Two-dimensional (2D) monolayer cell cultures represent a highly reductionist model of cancers due to the loss of physiological extracellular matrix (ECM) on artificial plastic surfaces. Consequently, cells under these conditions adapt by altering gene expression patterns and signaling networks, lose relevant properties such as differentiation, polarization and cell-cell communication, and artificially promote hyper-proliferation [1], [7]. The ease of drug uptake into a homogenous cell population on adherent 2D surfaces, also likely results in an overestimation of efficacy determined from in vitro cytotoxic screens, leading to ineffectiveness of candidate agents in vivo [8]. Moreover, 2D cultures lack complex 3D tissue structure and therefore hamper efforts to test the effectiveness of certain candidate compounds such as anti-angiogenic/vasculogenic agents prior to in vivo strategies [8]. Indeed 2D cytotoxic screens almost exclusively identify anti-proliferative candidate agents [9], [10].

There is a growing recognition conceptually that three-dimensional (3D) culture technologies may more accurately reflect tumor pathophysiology and recapitulate the in vivo tumor microenvironment in vitro. As pharmacological responses in xenograft studies do not consistently correlate with the clinical response, it is imperative that better human-specific predictive models are developed at a pre-clinical stage. Consequently, these models should better inform xenograft studies and early clinical trials in terms of predicting the pharmacological response to chemotherapeutic compounds [11], [12], [13] thereby additionally reducing the requirement of substantial numbers of animals for anti-cancer drug testing.

Multicellular spheroids are small in vitro aggregates ranging from 100–1000 µm in diameter and commonly cultured in suspension using spinner flasks, collagen gels and liquid-overlays [14]. The size of the cultured spheroid is limited to less than 1 mm by shear forces in the culture system and lack of internal cohesion/ECM within the spheroid. The geometry of the spheroid contributes to the formation of heterogeneous cell populations: the outer layer proliferates comparably to monolayer cultures; the inner layer of cells can become quiescent; and the core of spheroids can develop necrosis [15]. The proliferation of tumor cells grown in 3D spheroids is typically slower than that of monolayer cultures, exhibits different metabolic profiles and typically displays reduced sensitivity to chemotherapy and radiotherapy [7], [16], [17], [18], [19], [20]. For example, cytosine arabinoside (Ara-C) and Taxol which inhibit proliferation of glioblastoma cell lines in 2D cultures, but have failed clinical trials, show markedly reduced sensitivity in 3D cultures; an Ara-C dose 10 times higher than the effective dose in 2D cultures was necessary to reduce tumor cell proliferation [21], [22], [23]. A more recent 3D spheroid culture system from the Eccles and colleagues has provided automated, quantitative imaging capability that is potentially compatible with high-throughput preclinical targeting studies and may reduce the requirement of animal studies. Moreover functional assays of tumor migration and invasion are permitted in this system, thus facilitating drug screening that covers a broad range of compounds [24]. Consistently, gene expression signatures of primary cell cultures are maintained in spheroid cultures such as glioblastoma [25], [26].

Biological gels such as Matrigel have also been used as a substrate for spheroid growth. Matrigel is a basement membrane extract derived from the Engelbreth-Holm-Swarm mouse sarcoma that contains a diverse array of components, including collagen type IV, laminin and other ECM molecules, in addition to soluble signals such as cytokines and growth factors [27]. LaBarbera and colleagues demonstrated that tumor-related phenomena such as epithelial-mesenchymal transition in metastatic breast tumors can be recapitulated in 3D matrigel-based spheroid culture, a system additionally used as a high-throughput screening tool for small molecule inhibitors for breast carcinoma [28]. However, as Matrigel is a largely undefined and variable mixture of proteins, it has not been used widely for drug screening purposes. Porous collagen/hyaluronic scaffolds provide a more physiologically relevant in vitro model that permits the interactions between different cell types and ECM as described recently for mammary epithelial and adipocyte co-cultures [29]. Other breast cancerous 3D models have been described recently in which malignant breast cancer cells are embedded in reconstituted basement membrane with direct co-culture of endothelial cells, enabling investigation of the stimulatory effects of the latter cell type [30]. Suspension cultures such as spinner flasks have been investigated but are often hindered by shear stress and turbulence [31] effects experienced by the cells.

