Retinal organoid screening reveals ABT-737 and luminespib as potential agents against a cone- precursor-derived subtype of retinoblastoma

TL;DR

Researchers found that ABT-737 and luminespib may effectively target a specific type of retinoblastoma that starts in cone precursors, using retinal organoids and RPE models for drug testing.

Contribution

The study introduces a novel preclinical platform using retinal organoids and RPE models to screen drugs for a specific retinoblastoma subtype.

Findings

ABT-737 and luminespib showed strong cytotoxicity against proliferating cone precursors in retinal organoids.

Luminespib had moderate permeability across the outer blood-retinal barrier due to its lower molecular weight.

ABT-737 was more efficiently taken up by retinal organoids compared to luminespib.

Abstract

Retinoblastoma (Rb) features proliferating cone precursors, particularly those in the G2/M phase, as a major driver population revealed by recent single-cell transcriptomic studies, though other molecular subtypes with distinct cellular origins also contribute to tumor heterogeneity. In the current study, we utilized patient RB1−/− retinal organoids, which model a cone-precursor-derived subtype of Rb, to evaluate the cytotoxic efficacy and selectivity of candidate therapeutics targeting these tumor-initiating cells. A primary screen of 37 compounds identified 11 with significant cytotoxicity, which was subsequently refined to six candidates exhibiting activity against proliferating cone precursors. Among these, ABT-737 and luminespib emerged as lead compounds, demonstrating dose-dependent depletion of Ki-67+/RXRγ+ cone cells and strong apoptotic induction, evidenced by caspase-3 activation and Annexin V/7-AAD flow cytometry, without inducing necrosis. Single-cell RNA sequencing confirmed that both agents targeted the Rb-like proliferating cone precursors. Ocular pharmacokinetic modeling using pluripotent-stem-cell-derived retinal pigment epithelium (RPE) monolayers revealed limited permeability across the outer blood-retinal barrier for both drugs; however, luminespib showed moderate translocation, likely due to its lower molecular weight. In contrast, the uptake of ABT-737 by the retinal organoids was more efficient than that of luminespib. Together, these findings highlight ABT-737 and luminespib as potential therapeutic candidates against a cone-precursor subtype of Rb and demonstrate the utility of integrated retinal organoid and RPE models for preclinical drug screening and pharmacokinetic evaluation. However, validation in additional retinoblastoma subtypes beyond cone-precursor-derived tumors is essential to determine broader therapeutic applicability and efficacy. Future studies should prioritize testing these agents across diverse Rb genomic and phenotypic subtypes to address potential heterogeneity in treatment response.

Using patient-derived retinoblastoma retinal organoids, researchers identified ABT-737 and luminespib as potential therapeutics targeting tumor-initiating proliferating cone precursors in retinoblastoma. Integrated organoid and RPE models demonstrated high cytotoxic selectivity and drug permeability, providing a robust preclinical platform for evaluating targeted therapies against specific retinoblastoma molecular subtypes.

Figures

Cytotoxic screening in day 150 RB1^−/−^ (Rb) and RB1^+/+^ (control) organoids following 72-h drug exposure(A) Schematic representation of CRISPR-mediated gene editing to generate isogenic RB1^+/+^ and RB1^−/−^ iPSCs used to form Rb organoids and the methodology used in this study. (B) Primary drug screening results in RB1^−/−^ organoids. Statistical analysis was performed using one-way ANOVA. (C) Secondary drug screening comparing responses in RB1^+/+^ and RB1^−/−^ organoids across multiple drug concentrations. Two-way ANOVA with multiple comparisons was used to compare both genotypes and vehi

ABT-737 and luminespib reduce the proliferation and increase apoptosis of cone precursors in RB1^−/−^ organoids(A) Representative immunofluorescence images showing co-labeling of Ki67 and RXRγ used for quantification of proliferating cone precursors (RXRγ^+^Ki67^+^). (B) Representative immunofluorescence images showing co-labeling of caspase-3 and RXRγ used for quantification of apoptotic cone precursors (RXRγ^+^caspase-3^*+*^). (C) Quantification of proliferating cone precursors following 72-h drug treatment. Topotecan and sunitinib were included as positive controls. (D). Quantification of a

ABT-737 and luminespib selectively induce apoptosis in RB1^−/−^ organoidsOrganoids were treated for 72 h with each of the six candidate drugs (7.5 μM), vehicle control (0.1% DMSO), or positive controls (topotecan and sunitinib, both at 10 μM), then stained with PE-Annexin V and 7-AAD and analyzed by flow cytometry. (A) Representative density plots showing live (Annexin V^−^/7-AAD^−^), early apoptotic (Annexin V^+^/7-AAD^−^), late apoptotic (Annexin V^+^/7-AAD^+^), and necrotic (Annexin V^−^/7-AAD^+^) cell populations under selected treatment conditions (ABT-737, luminespib, topotecan, and vehi

Differential dose-dependent efficacy of ABT-737 and luminespib in reducing proliferating cone precursor populations in RB1^−/−^ retinal organoids(A) Representative immunofluorescence images showing RXRγ^+^Ki67^+^ cells following 72-h exposure to varying concentrations of ABT-737. (B) Representative immunofluorescence images showing RXRγ^+^Ki67^+^ cells following 72-h exposure to varying concentrations of luminespib. (C) Quantitative analysis of RXRγ^+^Ki67^+^ cells after ABT-737 treatment. (D) Quantitative analysis of RXRγ^+^Ki67^+^ cells after luminespib treatment. (A and B) White arrows indi

Dose-dependent cytotoxic effects of ABT-737 and luminespib on cone precursors in RB1^−/−^ retinal organoids(A) Representative immunofluorescence images used for quantification of RXRγ^+^caspase-3^+^ cells following 72-h exposure to varying concentrations of ABT-737. (B) Representative immunofluorescence images used for quantification of RXRγ^+^caspase-3^+^ cells following 72-h exposure to varying concentrations of luminespib. (C) Quantitative immunofluorescence analysis of RXRγ^+^caspase-3^+^ cells after ABT-737 treatment. (D) Quantitative immunofluorescence analysis of RXRγ^+^caspase-3^+^ cel

Single-cell transcriptomic analysis reveals a reduction in proliferating cone precursor cells in RB1^−/−^ retinal organoids following treatment with 7.5 μM ABT-737 and luminespib(Associated with Table S2) (A) UMAP visualization of integrated scRNA-seq data from all treatment groups, revealing 15 transcriptionally distinct cell clusters. (B) Feature plots showing expression of the cone precursor marker (*ARR3*), proliferation marker (*MKI67*), cone photoreceptor marker (*CNGB3*), and rod photoreceptor marker (*NRL*), highlighting co-expression of cone and proliferation markers in cluster 4. (C)

Drug efflux studies across the RPE barrier demonstrate differential permeation of ABT-737 and luminespib(A) Schematic representation of the drug flux experiment. ABT-737 and luminespib were added to either the apical or the basal side of the RPE monolayer cultured on Transwell inserts to assess directional permeability. (B) Quantitative analysis of drug concentrations over time in both apical and basal compartments, measured by LC-MS/MS. ABT-737 showed negligible translocation across the RPE barrier, whereas luminespib exhibited moderate permeability, particularly evident at the 360-min time p

Drug organoid permeability study demonstrates differential permeation of ABT-737 and luminespibQuantitative analysis of drug permeability measured both in medium and retinal organoids following 72-h exposure, measured by LC-MS/MS. ABT-737 appeared to show higher levels of absorption as compared to luminespib. Data are presented as mean ± SD; *n* = 6.

