Identification of tumorigenic retinal stem-like cells in human solid retinoblastomas



Retinoblastoma (RB) is the most common malignant tumor of the retina in human children. Although it has been hypothesized for a long time that RB derives from multipotent retinal stem cells (RSCs) or retinoblasts, the direct evidence that the presence of tumorigenic RSCs in RB tumors is still lacking. Some studies indicate that malignant tumors contain tumor stem cells similar to their normal tissue stem cell counterparts. With in vitro culture and differentiation method we demonstrate that tumorigenic retinal stem-like cells (RSLCs) indeed exist in RB lesions and that RB tumor-derived cultures encompass undifferentiated cells capable of extensive proliferation as clonal nonadherent neurospheres and can differentiate into different retinal cells in vitro. Interestingly, cultured cells expressed retinal development related genes including nestin, CD133, pax6, chx10 and Rx, and overexpressed Bmi-1, a gene required for self-renewal and proliferation of stem cells. Significantly, when these cultured cells were intraocularly transplanted into SCID mice, they gave rise to new tumors with histomorphological features and immunophenotypes similar to their parental primary RBs. The results show that RBs contain tumorigenic RSLCs that contribute to tumorigenesis. This study provides a new insight to investigate the histogenesis of RBs and establishes a model for other RB research. © 2007 Wiley-Liss, Inc.

Retinoblastom (RB) is the most common malignant tumor of the retina in children and has become an excellent study model for cancer biology and molecular mechanism.1, 2, 3, 4, 5, 6 It is categorized into differentiated and undifferentiated types by the World Health Organization (WHO) classification according to the presence of rosettes and the degree of atypia.2 Most of previous studies have been focusing on the cellular and molecular analysis of the bulk tumor mass to search cell origin of RB tumor and RB gene mutation. Although the origin of RB cells has been somewhat controversial, it has been generally agreed that RB cells originate from the multipotent primitive retinoblasts or retinal stem cells (RSCs).3, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16 Up to date, there is no direct evidence to support that tumorigenic RSCs exist in RBs. In recent years, brain tumor stem cells (BTSCs) have been isolated from human brain malignant tumor samples and cultured then transplanted into experimental animals to form new tumors, providing strong evidence that these BTSCs are the origin of the brain tumor.17, 18, 19 In 2005, Siegal et al. demonstrated that mouse and human RB cells contained a small subpopulation of cells exhibiting a cancer stem cell-like phenotype.20 Recently, our research group successfully cultured stem-like cells from human RB lesions using the methods culturing BTSCs and RSCs.19, 21 Our previous data showed that these stem-like cells exhibited RSC properties in vitro including extensive proliferation, self-renewal and multipotency.22 However, we did not know whether these retinal stem-like cells (RSLCs) were able to reproduce parental tumors in animal model.

In this study, we further characterized RSLCs cultured from human RB specimens in vitro and in vivo, and investigated whether these RSLCs can reproduce human RB tumors in animal model. Our results in this study showed that RSLCs had the capacities of self-renewal, extensive proliferation and multipotency in vitro, as well as expressed retinal development related genes. Additionally, the RSLCs isolated from the examined 4 primary RB tumors developed tumors during 2–12 weeks after intraocularly transplanted into NOD/SCID (nonobese diabetic, severe combined immunodeficient) mice. Moreover, histomorphological features and immunophenotypes of these xenograft tumors were very similar to their parental primary RBs. These findings demonstrate that RBs contain tumorigenic RSLCs, which could contribute to tumorigenesis.


ADTC, acutely dissociated tumor cell; BTSC, brain tumor stem cell; HANR, human adult neural retina; HENR, human embryonic neural retina; NTS, neurosphere-like tumor sphere; RB, retinoblastoma; RSC, retinal stem cell; RSLC, retinal stem-like cell; SFM, serum free medium; wg, week of gestation.

Material and methods

Tissue collection

Tumor samples were obtained from surgical patients with RBs. Tumors were graded at Zhongshan Ophthalmic Center (ZOC), Sen Yat-Sen University (SYSU), China, by ocular pathologists in accordance with the WHO established guidelines.2 Human embryonic neural retinas (HENRs) at the 9th week of gestation (wg) used as experimental control were obtained from legal routine therapeutic abortions at the Third Affiliated Hospital, SYSU. Human adult neural retinas (HANRs) were obtained from residual eye tissues after cornea transplantation performed at ZOC. All samples were collected with patients' consents and in accordance with protocols approved by institutional review boards at ZOC and the Third Affiliated Hospital, SYSU.

