Kaposi's sarcoma-associated herpesvirus (KSHV or HHV-8) infection of endothelial cells is an early event in the aetiology of the endothelial cell tumour Kaposi's sarcoma (KS). We have examined the effect of the KSHV latent protein viral FLICE-like inhibitory protein (vFLIP) on dermal microvascular endothelial cell (MVEC) survival as vFLIP is expressed in the KSHV-infected cells within KS lesions. To do this, we have used a lentiviral vector to express vFLIP in MVECs in the absence of other KSHV proteins. vFLIP activates the classical NF-κB pathway in MVECs and causes nuclear translocation of RelA/p65. This NF-κB activation prevents detachment-induced apoptosis (anoikis) of MVECs but does not inhibit apoptosis induced by removal of essential survival factors, including vascular endothelial growth factor (VEGF). vFLIP expression inhibits anoikis in part by inducing the secretion of an additional paracrine survival factor(s). The implications of these results for KS development are discussed.
Kaposi's sarcoma (KS) is considered to be a neoplasm of lymphatic endothelium infected with Kaposi's sarcoma-associated herpesvirus (KSHV or HHV-8) (Beckstead et al., 1985; Jussila et al., 1998; Weninger et al., 1999) as KS spindle cells express several lymphatic-lineage-specific proteins, including VEGFR-3 and podoplanin (Beckstead et al., 1985; Jussila et al., 1998; Weninger et al., 1999). However, gene expression microarrays of KSHV-infected endothelial cells showed that both lymphatic endothelial cell (LEC) and blood endothelial cell (BEC) markers are present in the KS signature. This suggests that KSHV infection induces a transcriptional reprogramming of both infected LECs and BECs (Wang et al., 2004). A recent study demonstrated that 50% of patients had circulating endothelial cells infected with KSHV (Pellet et al., 2006). The microvascular endothelial cells (MVECs) that we have used in this study represent a mixed population of LECs and BECs (Makinen et al., 2001) and are highly susceptible to KSHV infection (Wang et al., 2004).
In early KS lesions, only a small proportion (<10%) of spindle and endothelial cells are positive for KSHV (Dupin et al., 1999). Both spindle cells and the infiltrating inflammatory cells express high levels of interleukin 6 (IL-6), basic fibroblast growth factor (bFGF), vascular endothelial growth factor (VEGF), interleukin-1 beta (IL-1β), tumour necrosis factor alpha (TNF-α) and interferon gamma (IFN-γ) (Fiorelli et al., 1998; Miles et al., 1990; Salahuddin et al., 1988). IL-6 promotes growth of KS cells in vitro (Miles et al., 1990), and IFN-γ induces a spindle-cell-like phenotype in endothelial cells (Fiorelli et al., 1998) and also reactivates latent virus (Chang et al., 2000). VEGF functions in synergy with bFGF as a KS cell growth factor, enhancing the development of KS-like lesions when human acquired immune deficiency syndrome (AIDS)-KS cells were injected into mice (Ensoli et al., 1989). Therefore, it appears that, unlike classical tumour cells, KS spindle cells are highly dependent upon paracrine growth signals, and KS lesions are `cytokine-driven' tumours.
Several lytic viral genes encoded by KSHV have been associated with cytokine release in early KS lesions. The viral G-protein-coupled receptor (vGPCR) induces the expression of proangiogenic factors (VEGF, bFGF), proinflammatory cytokines (IL-1β, IL-6, IL-8, TNF-α) and adhesion molecules (Couty et al., 2001; Montaner et al., 2004; Pati et al., 2001; Schwarz and Murphy, 2001). vGPCR and vIL-6 were also shown recently to cause upregulation of angiopoietin-2 (Ang-2) in LECs infected with KSHV (Vart et al., 2007). Furthermore, expression of the viral CC chemokine viral macrophage inflammatory protein II (vMIP-II) results in the upregulation of the expression of multiple proangiogenic factors, including VEGF (Cherqui et al., 2007).
In later nodular KS lesions, >90% of the spindle cells contain latent KSHV (Boshoff et al., 1995; Dupin et al., 1999; Staskus et al., 1997) and express viral FLICE-like inhibitory protein (vFLIP) as one of a small number of latent proteins. vFLIP has been shown to upregulate IL-8 in human umbilical vein endothelial cells (HUVECs) (Sun et al., 2006), and vFLIP is responsible for the spindling phenotype observed in KSHV-infected primary ECs, as well as the production of IL-6 (Grossmann et al., 2006). Here, we examine the effect of vFLIP expression on MVEC survival. vFLIP prevents detachment-induced apoptosis (anoikis) of MVECs but does not inhibit apoptosis induced by removal of essential survival factors such as VEGF. We conclude that expression of vFLIP inhibits anoikis in part by inducing the secretion of an additional paracrine survival factor(s).