Full text

Introduction

Retinoblastoma (Rb) is a rare pediatric cancer of the developing retina, primarily caused by biallelic inactivation of the RB1 tumor suppressor gene located on chromosome 13q14 or by MYCN amplification in a susceptible retinal cell type.1 The incidence of Rb is approximately 1 in 15,000 live births. It accounts for 7%–17% of all tumors diagnosed during infancy and 4%–6% of all cancers in children under the age of 15 years.2^,^3^,^4 Globally, an estimated 8,600 to 9,000 children are affected each year, with around 40–50 new cases diagnosed annually in the United Kingdom alone.5 Rb typically presents with the presence of leukocoria, a white pupillary reflex with possible strabismus indicating foveal involvement. Proptosis is also commonly reported.6^,^7

Approximately 40% of Rb cases are bilateral and associated with germline mutations, while the remaining 60% are unilateral, of which roughly 15% have a germline as opposed to somatic origin. The majority of diagnoses occur in children under 5 years of age, with treatment strategies dependent on the number, location, and size of intraocular tumors. While excellent survival rates are reported in high-income countries, ranging from 85% to 97% for unilateral cases and 88%–100% for bilateral cases, treatment can lead to significant visual impairment and long-term quality of life challenges.5 These impacts are particularly pronounced in patients who require enucleation (surgical removal of the eye).

Although rare in high-income countries, metastatic Rb remains a major concern in low- and middle-income countries, where delayed diagnosis and inadequate healthcare infrastructure contribute to up to 50% of cases.8^,^9 Once the disease spreads beyond the eye to the optic nerve, extraocular tissues, or central nervous system, prognosis is poor even with aggressive treatment involving high-dose chemotherapy, stem cell transplantation, and radiotherapy. To reduce the global burden of Rb, it is essential to develop effective therapies that are not only safe and targeted but also affordable, easily stored, and accessible worldwide.

Treatment options for Rb include local therapies such as laser photocoagulation, cryotherapy, or plaque radiotherapy; chemotherapy administered systemically, intra-arterially, or intravitreally; external beam radiotherapy; and enucleation.10 For smaller tumors, local therapy alone may be sufficient, while more advanced cases often require a multimodal approach.11 Over the past decade, there has been a significant shift toward conservative treatments aimed at preserving both the eye (globe) and vision, while minimizing systemic toxicity. Intra-arterial and intravitreal chemotherapy have emerged as targeted delivery methods that reduce the complications associated with systemic chemotherapy, such as bone marrow suppression, alopecia, and nutritional compromise in the short term and nephrotoxicity and ototoxicity in the long term. However, these localized approaches are not without risks. They are associated with both ocular and systemic adverse effects, including vascular retinopathy, cystoid macular oedema, anterior segment toxicity, and orbital fat atrophy, which can limit their effectiveness as salvage therapies.12^,^13^,^14^,^15 One of the main reasons for the observed ocular toxicity is that current chemotherapy drugs are not specifically designed for the unique environment of the eye, particularly the retina. While these drugs can effectively kill cancerous cells, they do not target the specific pathways or molecular triggers responsible for the unchecked cone cell proliferation characteristic of Rb. This lack of specificity contributes to both suboptimal efficacy and significant side effects. Therefore, there is a clear and pressing need for novel therapeutics that are not only less toxic but also capable of effectively penetrating ocular lipid barriers and exhibiting a broad therapeutic window.

A two-step, biallelic inactivation of the RB1 tumor suppressor gene is required for tumor initiation in ∼98% of Rb patients.16^,^17^,^18^,^19^,^20^,^21 However, it has been suggested that biallelic RB1 inactivation alone leads to a non-proliferative retinoma, and progression to Rb requires additional genetic aberrations. Reported alterations in the RB1 gene are well characterized and heterogeneous, including single-nucleotide variants, deletions, rearrangements of the whole gene, as well as chromosomal deletions.16 Previous studies led to the identification of recurrent chromosomal copy-number alterations, including gains at 1q32, 2p24, 6p22, and losses at 16q22-24, most likely leading to activation of oncogenes or inactivation of tumor suppressor genes at these regions.17^,^19^,^20 More recent cytogenetic/molecular studies explicitly map these cytobands to candidate genes: gains at 1q32 (MDM4, KIF14), 2p24 (MYCN), 6p22 (DEK, E2F3), and loss at 16q22–24 (CDH11).22^,^23^,^24^,^25 Recent targeted next-generation sequencing studies have identified a high frequency (46%) of additional somatic and likely pathogenic alterations beyond RB1 biallelic inactivation, which correlate with aggressive histopathological features. These include focal high-level amplification of oncogenes such as MYCN and MDM4 and truncating mutations in tumor suppressor genes, including ARID1A, MGA, and BCOR.18

This growing understanding of the genetic complexity underlying Rb progression provides a crucial foundation for the development of targeted therapies. By identifying and characterizing secondary mutations that cooperate with RB1 loss, we can delineate tumor-specific pathways that drive malignancy. This insight enables the rational design of therapeutic strategies aimed at exploiting these molecular vulnerabilities. To advance this approach, we previously reported a comprehensive single-cell RNA sequencing (scRNA-seq) and assay for transposase-accessible chromatin using sequencing (scATAC-seq) on human Rb tumors.26 These analyses revealed key signaling pathways and upstream regulators that enabled the identification of 37 drug candidates as promising therapeutic targets.

In the current study, we utilized fully characterized patient-derived RB1^−/−^ organoids27 to screen 37 drug candidates tailored to the molecular context of Rb and conducted detailed molecular and pharmacokinetic investigations of two selected candidates. This approach demonstrates the power of leveraging tumor-specific pathways to identify effective, personalized therapies in a human-relevant in vitro model, significantly reducing the need for animal experimentation and accelerating the translation of findings toward clinical application.