Primary culture, passage, clonality and self-renewal

Primary cultures of retinal stem-like cells (RSLCs) were performed according to references.19, 21 A detailed protocol is provided in online supplementary materials. Briefly, tissue fragments obtained from either human retinoblastoma (RB) tumors or xenograft tumors were washed, minced, digested with trypsin, dissociated and passed through a series of cell strainers. Cells were seeded in serum free medium (SFM) containing basic fibroblast growth factor (bFGF, 20 ng/ml), epidermal growth factor (EGF, 20 ng/ml) and leukemia inhibitory factor (20 ng/ml) at a density of 100,000 cells per ml. Cells were cultured at 37°C in the atmosphere with 5%CO2 to 95% air and observed under invert microscope every other day. Media were exchanged twice a week with half of fresh SFM. EGF and bFGF (20 ng/ml final concentration respectively) were added to media at every change of media.

Seven to 10 days after primary culture, large neurosphere-like tumor spheres (NTSs) were harvested, digested with the same methods as the above section and passaged by 1 to 2 or 1 to 3 in SFM. For clonal cultures, cells disaggregated from tumor specimens or primary and sencondary NTSs were plated at a density of 1,000 live cells per milliliter in conditioned medium according to references.17, 23, 24 The conditioned medium consisted of a half of the old routine exchanged culture medium and a half of fresh SFM with 40 ng/ml EGF and 40 ng/ml bFGF. On day 1, 3, 5, 7 postplating, the number and volume of spheres were evaluated with LEICA IM50 Image Manager (Leica, Wetzlar, Germany) (see online supplemental materials). To verify the capability of self-renewal, serial subclonogenic analysis was performed as by Galli et al.25

Flow cytometry

Single cells from either 7–14 days NTSs or acutely dissociated tumor cells (ADTCs) were obtained by enzyme digestion, and then resuspended in 1× PBS with 0.5% BSA and 2 mM EDTA. Ten-microliter of CD133-2-phycoerythrin (fluorochrome-conjugated mouse antihuman monoclonal IgG1; Miltenyi Biotec, Bergisch Gladbach, Germany) was added, and incubated in the dark for 10 min, then washed with 2 ml PBS. Percentage of CD133+ was evaluated by flow cytometry (Coulter, Co., Fullerton, CA).

Differentiation assay

To analyze the differentiation potential of the RSLCs, early passage (passage 1 to 3) clonally derived NTSs were collected, washed with PBS, and then plated as whole spheres onto glass coverslips or 35 mm culture dishes (Cornings, Lowell, MA) precoated with 200 μg/ml poly-D-lysine (PDL, Sigma, St. Louis, MO) and 5 μg/ml laminin (Sigma) in differentiation medium composed of Neurobasal medium and B-27 (Gibco-Invitrogen, Grand Island, NY), without mitogens (EGF and bFGF). Cells were incubated in 37°C, humidified atmosphere with 5%CO2 to 95% air. Culture media were exchanged twice a week. Morphological changes were documented using reverse microscopy (Leica) each day. Coverslips were processed 7 days post-plating using immunocytochemistry.


NTSs cultured nonadherently in SFM for 7–14 days were collected, centrifuged and then cell pellets were washed with 0.1 M, pH 7.4 PBS and fixed with fresh 4% paraformaldehyde (PFA) for 15 min at 4°C. Cells were centrifuged, washed with PBS again and then smeared on precoated slides for immunocytochemistry analysis. After that, undifferentiated and differentiated cells were examined with the following retinal stem cells (RSC) markers including human specific nestin, pax6 and with the mature retinal cell markers including microtubule-associated protein 2 (MAP2), growth associated protein 43 (GAP43), glial fibrillary acidic protein (GFAP), syntaxin (Clone HPC-1), recoverin, rhodopsin, calbindin D 28 and protein kinase C-α (PKCα). A list of the antibodies with their dilutions and suppliers is provided in online supplementary materials. Immunofluorescence was performed with protocols as described previousely.21 In some cases, nuclei were counterstained with Hoechst 33342. Finally, staining was visualized by confocal microscope (LSM510, Carl Zeiss, Oberkochen, Germany). HENRs and HANRs were used as positive controls. PBS replaced first antibody as the negative control.