MVECs engineered to express vFLIP
Fig. 1 shows the dual promoter vector into which vFLIP was subcloned. vFLIP was inserted under the stronger spleen focus-forming virus (SFFV) promoter, and emerald GFP was expressed from the weaker human polyubiquitin C (Ub) promoter to serve as a marker for transduction. The vFLIP dual promoter vector (vFLIP_eGFP) was used to transduce MVECs for all the experiments, and a GFP vector was used as a control. MVECs were transduced with the vectors at a multiplicity of infection (MOI) of 30, then cultured for 48 hours before assaying for GFP expression. Fig. 1B shows that MVECs were efficiently transduced by lentivectors, with 76% of cells in the cultures staining positive for GFP using the GFP lentivector, and 53% of cells expressing GFP using the vFLIP_eGFP lentivector. Cells were also analysed for GFP expression using a confocal microscope after transduction with either vector at a lower MOI of ∼10. Fig. 1C shows that the population of MVECs that were transduced with the vFLIP_eGFP lentivector displayed elongation similar to the spindle cell shape characteristic of KS tumour cells. Both transduced and nontransduced cells are spindle shaped, suggesting that vFLIP might induce paracrine secretion of cytokine(s) that cause spindling. This result agrees with the findings of Grossman and colleagues, who observed that HUVECs, LECs and BECs expressing the vFLIP protein displayed elongation and spindling virtually identical to that observed with authentic KSHV latency (Grossmann et al., 2006).
vFLIP activates NF-κB pathways in MVECs
vFLIP of KSHV has been shown to be unique among the viral FLIP proteins in its ability to activate both the canonical (Chaudhary et al., 1999) and the alternative NF-κB pathways (Matta and Chaudhary, 2004). vFLIP-mediated activation of the classical NF-κB pathway results in upregulation of p100 expression [as it is encoded by an NF-κB-responsive gene (Dejardin et al., 2002)] and generation of its p52 subunit by activation of the alternative pathway, in tumour and haematopoietic cell lines (Matta and Chaudhary, 2004).
In order to investigate NF-κB induction by vFLIP in MVECs, we used quantitative confocal immunofluorescence microscopy to measure the relative amounts of the NF-κB subunits and their nuclear translocation in the presence and absence vFLIP. Fig. 2 (RelA/p65 staining) shows that expression of vFLIP induced the translocation of p65 into the nucleus; this can be seen as pink nuclei in the vFLIP-expressing cells. The inset histogram shows the intensity of nuclear RelA staining in vFLIP-expressing (green plot) and non-expressing cells; vFLIP clearly increases the level of nuclear RelA. However, p52 and RelB staining showed high nuclear expression even in the absence of vFLIP (see inset histograms).
The relative nuclear translocation of each subunit in MVECs with and without vFLIP expression was also assessed by comparison of the nuclear:cytoplasmic ratios of quantitative immunofluorescence staining (Fig. 3). NF-κB p52 and RelB subunits show principally nuclear staining irrespective of lentiviral transduction using either the GFP-only or the vFLIP_eGFP vector. This striking nuclear localization of RelB and p52 suggests high baseline levels of activation in untransduced MVECs that are not further increased by vFLIP. By contrast, the localization of RelA/p65 is mainly cytoplasmic in untransduced MVECs or cells transduced with the GFP-only vector and shows significant nuclear translocation in cells transduced with the vFLIP_eGFP vector.
vFLIP protects MVECs from anoikis
Endothelial cells die by apoptosis when detached from the extracellular matrix (Meredith, Jr et al., 1993). Transformed cells are very frequently characterised by their ability to grow in the absence of contacts with a solid extracellular matrix (Stoker et al., 1968), which might confer upon them the ability to metastasise. To examine the effects of vFLIP expression on detachment-induced cell death, adherent and detached MVECs were stained for annexin V, which measures the translocation of phosphatidylserine (PtdSer) to the outer layer or the external surface of the cell, a marker for early apoptotic events. In adherent cultures, the basal level for apoptosis was <10% for untransduced and transduced cells alike (Fig. 4A); anoikis was induced after 16 hours of detachment on polyHEMA-coated wells, and both uninfected and GFP-transduced cells entered apoptosis, with 37% and 48% of cells staining positive for annexin V expression, respectively (Fig. 4A). However, cells expressing the vFLIP protein were resistant to anoikis, with only 6% of cells staining positive for annexin V after detachment (Fig. 4A). Note that only 57% of the MVECs expressed vFLIP, as measured by GFP fluorescence (Fig. 4B), implying that anoikis is inhibited in both transduced and untransduced cells in the vFLIP cultures. This suggests protection of untransduced cells by a paracrine survival factor (see below). We obtained similar results when staining cells with propidium iodide (PI); 24% of uninfected cells and 26% of cells transduced with the GFP lentivector were PI positive, in contrast to only 5% of cells transduced with the vFLIP_eGFP lentivector staining positive for PI (data not shown). The results obtained from three independent experiments are summarised in Fig. 4C.
