Main

RB1-mutant retinoblastomas can originate from a cellular state found during retinal development in humans but not in other species9,10. Accordingly, to identify the cellular state and corresponding circuitry that sensitizes to RB1 inactivation, we examined the effects of Rb depletion on human fetal retinal cells. Samples were from post-fertilization weeks 17–19, when all retinal cell types and a range of maturation states are present.

Dissociated retinal cells were transduced with RB1-directed or control short hairpin RNAs (shRNAs), followed by co-staining for the proliferation-associated Ki67 and cell-type-specific markers. RB1 shRNAs abrogated Rb expression in long or medium wavelength (L/M)-opsin+ and thyroid hormone receptor β2+ (TRβ2+) cone precursors as well as in other cell types (Extended Data Fig. 1a). After 2 weeks, Ki67 was detected in cone-precursor-like cells co-expressing the photoreceptor marker CRX and the cone markers L/M-opsin, cone arrestin and RXRγ (Fig. 1a and Extended Data Fig. 1b–h). Ki67+ cone-marker+ cells were first detected 9 days after transduction, whereas clusters were routinely detected by day 23. Ki67 was not detected in cells expressing markers of rods (NRL), bipolar cells (strong CHX10), ganglion cells (BRN-3), or amacrine or horizontal cells (PROX1+ or PAX6+, nestin−) (Fig. 1a and Extended Data Fig. 1i, j). Ki67 was detected in cells expressing markers of retinal progenitor cells (RPCs) or Müller glia (nestin or CRALBP, SOX2), yet in similar proportions after RB1 shRNA or control shRNA (Fig. 1a and Extended Data Fig. 1j). RB1 shRNAs also induced incorporation of 5-ethynyl-2′-deoxyuridine (EdU), an indicator of S phase entry, increased expression of the mitosis marker phosphohistone H3, suppressed expression of the apoptosis marker cleaved caspase 3 (CC3), and induced proliferation in cells expressing cone but not other retinal cell markers (Fig. 1c, d and Extended Data Fig. 1k–n). By contrast, RB1 shRNAs induced CC3 and decreased the number of cells expressing markers of RPCs and glia (Fig. 1b, d and Extended Data Fig. 1n).

Figure 1: Proliferation of cone-like cells after Rb depletion in dissociated FW19 retina.
figure 1

a, b, Responses to RB1-733 shRNA (shRB1). D, day; SCR, scrambled control. a, Percentage Ki67+ among cells expressing the indicated retinal cell-type-specific markers. b, Prevalence of DAPI+ (4′,6-diamidino-2-phenylindole+) cells expressing cone (L/M-opsin, CRX), RPC (nestin), or Müller glia (CRALBP) markers. c, d, Responses to RB1-733 and RB1-737 shRNAs (shRB1-733 and shRB1-737, respectively). c, Proliferation of cells, of which >90% were cone marker+ at day 60. d, Percentage CC3+ among cells expressing the indicated markers at day 14. Values and error bars denote mean and s.d. of triplicate assays; P values are from unpaired Student’s t-test. All results replicated at least twice.

PowerPoint slide

Full size image

To assess whether the Rb-deficient proliferating cone-like cells derived from post-mitotic cone precursors, we examined the effects of Rb knockdown in prospectively isolated retinal cell populations. Populations were isolated by sorting for size, for CD133, which is expressed strongly in maturing photoreceptors and weakly in RPCs11, and for a CD44 epitope expressed by Müller glia and RPCs12 (Fig. 2a). Staining for cell-type-specific markers revealed populations enriched for cone precursors, for rod plus cone precursors, for RPCs plus glia, and for a mixture of rod, ganglion, bipolar, amacrine and horizontal cells (Fig. 2b and Extended Data Fig. 2a–g). In medium and large CD133hi, CD44− populations, 96–98% of cells co-stained for CRX and cone arrestin, which is cone-specific at post-fertilization week 19 (FW19) (Extended Data Fig. 2h). A similar enrichment was observed when cone precursors were identified using CRX and RXRγ (Extended Data Fig. 2h–k).