Semiquantitative RT-PCR

Total RNA was isolated from cells cultured on proliferation or differentiation conditions and from tissues of HENRs or HANRs by TRIzol (Gibco-Invitrogen). The extracted RNA was quantified using a Bio Photometer (Eppendorf, Hamburg, Germany). RNA (2.625 μg) from each sample was converted to cDNA with Revert Aid™ First Strand cDNA Synthesis Kit (Fermentas, Burlington, Canada), following the supplier's protocol. Human β actin expression was used as an internal control. PCR was performed with Easy-Do™ PCR PreMix Kit (SBS Genetech, Beijing, China) supplemented with 2 μl cDNA, 1μl 10 μM forward and reverse primer (see Table II of online supplementary materials) and 16 μl ultra-pure water on a thermocycler (GeneAmp PCR system 9700, ABI, Foster, CA). Five-minutes at 94°C was followed by 30 to 35 cycles that consisted of 30 sec at 94°C, 1 min at the annealing temperature corresponding to the primers used, and 1 min at 72°C. The final extension reaction was performed for 7 min at 72°C. Five-microliter of PCR products was visualized on GoldView-stained 2% agarose gel against a 100-bp ladder and documented with gel image analysis system (Vilber Lourmat, Cedex-1, France) under constant voltage electrophoresis for 30 min. The intensities of signals were quantified by percentage of integrated optical density of the specific gene to that of beta-actin. Negative controls included RNA, but no reverse transcriptase, to ensure that the PCR product was not amplified from genomic DNA.

Evaluation of tumorigenicity by intraocular injection

Tumorigenicity was determined by intraocularly (I.O.) injecting RSLCs cultured from RBs or injecting ADTCs into NOD/SCID mice. Four to 6 weeks old female mice were deeply anesthetized with an intraperitoneal injection of chloral hydrate (430 mg/kg) and the pupils dilated with topical 1% tropicamide to view the fundus. Under a binocular surgical microscope, we transsclerally injected 1 μl PBS containing about 105 cells into the vitreous cavity through 5 μl-microsyringe, and then monitored the survival of the mice and eyeball changes every 7 days after transplantation. Mice were killed at different times between 2 and 12 weeks postinjection according to the condition of the mice. The eyes were fixed with perfusion of 4% PFA, cryoprotected in 10–30% graded sucrose-PBS and sectioned at 8 μm on a cryostat. Haematoxylin and eosin (H&E) staining and immunohistochemistry were performed on 8 μm-thick cryostat sections. First antibodies used were similar to the above relative sections (see Table I of online supplementary materials). Immunodetection was performed using the Elite Vector Stain ABC System (Vector Laboratories, Burlingame, CA). Color visualization was performed using 3-3′-diaminobenzidine (Sigma) as the chromagen substrate. Nuclei were counterstained by haematoxylin. Patients' RB pathological specimens were used as controls. Five-percentage isotype control serum without primary antibody as negative controls.


Selection and expansion of RSLCs

To assess for the presence of RSLCs within human RB tumors, cells from postsurgery specimens of differentiated RBs (n = 2) and undifferentiated RBs (n = 6) (see Table I) were cultured at a density of 1 × 105 cells per ml in SFM, containing EGF and bFGF that favored stem cell growth, established previously for isolation of BTSCs or RSCs as neurospheres.17, 18, 19, 21 Three to 7 days after plating (AP), phase-bright tumor spheres resembling the classical neurospheres were detected in all of the cultures established from both differentiated and undifferentiated RBs. In the first 12 hr AP, disaggregated tumor cells remained in the single cell state in culture flasks (Fig. 1a). After that, these individual cells began to divide and form small floating spheres (≈2–5 cells in diameter) between 24 and 72 hr AP, then sizes and numbers of spheres gradually increased with time in culture. At 7 days, tumor spheres became very large in size, round and smooth in shape (Fig. 1b1d). Subsequently, the large tumor spheres were passaged by enzyme digestion and reseeded in fresh proliferative media every 7–10 days. In this study, tumor spheres could be passaged at least 6 passages and grew in vitro for 8 weeks (the longest sphere passaging). However, we found that cells after passage 3 grew markedly slower, consistent with the previous reports.21 Despite tumor spheres could be passaged multiple times, only passage 1 to 3 NTSs were used for all following study.