We also measured apoptosis using a DNA fragmentation enzyme-linked immunosorbent assay (ELISA), which quantitates mono- and oligonucleosomes (histone-associated DNA fragments) that are released from cells that die by apoptosis. Cell metabolism was measured using an MTT cell proliferation assay, which is based on the cleavage of the yellow tetrazolium salt MTT to purple formazan crystal by metabolically active cells. As shown in Fig. 5A, uninfected cells, and cells transduced with the GFP lentivector, exhibited a twofold and fivefold increase in DNA fragmentation, respectively, when they were detached. However, MVECs expressing vFLIP showed no increase in DNA fragmentation upon detachment. However, vFLIP could not protect MVECs from removal of the growth factors present in the medium (VEGF, bFGF, EGF and IGF-1), and, when we combined detachment from the matrix with growth factor removal, this removed the protection from anoikis by vFLIP (Fig. 5A). Upon detachment from the matrix, or removal of growth factor, the metabolic activity of the MVECs decreased, and expression of vFLIP did not prevent this (Fig. 5B). Thus expression of vFLIP can protect MVECs from anoikis by blocking the onset of apoptosis, but it is not sufficient to maintain cells in a fully active metabolic state.
vFLIP protects MVECs from anoikis through the NF-κB pathway
Blocking activation of NF-κB induces apoptosis in KSHV-infected primary effusion lymphoma (PEL) cells (Keller et al., 2000) and leads to the downregulation of NF-κB-inducible cytokines (Sun et al., 2006). We therefore investigated whether vFLIP protects MVECs against anoikis by activation of the NF-κB pathway. We inhibited NF-κB using Bay 11-7082, which irreversibly blocks the phosphorylation of IκBα and the subsequent release of active NF-κB into the nucleus (Cahir-McFarland et al., 2004). Fig. 6 shows that treatment of MVECs with Bay 11-7082 reversed the inhibition of anoikis by vFLIP. Interestingly, treatment of attached MVECs with Bay 11-7082 also induced apoptosis in untransduced, GFP- and vFLIP-transduced cells (Fig. 6), suggesting that activation of NF-κB is essential for the survival of MVECs.
The supernatant from vFLIP-expressing MVECs inhibits anoikis
The results of Figs 4, 5 and Fig. 6A suggested that expression of vFLIP not only protects cells that are positive for vFLIP expression but can also rescue neighbouring cells from anoikis. We therefore speculated that expression of vFLIP might induce the secretion of a survival factor. As shown in Fig. 6B, culture supernatant from uninfected or GFP-transduced cells did not protect untransduced or GFP-transduced MVECs from anoikis. However, when detached MVECs were grown in culture supernatant derived from vFLIP-expressing cells, untransduced and GFP-transduced cells were partially protected from anoikis. Indeed, untransduced and GFP-transduced cells exhibited an almost twofold decrease in the levels of DNA fragmentation and there was a significant difference (P<0.05) between these levels and the levels of DNA fragmentation of cells grown in the supernatant of untransduced or GFP-transduced cells. These data suggest that vFLIP-mediated NF-κB activation induces the secretion of one or more survival factors that partly protect cells against detachment-induced cell death.
Our data demonstrate that the vFLIP protein of KSHV activates NF-κB signalling in MVECs and that this activation protects MVECs from anoikis but not from withdrawal of VEGF. Supernatant from the vFLIP-expressing cells partly protects unmodified MVECs from anoikis, suggesting that vFLIP acts both by direct intracellular NF-κB signalling and also by the NF-κB-dependent secretion of a survival factor(s). In these cells, the major change in NF-κB activity that we have detected is the nuclear translocation of p65 induced by vFLIP. It therefore seems that this single NF-κB pathway is responsible both for an intrinsic survival signal and for inducing the secretion of a survival factor(s).
