Targeting of replicating CD133 and OCT4/SOX2 expressing glioma stem cells selects a cell population that reinitiates tumors upon release of therapeutic pressure

The existence of radio- and chemotherapy-surviving cancer stem cells is currently believed to explain the inefficacy of anti-glioblastoma (GBM) therapies. The aim of this study was to determine if a therapeutic strategy specifically targeting GBM stem cells (GSC) would completely eradicate a GBM tumor. In both the in vitro and the in vivo models, ganciclovir therapy targeting proliferating GSC promotes the survival of a quiescent, stem-like cell pool capable of reproducing the tumor upon release of the therapeutic pressure. Images of small niches of therapy-surviving tumor cells show organized networks of vascular-like structures formed by tumor cells expressing CD133 or OCT4/SOX2. These results prompted the investigation of tumor cells differentiated to endothelial and pericytic lineages as a potential reservoir of tumor-initiating capacity. Isolated tumor cells with pericyte and endothelial cell lineage characteristics, grown under tumorsphere forming conditions and were able to reproduce tumors after implantation in mice.

Different studies support the notion that GSCs can give rise to vascular structures by differentiating into endothelial cells that form blood vessels 26,27 and pericytes that support vessel function and tumor growth 28 or via vasculogenic mimicry, where tumor cells form a fluid-conducting matrix through the acquisition of plasticity that mimics endothelial function 29,30 .
We had determined previously that adipose mesenchymal stem cells (AMSCs) modified to express the HSV thymidine kinase gene when implanted in GBM tumors differentiate to the endothelial lineage and associate with tumor vasculature structures and with CD133+ cancer stem cells. Treatment with the prodrug ganciclovir (GCV) effectively targeted replicating tumor cells reducing tumor cell burden by a factor of 10.000 and increasing animal survival 23,31 . However, while tumors could become chronic by periodic therapeutic cell administration, complete tumor cure was infrequent and tumors eventually relapsed. Thus, we speculated that the therapeutic effectiveness resulted from targetting the GSC vascular niche.
In the current work, we seek to determine if a therapeutic strategy specifically designed to target GSCs in a GBM model could completely eradicate a glioblastoma tumor. We show that even after extended treatment against replicating GSCs, a pool of resistant tumor cells is selected that can reproduce the tumors upon release from the therapeutic pressure. Moreover, we identify in the tumor vascular compartment tumor cells that express endothelial and pericytic markers and are endowed with the capacity to recapitulate the tumor.

Results
CD133 and OCT4/SOX2 promoters regulated RFP and Renilla luciferase reporters of GBM stemmess in vitro. Human GBM U87 tumor cells were genetically modified by transduction with a lentiviral vector construct for stable expression of a chimeric reporter comprising Photinus pyralis luciferase (PLuc) and enhanced green fluorescence protein (eGFP) activities, controlled by the cytomegalovirus (CMV) constitutive promoter.
In addition, FACS-selected eGFP positive cells were transduced with a second trifunctional chimeric reporter for expression of Renilla reniformis luciferase (RLuc), the red fluorescent protein (RFP) and a truncated version of the herpes simplex virus thymidine kinase sr39tk (tTK) under control of either the CD133 or the OCT4/ SOX2 promoters (Fig. 1A). This strategy allowed independent monitoring by bioluminescence imaging (BLI) and confocal microscopy, of either the whole tumor population or the subpopulation of tumor cells with active GSC promoters CD133 or OCTA4/SOX2. Moreover, administration of GCV would allow the selective killing of replicating cells with active CD133 or OCTA4/SOX2 promoters.
Dually labeled U87 tumor cells were grown either in adherent plates or in non-adherent conditions to form tumorspheres and monitored by BLI. Quantification of images and evaluation of the RLuc/PLuc ratio, a measure of reporter-specific expression relative to cell number, showed significant increases in the activity of CD133 (P = 0.0015) and OCT4/SOX2 (P = 0.0006) promoters when cells were grown as tumorspheres (Fig. 1B), supporting their use as stem cell markers.