Results

Drug cytotoxicity screening in RB1−/− and control retinal organoids

In our earlier work, scRNA-seq and scATAC-seq revealed that Rb tumors are predominantly composed of cone precursors at various stages of the cell cycle. Among these, we identified the G2/M-phase cone precursors as the likely cell of origin.26 Focusing on Rb-enriched cone subclusters, we further identified upstream regulators and signaling pathways that enable these proliferating cone precursors to evade cell-cycle arrest and apoptosis.26 From these targets, one or more candidate drugs were selected, resulting in a total of 37 compounds (Table S1). Of these, 10 have already demonstrated >70% cytotoxicity in Rb cell lines.28 Patient-specific (RB1^−/−^) induced pluripotent stem cells (iPSCs)27 were differentiated into retinal organoids as a model of cone-precursor-derived subtype of Rb and treated with 10 μM of each drug for 72 h at day 150 of differentiation to ensure sufficient maturation and diversity of all retinal cell types of light-responsive retinal organoids29(Figure 1A). Topotecan, a standard chemotherapeutic used in Rb treatment, served as a positive control. The first screening identified 11/37 candidate drugs that exhibited marked cytotoxicity in RB1^−/−^ organoids compared to the vehicle control (Figure 1B). Secondary screening in both RB1^−/−^ and control RB1^+/+^ retinal organoids narrowed the list to six candidate drugs that induced selective cytotoxicity in RB1^−/−^ organoids at one or more tested concentrations (Figure 1C). Notably, ABT-737 and luminespib remained significantly effective at the lowest dose tested (2.5 μM) (Figure 1C).Figure 1. Cytotoxic screening in day 150 RB1^−/−^ (Rb) and RB1^+/+^ (control) organoids following 72-h drug exposure(A) Schematic representation of CRISPR-mediated gene editing to generate isogenic RB1^+/+^ and RB1^−/−^ iPSCs used to form Rb organoids and the methodology used in this study. (B) Primary drug screening results in RB1^−/−^ organoids. Statistical analysis was performed using one-way ANOVA. (C) Secondary drug screening comparing responses in RB1^+/+^ and RB1^−/−^ organoids across multiple drug concentrations. Two-way ANOVA with multiple comparisons was used to compare both genotypes and vehicle-treated controls (0.1% DMSO). (B and C) Data are presented as mean ± SD; n = 8. ∗p ≤ 0.05, ∗∗p ≤ 0.01, ∗∗∗p ≤ 0.001, ∗∗∗∗p ≤ 0.0001.

ABT-737 and luminespib decrease the proliferating cone precursors in the RB1−/− organoids

Given the highly proliferative nature of cone cells that drive Rb tumor formation, we further investigated the impact of selected six compounds by performing immunostaining with the proliferation marker Ki-67 in combination with the cone cell marker RXRγ. To assess the effects of potential therapeutic compounds, we treated day 150 RB1^−/−^ and RB1^+/+^ organoids for 72 h with 7.5 μM of the six candidate drugs identified in our secondary screen. For comparison, we included topotecan and the recently identified anti-Rb agent sunitinib,30 alongside a vehicle control (Figures 2A and 2C). This approach allowed us to evaluate the impact of each treatment on cone cell proliferation within the organoid model.Figure 2ABT-737 and luminespib reduce the proliferation and increase apoptosis of cone precursors in RB1^−/−^ organoids(A) Representative immunofluorescence images showing co-labeling of Ki67 and RXRγ used for quantification of proliferating cone precursors (RXRγ^+^Ki67^+^). (B) Representative immunofluorescence images showing co-labeling of caspase-3 and RXRγ used for quantification of apoptotic cone precursors (RXRγ^+^caspase-3^+^). (C) Quantification of proliferating cone precursors following 72-h drug treatment. Topotecan and sunitinib were included as positive controls. (D). Quantification of apoptotic cone precursors following 72-h drug treatment. Topotecan and sunitinib were included as positive controls. (C and D) Data are presented as mean ± SD; n = 5 organoids per condition. Statistical analysis was performed using two-way ANOVA with multiple comparisons to assess differences across genotypes and treatments relative to vehicle-treated controls. ∗p ≤ 0.05, ∗∗∗∗p ≤ 0.0001.

Of the six candidate drugs tested, only ABT-737 and luminespib (and topotecan) significantly reduced the fraction of proliferating cone precursors in RB1^−/−^ organoids to levels comparable to those observed in RB1^+/+^ control organoids (Figure 2C). Napabucasin, SB743921, serdemetan, and sunitinib led to a significant reduction in proliferating cone precursors relative to vehicle control (dimethyl sulfoxide [DMSO]) RB1^−/−^ organoids (Figure 2C). However, the fraction of remaining proliferating cone precursors was significantly higher compared to healthy RB^+/+^ controls, demonstrating that they did not lead to complete eradication of these cells. Parthenolide showed no effect, with RB1^−/−^ organoids treated with this drug displaying similarly elevated fractions of proliferating cone precursors compared to vehicle-treated RB1^−/−^ organoids. These findings were further supported by immunostaining for the apoptotic marker caspase-3 and the cone cell marker RXRγ (Figures 2B and 2D), which revealed a significantly increased apoptosis of cone precursors in RB1^−/−^ organoids treated with ABT-737, luminespib, and topotecan compared to vehicle-treated (DMSO) RB1^−/−^ organoids. Napabucasin, parthenolide, serdemetan, and SB743921 produced significant cytotoxic effect compared to healthy RB^+/+^ controls. However, when compared to vehicle control (DMSO) RB^−/−^ organoids, this effect was demonstrated to be nonsignificant (Figure 2D).

To further investigate the observed increase in apoptosis following treatment, flow cytometry was performed on RB1^−/−^ organoids 72 h post-treatment using PE-Annexin and 7-AAD staining to detect both early- and late-stage apoptosis (Figure 3A). Treatment groups included the six drugs identified during the screening process (Figure 1C) and the reference compounds sunitinib and topotecan. Of the six candidate drugs, only ABT-737 and luminespib, along with the positive control topotecan, induced a statistically significant increase in apoptotic cells (Figure 3B). While these treatments showed a clear overall rise in apoptotic cell populations, they also demonstrated an increase in late apoptotic cells and a corresponding reduction in the live cell population (Figure 3B). Notably, PE-Annexin^-^/7-AAD^+^ cells, indicative of necrosis, were undetectable in RB1^−/−^ organoids treated with ABT-737 and luminespib and reference control topotecan. This absence of necrotic cells suggests that the observed effects were not due to nonspecific cytotoxicity but rather reflect a more targeted induction of apoptosis.Figure 3ABT-737 and luminespib selectively induce apoptosis in RB1^−/−^ organoidsOrganoids were treated for 72 h with each of the six candidate drugs (7.5 μM), vehicle control (0.1% DMSO), or positive controls (topotecan and sunitinib, both at 10 μM), then stained with PE-Annexin V and 7-AAD and analyzed by flow cytometry. (A) Representative density plots showing live (Annexin V^−^/7-AAD^−^), early apoptotic (Annexin V^+^/7-AAD^−^), late apoptotic (Annexin V^+^/7-AAD^+^), and necrotic (Annexin V^−^/7-AAD^+^) cell populations under selected treatment conditions (ABT-737, luminespib, topotecan, and vehicle). (B) Quantification of total apoptotic cells (early + late) for all six candidates and both positive controls. Bars represent mean ± SD from three independent experiments (n = 3). Statistical significance was assessed by one-way ANOVA with multiple comparisons versus vehicle control: ∗∗∗p ≤ 0.001, ∗∗∗∗p ≤ 0.0001. DMSO, dimethyl sulfoxide; SD, standard deviation.

