Registered report: Tumour vascularization via endothelial differentiation of glioblastoma stem-like cells

The Reproducibility Project: Cancer Biology seeks to address growing concerns about reproducibility in scientific research by conducting replications of 50 papers in the field of cancer biology published between 2010 and 2012. This Registered report describes the proposed replication plan of key experiments from ‘Tumour vascularization via endothelial differentiation of glioblastoma stem-like cells’ by Ricci-Vitiani and colleagues, published in Nature in 2010 (Ricci-Vitiani et al., 2010). The experiments that will be replicated are those reported in Figure 4B and Supplementary Figure 10B (Ricci-Vitiani et al., 2010), which demonstrate that glioblastoma stem-like cells can derive into endothelial cells, and can be selectively ablated to reduce tumor progression in vivo, and Supplementary Figures S10C and S10D (Ricci-Vitiani et al., 2010), which demonstrate that fully differentiated glioblastoma cells cannot form functionally relevant endothelium. The Reproducibility Project: Cancer Biology is a collaboration between the Center for Open Science and Science Exchange, and the results of the replications will be published by eLife. DOI: http://dx.doi.org/10.7554/eLife.04363.001


Introduction
Glioblastoma multiforme (GBM) is a highly aggressive form of cancer characterized by an extensive network of vasculature that contributes to its invasiveness. However, the mechanisms of angiogenesis and the origin of the tumor vasculature remain poorly understood. While conventional theory suggests that GBM tumor vasculature derives from existing vessels or from bone marrow progenitor cells, recent studies have indicated that this is not the case (Purhonen et al., 2008). Indeed, the tumor endothelium may actually be derived from the tumor itself. GBM is maintained via a population of selfrenewing, tumorigenic cancer stem cells (CSCs), which have been implicated in tumor invasion and metastasis for a large variety of cancers (Vescovi et al., 2006;Fan et al., 2013). The progeny of these CSCs are not confined to a neural lineage but rather can differentiate into functional endothelium. Based on the work of Ricci-Vitiani et al., it appears that part of the vasculature in GBM originates from tumor cells, bypassing the normal mechanisms of angiogenesis. These findings offer insights into tumor self-renewal, and offer new options for cancer treatment by targeting the differentiation of tumor cells into endothelial progeny (Ricci-Vitiani et al., 2010;Kaur and Bajwa, 2014).
In order to demonstrate the CSC lineage of GBM tumor vasculature, Ricci-Vitiani et al. first traced the genetic lineage of the tumor endothelium. They analyzed the vasculature in 15 human glioblastoma patient samples and found that a large subset of endothelial cells harbored the same mutations and chromosomal aberrations as the tumors themselves. They also showed that in culture, glioblastoma stem-like cells (GSCs) could be differentiated to express multiple endothelial markers, and showed substantial tube-forming ability, whereas fully differentiated glioblastoma cell lines could not. Ricci-Vitiani et al. also traced the lineage of tumor vasculature in vivo, confirming the presence of human endothelial cells expressing green fluorescent protein (GFP) in a mouse xenograft of human GSCs expressing GFP. These experiments, and others, demonstrated that tumor xenografts obtained by injection of human glioblastoma neurospheres developed an intrinsic vascular network composed of tumor-derived endothelial cells (Ricci-Vitiani et al., 2010).
The authors next sought to investigate whether the GSC-derived endothelial cells contributed to tumor growth. This key finding is the focus of this replication study. The authors transduced glioblastoma neurospheres with a lentiviral vector containing the herpes simplex virus thymidine kinase gene (tk) under the control of the transcription regulatory elements of the endothelial marker Tie2. In this way, the tumor-derived endothelial cells could be selectively killed by exposure to ganciclovir. As negative controls, the authors used neurospheres transduced with an empty viral vector, as well as the differentiated glioblastoma cell line U87MG. As a positive control, they used neurospheres and U87MG cells transduced with a vector conferring constitutive expression of tk (PGK-tk). Upon administration of ganciclovir, selective targeting of endothelial cells generated by GSCs in mouse xenografts resulted in tumor reduction and degeneration, indicating the functional relevance of the GSC-derived endothelial vessels.
Prior to subcutaneous injection of transduced glioblastoma neurospheres, Ricci-Vitiani et al. first confirmed the lack of endogenous Tie2 expression in both the GSCs and the U87MG cells. This quality control step will be replicated in Protocol 1, the results of which will be compared to Figure S11C. Next, the viral transduction of GSCs and U87MG cells with expression constructs for Tie2-tk and PGK-tk, as well as empty viral vector, will be replicated in Protocol 2. The generation and analysis of xenografts using various cell lines in immunocompromised mice will be replicated in Protocol 3. This protocol will generate data that can be compared to original data presented in Figures 4B, S10B, S10C, and S10D.
Recently, multiple other studies have explored the phenomenon of tumor-derived vasculature in GBM. Two recent reports found results very similar to those of Ricci-Vitiani et al., showing that oncogeneinduced glioblastoma tumors gave rise to tumor-derived endothelial cells, as indicated by GFP expression. These studies also found that endothelial cells within tumors harbored the same genetic signature as the tumor itself (Wang et al., 2010;Soda et al., 2011). Similarly, Chiao et al. reported that GCSs formed vasculogenic mimicry in tumor xenografts and expressed pro-vascular molecules (Chiao et al., 2011). However, other groups have found that endothelial cells comprising GBM vasculature do not share the same genetic make-up as neoplastic tissues, and that GSCs routinely do not give rise to endothelial cells (Rodriguez et al., 2012;Cheng et al., 2013). Interestingly, Cheng et al. present an alternative hypothesis to that of Ricci-Vitiani et al. by showing that GSCs can give rise to vascular pericytes-which also express Tie2 (De Palma et al., 2005)-rather than endothelial cells. Targeting these GSC-derived pericytes disrupted vessel function and inhibited tumor size similarly as the results presented by Ricci-Vitiani et al. for targeting endothelial cells (Cheng et al., 2013). El Hallani et al. demonstrated that GSCs were capable of vasculogenesis in vitro, and that a fraction of GSCs could transdifferentiate into vascular smooth musclelike cells. However, their later work suggested that rather than transdifferentiating, the GSCs were fusing with endothelial cells to create a hybrid tumor vasculature (El Hallani et al., 2010;El Hallani et al., 2014). Conversely, using a GSC mouse xenograft, Lathia et al. did not observe the integration of tumor-derived cells into the vascular wall; however, this observation was only reported in the text. No data were shown (Lathia et al., 2011). Ghanekar et al. tried analogous experiments using hepatocellular carcinoma CSCs and did not find any evidence that the tumor cells gave rise to the endothelium (Ghanekar et al., 2013).

