinal organoids-derived tumors (SJRB-iPSC-4 and SJRB-iPSC-8), and two CRISPR-modified retinal organoids (SJRB-iPSC-4CR, SJRB-iPSC-8CR). Each dot is the mean of technical duplicates, the bar is the mean and standard deviation between replicates. All data are normalized to GAPDH and plotted relative to H9 ESCs (dashed line). F, GCircos plot of representative organoid-derived retinoblastoma (SJRB-iPSC-4CR-T-E and SJRB-H9CR-T-C) showing somatic mutations acquired in the tumor relative to the iPSC/ESC line. The copy number changes across the genome are shown below each tumor. H Principal component analysis (PCA) of RNA-seq of organoid-derived retinoblastomas, O-PDXs, patient tumors, iPSC/ESCs, and retinal organoids. I, J Hematoxylin and eosin (H&E)-stained organoid-derived tumor showing rosettes and IHC for SYK (brown) which is not present in the normal retina but is upregulated in retinoblastoma. Staining was completed on three tumors with similar results. Scale bars: 25 μm.

Tumors were propagated and cryopreserved as done previously6,21,28. Next, we performed whole-genome sequencing and RNA-seq of each independent iPSC-RB when sufficient tissue was available after initial propagation and screening. The patient-derived iPSC-RBs showed inactivation of the 2nd allele of the RB1 gene and no other mutations in known cancer genes (Fig. 3F, G, Supplementary Data 4, and Supplementary Fig. 3). Importantly, there were also copy number gains in MDM4 and MYCN which are common in retinoblastomas (Fig. 3F, G, Supplementary Data 4, and Supplementary Fig. 3). In addition, the RNA-seq from iPSC-RBs most closely matched retinoblastomas in principal component analysis (Fig. 3H and Supplementary Data 4). The difference in O-PDX and organoid-derived retinoblastomas are primarily due to differences in the tumor microenvironment. Vascular endothelial cells, macrophages, and other normal human cells present in the patient tumors are murine in O-PDX and organoid tumors and therefore filtered out in RNA-sequencing analysis. One of the hallmarks of retinoblastoma is the epigenetic deregulation of the SYK oncogene which is required for tumorigenesis6. SYK RNA and protein were upregulated in iPSC-RBs as found in patient tumors (Fig. 3E, I, J and Supplementary Data 4). Taken together, these data suggest that spontaneous retinoblastoma can form from patient-derived iPSCs differentiated into retinal organoids or be induced by CRISPR-Cas9 targeting of the RB1 locus in both hESCs and patient-derived iPSCs.

iPSC-RBs recapitulate the epigenetic and clonal features of retinoblastoma

High-density DNA methylation arrays (Illumina Infinium 850 K) have been extensively used to classify tumors based on their genome-wide methylation signatures and copy number variations29,30,31. The assay is robust even with small amounts of formalin-fixed paraffin-embedded (FFPE) material. This has been particularly useful for pediatric tumors of the central nervous system29. To establish a baseline for retinoblastoma, we profiled 53 retinoblastoma patient tumors, including tumors with histopathological features of differentiation (rosette formation) as well as less differentiated tumors (Fig. 4A, B). Unsupervised hierarchical clustering of the patient tumors revealed that they separate based on differentiation evaluated from the histopathology and DNA copy number variation from the Infinium 850 K array (Fig. 4C). Indeed, the more differentiated tumors had fewer copy number alterations than those with more aggressive undifferentiated histopathologic features and some patient tumors were heterogeneous between regions of differentiated and undifferentiated tumor cells (Fig. 4D–F).

Fig. 4: DNA methylation profiling of retinoblastoma.

A, B Hematoxylin and eosin (H&E) staining of differentiated and undifferentiated patient retinoblastoma. C Unsupervised hierarchical clustering of retinoblastoma tumors showing separation of the differentiated and undifferentiated samples. DF Copy number variation in the differentiated retinoblastomas (D) and undifferentiated patient retinoblastomas (F) and H&E staining (E) of a patient eye that has regions of both undifferentiated (region 1) and differentiated (region 2) tumor, suggesting intratumor cellular heterogeneity. G Copy number variation and clustering of the retinal organoid-derived tumors. H t-distributed stochastic neighbor embedding (tSNE) plot of the St. Jude retinoblastomas, retinal organoid-derived tumors, retinal organoids, and iPSC/ESCs integrated with the Molecular Neuropathology (MPN) database of brain regions and pediatric brain tumors. I tSNE plot of the boxed region in (H) showing clustering of the organoid-derived tumors with patient retinoblastomas. J tSNE plot of the patient, O-PDX, and organoid-derived retinoblastomas with iPSC/ESCs and normal retinal organoids. K tSNE plot of O-PDX and organoid-derived retinoblastomas with iPSC/ESCs and normal retinal organoids. Scale bars: 25 μm.