Figure 2: Cone precursor response to Rb depletion.
figure 2

a, Dissociated FW18 retinal cells sorted by size, CD133 and CD44, with major populations designated. FSC, forward scatter; misc, miscellaneous; SSC, side scatter. b, Percentage of cone arrestin+(cArr+), CRX+ cones, NRL+ rods, nestin+ RPCs and glia, and PAX6+, nestin− horizontal, amacrine or ganglion cells in each population. Lg, large; med, medium; sm, small. c, Responses to RB1-733 shRNA. Percentage of Ki67+ or CC3+ cells at 14 days (top), and cell numbers at days 14 and 23 (bottom). *P < 0.05, **P < 0.01 (compared to scrambled control). Results in a–c are representative of three independent experiments. d, Ki67+, L/M-opsin+ cone precursors (white arrows) in FW19 fovea 15 days after transduction with RB1-733 and RB1-737 shRNAs; and Ki67+, L/M-opsin− cells probably representing RPCs or glia (yellow arrows) after transduction with RB1 shRNA or scrambled control. Scale bar, 20 μm. Values and error bars denote mean and s.d. of triplicate assays; P values are from unpaired Student’s t-test.

PowerPoint slide

Full size image

RB1 shRNAs induced similar RB1 knockdown in each retinal cell population (Extended Data Fig. 3a). After 2 weeks, Ki67 was detected in 80% of cells in the cone-enriched population (Fig. 2c), probably reflecting a high ratio of shRNA-expressing lentivirus to target cells and cone precursor proliferation. After 3 weeks, cone precursor numbers increased (Fig. 2c). Rb depletion did not induce proliferation in RPCs and glia, but increased the proportion of CC3+ cells entering apoptosis (Fig. 2c). Sorted populations transduced with the scrambled control had higher CC3+ rates than unsorted cultures, potentially reflecting separation of RPCs and glia from neurons13,14. Nevertheless, Rb knockdown induced proliferation and apoptosis in cells with the same immunophenotypes as in unsorted cultures. Notably, Rb depletion induced the cell-cycle-related genes CCNE1, SKP2, E2F1, RBL1, CCNB1 and CDK1 in cone precursors, and induced p53-responsive genes in sorted RPCs and glia (Extended Data Fig. 3). Cell-cycle-related genes were induced several days before Ki67, suggesting that further reprogramming was needed for cell cycle entry.

RB1 shRNAs also induced cone precursor proliferation in intact retinas. Ki67 was detected in L/M-opsin+ cone precursors in the fovea, demarcated by cones but not rods, 15 days after transduction (Fig. 2d). Ki67 was not detected in cells expressing rod, amacrine, horizontal or ganglion cell markers (Extended Data Fig. 4a, d). Ki67 was detected in RPCs and glia marked by PAX6+, nestin+ or by CHX10+, CRX−, yet in similar proportions after transduction with RB1-directed and control shRNAs (Extended Data Fig. 4b–d). Moreover, a yellow fluorescent protein (YFP)-expressing RB1 shRNA vector selectively induced Ki67 in YFP+ cones, although all cell types were transduced (Extended Data Fig. 4e–h).

We next determined whether Rb-depleted cone precursors and retinoblastoma cells depend on similar signalling circuitry. Retinoblastoma cell proliferation requires several proteins that are prominent in cone precursors, including TRβ2, RXRγ, MYCN and MDM2 (ref. 7). Depletion of these factors suppressed Ki67 expression and cone precursor proliferation both in dissociated retinal cultures (Extended Data Fig. 5a, b) and in isolated populations (Fig. 3a). Retinoblastoma cell proliferation also requires SKP2-mediated degradation of Thr 187-phosphorylated p27 (ref. 15). Concordantly, SKP2 depletion suppressed cone precursor proliferation and increased CC3 (Fig. 3a and Extended Data Fig. 5a). Notably, maturing cone precursors had exceptionally high Thr 187-phosphorylated p27 (Extended Data Fig. 5c), coincident with a maturation-associated decrease in total p27 (ref. 16), suggesting that SKP2-mediated p27 degradation might enable cone precursor proliferation. Consistent with this view, cone precursor proliferation was suppressed by ectopic p27 and enhanced by ectopic SKP2 or p27 knockdown (Fig. 3b and Extended Data Fig. 5b), as in retinoblastoma cells15. Thus, Rb-depleted cone precursors and retinoblastoma cells had similar signalling requirements.

Figure 3: Effects of cone precursor circuitry on response to Rb depletion.
figure 3

a, Prevalence of Ki67+ or CC3+ cells (top) and cell numbers (bottom) after shRNA transduction of isolated cone precursors. b, Isolated cone precursor response to co-transduction with RB1 shRNA and BE-Neo (BN) vector, BN-p130 or BN-SKP2. c–e, Effect of p130 overexpression or p107 knockdown on Y79 cell proliferation (c, d) and protein expression (e). *P < 0.05, **P < 0.01 (compared to scrambled and vector controls); †P < 0.05, ††P < 0.01 (compared to RB1 shRNA plus SCR (a), or to RB1 shRNA plus BN vector (b)). Results represent at least two independent experiments. Values and error bars denote mean and s.d. of triplicate assays; P values are from unpaired Student’s t-test.