One method of generating 3D aggregates in a non-turbulent environment and where cells secrete endogenous ECM and growth factors is the Rotary Cell Culture System (RCCS™), originally developed by the National Aeronautics and Space Administration (NASA) to study the effects of microgravity on cells and tissues [32]. The system has been successfully used in tissue engineering [33], developmental biology [34] and microbiology [35]. The RCCS method has been extended to tumor cell propagation (e.g. in prostate cancer [36] and colorectal cancer [37]) where minimal shear force and continually suspended ‘free-fall’ culture encourages cell aggregation. This permits biological/biochemical processes to develop under closely monitored environmental and operational conditions and encourages cells to form 3D tissue-like aggregates which display physiologically relevant phenotypes such as lower proliferation rate, hypoxic regions and vasculature [1]. No previous publications have specifically examined brain tumor growth in the RCCS or made comparison between 2D culture, 3D culture and actual brain tumor specimens. Analysis of gene expression, microRNA and metabolic profiles demonstrates that our RCCS aggregates (in comparison to 2D monolayer cultures) display a genotype significantly more similar to actual brain tumors and may thus replicate tumor behavior (e.g. response to therapeutic agents) more faithfully. In addition, 3D aggregates can be cultured for at least up to 4 weeks, compared to approximately one week in 2D monolayer culture (due to cell confluency), providing a means to assess drug efficacy and repeated dosing over a longer time period. No artificial matrix or ECM components are introduced, allowing cells to produce their own structures and with no artificial barriers to drug diffusion. The increased size and heterogeneity of RCCS aggregates compared to other culture methods e.g. neurospheres or multicellular spheroids, increases their physiological relevance, allowing drug testing against a simulation of an entire tumor rather one selected cell line sub-clone as in 2D or neurospheres. Critically, the RCCS aggregates display marked internal heterogeneity with separate populations of proliferative, senescent and necrotic cells, influenced in part by the changing oxygen concentration from rim to core of the aggregate. This uniquely recapitulates the variety of in vivo environmental niches within which cancer cells reside and which are now thought critical to tumor development and progression.

It is increasingly clear that cancer is a process involving complex interactions between tumor cells and the surrounding microenvironment, including non-neoplastic cells, ECM and environmental conditions such as oxygen tension [54], [55]. Critically, these conditions vary within different regions of high grade brain tumors with a characteristically hypoxic necrotic core; an intermediate zone where cells survive but under hypoxic strain and the periphery of the tumor where cancer cells invade into relatively well oxygenated brain. Models have been proposed whereby this heterogeneity of conditions gives rise to different sub-populations of tumor cells which fulfill different functions and niches within the tumor as a whole. Much interest has focused on the immediate vicinity of small tumor associated blood vessels within the brain, termed the microvascular niche, as a possible favorable habitat for the stem-like cells which may be critical in tumor propagation and resistance to treatment [56], [57]. It may be that different areas of the tumor may favor other sub-clone populations such as proliferative or invasive cells. Monolayer cell culture creates a relatively homogeneous population of cells that are all subject to the same selection pressures. Heterogeneity between different areas within the same tumor has long been recognized histologically, but has also now been demonstrated molecularly with different driving genes amplified in different areas of a glioblastoma [6]. This heterogeneity may partially explain the failure of targeted therapies, with only a subset of tumor cells being potentially sensitive to a particular compound. The RCCS model recapitulates heterogeneity histologically, allowing more representative drug testing against the entirety of a tumor rather than solely the proliferative fraction present in a 2D monolayer.