Immunohistological and flow cytometric analyses indicate that both ABT-737 and luminespib exhibit promising efficacy in selectively eliminating proliferating cone precursor populations. Consequently, the subsequent phase of the study focused on evaluating the dose-response relationships of these compounds to further characterize their therapeutic potential. RB1^−/−^ organoids were treated for 72 h with serially decreasing concentrations of ABT-737 and luminespib (7.5, 5, 2.5, 1, 0.1, 0.01 μM) and subsequently assessed for the presence of proliferating cone precursor cells (Figure 4). The results indicate that ABT-737 significantly reduces the proliferative population only at the highest concentration tested (7.5 μM; Figures 4A and 4C). In contrast, luminespib demonstrates a broader effective range, with concentrations between 2.5 and 7.5 μM yielding a marked reduction in proliferating cone precursors relative to the vehicle control (Figures 4B and 4D).Figure 4. Differential dose-dependent efficacy of ABT-737 and luminespib in reducing proliferating cone precursor populations in RB1^−/−^ retinal organoids(A) Representative immunofluorescence images showing RXRγ^+^Ki67^+^ cells following 72-h exposure to varying concentrations of ABT-737. (B) Representative immunofluorescence images showing RXRγ^+^Ki67^+^ cells following 72-h exposure to varying concentrations of luminespib. (C) Quantitative analysis of RXRγ^+^Ki67^+^ cells after ABT-737 treatment. (D) Quantitative analysis of RXRγ^+^Ki67^+^ cells after luminespib treatment. (A and B) White arrows indicate cells co-expressing RXRγ and Ki67. (C and D) Statistical comparisons were performed using one-way ANOVA with multiple comparisons against vehicle control (0.1% DMSO). Data are presented as mean ± SD (n = 3–8). ∗p ≤ 0.05, ∗∗p ≤ 0.01.

To further investigate the dose-dependent cytotoxic effects of ABT-737 and luminespib on cone precursors, activated caspase-3 was used as a marker of apoptosis in combination with RXRγ to identify apoptotic cone photoreceptors (Figures 5A and 5B). Analysis of ABT-737-treated RB1^−/−^ organoids revealed a significant increase in RXRγ^+^caspase-3^+^ cells at higher concentrations (7.5, 5, and 2.5 μM) compared to vehicle-treated controls, while lower concentrations (≤1 μM) did not show a statistically significant difference (Figure 5C). Bearing in mind that proliferating cone precursors are only markedly reduced at the 7.5 μM treatment dose (Figure 4C), these findings together suggest that ABT-737 may also exert cytotoxic effects on other retinal cell types beyond the intended proliferating cone precursors. Analysis of luminespib-treated organoids revealed a dose-dependent increase in RXRγ^+^caspase-3^+^ cells within the effective concentration range that also reduces markedly the fraction of proliferating cone precursors (2.5–7.5 μM) (Figures 4D and 5D), suggesting that luminespib may exert a more targeted cytotoxic effect, preferentially affecting proliferating cone precursors only.Figure 5. Dose-dependent cytotoxic effects of ABT-737 and luminespib on cone precursors in RB1^−/−^ retinal organoids(A) Representative immunofluorescence images used for quantification of RXRγ^+^caspase-3^+^ cells following 72-h exposure to varying concentrations of ABT-737. (B) Representative immunofluorescence images used for quantification of RXRγ^+^caspase-3^+^ cells following 72-h exposure to varying concentrations of luminespib. (C) Quantitative immunofluorescence analysis of RXRγ^+^caspase-3^+^ cells after ABT-737 treatment. (D) Quantitative immunofluorescence analysis of RXRγ^+^caspase-3^+^ cells after luminespib treatment. (A and B) White arrows show cells with co-localized marker expression. (C and D) Statistical comparisons were performed using one-way ANOVA with multiple comparisons relative to vehicle control (0.1% DMSO). Data are presented as mean ± SD (n = 3–8). ∗p ≤ 0.05, ∗∗∗∗p ≤ 0.0001.

To assess the cellular composition of RB1^−/−^ organoids, samples treated with ABT-737 (7.5 μM), luminespib (7.5 μM), or vehicle control (0.1% DMSO) were subjected to scRNA-seq. After quality control and filtering, transcriptomes from 10,528 cells across all treatment groups were retained for downstream analysis. Data were integrated using the Seurat package to increase statistical power and enable joint clustering. Uniform manifold approximation and projection (UMAP) visualization identified 15 transcriptionally distinct cell clusters (Figure 6A). Differentially expressed marker genes defining each cluster are listed in Table S2. Cluster 4 was characterized by high expression of cone precursor markers (e.g., ARR3 and RXRG) and proliferation-associated genes (e.g., MKI67, HELLS, and CDCA7; Figure 6B) and was designated as the Rb-like cluster. Notably, treatment with ABT-737 or luminespib reduced the proportion of cells within this cluster compared to the DMSO vehicle control (Figure 6C), suggesting the depletion of proliferating cone precursor cells. Analysis of other clusters indicated a decrease in the amacrine and horizontal cells and Muller glia in the ABT-737- and luminespib-treated RB1^−/−^ organoids, respectively (Figure S1), but in both cases, the significance of these changes could not be assessed, as the scRNA-seq experiment was performed only once. Together, these data demonstrate that both ABT-737 and luminespib reduce the population of proliferating cone precursors within the RB1^−/−^ retinal organoids.Figure 6. Single-cell transcriptomic analysis reveals a reduction in proliferating cone precursor cells in RB1^−/−^ retinal organoids following treatment with 7.5 μM ABT-737 and luminespib(Associated with Table S2) (A) UMAP visualization of integrated scRNA-seq data from all treatment groups, revealing 15 transcriptionally distinct cell clusters. (B) Feature plots showing expression of the cone precursor marker (ARR3), proliferation marker (MKI67), cone photoreceptor marker (CNGB3), and rod photoreceptor marker (NRL), highlighting co-expression of cone and proliferation markers in cluster 4. (C) UMAP plots displaying the distribution of cluster 4 cells across treatment conditions, illustrating a reduction in this population in response to ABT-737 and luminespib treatment compared to vehicle control.

Ocular pharmacokinetics of ABT-737 and luminespib

Ocular pharmacokinetic studies examining drug absorption, distribution, metabolism, and clearance are a critical component of the drug prioritization process preceding preclinical evaluation. While absorption, distribution, and metabolism can be effectively studied using RB1^−/−^ retinal organoids alongside control organoids, assessing drug clearance requires consideration of the blood-retinal barrier (BRB), which plays a pivotal role in regulating drug movement. Specifically, the BRB facilitates drug elimination from the eye into systemic circulation following intravitreal administration and governs drug entry into the retina from systemic circulation during intravenous or intra-arterial delivery.

The BRB comprises two main components: the endothelial cells of retinal blood vessels and the retinal pigment epithelium (RPE). The RPE forms a tightly regulated, polarized monolayer situated between the retina and the fenestrated choroidal vasculature. These fenestrations allow for passive diffusion of drugs into the extravascular space of the choroid. The RPE mediates bidirectional drug transport outward from the vitreous to the choroid and inward from the choroid to the retina.