PowerPoint slide

Full size image

We also assessed the roles of the Rb-related p107 and p130 proteins (also known as RBL1 and RBL2, respectively). In mouse models, retinal tumorigenesis required loss of Rb combined with loss of p107, p130 or p27 (refs 10, 17). However, in human retinoblastomas, p130 (also known as RBL2) losses are common, whereas p107 (also known as RBL1) losses are rare18 (Extended Data Fig. 6a). Moreover, whereas maturing cone precursors had abundant p130 and minimal p107, retinoblastomas had barely detectable p130 yet prominent p107 (Fig. 4b and Extended Data Fig. 6b), implicating p130 but not p107 in retinoblastoma suppression. Concordantly, co-knockdown of p130 with Rb increased cone precursor proliferation (Fig. 3a and Extended Data Fig. 5a, d) and p130 overexpression suppressed cone precursor and retinoblastoma cell proliferation (Fig. 3b, c, e). Meanwhile, p107 knockdown suppressed proliferation both in Rb-depleted cone precursors (Fig. 3a and Extended Data Fig. 5a, d) and in retinoblastoma cells (Fig. 3d, e and Extended Data Fig. 5e, g). In Y79 cells, p107 knockdown decreased expression of MYCN and SKP2, while it increased the SKP2 target, p27 (Fig. 3e). These effects were seen with two shRNAs and were rescued by p107 restoration (Fig. 3d, e and Extended Data Fig. 5d–i). Furthermore, p107 overexpression enhanced proliferation of retinoblastoma cells while suppressing that of neuroblastoma cells (Extended Data Fig. 5h–j). Thus, both in Rb-depleted cone precursors and in retinoblastoma cells, p130 suppressed proliferation whereas p107 had a proliferative role distinct from its function in mouse models.

Figure 4: Rb-depleted or Rb/p130-depleted cone precursor tumours.
figure 4

a, Haematoxylin-and-eosin-stained Rb-depleted and Rb/p130-depleted cone xenograft tumours and human retinoblastoma (n = 4). KD, knockdown. Dashed lines denote Flexner–Wintersteiner rosettes; solid lines denote fleurettes; asterisks mark rosette cavities. b, Cone- and cell-cycle-related protein expression in human retinoblastoma and cone xenografts (n = 6). Scale bars, 40 μm (a, b). c, Transmission electron microscopy of Flexner–Wintersteiner rosettes in a human retinoblastoma and a cone-derived tumour, with mitochondria (arrows) between nuclei and rosette cavity (n = 2). The ×25,000 images are from the boxed area (top) or from a rosette not shown (bottom). Results are representative of at least two experiments. d, Model of cone-precursor retinoblastoma origin highlighting proteins that suppressed (blue) or promoted (red) the proliferative response.

PowerPoint slide

Full size image

After several months, some cone precursor cultures depleted in Rb or in both Rb and p130 (Rb/p130-depleted) formed suspension aggregates resembling retinoblastoma cells (Extended Data Fig. 7a). Rb/p130-depleted cultures proliferated more robustly and longer than those with Rb depletion alone, consistent with p130 losses in many retinoblastoma cell lines (Extended Data Fig. 6a). The cultures had properties consistent with Rb/p130-depleted cone precursors (Extended Data Fig. 7b–h). When engrafted either 3 months or within 1 week after knockdown, Rb- or Rb/p130-depleted cone precursors formed retinoblastoma-like tumours in subretinal xenografts (Fig. 4a and Extended Data Figs 8 and 9). For cells engrafted within 1 week, tumours appeared within 6–14 months (Extended Data Fig. 8b), similar to the time needed to form tumours in children.