Materials and methods
Unless otherwise noted, all protocol information was derived from the original paper, references from the original paper, or information obtained directly from the authors.

Protocol 1: Evaluation of Tie2 expression in various cell lines using qPCR
This protocol evaluates the expression of the endothelial marker Tie2 in three cell lines using semi-quantitative PCR: patient-derived glioblastoma neurospheres (GSC83), human glioblastoma cell line U87MG, and normal human dermal microvascular endothelial cells (HMVEC-d). The expression of Tie2 will be normalized against the endogenous expression of 18S rRNA. Expression of Tie2 is expected to be very low in GSC83 and U87MG cells, and robust in the endothelial cell line HMVEC-d, as depicted in Figure S11C. This protocol serves as a quality control step to ensure the lack of Tie2 expression in the glioblastoma cell lines used later in the study. Sampling 1. This experiment will be performed three times (biological replicates) with each run using two technical replicates, for a final power of at least 80%. A. Test conditions: i. qRT-PCR of Tie2 (and 18S rRNA) from GSC83 glioblastoma neurospheres.

Materials and reagents
Procedure Note: All cell lines will be sent for STR profiling and mycoplasma testing. A. GSC83 cells should be plated at 20,000 cell/ml and subcultured once every 7 days at the same plating number. One week is sufficient time for two doublings to occur. i. Cells should be cultured in stem cell medium consisting of serum-free Dulbecco's Modified Eagle's Medium (DMEM)/F-12 containing: a. 20 ng/ml human recombinant epidermal growth factor (EGF). b. 10 ng/ml human recombinant basic fibroblast growth factor (FGF2). c. 2 mM glutamine. d. 0.6% glucose. e. 9.6 μg/ml putrescine. f. 6.3 ng/ml progesterone. g. 5.2 ng/ml sodium selenite. h. 0.025 mg/ml insulin. i. 0.1 mg/ml transferrin. B. U87MG cells should be cultured in 25 cm 2 tissue culture flasks in DMEM with 10% FBS. i. Subculture cells at a ratio of 1:2 to 1:5; renew medium 2-3 times per week. C. HMVEC-d (normal human dermal microvascular endothelial cells) should be cultured in 25 cm 2 tissue culture flasks in Endothelial Growth Medium-2 Microvascular (EGM-2MV). i. Subculture cells when they are 70-80% confluent; change growth media every other day. 2. Split each cell line into three separate 25 cm 2 flasks. These separate flasks constitute biological replicates for eventual downstream gene expression analysis.
A. Allow cells to grow to log phase. 3. Isolate total RNA from the cells in each 25 cm 2 flask (nine flasks in total) according to the manufacturer's instructions for TRI reagent. For each sample, harvest the entire population of cells in the flask. A. Report total concentration and purity of isolated total RNA. 4. Reverse transcribe mRNA to cDNA according to the manufacturer's protocol.