Next, we performed the same DNA methylation array profiling on our collection of iPSC/ESC-derived retinal organoids, retinoblastoma O-PDXs, and organoid-derived retinoblastomas described above. Four of our organoid-derived tumors had sufficient DNA for methylation array profiling and passed our quality control metrics (Supplemental Information). The organoid-derived retinoblastomas clustered more closely with the undifferentiated patient tumors using the same unsupervised methods (Fig. 4G and Supplemental Information). Based on copy number variants, the organoid-derived tumors were intermediate between the differentiated and undifferentiated patient retinoblastomas (Fig. 4G). In unsupervised methylation analysis, the O-PDXs and iPSC-RBs overlapped with retinoblastomas/pineal tumors from a reference collection of 2901 brain tumor methylation profiles29 and they were clearly separated from the normal retinal organoids and iPSCs in tSNE plots (Fig. 4H–K and Supplementary Data 5).

To further validate the identity and heterogeneity of the retinoblastomas derived from iPSCs, we performed single-cell RNA sequencing on 11 retinoblastoma O-PDXs in biological duplicate for a total of (114,167 cells), 5 healthy adult human retina (24,445 cells), 6 patient retinoblastomas (27,825 cells), and 2 organoid-derived tumors (13,864 cells). We also included human retinal progenitor cells from a publicly available scRNA-seq dataset on human fetal retina32. We combined the five healthy adult retina and proliferating fetal retinal progenitor cells and created a reference dataset for all retinal cell types (progenitors, rods, cones, ganglion, horizontal, amacrine, bipolar cells, and Müller glia) as well as non-retinal cell types (vascular endothelial cells and immune cells) using Seurat (v3)33 (Fig. 5A). In the retinoblastoma samples, the tumor cells were distinguished from normal cells based on inferred copy number alterations (Supplementary Fig. 4A and Supplemental Information). For each tumor sample, after pairs of cell correspondences between the reference and tumor dataset (anchors) were identified, the cell-type classification was projected and transferred onto each tumor dataset to determine if there were cells with expression profiles similar to specific retinal cell types in the tumors (Fig. 5B). Importantly, the cell cycle genes were removed before the label transfer to prevent bias toward proliferating retinal progenitor cells. Despite excluding the cell cycle genes, the most common cell identity from the label transfer was retinal progenitor cells at 52% (81,283/156,244) followed by rod photoreceptors at 31% (48,591/156,244) (Supplementary Data 6). Indeed, for every patient tumor, O-PDX, and organoid-derived tumor, the most common cell identity was retinal progenitor cells (range 39–63%) and they were enriched in cell cycle genes even though the cell cycle genes themselves were not used for the label transfer (Fig. 5C, D and Supplementary Data 6). To determine if there was evidence of tumor cells with progenitor signatures giving rise to more differentiated tumor cells with photoreceptor gene expression signatures, we performed RNA velocity analysis (Supplemental Information). While some tumors showed evidence of the transition from progenitors to photoreceptors, others showed the opposite pattern (Supplementary Fig. 4). A more definitive clonal analysis will be required to determine the relationship between the cell populations in retinoblastoma.

Fig. 5: Cellular heterogeneity of retinoblastomas.

A Uniform manifold approximation projection (UMAP) plot of scRNA-seq of normal human retinal cells including retinal progenitor cells and all adult retinal cell types. B UMAP plot of scRNA-seq of human retinoblastoma from patient tumors, O-PDXs, and organoid-derived tumors. The label transfer for cell identity is displayed, and the overall numbers are represented in the piechart in the upper left corner. C Representative UMAP plot of one of the organoid-derived tumors showing cells with the progenitor cell signature and rod signature and the relative distribution of gene expression profile for proliferation. D Barplot of the proportion of single cells with the retinal progenitor cell gene expression signature and the rod signature for each tumor (upper panel) and the proportion of those cells that are proliferating (S/G2/M, lower panel) for each tumor analyzed by single-cell RNA sequencing (n = 19). E, F UMAP plot of gene expression for normal retina and retinoblastoma for a representative retinal progenitor cell gene (HES6) and a photoreceptor gene (AIPL1). G, H, IViolin plot of the distribution of expression of a progenitor gene (HES6), a photoreceptor gene (AIPL1) and a cone gene (PDE6H) across retinal cell types and retinoblastomas. The colors match the colors in (A, B).