Cone-precursor-derived tumours had differentiated histology, little Rb or p130, many Ki67+ cells, and prominent p107 and SKP2, consistent with robust proliferation (Fig. 4a, b and Extended Data Fig. 9). They expressed the photoreceptor-related CRX, CD133 and IRBP and the cone-specific cone arrestin, L/M-opsin and RXRγ; all at levels similar to retinoblastomas and developing retinas (Fig. 4b and Extended Data Fig. 9). This is consistent with the many cone-specific proteins in retinoblastoma tumours7 (Supplementary Table 1). As in human retinoblastomas7, cone-precursor-derived tumour cells lacked numerous markers of other retinal cell types and had rare S-opsin and rhodopsin expression (Extended Data Fig. 10). Rb-depleted and Rb/p130-depleted cone precursor tumours also had structures resembling Flexner–Wintersteiner rosettes and fleurettes (Fig. 4a), which are retinoblastoma hallmarks19. Transmission electron microscopy confirmed the rosettes, with mitochondria positioned between the nuclei and rosette lumens (Fig. 4c). Dense core vesicles were not seen in two Rb-depleted cone precursor tumours nor in two retinoblastomas, consistent with the reported rarity of such structures20,21. Finally, single nucleotide polymorphism (SNP)-array analyses of two tumours revealed no megabase-size gains or losses, whereas quantitative PCR (qPCR) analyses revealed a partial RB1 loss but no other frequently reported changes (Extended Data Fig. 8c–e), consistent with the lack of DNA copy number alterations in some retinoblastomas22,23,24. Thus, cone precursor tumours resembled human retinoblastomas at the histological, ultrastructural, retinal marker and molecular cytogenetic levels.

This study examined collaboration between Rb loss and retinal cell-type-specific circuitries. We found that the circuitry of maturing L/M-cone precursors was uniquely conducive to proliferation and development of retinoblastoma-like tumours. Although we cannot exclude the possibility that Rb loss could induce a cone program and proliferation in other cell types, the robust responses of the most highly enriched cone precursor populations and of cells in an intact fovea suggest that cone precursors are the primary if not the sole responding cell type. Cone precursor features that collaborated with Rb loss included cone lineage factors (TRβ2 and RXRγ), highly expressed oncoproteins (MYCN and MDM2), and p27 downregulation probably mediated by SKP2. Some of these features may be interdependent, as RXRγ promoted MDM2 expression7, yet the larger program encompassing these features and its developmental purpose are unknown. Importantly, Rb-depleted cone precursor tumours had differentiated histology and lacked gross DNA aberrations, similar to putative early retinoblastoma elements25. These findings support a model in which Rb-deficient cone precursors form differentiated retinoblastomas, then dedifferentiate (Fig. 4d) and possibly acquire non-cone features8,22. Much of the circuitry implicated in cone precursor tumour initiation was also needed for retinoblastoma cell proliferation7,15, suggesting that tumour cells can be addicted to the cancer-predisposing circuitry of their originating cell types.

Methods

Retinoblastoma and retinal cell culture

Retinoblastoma cell lines Y79 and Weri-RB1 were obtained from the ATCC. RB177 was from an early passage culture and its identity confirmed by RB1 mutation sequencing. Retinoblastoma cells were confirmed free of mycoplasma and cultured in RB culture medium (IMDM, 10% FBS, 55 μM β-mercaptoethanol, with glutamine, penicillin, streptomycin, fungizone and 10 μg ml−1 plasmocin (Invivogen)7,26). Fetal eyes were obtained with informed consent from the Human Fetal Tissue Repository of the Albert Einstein College of Medicine and from Advanced Bioscience Resources under protocols approved by the Memorial Sloan-Kettering Cancer Center (MSKCC) Institutional Review Board, the Albert Einstein College of Medicine Institutional Review Board, and the Children’s Hospital Los Angeles Committee on Clinical Investigations. After transport in IMDM with 10% FBS on ice, eyes were rinsed in 70% ethanol for 3 s and washed in sterile PBS. Eyes were opened using a sterile scalpel and lens removed. Retinas were detached using forceps and incubated in papain solution (Worthington Tissue Dissociation Kit) for 10–30 min at 37 °C and 5% CO2, with pipette mixing every 5 min. After dissociation to ∼20-cell clusters, cells were diluted with 10 volumes of PBS and collected by centrifugation at 2,000 r.p.m. (Sorvall, Legend RT), re-washed in PBS (all centrifugations at 2,000 r.p.m. unless otherwise stated), suspended in RB culture medium as above, incubated at 37 °C with 5% CO2 overnight, and frozen in RB culture medium containing 10% dimethylsulphoxide (DMSO). Supernatant was transferred into a sterile container after each centrifugation and re-spun to prevent retinal cell losses. For lentivirus infections, cells were recovered from liquid nitrogen, cultured overnight, washed with PBS, suspended in 0.05% trypsin/EDTA (Cellgro) for 3–10 min with gentle pipetting, re-centrifuged, suspended in RB culture medium as above, and immediately infected. Cultures were maintained at high density, typically 50,000 cells per well (24-well dish) for unsorted cultures, with media changes every 3 days.