Confirmatory analysis plan
This replication attempt will perform the statistical analyses listed below, compute the effect sizes, compare them against the reported effect size in the original paper and use a meta-analytic approach to combine the original and replication effects, which will be presented as a Forest plot.
1. Statistical analysis of the replication data: A. One-way ANOVA to analyze the means of GSC83, U87MG, and HMVEC. i. We will then perform a Fisher's LSD test to perform multiple pairwise comparisons: a. GSC83 compared to HMVEC. b. U87MG compared to HMVEC. c. GSC83 compared to U87MG (sensitivity).

Known differences from the original study
In the original study, multiple human glioblastoma neurospheres were screened for Tie2 expression. The human glioblastoma cell lines U251 and T98G were also analyzed, as well as the human endothelial cell line HUVEC. This replication study will be using a single established glioblastoma neurosphere cell line (GSC83) provided by the authors. The authors will also provide their U87MG and HMVEC cell lines. All known differences in reagents and supplies are listed in the 'Materials and reagents' section above, with the originally used item listed in the 'Comments' section. All differences have the same capabilities as the original and are not expected to alter the experimental design.

Provisions for quality control
The cell lines used in this experiment will undergo STR profiling to confirm their identity and will be sent for mycoplasma testing to ensure there is no contamination. The sample purity (A 260/280 and A 260/230 ratios) of the isolated RNA from each sample will be reported. All data obtained from the experiment-raw data, data analysis, control data, and quality control data-will be made publicly available, either in the published manuscript or as an open access dataset available on the Open Science Framework project page for this study (https://osf.io/ mpyvx/).

Protocol 2: Lentiviral infection of glioblastoma cells and stable cell generation
This protocol describes the methods necessary to virally transduce GSC83 glioblastoma neurospheres, as well as U87MG cells, with thymidine kinase expression constructs. The protocol first details the production of three different lentivirus strains (PGK-tk, Tie2-tk, and an empty viral vector), and then explains the techniques necessary to transduce the two human glioblastoma cell lines. Finally, the protocol includes methodology associated with assessing the transduction efficiency of glioblastoma cell lines via flow cytometry analysis as a quality control check. A. Remove an aliquot of cells (20,000-50,000 cells) from each flask. Untransduced cells (both GSC83 and U87MG) should also be prepared for use as a negative control. i. Pull down the cells by centrifuging each flask at ≤1000 rpm. ii. Remove the supernatant, leaving approximately 150-200 μl of media in the flask. iii. Use a P200 pipette to gently dissociate the cells. Pipette up and down several times to obtain a single cell suspension. iv. Save an aliquot for flow analysis, and passage the remaining cells into a new flask to expand them for further experiments. B. Incubate the freshly dissociated cells for 5 min with 7-amino actinomycin D (7-AAD; final concentration 5 μg/ml). C. Analyze cells for GFP expression using a FACSCalibur instrument. Apply the following sequential gates to the dot plots to select viable cells:

Materials and reagents
i. FSC area/SSC area. ii. SSC width/SSC area to exclude aggregates. iii. FSC area/7-AAD area to select viable cells. iv. Untransduced cells serve as a negative control. D. Plot fluorescent protein expression in gated cells using bivariate plots. 11. Determine the percentage of transduced cells positive for GFP reporter expression in each population of cells.
A. Exclusion criteria: Expression should be ≥80% positive in each transduced population in order for cells to be used for xenograft injection. 12. Continue to expand and culture cells until ready for injection into immunocompromised mice.

Confirmatory analysis plan
1. Statistical analysis of the replication data: Not applicable.

Known differences from the original study
All known differences in reagents and supplies are listed in the 'Materials and reagents' section above, with the originally used item listed in the 'Comments' section. All differences have the same capabilities as the original and are not predicted to alter experimental outcome.

Provisions for quality control
Endotoxin-free plasmid DNA for expression constructs will be analyzed for concentration and purity. In order to verify the construction of Tie2-tk and PGK-tk constructs, restriction digestion mapping will be performed. Banding pattern will be compared to expected band sizes based on plasmid maps received from the original authors. Flow cytometry data will be analyzed using the software package FlowJo and the achieved transduction efficiency (%GFP + cells) will be calculated for each infected cell population. Untransduced, wild-type cells (both GSC83 and U87MG) will serve as a negative control for flow cytometry. 7-AAD will be used to exclude dead cells from flow analysis. All data obtained from the experiment-raw data, data analysis, control data, and quality control data-will be made publicly available, either in the published manuscript or as an open access dataset available on the Open Science Framework project page for this study (https://osf.io/mpyvx/).