Adhesion of ECs to the extracellular matrix and intercellular adhesion are essential for EC survival and angiogenesis. In the absence of any extracellular matrix (ECM) interactions, ECs rapidly undergo apoptosis (Meredith, Jr et al., 1993). Attachment of ECs on vitronectin or osteopontin (ECM proteins that are known ανβ3-integrin ligands) induces NF-κB activity. The ανβ3-integrin-mediated EC survival effect depends on the osteopontin-induced, NF-κB-dependent gene encoding osteoprotegerin (Malyankar et al., 2000). The phosphoinositide-3-kinase–Akt pathway is also involved in the anti-apoptotic signalling promoted by integrin–cell-matrix interactions (Khwaja et al., 1997; Wary et al., 1996). We propose that vFLIP, in the presence of VEGF and other growth factors, can substitute for the ECM-induced NF-κB signal.
The expression of numerous cytokines necessary for endothelial cell survival is induced by KSHV infection of MVECs, including Ang-2, monocyte chemoattractant protein 1 (MCP-1) and VEGFC (Wang et al., 2004). KSHV infection of primary human umbilical vein endothelial cells (HUVECs) induces the expression of Ang-2 (Ye et al., 2007). Acute productive infection of endothelial cells by KSHV infection leads to potent activation of the NF-κB pathway, which results in the production of high levels of MCP-1 (Caselli et al., 2007). Latent KSHV infection of HUVECs has been shown to result in the upregulation of several chemokines, such as MCP-1, NAP 2 and RANTES, as well as CXCL16 (Xu and Ganem, 2007), which has been associated with endothelial cell chemotaxis, growth and proliferation (Zhuge et al., 2005). All of these cytokines, or perhaps a cytokine combination, are candidates for the survival factor induced by vFLIP.
Studies of X chromosome inactivation patterns in nodular KS lesions that contain latently infected spindle cells suggest that both monoclonal and polyclonal patterns of inactivation exist (Delabesse et al., 1997; Rabkin et al., 1995; Rabkin et al., 1997). A study of size heterogeneity in KSHV terminal repeats in nodular lesions also demonstrated monoclonal, oligoclonal and polyclonal patterns of infection, implying that KSHV infection precedes tumour expansion (Judde et al., 2000). As inhibition of anoikis is a prerequisite for metastasis (Douma et al., 2004), we suggest that one of the actions of vFLIP in these latently infected cells in nodular lesions is to inhibit anoikis. This will allow the detachment and spread of single cells, thereby contributing to the metastasis that is recognised as a hallmark of epidemic KS, resulting in the development of larger monoclonal lesions (Buchbinder and Friedman-Kien, 1992).
Materials and Methods
Virus and plasmid constructs
HIV-1 packaging and VSV-G envelope plasmids were kindly provided by D. Trono (Geneva, Switzerland) and are described elsewhere (Zufferey et al., 1997). The vector plasmid pHR′-CSIW-pUb-Em (double promoter vector) is based on the plasmid pHSIN-CSGW (Demaison et al., 2002) and contains the reporter gene encoding eGFP under the control of the SFFV promoter. This plasmid was modified to express vFLIP by placing it in the position of the original gene encoding eGFP, under the control of the SFFV promoter, and also to express emerald GFP (EmGFP), by cloning the gene encoding EmGFP under the control of the ubiquitin promoter. The vFLIP cDNA was amplified from cDNA synthesised from the BC3 primary effusion lymphoma cell line.
Lentiviral vector production
Lentivirus was produced using a three-plasmid transient-transfection system, as described previously (Besnier et al., 2002; Zufferey et al., 1997). For each plate, DNA plasmids were mixed as follows: 2.5 μg of the packaging plasmid (phCMVΔR8.9) were mixed with 2.5 μg of the envelope plasmid (pMD-G) and 3.75 μg of the transfer vector plasmid, and the volume was made up to 37.5 μl with TE buffer. 45 μl of the transfection reagent Fugene-6 (Roche) were diluted in 500 μl OptiMEM (serum-free medium, GibCo), and this mixture was added to the DNA mix and incubated at room temperature for 15 minutes. Meanwhile, the medium on the cells was changed, and then the transfection mixture was added drop-wise, at the same time swirling the plates to ensure even distribution. 24 hours later, the medium on the cells was changed again to remove Fugene-6. The virus-containing supernatant was harvested 48-72 hours post transfection and passed through a 0.45 μm filter to remove any cells. The virus was concentrated by ultracentrifugation at 120,000 g for 1.5 hours and resuspended in 1 ml OptiMEM, aliquoted and stored at –80°C. Serial dilutions of the virus were made in OptiMEM and used to infect 293T cells that were plated at 2×105 cells/well in 24-well plates. The cells were assayed 2 days post infection for the expression of the reporter gene encoding GFP by flow cytometry, in order to determine the titer.