To verify the functionality of the tTK gene, U87 cells grown in adherent plates or under tumorsphere forming conditions were treated with 4 μg/ml GCV for a 10-day period, during which PLuc activity was monitored. In both conditions, treatment with GCV resulted in a significant decrease in cell number, as compared with non-treated cells ( GCV treatment in vitro selects/induces a non-proliferating population of GSCs. While GCV treatment in vitro directly targets dividing CD133+ and OCT4/SOX2+ U87 cells, other replicating neighboring tumor cells can also be indirectly killed by a bystander effect. In our in vitro tumorsphere experiments, GCV treatment did not eliminate all the cells in culture but left a pool of U87 cells that remain alive after the treatment. Quantification of the fraction of RFP positive cells before and after GCV treatment demonstrated an increase in the proportion of CD133 (P = 0.0003) and OCT4/SOX2 (P = 0.0002) positive cells relative to the total tumor cell population (Fig. 1G,H), a fact that was also accompanied by an increase in the RLuc/PLuc ratio (P = 0.05 and P = 0.0022, respectively) (Fig. S1). Since GCV is toxic for replicating cells, these observations strongly suggest the existence in vitro of a CD133+ and OCT4/SOX2+ pool of GCV surviving tumor stem cells.
To further confirm this hypothesis, replication of RFP negative and positive cells in tumorspheres was analyzed using an independent procedure before and after 10 days of GCV treatment (Figs 1I,J and S2). Our results showed that the pool of CD133 and OCT4/SOX2 RFP expressing U87 cells was essentially insensitive to GCV treatment and it comprised a very low proportion compared with that of replicating cells. Conversely, there was a significantly larger proportion of replicating cells within the RFP negative pool and this pool was effectively reduced by GCV treatment.
Images of a representative mouse for each group ( Fig. 2A,B,E, F) and Plots of total light events (PHCs) recorded in the acquired images (Fig. 2C,G) showed that GCV targeting of tumor stem cells resulted in a significant inhibition of tumor growth as compared to non-treated controls. In agreement with this, Kaplan Meier graphs (Fig. 2D,H) showed significant increases in animal survival in response to GCV treatment (P < 0.0001 www.nature.com/scientificreports www.nature.com/scientificreports/ In spite of the increase in median survival, neither PLuc nor RLuc activity was completely eliminated by GCV treatment. A stable and measurable non-zero baseline of luciferase activity persisted even after prolonged GCV treatment, emphasizing the existence of a pool of tumor cells that survives the therapy and did not proliferate. www.nature.com/scientificreports www.nature.com/scientificreports/ The GCV surviving pool comprised cells that expressed RLuc regulated by GSC specific promoters CD133 and OCT4/SOX2 (Fig. 2B,F) respectively.
In order to determine if GCV treatment had effectively eliminated all tumor-initiating cells or, alternatively if remaining cells could restart tumors, mice were inoculated with U87-G-P/CD133-R-R-tTK or U87-G-P/ OCT4/SOX2-R-R-tTK cells and treated or not with GCV (n = 5/group). When mice arrive at the median survival day (determined for non-treated animals at days 39 and 54 for U87-G-P/CD133-R-R-tTK and U87-G-P/ OCT4/SOX2-R-R-tTK tumors, respectively), GCV administration was withdrawn in half of the mice of each group. Monitoring of PLuc activity showed an increase in tumor growth by days 63 (CD133) and 77 (OCT4/ SOX2), while tumor growth remained inhibited in animals under continued GCV treatment (Fig. 3A,B,D,E). Kaplan Meier survival graphs of U87-G-P/CD133-R-R-tTK and U87-G-P/OCT4/SOX2-R-R-tTK injected mice (Fig. 3C,F) showed that the median survival time was significantly superior in the animals that received GCV daily (P = 0.0012, P = 0.0003, respectively).
Thus, similarly to the in vitro model experiments, in vivo therapy targeting proliferating GSCs promotes the survival of a quiescent, stem-like luciferase-expressing cell pool capable of reproducing the tumor upon release of the therapeutic pressure.
Imaging of the GCV surviving GSC niche. To further understand the therapy mechanism, we used the CLARITY procedure to remove light dispersing myelin lipids from fixed brain tissue rendering it transparent to light. This approach allows confocal microscope 3D-imaging of fluorescent tumor cells embedded in the transparent brain tissue.
Confocal images of U87-G-P/CD133-R-R-tTK or U87-G-P/OCT4/SOX2-R-R-tTK tumors following the CLARITY procedure showed red fluorescent tumor stem cells forming dispersed aggregates within a complex structure comprising green fluorescent tumor cells (Fig. 4A,B, top panels).