Protocol 3: Monitoring xenograft tumor size after selective targeting of cells with ganciclovir
This protocol is designed to test whether GSC-derived endothelial cells can contribute to tumor growth in vivo. Virally transduced glioblastoma neurospheres expressing the herpes simplex virus thymidine kinase gene (tk) under the control of the endothelial marker Tie2 are subcutaneously injected into immunocompromised mice. Following tumor formation, mice are treated with ganciclovir, which selectively kills any cells expressing tk. Negative controls include neurospheres transduced with an empty viral vector, as well as the differentiated glioblastoma cell line U87MG, which should not give rise to endothelial cells. Positive controls include neurospheres and U87MG cells transduced with a vector conferring constitutive expression of tk (PGK-tk), which should target all tumor cells. Selective targeting of tk-expressing tumor cells should result in tumor reduction and degeneration, indicating the functional relevance of the GSC-derived endothelial vessels (as shown in Figures 4B and S10B).

Sampling
1. These experiments will utilize at least three mice per treatment group, for a minimum power of 80%.
A. See 'Power calculations' section for details. D. Untransduced (wild-type) versus empty vector (sensitivity). b. Following the one-way ANOVA, the following planned pairwise comparisons will be made using the Bonferroni correction to account for multiple comparisons: 1. For mice implanted with U87MG cells: A. Tie2-tk versus PGK-tk. B. Tie2-tk versus empty vector (sensitivity). C. PGK-tk versus empty vector. D. Untransduced (wild-type) versus empty vector (sensitivity). ii. The authors originally examined the percent changes in tumor diameter between mouse treatment groups using multiple uncorrected two-tailed t-tests. We will replicate their t-tests, but also use Bonferroni-corrected t-tests within the framework of the ANOVA. B. Comparison of percent volume change in control versus tk-expressing tumors.
i. Differences in percent volume change of tumors before and after GCV treatment will be analyzed as described for percent diameter change above. C. Comparison of tumor growth rates.
i. We will measure tumor growth rates across all mouse cohorts over the length of the study, both before and after GCV treatment. These data were not analyzed in the original study, so we consider them exploratory data. We will plot tumor growth curves for each animal and calculate the area under the curve (AUC) before and after GCV treatment. We will perform an ANCOVA on the different treatment groups to evaluate the AUC after GCV treatment, with the baseline (AUC before GCV treatment) included as the covariate. Further, we will perform Bonferroni corrected t-tests for pairwise comparisons between controls and tk-expressing tumors.

Known differences from the original study
The methods section in the original paper stated that mice were dual-injected with both control and Tie2tk expressing neurospheres into the right and left flanks, respectively, and bilateral tumors were allowed to form. However, subsequent dialogue with the authors clarified that mice actually only received a single injection, as dual injections often led to problematic differences in tumor growth rates. Therefore, we will be using a single-injection model, where mice will be either injected with tk vectors or controls, but not both simultaneously. We will only be comparing GSC83-derived cell lines and U87MG-derived cell lines, excluding the other GSC lines used in the original study. Along with measuring differences in tumor diameter, we will also be measuring tumor volume throughout the course of the study. In the original study, it was not specified how many tumors were harvested and histologically analyzed. We have elected to harvest a random subset of tumors that represent all treatment groups. All known differences in reagents and supplies are listed in the 'Materials and reagents' section above, with the originally used item listed in the 'Comments' section. All differences have the same capabilities as the original and are not expected to alter the experimental design.

Provisions for quality control
The genetic integrity, mycoplasma-free and rodent pathogen-free purity, and efficient viral transduction of each cell line used in this experiment have been previously validated in Protocols 1 and 2. All mice will be handled and housed in accordance with the Institutional Animal Care and Use Committee (IACUC). All data obtained from the experiment-raw data, data analysis, control data, and quality control data-will be made publicly available, either in the published manuscript or as an open access dataset available on the Open Science Framework (https://osf.io/mpyvx/).

Power calculations
Unless otherwise stated, all data values are derived from the original paper, or were provided by the original authors.

Protocol 1
Summary of original data provided by Ricci-Vitiani et al.
Normalized Tie2 expression across cell lines ( Figure S11C) Mean SD n *This is a sensitivity calculation. There is no original effect size. #This is a sensitivity calculation. The original effect size is 0.226838.