Figure 1.

Cells derived from human RB specimens can form NTSs in SFM media and express stem cell markers. ad: An example of growth kinetics of NTSs from RB tumor (RB6) (a: 12 hr; b: 72 hr; c: 7 days; d: Growth curve of NTSs). e–h: The NTSs expressed RSC markers nestin (e, green) and pax6 (f, red) as did the HENRs at 9th wg (g, nestin, green; h, pax6, red). i and j: Some cells in RB specimens also expressed nestin (i, brown) and pax6 (j, brown). k and l: flow cytometric quantification of CD133 expression in the NTSs (k) and ADTCs (l) from RB6 respectively. (Scale bar = 20 μm in ac, ej).

Table I. Clinic Characteristics of RB Solid Tumors
Tumor no.Patient age (yrs)SexLateralPathological subtypeFamily history
  1. LD, low differentiation; HD, high differentiation.


Proliferation, clonality and self-renewal

Because capabilities of proliferation and self-renewal are key features of neural and other stem cells, we used cloning and subcloning culture assays17, 24 to assess whether individual cells derived from tumor spheres had the ability to divide and form new NTSs. As a result, clonally derived NTSs from all of 8 cases were visible within 7 days AP. Moreover, when the secondary NTSs were subjected to serial subcloning assays, the generation of multiple secondary clones could be seen in all of tumors tested (n = 3). These findings suggest that NTSs contain individual stem-like cells with the ability to self-renew and form new neurosphere colonies. Clonal frequency was between 0.4 and 2.3% of the primary tumor cell population (n = 3).

Expression of retinal development related molecular markers of RSLCs

We next asked whether RB-derived RSLCs grown under proliferative conditions express specific molecular markers enriched in RSCs or early stage development of retinas.26, 27, 28, 29 The results from immunofluorescence study indicated that majority of cells from NTSs in proliferation media expressed RSC markers nestin and pax6 (Figs. 1e and 1f) whereas relatively few cells expressed markers of mature retinal cells. Therefore, immunophenotypes of these cells were similar to those of HENRs at 9th wg in which almost all cells expressed RSC markers (Figs. 1g and 1h), but not the markers of most mature retinal cells.

In addition, we examined immunoreactivity of RSC markers nestin and pax6 in RB tumor samples (n = 8) to determine whether they also contain some cells with stem cell characteristics. Surprisingly, immunoreactivity of nestin or pax6 (Figs. 1i and 1j) was observed in a subpopulation of cells from both undifferentiated and differentiated RBs although percentages of nestin or pax6 positive cells in tumor cell population varied widely in each sample. This indicates that a subpopulation of cells in human RB tissues has characteristics of RSCs.

Furthermore, flow cytometric analysis demonstrated that 30–86.3% of cells cultured in SFM for 8 days from primary RB tumors expressed CD133, while 2.8–28% of ADTCs were CD133 positive (Figs. 1k and 1l), suggesting that patient's tumor samples contained a subset of cells with features of neural stem cells (NSCs) and SFM can efficiently amplify and enrich stem-like cells from mixed RB tumor cells.

Meanwhile, we also compared gene expression signatures between undifferentiated and differentiated cells respectively derived from 5 primary RB samples (RB1 to RB5) by semiquantitative RT-PCR analysis. Primers were designed for the following human genes: chx10, pax6, Rx and nestin, genes required for normal retinal development;26, 27, 28, 29 CD133, a cell surface protein expressed on all fetal human NSCs30; and Bmi-1, a polycomb group gene required for self-renewal and proliferation of normal and leukemic hematopoietic stem cells.31 Interestingly, our data showed that all of the examined NTSs cultured in SFM media had intensive gene expression of nestin, CD133 and Bmi-1 which were similar to 9th wg HENRs, suggesting that these NTSs contain RSLCs having a strong capacity of self-renewal and proliferation. NTSs from RB1 to RB4 had middle level expression of pax6 and weak to middle level expression of chx10 as well as Rx. Conversely, the NTSs from RB5 had high expression of Rx and weak to middle expression of pax6 and chx10 (Fig. 2). By contrast of undifferentiation tumor spheres, expression of the above specific genes in differentiated spheres was markedly reduced or negative (Fig. 2), which is consistent with our prediction that tumor-derived stem cells would produce differentiated cells at the expense of multipotent progenitor cells under differentiation conditions.