Monolayer cell culture typically takes place in room air (21% oxygen content) with supra-physiological concentrations of glucose and other metabolites in the feed medium. The normal brain oxygen saturation is around 7% with tumor regions ranging from 5% to less than 1% [58]. Cell culture can be performed in a hypoxic chamber, though this still does not replicate the gradient in oxygen concentration seen in tumors in vivo. Previous studies have demonstrated that the oxygen concentration to which cells are exposed is of critical importance in defining the behavior of those cells in high grade brain tumors [59], [60], [61]. This may have direct therapeutic implications in areas such as the derivation of dendritic cell based anti-cancer vaccines. It also seems that the gene expression profile of tumors is affected by region, with biopsies taken from more ischemic/necrotic regions displaying a more mesenchymal genotype [62]. The behavior of cancer stem-like cells is also known to be profoundly impacted by their environmental oxygen levels [63], [64], [65]. Many processes involved in the development of CNS tumors are also by definition 3D such as angiogenesis and interaction with the ECM. Again, these processes are not re-capitulated in 2D culture which is a highly reductionist model. We have demonstrated that an oxygen gradient exists across the RCCS aggregates replicating a key feature of the environmental conditions to which brain tumors are exposed. This seems likely to have implications for cell behavior within the culture model, crucially replicating the exposure of different areas of the cell population to different conditions within the same tumor. Members of our group have previously demonstrated similar findings in breast cancer where the hypoxic, relatively harsh micro-environment seems to select for a more aggressive metastatic phenotype [66].

In vivo animal models typically consist of xenotransplantation of a human cancer cell line into an immunocompromised rodent host. Orthotopic grafting replicates many facets of the CNS environment in which human tumors develop and some processes such as development of a vascular network. However, these models all have a rodent host as their basis and the disappointing record of therapeutic agents taken into phase I/II trials following animal studies demonstrates how specific tumor/host environment interactions are critical in governing tumor behavior. It is as yet unclear whether, for example, interactions between tumor and endothelial cells are species specific. The immune system (known to have a key role in human brain tumors) in these animals also behaves very differently to that in a normal human, further reducing the strength of animal models. Animal experiments have other drawbacks such as their cost and the ethics of utilizing large numbers of animals in research with an increasing pressure to reduce the use of animals in research and drug development.