Published studies indicate that the choroid and Bruch’s membrane contribute minimally to the overall barrier function, highlighting the RPE as the primary determinant of permeability at the outer BRB.31 Human embryonic stem cell (hESC)-derived RPE models have demonstrated drug-restrictive properties comparable to those of ex vivo bovine RPE-choroid tissues. Consequently, the hESC-RPE model serves as a key platform for evaluating candidate drug permeability across the outer BRB.32

To evaluate the permeability of the RPE to ABT-737 and luminespib, iPSC-derived RPE cells (iCell RPE, Fujifilm Cellular Dynamics) were cultured on Transwell inserts, where they formed a tight transepithelial barrier, as indicated by TEER values of approximately 1,500 Ω⋅cm^2^. Drugs were administered to both the apical and basal compartments, and medium was collected from each side at various time points over a 6-h period. Drug concentrations were quantified using liquid LC-MS/MS analysis to assess directional permeability across the RPE monolayer. When the RPE monolayers were exposed to 7.5 μM ABT-737 on the apical side, the drug remained predominantly in the apical compartment, with negligible detection in the basal compartment (Figure 7B). A similar distribution was observed when ABT-737 was added to the basal compartment, indicating that the compound does not traverse the RPE monolayer under these conditions. Luminespib exhibited a similar permeability profile to ABT-737, with one notable exception at the final time point (360 min). Concentrations increased from 0.07 and 0.2 ng/mL at 0 min to 92.5 and 121.8 ng/mL at 360 min following apical and basal administration, respectively (Figure 7B). These results suggest that luminespib possesses a moderate ability to traverse the RPE barrier. This permeability may be influenced by its lower molecular weight (465.54 Da) compared to ABT-737 (813.43 Da), potentially facilitating limited but detectable translocation across the RPE monolayer.Figure 7. Drug efflux studies across the RPE barrier demonstrate differential permeation of ABT-737 and luminespib(A) Schematic representation of the drug flux experiment. ABT-737 and luminespib were added to either the apical or the basal side of the RPE monolayer cultured on Transwell inserts to assess directional permeability. (B) Quantitative analysis of drug concentrations over time in both apical and basal compartments, measured by LC-MS/MS. ABT-737 showed negligible translocation across the RPE barrier, whereas luminespib exhibited moderate permeability, particularly evident at the 360-min time point. Data are presented as mean ± SD; n = 2.

Drug stability was assessed in RB1^−/−^ retinal organoids treated with the two selected final compounds, namely ABT-737 and luminespib. Concentrations of the two drugs were quantified in both extracellular culture medium and intracellular concentrations in organoids. Analysis was carried out by LC/MS/MS, following appropriate extraction procedures. Samples were analyzed alongside standard curves across a linear range appropriate for the drug and experimental conditions. Following analysis, ABT-737 can be seen to permeate into the organoid from the medium at a higher rate than luminespib (Figure 8), with a percentage absorption from the medium of 41% ± 10.6% compared to luminespib at 9.8% ± 5.8%. However, both candidate drugs demonstrated a higher level of absorption as compared to topotecan, which demonstrated absorption of 6.48% ± 4.69%. This demonstrates that while ABT-737 has higher propensity to permeate into the retina as measured by retinal organoids, both ABT-737 and luminespib demonstrate increased permeative ability as compared to topotecan.Figure 8. Drug organoid permeability study demonstrates differential permeation of ABT-737 and luminespibQuantitative analysis of drug permeability measured both in medium and retinal organoids following 72-h exposure, measured by LC-MS/MS. ABT-737 appeared to show higher levels of absorption as compared to luminespib. Data are presented as mean ± SD; n = 6.

Discussion

This study employed Rb-genotyped retinal organoids (RB1^−/−^ vs. RB1^+/+^) as a model of cone-precursor-derived subtype of Rb, to identify potential therapeutics. A sequential screening strategy narrowed 37 initial candidates to 6, evaluated for selective cytotoxicity against proliferating cone precursors (RXRƴ+/Ki67+) using apoptosis (CASP3+, Annexin/7-AAD) and proliferation (Ki67+) markers. ABT-737 and luminespib emerged as lead candidates, demonstrating significant apoptosis induction and reduction of proliferating cones comparable to vehicle control. Immunofluorescent staining and scRNA-seq confirmed their efficacy against the target cell population, positioning both as potential candidates for clinical development. ABT-737 demonstrated broader cytotoxicity in RB1^−/−^ organoids compared to luminespib, suggesting potential activity against other cell types such as amacrine and horizontal cells, as suggested by the scRNA-seq analysis.

Rb arises from cone precursor cells that, like many cancer cells, exhibit rapid proliferation and often overexpress anti-apoptotic Bcl-2 family proteins, promoting cell survival and resistance to apoptosis.33 ABT-737, a BH3 mimetic small-molecule inhibitor, targets the anti-apoptotic proteins Bcl-2 and Bcl-xL by binding to their hydrophobic groove, thereby disrupting function and tipping the balance toward apoptosis.34^,^35^,^36 This shift in the Bax/Bcl-2 ratio has been associated with improved clinical outcomes in various cancers.37 In our study, ABT-737 significantly reduced the population of proliferating cone cells in iPSC-derived RB1^−/−^ organoids; however, its efficacy was limited in human Y79 Rb cells, which are resistant, likely due to the absence of Bax.38 This resistance highlights the potential benefit of combination therapies, as ABT-737 can act synergistically with agents including topotecan and cisplatin to enhance apoptosis even in resistant cell lines.39 Importantly, ABT-737 demonstrated no significant toxicity at clinically relevant doses, further supporting its suitability for combination regimens.40

Due to the hostile nature of the cancer cell microenvironment, tumor cells must frequently adapt to promote their survival. A key player in this adaptive response is heat shock protein 90 (Hsp90), which functions as a central molecular chaperone, orchestrating a wide range of downstream signaling pathways that support cell survival and proliferation. Notably, Hsp90 expression is significantly elevated in tumorigenic populations.41^,^42 Similar to the anti-apoptotic protein Bcl-2, Hsp90 presents a therapeutic target due to its overexpression, reported to be 2–10 times higher in cancer cells compared to normal cells, and its essential role in maintaining malignant cell viability.43 Luminespib, an Hsp90 inhibitor, disrupts this chaperone’s interaction with its client proteins, leading to apoptotic cell death primarily through activation of the caspase cascade.44 By exploiting cancer cells’ heightened dependence on Hsp90, this strategy selectively impairs tumor cell survival, offering potential therapeutic benefits in cancers such as Rb.

scRNA-seq is a powerful tool for dissecting tumor heterogeneity, making it particularly valuable in the study of Rb. By enabling the identification of distinct cellular subpopulations involved in tumorigenesis, scRNA-seq offers mechanistic insights not achievable with bulk transcriptomic approaches. Prior studies using retinal organoids have demonstrated the utility of this technique, revealing the presence of Rb-like population marked by a significant increase in MKI67^+^ cone photoreceptors.45 Importantly, scRNA-seq also allows for direct comparison between untreated and drug-treated conditions, providing insight into the mechanisms underlying therapeutic efficacy. The observed elimination of the proliferative cone population following treatment with ABT-737 and luminespib supports the reactivation of apoptotic pathways in cone-derived tumor cells, further validating BCL-2 and HSP90 as actionable targets in Rb.