Figure 2.

Gene expression profiles on undifferentiated and differentiated NTSs from two primary RBs (RB4 and RB5) and on two controls from HENRs at 9th wg and HANRs by RT-PCR. a: RT-PCR products on 2% agarose gel electrophoresis; b: Gray scale analysis of figure a. Undifferentiated NTSs (RB4U, RB5U) expressed many retinal developmental genes, similar to that of HENRs. However, expression of the above specific genes in differentiated NTSs (RB4D, RB5D) was markedly reduced or negative, similar to HANRs. M, marker, 100-bp ladder.

Multipotentiality of RB-derived RSLCs

Multipotency is a critical feature distinguishing stem cells with differentiated cells. Therefore, we argued whether RSLCs derived from RBs maintain the differentiative potential of RSCs that can differentiate into 7 main cell types of mature retina under specific differentiation conditions.21, 27 When single spheres from primary RB tumors were cultured on differentiation conditions, cellular morphologic changes appeared soon after plating. On the first day, almost all tumor spheres attached to precoated substrate, and a few cells migrated out of the margin of NTSs, and then extended a few of cell processes (Fig. 3a). On the third day, a vast majority of cells spreaded out of NTSs and developed typical characteristics of neural differentiation with neurite processes. Majority of migrating cells became flatten in fusiform shape or polygon with short and branched cellular processes (Fig. 3b). On the seventh day, neurite-like processes grew longer, ramified more. Cells differentiated in highest degree and formed typical retinal neural network-like structure (Fig. 3c). In immunocytochemistry study, a majority of differentiated cells expressed markers of retinal neurons (MAP2 and GAP43), glial cells (GFAP) and photoreceptors (recoverin); while a minority of differentiated cells expressed markers of amacrine cells (syntaxin) and bipolar cells (PKCα) (Figs. 3d3i). However, during the 7-day differentiation, we did not observe calbindin D 28 (for horizontal cells) and rhodopsin (for rod cells) positive cells. Therefore, the in vitro differentiation potential of the NTSs roughly matched the phenotypic signature observed in the patients' original tumors,10, 11, 12, 13, 14, 15 but was different from that of human RSCs.21 Such multipotency was stably maintained even after extensive culturing, up to passage 6.

Figure 3.

RB-derived RSLCs are multipotential. When clonal spheres from a primary RB tumor, RB6, were transferred to differentiation condition, cellular morphologic changes appeared soon after plating (a: day 1; b: day 3; c: day 7). Immunocytochemistry showed that differentiated cells expressed markers of most of mature retinal cells, such as MAP2 (d), GAP43 (e), recoverin (f), syntaxin (g), PKCα (h) and GFAP (i). Scale bar, a,b = 100 μm, c,f,i = 50 μm, d,e,g,h = 20μm. dh, nuclei were counterstained with Hoechst 33342 (5 μg/ml).

Xenograft tumorigenicity of RB-derived stem-like cells

The above findings showed that RB contained multipotent, self-renewing and population-expanding cells that satisfied all of the defining criteria expected from tumor stem cells in vitro. Nonetheless, given that the most critical attribute of tumor stem cells is their capacity to generate and perpetuate their tumor of origin,32 we sought to verify whether our RB-derived RSLCs could serve as tumor-initialing cells by assessing their actual ability to reproduce tumors in vivo. Strikingly, all of the examined RSLCs cultured from the 4 primary RB tumors (RB3, RB5, RB6 and RB8) reproducibly developed tumors during 2 weeks to 12 weeks after transplanted into the vitreous cavity of NOD/SCID mice, with a take efficiency of 100%. In the contrast, 1 of the 4 ADTCs developed tumor, with a take efficiency of 25%. These results also demonstrated that the present in vitro culture system could selectively enrich tumorigenic RSLCs. Histopathological examination showed that cells of these xenografts were small and round with scanty cytoplasm, karyomegaly and anachromasis. Interestingly, tumor cells from a human differentiated RB tumor (RB3) formed several Homer-Wright rosettes structures, which are typical of features for middle or high RBs. Importantly, an xenografts from the RSLCs cultured from RB tumor (RB8) with the optical nerve invasion not only grew in the vitreous cavity, but also infiltrated the adjacent retina, choroids, sclera, optical nerve and then extended into intraorbital tissues, which showed many mitotic figures and necrotic areas (Figs. 4a4c). Therefore, histomorphological signatures of the xenograft tumors were very similar to those of the parental human RBs.