Other 3D models have been used to try and overcome these issues and the most widely used of these in cancer and stem cell research is the non-adherent neurosphere culture. It is well recognized that this culture method increases the proportion of cells displaying stem-like properties and markers [67]. Other advanced culture methods used to model tumor growth include the hanging drop method (creating similar entities to neurospheres) [68], hollow fibers [69], microfluidic chambers and multicellular layer constructs [70]. The latter three methods still utilize support membranes or fibers which provide an artificial physical contact point for cells and do not replicate true unrestricted growth in 3D. Neurospheres range in size from around 50 µm to a maximum of around 300 µm in diameter (compared to 1–9 mm for RCCS aggregates). At their largest size they begin to exhibit a necrotic core [71] in a similar fashion to RCCS aggregates, though the effect is significantly less pronounced. Neurosphere culture generally involves culture media that differ from the standard preparations used for monolayer culture; for example neurosphere medium is usually serum free and supplemented with epidermal growth factor and basic fibroblast growth factor. The RCCS aggregates are formed in the same medium as used for 2D cell culture, thus removing the influence of media composition on comparisons between RCCS and 2D culture using the same cell line. The recent demonstration of a 3D culture system with automated and quantitative analysis, similarly generates tumor spheroids with a size range of 300–500 µm in diameter [24]. Although matrix invasion and angiogenic differentiation was induced when tumor cells were co-cultured with embryoid bodies, we believe that the ~10-fold larger RCCS aggregates display more acute cellular heterogeneity with clearly delineated proliferating rim, necrotic core and peri-necrotic rim with direct observation of hypoxic stress and senescent cells in this region. Indeed we have also shown that a proportion of RCCS brain tumor aggregates exhibit a vasculogenic phenotype and angiogenic gene expression profile in the absence of any form of co-culture, a phenomenon termed ‘vasculogenic mimicry’ (Smith et al, manuscript in preparation). Moreover as the immediate tumor microenvironment is generated by the secretion of endogenous ECM rather than exogenous and artificial substrates, the resulting paracrine signaling and resulting global gene expression changes in the RCCS may be more informative than other 3D methods. In support of the importance of cancer cell-ECM interactions with regards to mimicking physiological aspects of tumor biology, Loessner and colleagues describe a 3D culture consisting of a sophisticated synthetic hydrogel matrix with incorporated biomimetic features. This method permitted exposure to similar biochemical and biomechanical stimuli in all directions due a homogenous dispersion of ovarian tumor cells. Importantly, increasing matrix stiffness inhibited spheroid growth resulting in a more compact spheroid. A recent high-throughput automated 3D culture system based upon microfluidic channels similarly permits the generation of homogenous and densely packed tumor cell aggregates in a reproducible manner, resulting in increased drug resistance compared to 2D cultures. These findings implicate cell density, cell-cell interactions and cell-ECM interactions as contributing mechanism by which the proliferation rate of cells in 3D may be relatively slower than corresponding 2D cultures and additionally why 3D cultures may be relatively more drug-resistant [72], [73]. We have profiled the same cell lines after culture in 2D and 3D RCCS, demonstrating statistically significant differences in the expression levels of many key genes known to be involved in the neoplastic process. Profound changes take place between 2D and 3D in the expression of genes involved in areas such as ECM, cell proliferation and stem-like characteristics. These changes alter the overall genetic profile of the sample, rendering them significantly more similar to the profiles exhibited by our cohort of primary pediatric high grade gliomas, though there are clearly still differences between 3D cultures and clinical tumors. Investigation of agents targeting specific molecular pathways is again likely to be more efficacious in model systems such as the RCCS which more closely replicates gene expression and protein activity levels observed in clinical tumors. Sensitivity to drugs will depend upon the exact expression levels of many genes and it is clear that radically different and less meaningful results are likely to be achieved in drug testing using 2D compared to 3D culture.

Similar results are observed when considering the metabolic profiles generated by the in vitro MRS analysis performed. Culture method was found to be the most important factor in generating the metabolic profile and samples cluster accordingly. Lipids are consistently increased in 3D compared to 2D culture, which typically represents increased levels of cell death, accurately reflecting the changes observed in the histology of aggregates with their necrotic centers. Elevated lipid peaks are also typically observed in necrotic regions of actual high grade brain tumors, especially glioblastomas [74]. The decreased phosphocholine observed in 3D culture is consistent with slower growing cells such as observed in vivo after chemotherapy [75] which provides metabolic corroboration of the proliferation rates observed in our study. Moreover a recent study provides evidence that homoeostasis in 3D tumor aggregates is achieved through a balance between cell proliferation, growth arrest and cell death [76], supporting the recapitulation of heterogeneous tumor sub-populations within 3D systems as described here. Increased myo-inositol and glycine may correspond to higher grade behaviors within tumors [77] and their elevation in 3D compared to 2D culture is intriguing and warrants further investigation.