In the case of GCV-treated U87-G-P/CD133-R-R-tTK or U87-G-P/OCT4/SOX2-R-R-tTK tumors, the CLARITY procedure allowed us to visualize small, approximately 500 µm in diameter, niches of therapy-surviving tumor cells. Magnification of images from these cell reservoirs revealed their organization into a network of vascular-like structures formed by eGFP tumor cells as well as cells expressing both RFP and eGFP reporters (resulting in yellow fluorescence); thus, putative tumor stem cells (Fig. 4A,B, bottom panels).
Conventional immunohistochemical analysis of the treated tumors confirmed that after GCV treatment, the remaining small tumor area consisted of cellular structures and voids reminiscent of capillary networks (Fig. 4C). We thus propose that in response to anti-proliferative treatment, the regeneration capacity of a tumor resides in a surviving niche of specialized, non-replicating tumor cells that express the CD133 and OCT4/SOX2 stem cell markers and form tubular-like structures.
To determine whether the tubular structures found after GCV treatment were tumor cells differentiated into endothelial cells or pericytes, we analyzed them using immunohistochemistry procedures for the presence of specific lineage markers. However, we rarely observed colocalization of eGFP-expression and the specific endothelial or pericytic cell markers lectin and desmin, respectively ( The capacity of differentiated tumor endothelial cells and pericytes to recapitulate tumor growth. The remarkable therapy resistant tubular-like structures suggest the possibility that cell types with GSCs properties may be preexistent within the endothelial and/or pericytic compartments of the tumor, and are selected or enriched for by the therapeutic pressure. Alternatively, the tubular structures may be remnants of vasculogenic mimicry, a process whereby tumor cells can form fluid-conducting channels mimicking a vascular function, being in this case enriched in GSCs as a result of therapy. In order to verify these hypotheses, and its generality, SCID mice, 20 per group, were inoculated with either U87-eGFP-PLuc, NCH644-eGFP-PLuc or NCH421k-eGFP-PLuc cells (the latter two are aggressive diffusely growing tumors, see Fig. S3) and allowed to grow. When the tumor reached the appropriate size; tumors were harvested and disaggregated into individual cells. Cells were sorted for simultaneous expression of eGFP (tumor cells) and two different endothelial cell markers (CD31 and CD105) and collected as double positive or double negative cell pools. Alternatively, cells were sorted for the expression of eGFP and two pericyte markers (CD146 and CD248) and also collected into double positive and double negative cell pools. The four types of cell pools were grown under tumorsphere forming conditions to evaluate their tumorsphere generating capacity.
The first relevant observation was that double negative and double positive cell pools were capable of producing tumorspheres.
In vivo results showed that, regardless of the tumor type of origin, tumorspheres from both CD31+ CD105+ cells and CD146+ CD248+ cells were able to generate tumors that killed the host.
As expected, tumorspheres from cell pools negative for either endothelial or pericytic markers that also comprise "non-vascular" GSCs, also recapitulated tumors and killed their hosts (Fig. 6A-F, left and middle panels).
Kaplan Meier plots of animal survival show no statistical difference in the capacity of tumors from the different tumorsphere types to kill mice.
Thus, these data indicate that both, double positive cells of endothelial and pericyte lineages derived from tumors are able to dedifferentiate and recapitulate tumors, pointing at the tumor vascular system as a reservoir of tumor-initiating cells.
Moreover, it appears that the abundance of tumor cells with the capacity to recapitulate tumors is similar in the tumor vascular compartment (endothelial and pericytic lineages) and the rest of the tumor.

Discussion
In previous work we observed that cytotoxic tTK expressing hAMSCs that targeted the GSC niche differentiated to vascular components in close proximity to CD133 expressing GBM cells and were capable of reducing tumor burden by a factor or 10 4 in the presence of GCV. These results led us to speculate that cell therapy was targeting the GSC niche. However, even after prolonged treatment, release from the therapeutic pressure would result in tumor recurrence 31 .  www.nature.com/scientificreports www.nature.com/scientificreports/ In the current work, we test if a therapeutic strategy directly targeting GSCs could eliminate a GBM U87 tumor completely. We designed a strategy to express in all tumor cells PLuc and eGFP reporters and, RLuc, RFP and tTK genes in the subpopulation of GSCs with active CD133 or OCTA4/SOX2 promoters. In this manner, the tTK gene would allow us to selectively "kill at will" replicating tumor cells that activated the GSC specific  www.nature.com/scientificreports www.nature.com/scientificreports/ promoters. CD133 and OCT4/SOX2 promoter functionality and responsiveness were validated by demonstrating in vitro cell killing capacity in the presence of GCV and production of renilla luciferase activity in response to induction of stemness in tumorspheres.