FACS

Approximately 10 million dissociated retinal cells (∼5 million per retina) were cultured for 18 h after thawing in RB culture medium, collected by centrifugation, washed with PBS, digested with 5 ml warm 0.05% trypsin/EDTA for 5–15 min while triturating in a 24-well culture plate 20–30 times per minute using a 1,000-μl tip and checking every ∼3 min, to produce 90–95% single cells, centrifuged as above (retaining the supernatant to prevent cell loss), suspended in 400 μl 5% FBS in PBS, and incubated at room temperature for 10 min. Cells (100 μl) were combined with 100 μl 4 μg ml−1 of mouse IgG (Sigma, I-8765), 300 μl of cells were combined with 300 μl of pre-mixed anti-CD133-phycoerythrin (PE) (Miltenyi Biotec, 130-080-801) at 1:6 and anti-CD44 fluorescein isothiocyanate (FITC) (clone IM7, Abcam ab19622 or BD Biosciences, BDB553133) at 1:25, to give 1:12 CD133 and 1:50 CD44 final dilutions. After 1 h at room temperature, cells were diluted with 900 μl 5% FBS in PBS, centrifuged as above, suspended in 500 μl 5% FBS/PBS with 300 ng ml−1 DAPI and held on ice until sorting. Cells were sorted using a Becton-Dickenson FACSAria SORP with 100 mW 488-nm laser, the triple bandpass filter removed in the FITC channel, FACSDiva v8.0 software, and selecting live single cells based on FSC width, SSC width, and DAPI exclusion. On FSC/SSC plots, cells were divided into small, medium, and large size groups and evaluated for CD133-PE and CD44-FITC. Eight populations collected into 500 μl complete medium as above were small, medium and large CD133hi, CD44−; small, medium, and large CD133lo, CD44+; small CD133−, CD44−; and ungated live single cells. Each population was cultured in 50% Y79-conditioned medium with fungizone (50% fresh RB culture medium combined with 50% filtered Y79-conditioned RB culture medium), and half of the volume changed with fresh 50% Y79-conditioned medium every 3 days. Sorted populations were characterized by adhering cells to poly-l-lysine-coated coverslips (1,000–2,000 cells each) for 3 h, fixing in 4% paraformaldehyde (PFA) for 5 min, washing in PBS four times, and storing at −20 °C until immunostaining. Lentivirus infection was performed within 24 h after sorting.

Lentiviral shRNA and cDNA expression constructs

pLKO lentiviral shRNA vectors from the TRC library (Open Biosystems/Thermo Scientific or MSKCC SKI High-Throughput Drug Screening and RNAi Core Facility)27 were designated by ‘sh’ followed by the name of the target gene and last 3–4 digits of the TRC or SKI identification numbers (Supplementary Table 2). shRNAs directed specifically against THRB variant 2 (also known as TRβ2) were designed using Invitrogen BLOCK-iT RNAi Designer (http://rnaidesigner.invitrogen.com/rnaiexpress/) and siDirect (http://sidirect2.rnai.jp/doc/) and cloned using the TRC cloning strategy (http://www.addgene.org/pgvec1?f=v&cmd=showfile&file=protocols) with deoxyoligonucleotides for DNA-directed RNAi (Integrated DNA Technologies). They are designated according to the position of the first shRNA target nucleotide after the translation initiation site (Supplementary Table 2). The pLKO scrambled control was Addgene plasmid 1864 (ref. 28). pLKO-YFP-shRB1-733, pLKO-YFP-shRB1-737 and pLKO-YFP-SCR control virus were produced by replacement of the puromycin resistance gene with YFP complementary DNA using In-Fusion cloning (Clontech), and provided by Z. Li. The lentiviral cDNA expression BN vector was created by replacing the EGFP gene of the BE-GFP lentiviral vector29 with the neomycin resistance gene between the EcoRI and BamHI sites (with assistance of S. Puranik). BN-p130 was produced by inserting human p130 cDNA between the BE-Neo PshAI and XbaI site. BN-SKP2 and BN-p107 were produced by inserting human SKP2 and p107 cDNA, respectively, between the BsiWI and PspXI sites of BE-Neo. Because shRBL1-2621 (shp107-1) targets the 3′ untranslated region, only the RBL1 open reading frame was cloned into BE-Neo to produce shRBL1-2621-resistant BN-p107. To produce shRBL1-2623 (shp107-2)-resistant BN-p107-2r, the shRBL1 target sequence GCAGTGAATAAGGAGTATGAA was mutated to GCAGTAAACAAAGAATATGAA without amino acid sequence changes using In-Fusion cloning (Clontech). BE-p27 was as described15.