Beyond validating known pathways, scRNA-seq can uncover additional therapeutic and diagnostic opportunities. While this study employed CRISPR/Cas9-mediated homozygous RB1 knockout in iPSCs from a single-patient-derived line, this approach does not capture the full clinical and molecular heterogeneity of retinoblastoma, representing a key limitation. Previous research has identified at least two distinct molecular subtypes of retinoblastoma: subtype 1, characterized by strong cone photoreceptor gene expression, and subtype 2, enriched for neuronal and ganglion cell markers.46 Notably, nearly one-third of genes differ between these subtypes, with subtype 2 exhibiting elevated expression of TFF1, a gene linked to enhanced tumorigenicity and increased metastatic potential.46^,^47^,^48 Expanding the diversity of genotypes tested would address a major limitation of this study, particularly by evaluating ABT-737 and luminespib against subtype 2 tumors, and additional mutations such as BCOR (found in 7%–13% of cases) or high copy-number RB1 alterations associated with later-onset disease.

While the data in this study suggest that ABT-737 and luminespib effectively reduce the proliferative cone population driving Rb, several considerations remain before clinical translation is feasible. As previously discussed, the presence of distinct Rb subtypes, including those associated with a higher risk of metastasis, underscores the need for broader studies involving diverse patient-derived samples to fully capture tumor heterogeneity.46 Expanding the dataset to include a wider range of Rb cases would improve the generalizability and clinical relevance of these findings. To further bridge the gap to clinical application, phase 0 trials could be conducted using tissue from patients who have undergone enucleation. These early-stage trials would help assess the selectivity of treatment by verifying that non-target cell populations are unaffected. However, a key challenge remains: the poor ability of ABT-737 to penetrate the RPE barrier, likely due to its high molecular weight (813.43 g/mol), which also suggests limited capacity to cross the blood-brain barrier (BBB).49^,^50 In contrast, luminespib, with a lower molecular weight (465.54 g/mol), may have a modest capacity to cross these barriers, although current evidence for its permeability across the RPE or BBB is limited. The inability of drugs to cross the BRB could be circumvented using intravitreal injection. This method has the advantage of being able to administer the drugs directly but does introduce potential increased risk. However, when analyzed for their ability to permeate into RB1^−/−^ organoids, ABT-737 demonstrated the highest internal detection, significantly higher than luminespib and topotecan, suggesting that while permeation across the RPE barrier is poor, ABT-737 could be readily absorbed through the retinal layers.

While the retinal organoid model used in this study offers a robust and reproducible 3D system to study Rb, further refinements could enhance its utility for drug testing. Incorporating vascularized organoids would allow investigation of drug pharmacokinetics and accumulation, building upon prior models that have demonstrated chemotherapeutic retention within vascular networks.51 Additionally, because nutrient and oxygen diffusion become insufficient beyond organoid sizes of 1–2 mm^3^, integrating vasculature is necessary to maintain viability and to better recapitulate the tumor microenvironment, which is known to be distinct and self-sustaining.52 To advance this even further, organ-on-a-chip technology could be employed to model both the RPE barrier and the organoid itself. Such a system would allow simultaneous assessment of drug permeability across the RPE and diffusion into retinal tissue. Although these models are technically complex and resource-intensive, early studies have demonstrated their value as microfluidic platforms with broad potential applications in drug screening and disease modeling.53 While reduction in reliance on animal models is an advantage of organoid studies, the use of xenograft models could help address some limitations, including the ability to measure additional cytotoxic effects.54

In summary, this study demonstrates that targeted inhibition of anti-apoptotic and chaperone pathways using ABT-737 and luminespib effectively reduces the proliferative cone precursor population driving Rb in a human retinal organoid model. Both agents induced significant apoptosis and loss of proliferating cone cells, with scRNA-seq confirming their specificity for the tumorigenic population. While the efficacy of ABT-737 may be limited by its inability to cross the RPE barrier and its broader cytotoxicity, luminespib shows greater translational promise due to its lower molecular weight and potential for improved tissue penetration. These findings highlight the importance of addressing tumor heterogeneity and drug delivery challenges in future studies and highlight the utility of advanced organoid and single-cell approaches for preclinical therapeutic evaluation. To aid the translation into clinical applications, further approaches such as xenografts could prove useful, being able to demonstrate the ability to cross BRB in vivo, selectivity in removing proliferating cone precursors, as well as long-term impact. Ultimately, ABT-737 and luminespib represent potential leads for the development of more effective, targeted therapies for Rb, warranting further investigation in diverse patient-derived models and early-phase clinical trials.

Materials and methods

Culture of pluripotent stem cells

RB1^−/−^ and RB1^+/+^ human induced pluripotent stem cells (hiPSCs)27 were cultured using mTeSR Plus medium (STEMCELL Technologies 100-0276) on Matrigel (Corning 354230)-coated 6-well tissue culture plates (TPPs). Cells were grown at 5% CO_2_, 20% O_2_ in a humidified incubator at 37°C with media changes every second day. Cells were passaged twice weekly at a ratio of 1:3 using Versene Solution, EDTA (15040033).

Differentiation of hiPSCs into retinal organoids

Retinal organoids were developed using a modified version of the previously described Hallam protocol.29 Briefly, cells were cultured to 90% confluency and then dissociated using Accutase (Thermo Fisher Scientific, Waltham, MA) to generate a single-cell suspension, which was centrifuged at 300 rcf for 4 min. Cells were resuspended in mTeSR Plus medium (STEMCELL Technologies 100-0276) with 10 μM ROCK inhibitor (Chemdea Y-27632), and 7,000 cells in 100 μL were seeded into Lipidure (NOF CM5206)-coated 96-well U-bottom culture plates (TPP 92406). After 2 days, 100 μL of RO differentiation medium was added, comprised of 45% IMDM (Gibco 21980-032), 45% HAM’s F12 (Gibco 21765-029), 10% KOSR (Gibco 10828-028), 1% GlutaMAX (Gibco 35050-038), 1% Chemically Defined Lipid Concentrate (Gibco 11905031), 1% Pen/Strep (Gibco 15140-122), and 450 μM 1-thioglycerol (Sigma, UK M6145), with 50% medium changes carried out every 2 days. On day 6, the medium was supplemented with 1.5 mM Recombinant Human BMP-4 (Gibco PHC9534). At day 18, maintenance medium was introduced consisting of DMEM/F12 (Gibco 31330-038), 10% FBS (Gibco A5256801), 1% GlutaMAX (Gibco 35050-038), 1% N2 (Gibco 17502048), 1% Pen/Strep (Gibco 15140-122), 0.1 mM Taurine (Merck T8691), 40 ng/mL T3 (Sigma, T6397), 0.25 μg/mL Amphotericin B (Gibco 15290-018), with fresh supplementation of 0.5 μM Retinoic Acid (Sigma, R2625) at every medium change up to day 120. Media changes were performed three times a week.