Figure 4.

RB-derived RSLCs generate new tumors xenografted I.O. in NOD/SCID mice. a–l, 12 weeks after injection into vitreous cavity, RSLCs from RB8 developed a large tumor which recapitulated the features of the original tumors. a: slit-biomicroscope (Down-left, magnification of right eyeball); b–c: H&E examination showed that the tumor consisted of highly packed undifferentiated small rounded cells, had several Homer-Wright rosettes (see star) (b), infiltrated the adjacent retina (r), choloid, sclera (s), then extended into intraorbital (o) tissues (c). Immunostaining indicated that graft tumor widely expressed nestin (d), Ki67 (e), NSE (f), MAP2 (g) and recoverin (h), but only a few of tumor cells expressed PKCα (arrow) (i) and syntaxin (arrow) (j), and no cells expressed calbindin D 28 (k), GFAP (l) and rhodopsin. m–o: The human RB tumor (RB8) consisting of small round tumor cells (m, H&E), mainly expressed NSE (n), MAP2 and recoverin, but not express GFAP (o). Scale bar, c = 200 μm; b, d–o = 20 μm. CO, cornea.

In immunohistochemistry study, many cells in xenografts expressed markers of RSCs (human specific nestin), neural cells (MAP2, NSE), photoreceptors (recoverin) and proliferation marker Ki67 while only a few of cells expressed markers of the other retina cells, such as amacrine cells (syntaxin), bipolar cells (PKCα). Conversely, we did not observe the expression of markers of horizontal cells (calbindin D 28), glial cells (GFAP), nor rod cells (rhodopsin) (Figs. 4d4l). These features resembled their parental primary RBs (Figs. 4m4o). Although it is true that GFAP+ glial cells present in primary intraocular RBs, there is much controversy on their cell origin.10, 11, 12, 13, 14 Some researchers suggest that they are tumorigenic while others believe that they are responsive proliferation of normal glial cells in retina tissue. Our data support the latter opinion because we did not observe GFAP+ glial cells in human RB metastasis lesions of choroid and ocular orbit as well as in xenograft tumors of RBs. Accordingly, our results are consistent with Galli's who reported that tumor neural stem-like cells in glioblastomas were multipotent in vitro but monopotent or bipotent in vivo.17 The differences of immunoreactivity between in vitro and in vivo tumor cells might be relative to the differences of cell microenvironment.

To further confirm the capacity of tumor-initializing of RB-derived stem-like cells, we performed sequential transplantation experiments according to Galli's method with some modifications.17 Using the same approaches in the above related section, single tumor cells, isolated from original xenograft tumors, were expanded, cloned, then reinjected I.O. into new recipients. As a result, they were still capable of developing new RB-like tumors, with characteristics resembling the original RBs.


It has been hypothesized that only a small subpopulation of cells in the tumor, termed tumor stem cells, possesses the ability to self-renew and drive the progress, metastasis and recurrence of the tumor. Most of the tumor cells lose the ability to proliferate and self-renew, and they differentiate into tumor cells that become the phenotypic signature of the tumor.33, 34 However, up to date, the tumor stem cells were isolated, propagated and characterized only in a few malignant tumors.17, 18, 19, 35, 36, 37 It is still unclear whether the tumor stem cells also exist in human RB, although human RB has widely served as a model for studying cell origin and molecular pathogenesis of malignant tumors.3, 4, 5, 6, 7, 8