Resistance to chemotherapeutic drugs has been shown to differ significantly for cell lines grown in vivo, in neurospheres or in monolayers [78], [79], [80], [81], [82]. Resistance is generally highest in vivo and lowest in 2D, varying by at least two orders of magnitude. Through exposure to the histone deacetylase inhibitor Vorinostat, our cultures in the RCCS confirm the markedly enhanced resistance conferred by 3D culture (2.5 fold and 13.5 fold for U87 and KNS42 cells respectively) which has significant implications for pre-clinical drug testing strategies and dosage regimes to be utilized in vivo. A significant over-estimate of agent potency is likely using solely 2D data. Even when considering Vorinostat exposure in brain tumor neurosphere cultures (IC₅₀ 0.5 µM) that enhance for cancer stem/progenitor cells [83], the RCCS aggregates are relatively more resistant indicating that the RCCS may provide a better in vitro system to evaluate agents targeting stem/progenitor pathways. A key feature of RCCS culture is the creation of a tissue barrier between agents applied exogenously and cells deep within the aggregate. This creates the necessity for agents to penetrate significant depths into tissue, replicating in vivo conditions. This may be further compounded by the formation of cell-secreted ECM within aggregates that may limit drug penetrance and enhance integrin-mediated pro-survival signaling [73], [84]. Such considerations are of key importance when considering the interaction between drugs and also the genetic and epigenetic profiles emerging from various methods of culture. The more resistant phenotype of RCCS aggregates may be directly associated with a lack of homogenous drug exposure within the culture. However, our results do not exclude the possibility that the RCCS microenvironment facilitates the creation of a niche that enriches for and/or harbors brain tumor sub-populations that are intrinsically resistant to Vorinostat. It is likely that both phenomena contribute to the relatively more resistant phenotype in 3D compared to 2D cultures. The increased resistance shown by aggregates containing areas of hypoxic cells may be consistent with enrichment of glioma stem-like cells in hypoxic and vascular niches that may be intrinsically resistant to chemotherapy [15], [57], [63].

Importantly, our findings contrast to a recent study assessing the efficacy of Vorinostat against ependymoma brain tumor stem cells propagated using the neurosphere assay. Witt and colleagues report an IC₅₀ value of 0.78 µM when ependymoma neurosphere cultures were treated to a clinically achievable concentration range of Vorinostat and abolition of neurosphere initiating capacity (i.e. self-renewal capacity) at an IC₅₀ of 0.5 µM [83]. Based solely on IC₅₀ values, the ependymoma neurosphere cultures were 16–17 fold more sensitive than our RCCS aggregate cultures, supporting the notion that the RCCS tumor microenvironment induces greater intrinsic tumor cell resistance and/or better reflects difficulties in achieving effective drug dose penetration in vivo. However, we cannot exclude the possibility that differences in Vorinostat efficacy is at least in part due to the different cell types and lines used in each study. In a similar finding, treatment of GBM neurospheres with the telomerase antagonist Imetelstat, led to an IC₅₀ value of 2 µM after prolonged 4-week exposure [85] enforcing the hypothesis that neurospheres are relatively more sensitive than RCCS aggregates. IC₅₀ values detected in the micromolar range in our study highlight a caveat of standard 2D drug screens where an agent is typically discarded if its potency is not in the nanomolar range. Conversely there are instances of clinically failed anti-cancer compounds with nanomolar potency in vitro [86]. We propose introducing the RCCS as a second tier preclinical efficacy/toxicity testing tool, whereby targets forwarded from 2D drug screens will be further scrutinized in the RCCS prior to the uptake of lead compounds for in vivo trials. Furthermore it will also be important to consider sequential exposure of RCCS aggregates to drug combinations as enhanced doxorubicin accumulation and toxicity has recently been shown when tumor spheroids were pre-treated with mitoxantrone or paclitaxel [87].

We have also demonstrated that the RCCS can be used as a system for maintaining viable explants of primary tumor direct from surgery for at least 3 weeks, a distinct advantage over alternative 3D spheroid culture systems [24], [28], [29], [30]. Viable dividing cells are maintained and internal architecture is preserved including the presence of endothelial cells. The centre of explants develops necrotic areas in a similar fashion to cultured aggregates, indicating a similar tolerance level to hypoxia. Further investigations will encompass more clinically-relevant drug testing against these primary explants which may allow prediction of the response of the tumor in the patient to the same agent. Array real-time-PCR shows clear differences in gene expression between early passage cell lines derived from the tumor and subsequently cultured in 2D and tumor explants cultured in the RCCS.

Figure S1

Table S1

Table S2

Table S3

Table S4

Table S5

Table S6

Table S7