Lentivirus production and infections

Lentiviruses were produced by reverse transfection of suspended 2 × 107 293T cells using 20 μg lentiviral vector, 10 μg pVSV-G, 20 μg pCMV-dR8.91 (ref. 30) and 100 μl Polyjet (SignaGen) or Lipofectamine 2000 (Life Technologies) in 15 cm3 dishes. The 3-ml plasmids–Polyjet complex and 1.5 ml 293T cell suspension were mixed in 50 ml centrifuge tubes and shaken for 30 min before being transferred to dishes. Virus collected 48 and 72 h after transfection was combined, concentrated 50–100-fold by centrifugation at 25,000 r.p.m. for 90 min, and suspended in RB culture medium. Concentrated virus (500–2,000 μl) was used to infect 5 × 105 Y79, Weri-1 or RB177 retinoblastoma cells, or to infect 5 × 105 total retinal cells or 1 × 105 of each sorted retinal cell population in 500 μl of filtered conditioned RB culture medium in the presence of 4 μg ml−1 polybrene (Sigma-Aldrich) followed by gentle pipetting 25 times and shaking for 10 min in the hood. After 18 h, cells were diluted in an equal volume of conditioned RB culture medium and maintained at 37 °C with 5% CO2. For co-infections, 100 μl of each concentrated virus was used to infect either 1 × 104 total retinal cells or 1 × 103 sorted retinal cells suspended in 100 μl of conditioned RB culture medium with 4 μg ml−1 polybrene in a total volume of 300 μl, and medium was replaced with 150 μl 50% Y79 and other RB cell conditioned medium 24 h after infection. Infected cells were selected starting 48 h after infection with 1.4–3 μg ml−1 puromycin for 48–72 h or with 50–100 μg ml−1 G418 for 4–7 days, and fed every 2–3 days by replacing two-thirds of the media with 50% Y79 and other RB cell-conditioned medium.

Intact FW19 retinas were either infected within the globe, by cutting a cross-section through the cornea, removing the lens and most of the vitreous, and pipetting 500 μl of concentrated pLKO versions of shRB1-733 and shRB1-737 or scrambled control lentivirus into the sub-retinal space and vitreous (causing retinal detachment) in a 24-well plate with the globe submerged in RB culture medium with 1 ml lentivirus suspension, followed after 2 days by addition of 2 ml of freshly prepared concentrated lentivirus; or infected after removal of the intact retina and residual vitreous in a 12-well plate, by addition of 1 ml of 80× concentrated pLKO-YFP-shRB1-733 or scrambled control lentivirus, reinfection with the same viruses 1 and 3 days later, and changing 50% of medium with a 1:1 mixture of fresh and ocular globe-conditioned medium daily thereafter. Displaced retinal tissue was fixed with 2% PFA in PBS for 2 h at 4 °C, and eyes with remaining tissue were fixed in 2% PFA in PBS overnight at 4 °C. Tissue samples were washed with PBS, transferred to 30% sucrose in PBS, and embedded in 30% sucrose/PBS:OCT at a 2:1 ratio, and cryosectioned at 8–10 μm.

Real-time quantitative PCR

Total RNA was isolated using StrataPrep total RNA microprep kit (Stratagene) for <1,000 cells (in FACS isolated populations) or GenElute Mammalian Total RNA Miniprep Kit (Sigma) for all other analyses. cDNA was synthesized using ImProm-II Reverse Transcription System (Promega). Primers were designed by Beacon Designer software (Premier Biosoft International) or Primer3 (http://frodo.wi.mit.edu/primer3/) (Supplementary Table 3). Relative messenger RNA levels were determined by qPCR using QuantiTect SYBR Green PCR Kit (Qiagen) or Maxima SYBR Green qPCR Master Mix (Fermentas) on an Applied Biosystems ABI 7900HT Sequence Detection System or ViiA 7 Real-Time PCR System using 95 °C 10 min followed by 40 cycles of 95 °C 20 s, 54 °C 30 s, 72 °C 30 s. Each sample was evaluated in triplicate and normalized to ACTB and GAPDH. Values represent the averages of both normalized results and error bars the standard deviation.