In this study, we reported for the first time that the tumor RSCs were isolated and identified from human RB samples using similar methods for isolation of BTSCs or human RSCs.19, 21 Our data indicated that RSLCs had the ability to form neurosphere-like clusters in SFM and could be passaged multiple times even under clonal cell density. This feature is shared by all of the 8 RB tumors examined in this study, suggesting that these cells forming spheres had the ability to proliferate and self-renew. Nevertheless, we observed that sphere cells grew and proliferated quickly during the 1st ∼3rd week of primary culture and could be passaged once every 7 to 10 days, while cells after 3 weeks of primary culture grew markedly slower. The growth features of RB stem-like cells are consistent with those of human RSCs and medulloblastoma stem cells.17, 18, 19, 21, 27 Coles et al. and Yang et al. demonstrated that human retinal stem or progenitor cells cultured in suspension were difficult to be passaged, whereas they could be passaged for a longer term when they were cultured in attachment.21, 27 Similarly, Hemmati et al. and Galli et al. reported that cells forming spheres of medulloblastomas grew and divided fast in early passages (<4 passages) of proliferation culture, but slowly in later passages and could not be cultured for long term.17, 19 Conversely, Galli et al. found that cells forming spheres of glioblastoma cultured in the same media grew slower than medulloblastoma in early passages, but could be cultured for longer term.17 Therefore, our data together with previous studies indicated that present culture conditions used to propagate at least RB or medulloblastoma stem cells should be improved in future studies.

The present reports have indicated that both tumor stem cells and their normal tissue stem cell counterparts share many common features including molecular markers and in vitro and in vivo multipotency except for the capacity of proliferation and self-renewal.33, 34, 35, 36, 37 Genes such as nestin, pax6, Rx and chx10 play a pivotal role in development, differentiation and maturation of retinal cells. Besides, they are also molecular markers of RSCs or retinal early stage development.28, 29 Interestingly, we found that undifferentiated NTSs from RB samples expressed stem cell markers as did HENRs or RSCs, while differentiated NTSs had yet significantly less or negative expression of retinal or neural stem cell markers and proliferation marker Bmi-1 as did HANRs, regardless of the variability from tumor to tumor. Notably, when the NTSs were transferred to differentiation conditions, they could differentiate to produce a phenotype similar to the RB tumor samples mainly including neurons, photoreceptors and glial cell.10, 11, 12, 13, 14, 15 Furthermore, when they were I.O. transplanted into immunodeficient mice, they developed xenograft tumors recapitulating histological signatures and immunohistochemical features of the original RB tumors, such as their rosette structures and invasion. These evidences suggest that RB tumors contain tumorigenic RSLCs with characteristics similar to, but not identical to RSCs.

For many years, researchers tried to identify the origin of the RB cell with advanced biotechnology.3, 7, 8 Two decades ago, in vitro differentiation assays documented that treatments of the bulk RB mass or RB cell lines with certain agents could induce a partial differentiation of cell types resembling those of the mature retina, such as photoreceptors, glia, conventional neurons as well as pigment epithelia, regardless of controversy of differentiated cell signatures.16, 38, 39, 40, 41 In addition, immunohistochemical examination on RB pathological specimens demonstrated that RBs exhibited phenotypic heterogeneity, being composed of cells expressing both undifferentiated and differentiated markers.10, 11, 12, 13, 14, 15, 20, 42, 43, 44 Hence, these investigators inferred that RB might come from multipotent or bipotent retinal cells. In the present study, we established a link between normal retinal genesis and tumorigenesis of RB with the application of principles for study of normal neural or retinal stem cells to RB tumor cell populations. Our data at least indicate that there is a close relationship between normal RSCs and tumorigenic RSLCs, as well reinforced the hypothesis that RBs might derive from the transformation of RSCs or retinoblasts.

In conclusion, the study of the developmental origin of RBs has important impaction for diagnosis, therapy and prevention of the tumor. To our knowledge, for the first time we demonstrated that RBs contained a subset of tumor stem cells in that these cells could be propagated in vitro as nonadherent neurospheres with SFM, express stem cell markers, and reproduce parental tumors in immunodeficient mice. This experimental system may provide an ideal in vitro model for studying RB-initiating cells and discovering new drug targets for RBs. These findings provide further evidence supporting the hypothesis that RBs are derived from RSCs or retinoblasts. But it remains unclear why RSLCs differ from RSCs. Future studies are needed to search for specific cellular surface markers and do cell sorting as well as in vivo limiting dilution assay to testify the phenotypically and functionally heterogeneity of RB cells. In addition, the studies on the differences between RSLCs and RSCs at molecular and cellular levels are also needed to clarify the pathogenesis of RB tumors.


We thank Dr. Zhongyao Wu, Dr. Jianhua Yan and Dr. Huasheng Yang (Ocular Tumor Wards, ZOC, SYSU) for providing RB specimens.