RPE differentiation

The commercial iCell Retinal Pigment Epithelium (RPE) Cells (Cellular Dynamics International R1101) were used to assess the ability of candidate drugs to permeate the RPE barrier. Cells were plated onto Matrigel-coated 24-well 0.4 μm ThinCert cell culture inserts (Greiner 662641) as per the manufacturer’s instructions. Medium composition was as follows: 91.3% MEM alpha (Thermo Fisher Scientific, #12571-063), 5% Fetal Bovine Serum, Value (Thermo Fisher Scientific, # A5256801), 1% N-2 Supplement (Thermo Fisher Scientific, #17502-048), 55 nM Hydrocortisone (Sigma, #H6909), 250 μg/mL Taurine (Sigma, #T0625), 14 pg/mL Triiodo-L-thyronine (T3) (Sigma, #T5516), and optionally 25 μg/mL Gentamicin (Thermo Fisher Scientific, #15750-060). Briefly, one vial of ≥5 × 10^6^ cells/vial was thawed by hand and used to resuspend cells into a total of 8 mL medium. The cell suspension was then centrifuged at 300 rcf for 5 min, and supernatant was then aspirated. Cells were resuspended in pre-warmed medium at a value of ∼0.5 × 10^6^ cells/mL. This cell suspension was then used to plate cells at the recommended value of 1.58×10^5^ cells/cm^2^ (≈54,000 cells/ThinCert 0.33cm^2^). Cells were then maintained for 28 days before use, with medium changes being conducted every 2 days.

RPE TEER measurements

Transepithelial electrical resistance (TEER) measurements were conducted using a Millicell ERS-2 Voltohmmeter (Millipore, MERS00002). RPE Transwells were removed from the incubator and left to reach room temperature for around 15 min prior to measurements being taken. The electrode was sterilized in 70% ethanol for 15 min, then rinsed with sterile PBS. The electrode was inserted into the well at a 90° angle, ensuring the shorter tip is inserted into the Transwell and the longer tip into the well. The unit area resistance (Ω·cm^2^) was calculated by subtracting the resistance reading from the blank well from the sample then multiplying by the effective membrane area of the Transwell insert (0.33 cm^2^).

Immunohistochemistry

Immunohistochemistry was performed following a previously established protocol.27 Briefly, organoids were washed with PBS and fixed in 2% paraformaldehyde for 15 min. After three PBS washes, they were incubated in 30% sucrose at 4°C overnight, then embedded in OCT compound using pre-filled molds.

Cryosections of 10 μm thickness were prepared using a Leica cryostat (CM1850). Sections were blocked and permeabilized for 1 h in a solution containing 0.3% Triton X-100 (Sigma) and 10% normal goat serum (NGS; Thermo Fisher Scientific) in PBS. Primary antibodies (Table S2) were diluted in 0.1% Triton X-100 and 1% NGS in PBS and incubated with the sections overnight at 4°C.

Following incubation, sections were washed three times with PBS and then incubated with secondary antibodies (Table S2) diluted in 50% antibody diluent (composed of 4.5 mL PBS, 500 μL NGS, 15 μL Triton X-100, and PBS to 50%) for 1 h at room temperature. After final PBS washes, sections were mounted using VectaShield mounting medium with Hoechst (1:1,000; Thermo Fisher Scientific H3570).

Fluorescent images of 10 μm sections were acquired using a Zeiss Axio Imager Z2 microscope equipped with an Apotome 2 (Zeiss, Germany). Quantitative analysis was performed using ZEN (Blue Edition; Zeiss) and MATLAB (MathWorks). A minimum of 5–6 organoids were imaged per experiment.

Organoid drug exposure

Stock solutions (1,000× exposure concentration in DMSO) of drugs were prepared prior to exposure and maintained at −20*°C.* A 1:500 dilution of drugs was prepared to account for a 50% media change, resulting in a final dilution of 1:1,000. Exposed organoids were returned to the incubator where they remained for 72 h until further analysis was conducted.

Retinal organoid cytotoxicity screening

Briefly organoids were exposed to candidate drug concentrations (Table S1) over a 72-h period. During this time, the medium contained IncuCyte Cytotox Green dye (Sartorius 4633) at a 1:8,000 dilution or NIR (Sartorius 4846) at a 1:4,000 dilution. Images were captured every 6 h using a 4× objective with 300 ms green and 400 ms NIR exposure on the IncuCyte SX5 Live-Cell Analysis System (Sartorius). Changes in fluorescent intensity over the time course were determined by calculating the mean green intensity/μm using the IncuCyte software.

Organoid dissociation

Papain activation

Organoids were dissociated using Papain (Stem Cell Technologies 7466) reconstituted in a filter-sterilized activation solution consisting of 1.1 mM EDTA (Invitrogen 15575020), 0.067 mM mercaptoethanol (Gibco 21985-023), and 5.5 mM L-cysteine HCl (Thermo Fisher Scientific, L06328.14) to a stock concentration of 250 units/mL, then activated at 37°C for 30 min.

Dissociation

Immediately before use, the dissociation solution was prepared by diluting papain stock solution to a final concentration of 30 units/mL, and DNase I (STEMCELL Technologies, #07900) to 125 units/mL, in HBSS (STEMCELL Technologies, #37150). Concurrently, the inhibitor solution was prepared using 10 mg/mL trypsin inhibitor from chicken egg white (Sigma, T2011), also diluted in HBSS. Up to 10 organoids were harvested and washed once with HBSS. Following the wash, organoids were resuspended in 500 μL of dissociation solution and incubated at 37°C on an orbital shaker set to 90 rpm or higher for 30 min. After incubation, each sample was gently triturated by manual pipetting 8–10 times, taking care to avoid bubble formation. The samples were then returned to the orbital shaker for an additional 30 min. After a total of 60 min, the samples were visually inspected to confirm complete dissociation. The dissociated organoid suspensions were passed through a 37 μm reversible cell strainer (STEMCELL Technologies, #27215) directly into the inhibitor solution (10 mg/mL trypsin inhibitor in HBSS).

Flow cytometry

Apoptosis assay

Apoptosis was assessed using the BD Pharmingen PE Annexin V – 7AAD Apoptosis Detection Kit I (BD Biosciences, catalog number S59763) according to the manufacturer’s instructions with minor modifications. Briefly, organoids were dissociated using papain as previously described, cells were then washed twice with ice-cold DPBS and centrifuged between washes at 300 rcf for 4 min at 4°C. The resulting cell pellet was resuspended in 1× Annexin V Binding Buffer (prepared from 10× stock solution with deionized water), and cell counts were conducted using a hemocytometer. A minimum of 1 × 10^5^ cells were transferred to 15 mL flow cytometry tubes. Cells were then incubated with 5 μL of PE Annexin V and 5 μL of 7-Aminoactinomycin D (7-AAD) for 15 min at room temperature (approximately 22°C–25°C) ensuring protection from light. Samples were analyzed using a BD FACSAria III flow cytometer, acquiring at least 10,000 events per sample. Live cells were identified by gating on FSC-A/SSC-A, followed by exclusion of doublets using FSC-A/FSC-H. Apoptotic cell populations were identified based on Annexin V and 7-AAD staining profiles. At least 10,000 events were recorded from each sample. Data analysis was performed using FCS Express software (De Novo Software, v.7.12).

Cell-cycle assay

Following organoid dissociation, cells were fixed in 70% ethanol overnight, followed by two PBS washes. Samples were then incubated for 30 min in staining solution consisting of 200 μg/mL RNase A (Qiagen, 19101) and 20 μg/mL Propidium Iodide (Sigma Aldrich, P4864) in PBS. These samples were taken through flow cytometry within 1 h of staining with at least 10,000 events per determination and analyzed using FCS Express (De Novo Software).