Immunostaining

Antibodies are described in Supplementary Table 4. Eyes were prepared and cryosectioned as described7,16. Cultured retinal cells were dissociated by gentle triturating, spread on poly-l-lysine-coated slides, incubated in a humidified incubator at 5% CO2 and 37 °C for 3 h, fixed in 4% PFA and PBS for 5 min, gently rinsed with PBS four times, vacuum-dried for 5 min, and stored at −20 °C.

The following co-staining combinations and orders were used to assess Ki67 expression in different retinal cell types. For cones: 1a. Mouse anti-cone arrestin31, anti-mouse-biotin, streptavidin-FITC, rabbit anti-Ki67, anti-rabbit-Cy3, rabbit anti-CRX and anti-rabbit-Cy5. 1b. Mouse anti-Ki67, anti-mouse-Cy3, rabbit anti-CRX, anti-rabbit-FITC, rabbit anti-human cone arrestin32 and anti-rabbit-Cy5. 2. Mouse anti-Ki67, anti-mouse-biotin, streptavidin-FITC, rabbit anti-CRX, anti-rabbit Cy3, rabbit anti-L/M-opsin and anti-rabbit-Cy5. 3. Mouse anti-RXRγ, anti-mouse-biotin, streptavidin-FITC, rabbit anti-Ki67, anti-rabbit-Cy3, rabbit anti-CRX and anti-rabbit-Cy5. For progenitors, Müller, and horizontal amacrine cells: mouse anti-Pax6, anti-mouse-biotin, streptavidin-FITC, rabbit anti-Ki67, anti-rabbit-Cy3, rabbit anti-nestin and anti-rabbit-Cy5. For other retinal cell types: mouse anti-human Ki67, anti-mouse-biotin, streptavidin-FITC, and rabbit antibodies for retinal specific markers, anti-rabbit Cy3. For BrdU labelling, 10 μM BrdU was added to medium for 2 h on day 23 after Rb knockdown and cells were stained with rat anti-BrdU, anti-rat-FITC, rabbit anti-CRX, anti-rabbit-Cy3, rabbit anti-L/M-opsin and anti-rabbit-Cy5.

For co-staining with mouse antibodies, sections or cells were treated with 1 mM EDTA/PBS for 5 min at room temperature and washed with PBS. Sections were treated with ABC kit reagent A (Vector Laboratories) in PBS for 15 min, washed in PBS, treated with ABC kit reagent B (Vector Laboratories) in PBS for 15 min, washed in PBS, blocked and permeabilized for 20 min in super block (2.5% horse serum, 2.5% donkey serum, 2.5% human serum, 1% BSA, 0.1% Triton-X-100 and 0.05% Tween-20 in PBS; filtered with 0.22 μm filter), incubated in mouse primary antibody in super block overnight at 4 °C, washed in PBS, incubated in biotinylated horse anti-mouse antibody in super block for 30 min, washed in PBS, incubated with FITC-conjugated streptavidin in PBS, and washed with PBS.

For co-staining with antibodies of other species, after completing the first staining reaction as above, sections were incubated in super block for 20 min, incubated overnight with primary antibody in super block, washed in PBS, incubated with Cy3- or Cy5-conjugated secondary antibody in super block for 30 min, and washed in PBS. Sections were then stained with 1 μg ml−1 DAPI in PBS, dried, mounted in Vectashield (Vector Labs) and imaged using an Axioplan2 (Carl Zeiss MicroImaging, LLC) or confocal DMIRE2 (Leica) microscope. Antibody specificity was confirmed by staining in parallel with control IgG or no primary antibody.

Antibody-dependent immunofluorescence signals were distinguished from autofluorescence by virtue of signal detected in only one colour channel. Cells with autofluorescence in multiple channels or with DNA condensation, fragmentation, or degradation were excluded. Cytoplasmic autofluorescence common in astrocytes, Müller glia, and ganglion cells was distinguished from authentic antigens by its detection at multiple wavelengths in cells with characteristic glial cytoplasmic and nuclear morphology. Nonspecific cytoplasmic staining by concentrated nestin antibody was distinguished from authentic nestin staining by its homogeneous rather than fibre-like structure. Nonspecific nuclear staining of L/M-opsin was avoided by using reduced antibody concentration.

EdU labelling and detection

Click-iT EdU Alexa Fluor 488 Imaging Kit was used for EdU labelling to detect proliferation. Dissociated or sorted retinal cells were infected with shRB1 or control lentivirus. After 14 days, 20 μg ml−1 of EdU was added into medium and incubated for 1 h; the cells were attached on poly-l-lysine-coated coverslips for 2 h and fixed for 5 min.