Following we implanted the targeted cells in live animal brains and the resulting tumors were subject to GCV therapy. In the in vivo experiments, this strategy was very effective at reducing tumor growth. However, it should be noted that targeting tumor cells with an active OCT4/SOX2 promoter had a more potent effect than targeting those with an active CD133 promoter; probably due to the higher frequency of OCT4/SOX2 active cells in our GBM model. Interestingly, the therapeutic effectiveness of targeting GSCs was comparable to that of using tTK expressing hAMSCs in previous work 31 .
Also, as in the case of hAMSC therapy, neither one of the GSC targeting strategies was capable of completely eliminating tumors, which in the current case survived during long treatment periods at nearly undetectable levels, as would be indicated by background levels of PLuc and CD133 and SOX2/OCTA4 regulated RLuc activities. Release of the therapeutic pressure resulted invariably in the increase of luciferase activity, growth of the tumors and death of the animals.
Since the TK/GCV strategy is toxic for replicating cells, we concluded that our therapy might be selecting or inducing a stable, possibly non-replicating pool of GSCs expressing CD133 and OCT4/SOX2 regulated reporters that survived to the cytotoxic agent; a conclusion borne out by our in vitro experiments.
Using the CLARITY procedure in combination with confocal microscopy we were able to observe a scattered distribution of red-fluorescent GSCs within the 3D organization of green-fluorescent tumor cells in undisturbed tumors. Surprisingly, in GCV treated animals, we found very small, therapy-resistant structures made of networks of tubular-like green fluorescent cells also expressing the red fluorescent GSC tracer.
While it is widely accepted that GBM cells can transdifferentiate into endothelial cells 26,27,32 or pericytes 28 , except for very infrequent instances, we failed to immunohistochemically detect endothelial and pericyte markers in the therapy-resistant structures. Thus, suggesting that expression of GSC characteristics could have arisen from "dedifferentiation" of tumor endothelial or pericytic cells.
An alternative explanation could be related to vasculogenic mimicry, a mechanism by which tumor cells form functional vascular-like structures without vascular differentiation 33,34 . In such context, vascular mimicking structures could be a reservoir in which special non-dividing GSCs would become enriched as dividing tumor cells are killed by therapy.
The therapy resistant tubular structures suggested that the tumor vascular compartment could be a, so far undetected, reservoir of GSCs. To explore such hypothesis, we use U87 and 2 other glioblastoma cell lines NCH644 and NCH421k that were isolated from brain tissue and were cultured in serum-free spheroids, making them a good model of glioblastoma explants. We isolated from growing tumors of the 3 different cell lines, green fluorescent tumor cells that simultaneously expressed two endothelial cell markers (CD31 and CD105) or two pericyte markers (CD146 and CD248) and used them, as well as the corresponding depleted cell pools, to produce tumorspheres. Our in vitro tests indicated that, as expected, the tumor cell pools depleted of endothelial or pericytic markers retain the capacity to produce tumorspheres. Surprisingly, cells double positive for two endothelial or two pericyte markers were also capable of producing tumorspheres. Moreover, tumorspheres from the endothelial and pericytic compartments of the three tumor types tested were as capable to produce mice killing tumors as those derived from the depleted controls representing the rest of the tumor.
Our findings suggest reasons for the failure of some therapeutic strategies. We provide first evidence showing that some cytotoxic therapies targeting replicating tumor-initiating cells can dramatically reduce tumor burden but fail to eradicate tumors, inducing instead the formation of vascular-like structures comprising therapy-surviving tumor-initiating cells that will recapitulate the tumor. Moreover, tumor cells positive for endothelial and pericytic markers form a compartment that comprises tumor cell types that are or can be induced to form tumor-initiating cells with the capacity to recapitulate tumors. Lentiviral particle production and cell infection. Lentiviral particles were produced by transfecting HEK293T cells with pMD-G-VSV-G, pCMV DR8.2 and either CMV-PLuc-IRES-eGFP, CD133-RLuc-RFP-tTK or OCT4/SOX2-RLuc-RFP-tTK respectively to HEK293T cells. U87, NCH644, and NCH421K cells were