Single-cell RNA sequencing

scRNA-seq was conducted using the Parse Evercode WT Mini V3 kit; all steps were conducted as per manufacturer’s instructions. Samples were split evenly for library preparation to ensure even representation in final solution. The two samples were run using a NovaSeq 6000 with SP 100 cycle v.1.5 kit. Run parameters were 64/8/8/58 cycles, and 5% PhiX control was spiked in. The sequencing data were aligned to GRCh38, and the reads quantified using Trailmakers (v.1.4.0). The filtered Trailmakers outputs were quality controlled to remove dead cells and debris prior to downstream analysis. The following filtering thresholds were applied: a minimum of 5,000 counts per cell, a minimum of 3,000 detected genes per cell, and a maximum of 5% mitochondrial gene content. Each sample was processed and clustered individually using Seurat (v.5.3.0). Cells were normalized and scaled, and highly variable features were identified. Principal-component analysis (PCA) was performed for dimensionality reduction, and the top 20 principal components were used to construct a shared nearest neighbor (SNN) graph for clustering using the Louvain algorithm. Following individual clustering, samples were integrated using the Harmony (v.1.2.3) batch correction method to account for technical variation across datasets. To minimize confounding effects, gene counts and read counts were regressed out during the integration process. The integrated dataset was clustered using the same approach as for individual samples, and marker genes for each cluster were identified. Cell-type annotations were assigned based on the expression of known retinal marker genes.

Statistical analysis

All statistical analysis was performed using the GraphPad Prism Software (Dotmatics). Preliminary and secondary drug screening data to compare cytotoxic effect to vehicle control were analyzed using two-way ANOVA with Tukey’s multiple comparisons comparing both RB1^−/−^ & RB1^+/+^ vehicle controls against treatment groups and RB1^−/−^ & RB1^+/+^ against each other within treatment groups. Changes in Ki67/CASP3 for different doses of ABT-737 and luminespib application were assessed with one-way ANOVA with Dunnett’s multiple comparisons to compare the RB1^−/−^treated organoids to the vehicle (DMSO). To measure changes in the apoptotic cell population via flow cytometry, one-way ANOVA with multiple comparisons was used to compare the RB1^−/−^ treated organoids to the vehicle (DMSO).

LC/MS/MS analysis

Mass spectrometer settings

The QTRAP 5500 mass spectrometer (AB SCIEX, Macclesfield, UK) was operated in electrospray positive mode using the TurboIonSpray source. Quantitative ion transitions were optimized for each drug using a multiple reaction monitoring scan. The selected ion transitions were 813 > 389, 467 > 379, 307 > 238, and 423 > 377 for ABT-737, luminespib, fluconazole, and topotecan, respectively. Curtain gas was 30 psi, collision gas was medium, IonSpray voltage was 5,500 V, source temperature was 300°C, and ion source gases one and two were 15 psi.

Chromatography conditions

Samples were injected onto a ZORBAX Eclipse Plus C18 RRHD 1.8 μm, 2.1 × 50 mm column (Agilent Technologies, Cheadle, UK) using an ACQUITY Premier Ultra Performance Liquid Chromatography system (Waters Ltd, Wilmslow, UK). The aqueous mobile phase A was 0.1% formic acid (Thermo Fisher Scientific, Loughborough, UK) in Milli-Q water, and the organic mobile phase B was 0.1% formic acid in HPLC-grade acetonitrile (Thermo Fisher Scientific, Loughborough, UK). Gradient elution was performed over 6 min at a constant flow rate of 0.3 mL per minute. The gradient program consisted of 5%–100% B (0–2 min); 100% B (2–4 min); 100%–5% B (4–4.5 min); and 5% B (4.5–6 min).

Standard curves containing a combination of all drugs were prepared in relevant control matrix (RPE medium, RO medium, or 1% formic acid + 1% Triton X [Sigma Aldrich, Gillingham, UK] in water for organoid analysis) across the ranges of 10–6,000 ng/mL for ABT-737, 1–1,000 ng/mL for fluconazole, and 10–4,000 ng/mL for luminespib and topotecan. Quality control (QC) samples were also prepared at 3× the lower limit of quantitation and at 50% and 75% of the upper limit of quantitation.

RPE and retinal organoids medium

Frozen samples were removed and allowed to thaw to room temperature under light-protected conditions. Following thawing, 25 μL of each sample, standard, QC, or blank medium were transferred to a 0.5 mL Eppendorf tube. Protein precipitation was performed by the addition of 225 μL of 100% acetonitrile to each tube. Samples were then vortex-mixed for 5 s and centrifuged at 20,817 rcf for 5 min at 4°C. The resulting supernatant (200 μL) was transferred to a polypropylene insert within a 1.5 mL glass vial and submitted for liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis with an injection volume of 1 μL. Two replicates of each standard, three replicates of each QC, two replicates of 100% acetonitrile (as a procedural blank), and five replicates of blank medium were included in the analytical batch alongside the unknown samples.

Retinal organoid cell pellets

Frozen samples were reimoved and allowed to thaw to room temperature under light-protected conditions. Cells were lysed by adding 300 μL of 1% formic acid +1% Triton X in water, gently vortex-mixing for 5 s, then sonicating at ambient temperature for 3 × 30 s with 2-min intervals. Following lysis, 50 μL of cell suspension, standard, QC, or blank buffer were transferred to a 1.5 mL Eppendorf tube. Liquid-liquid extraction was performed by the addition of 250 μL of 100% acetonitrile to each tube. Samples were then vortex-mixed for 5 s and centrifuged at 14,000 g for 5 min at 4°C. The resulting supernatant (200 μL) was transferred to a polypropylene insert within a 1.5 mL glass vial and submitted for LC-MS/MS analysis with an injection volume of 1 μL.

Data availability

The scRNA-seq data are submitted to the Gene Expression Omnibus under the accession number GSE305503.

Acknowledgments

We thank the 10.13039/501100022850Little Princess Trust (CCLGA 2022 18), Fight for Sight, and EPSRC/ERC (EP/Y031016/1) for funding this work. The purchase of IncuCyte used in this study was supported by a UKRI MRC Capital Funding for World Class Labs Award (MR/X012360/1). We would also like to thank the 10.13039/501100000774Newcastle University FMS Bioimaging Unit for their continual support during this project. We would also like to thank Ute Jungwirth for her support during this project.

Author contributions

J.K. and A.R. designed and conducted experiments, data analyses, figure preparations, and manuscript writing. E.S., D.A., R.H., A.U., A.G., A.R., R.Q., and J.C. conducted experiments and data analyses. D.S., M.P., L.A., M.J., I.H., and G.V., study and experimental design and fund raising. M.L., study and experimental design, fund raising, data analyses, figure preparations, manuscript writing, and data dissemination. All authors approved the final version of this manuscript.

Declaration of interests

I.R.H. receives rewards to inventors for abiraterone from the Institute of Cancer Research, and research funding from Astex Pharmaceuticals.