The cells were blocked and permeabilized for 20 min in super block as above and EdU was detected by addition of Click-iT reaction cocktails containing 2 μM Alexa Fluor 488 azide for 1 h. Co-staining was performed after EdU labelling with different combination of antibodies for retinal cell markers and secondary antibodies conjugated with Cy3 or Cy5, described as above. For cones, cone arrestin-Cy3 plus CRX-Cy5, RXRγ-Cy3 plus CRX-Cy5, and L/M-opsin-Cy3 plus CRX-Cy5 were used for co-staining with EdU.

Immunoblotting

Cells were washed in PBS, lysed in ELB+ (150 mM NaCl, 50 mM HEPES, pH 7.4, 0.1% NP40, 5 mM EDTA, 2 mM dithiothreitol (DTT), 1 mM phenylmethylsulfonyl fluoride, 10 mM NaF, 1 mM sodium orthovanadate, Thermo Scientific Halt phosphatase inhibitor cocktail and protease inhibitor cocktails), separated on 4–20% Ready Gel polyacrylamide gels (Jule Biotechnologies INC) or 8% polyacrylamide (for Rb western), and transferred to Hybond-ECL nitrocellulose membrane (Amersham). Membranes were probed with antibodies (Supplementary Table 4) and developed using horseradish peroxidase-conjugated anti-mouse or anti-rabbit antibodies and the ECL Advance Western Blotting Detection Kit (Amersham Biosciences) or Thermo Scientific SuperSignal-West Femto Chemiluminescent Substrate, and HyBlot CL X-Ray film (Denville Scientific).

Xenografts

All animal experiments complied with ethical regulations and were approved by the MSKCC Institutional Animal Care and Use Committee. Xenografts were performed on 6-week-old male athymic (Foxn1−/−) mice (Taconic) or 6-week-old male non-obese diabetic–severe combined immunodeficient (NOD–SCID) Il2rg−/− mice (Jackson Laboratories). Cultured cells were dissociated by pipetting, suspended in RB growth medium at 5 × 104 cells per microlitre (day 90) or 2.5 × 103 cells per microlitre (days 3 or 7), held on ice, and 2 μl injected into the subretinal space as described7. Irradiated 5053 rodent diet with amoxicillin was provided from 2 days before to 2 weeks after injection to prevent infection. Some tumour-bearing eyes were fixed and embedded as described7,16.

DNA copy number analyses

Genomic DNA from retinoblastomas, cone-derived retinoblastoma-like cells, and cone-derived xenograft tumours were isolated with QiaAMP DNA Mini kit (Qiagen). Genomic DNA of cone-derived cells was digested with XhoI to separate pLKO DNA hairpin structures. Relative DNA levels were determined in triplicate by qPCR using QuantiTect SYBR Green PCR Kit (Qiagen) on an Applied Biosystems ABI 7900HT Sequence Detection System or ViiA 7 Real-Time PCR System, using primers listed in Supplementary Table 5 and normalizing to the average of the HNF4A and BRCA1 genes. Integrated pLKO-shRB1-733, shRB-737 and shRBL2-923 copy numbers were analysed using primers corresponding to the pLKO.1 U6 promoter and RB1- or RBL2-specific shRNA sequences. High resolution SNP-array DNA copy number analyses were performed using CytoScan HD (Affymetrix, 901835). Data were analysed using Chromosome Analysis Suite 2.0 (Affymetrix).

Statistical analyses

Measurements were performed in triplicate and differences between means assessed for significance using unpaired Student’s t-tests. Sample sizes were chosen based on the maximum cell numbers that could be used for individual experiments given sample availability.

Transmission electron microscopy

Human retinoblastomas and cone-derived xenograft tumours were fixed with 4% PFA in PBS, rinsed in 0.1 M sodium cacodylate buffer, post-fixed in 2% osmium tetroxide for 1 h, rinsed in distilled water, dehydrated in a graded series of 50%, 75%, 95% and 100% ethanol, followed by two 10-min incubations in propylene oxide and overnight incubation in 1:1 propylene oxide/Poly Bed 812. The samples were embedded in Poly Bed 812 and cured at 60 °C. Ultra-thin sections were obtained with a Reichert Ultracut S microtome. Sections were stained with Uranyl Acetate and Lead Citrate and photographed using a Jeol 1200EX transmission electron microscope.