Previous retinoblastoma single-cell gene expression array analysis of a single O-PDX suggested that tumor cells may have a hybrid gene expression signature of multiple cell types14. Consistent with those data, we found that genes that are normally expressed in a mutually exclusive pattern in the normal retina such as HES6 and AIPL1 are co-expressed in retinoblastoma tumor cells (Fig. 5E–H). This was also true for cone, rod, and amacrine genes (Fig. 5I and Supplementary Fig. 4E, F). Therefore, individual retinoblastoma tumor cells express a hybrid gene expression signature that does not normally occur during retinal development. Taken together, our scRNA-seq analysis showed that our organoid-derived tumors are indistinguishable from the O-PDXs and the patient tumors in terms of cell identity and proliferation (Fig. 5G and Supplementary Data 6) and all data are available in a Cloud-based viewer (


We have developed iPSCs from 15 participants with germline RB1 alterations and we have optimized a 3D retinal organoid culture system for producing human retinoblastoma in the laboratory. We also developed the tools to induce RB1 mutations in a wild-type human stem cell and produce retinoblastomas indistinguishable from those of patient-derived iPSCs. The organoid-derived retinoblastomas have molecular, cellular, histopathologic, genetic, epigenetic, and clonal features that are indistinguishable from patient tumors and O-PDX models. In contrast with O-PDXs, this model is not reliant on patient tumor tissue, which in the case of retinoblastoma is usually only available after enucleation. In addition, our model can be used to derive multiple tumors from the same patient and can be used to generate tumors from carriers who never developed retinoblastoma. The process of producing retinoblastoma in our system was inefficient (<5%) and time-consuming (12–18 months per tumor for engraftment and propagation). Still, this is only slightly longer than time to engraftment of a retinoblastoma O-PDX. Subsequent rounds of differentiation and injection after screening organoids for high percentages of retinal tissue have greatly increased efficiency, including the formation of at least one tumor from all CRISPR-mutated SJRB-iPSC lines that were selected for injection. The introduction of additional perturbations such as ectopic expression of MDM2/4, MYCN, or SYK may accelerate tumorigenesis. Previous attempts to generate retinoblastoma from genetically modified H9 ESCs may have failed because they did not allow enough time for tumors to grow (60–90 days) or the lack of normal retinal development in the absence of RB134. Our approach using a mosaic approach with the same H9 ESC produced multiple independent tumors in 200–300 days. None of our tumors was teratomas, indicating that retinal organoid differentiation prior to intravitreal injection was sufficient to eliminate any residual stem cells. However, we did identify two tumors using this method that had neuronal features but were not retinoblastomas. Therefore, it is essential to implement robust diagnostics for retinoblastoma from retinal organoids that include molecular, cellular, genetic, and epigenetic features. This study also provided new insights into the cellular identity within retinoblastoma demonstrating a bias toward retinal progenitor cells and rods. It is possible that the highly proliferative tumor cells with the retinal progenitor cell identity give rise to more differentiated tumor cells that have features rods and other neurons but lineage tracing will be required to test that hypothesis. This complex process of the tumor cells progressing through some aspects of retinogenesis may have confounded previous attempts to identify the retinoblastoma cell of origin from gene expression analysis of bulk tumors. Indeed, scRNA-seq shows a hybrid cellular phenotype, which may simply reflect the multipotency of retinal progenitor cells. Our retinal organoid system will be an important tool for determining if those tumor cell populations are distinct clones or if they represent the dynamic changes in gene expression of individual clones over time. The tumor modeling described here may also be useful for testing novel therapeutic combinations for individual patients.

Article classification: Biological abstract
Share to:
Add:Room A11-329, 1st Floor, No.1, SBI Venture Street, Optics Valley, East Lake
New Technology Development Zone, Wuhan, China.
Certificate NO.:U18Q28010569R0S