Cell culture medium and fetal calf serum were obtained from GibCoBRL. Human embryonic kidney (HEK) 293T cells were cultured in Dulbecco's modified Eagle's medium (DMEM) with 10% fetal calf serum (FCS), penicillin and streptomycin in a 10% CO2 humidified incubator at 37°C. Human adult dermal microvascular endothelial cells (HMVEC-d) were purchased from Lonza (Clonetics – Primary Cell and Media Systems) and cultured in EGM-2MV Bulletkit medium (Clonetics), in a 5% CO2 humidified incubator at 37°C. For infection of MVECs, 1×105 cells were plated in a 24-well plate, and virus supernatant was diluted and added to the wells at a MOI of 30. After 6 hours, the medium on the cells was topped up to dilute the concentrated virus, and cells were cultured as described above. Transduction efficiency was measured by flow cytometry.
The protocol used was based on one described previously (Frisch and Francis, 1994). MVECs were grown to ∼70% confluency in 24-well plates and were either not transduced or transduced with a lentivirus encoding GFP alone or a lentivirus encoding GFP and vFLIP. 48 hours post transduction, the cells were trypsinised and equal numbers either replated immediately and allowed to adhere or maintained in suspension for 16 hours by plating them in wells that had been coated with the anti-adhesive polymer polyHEMA (Fukazawa et al., 1995; Fukazawa et al., 1996). Briefly, polyHEMA plates were made by applying 200 μl of a 12 mg/ml solution of polyHEMA (diluted 1:10 in 95% ethanol from a 120 mg/ml stock), and the plates were then left to dry in the tissue culture hood. This process was repeated three times for sufficient coating of the wells. Following coating, the wells were washed twice with PBS and once with Hank's buffered salt solution (HBSS) and left to air dry. After 16 hours, cells in suspension were collected by pipetting, and adherent cells were trypsinised. All cells were then replated in 96-well plates for 4 hours and cell metabolism was measured by an MTT assay, while apoptosis was measured using a Death Detection ELISA (Roche).
Cell viability and apoptosis assays
The metabolism of MVEC populations was measured using a colorimetric method based on the conversion of 3-[4,5-dimethylthiazol-2-yl]-2,5-diphenyltetrazolium bromide (MTT; Sigma-Aldrich) to a formazan product (Hansen et al., 1989). Briefly, 100 μl of EGM-2MV medium containing 5×104 cells was placed into each well of a flat-bottomed 96-well plate in triplicate. The MTT reagent (diluted from a 5 mg/ml solution in PBS) was then added to all the wells at a final concentration of 0.5 mg/ml and the cells were incubated with the MTT reagent for 4 hours at 37°C. The reaction was terminated by adding 100 μl/well of 10% SDS. The plates were left overnight in the dark at room temperature, following which the absorbance was read at 570 nm.
For the detection of apoptosis, cells were stained with the annexin V reagent. 1×106 cells were collected by centrifugation and washed once in cold (4°C) phosphate-buffered saline (PBS) before staining with TACS™ Annexin V-Biotin Apoptosis detection kit (Trevigen), as per the manufacturer's instructions. Analysis was performed by FACScan using CELL QUEST software. Light scatter was used to exclude cell debris, and 1×104 cells within this region were recorded to determine the percentage of annexin V staining. Apoptosis was also measured using a Cell Death Detection ELISAPLUS kit (Roche). Briefly, 100 μl of medium containing 1×104 cells were plated into each well of a round-bottomed 96-well plate in triplicate. Cells were then lysed and the supernatants analysed for the presence of fragmented DNA by the addition of the antibody responsible for capturing cytoplasmic histone-associated DNA fragments, as per the manufacturer's instructions. After 2 hours of incubation, the reactions were developed by the addition of the ABTS substrate, and optical density was measured at 405 nm against ABTS solution as a blank using an ELISA plate reader.
Immunofluorescence assay for NF-κB
Quantitative confocal microscopy was used to compare the subcellular localisation of NF-κB subunits in MVECs. 60% confluent cells, grown on glass cover slips (N°1.5, VWR), were either not transduced or transduced with a lentivector encoding GFP alone or a vFLIP_eGFP vector and cultured for a further 48 hours to allow expression of the vector-encoded genes. Cells were then fixed with 3.7% paraformaldehyde and permeabilised with 0.2% Triton-X100 (Sigma Aldrich) before immunostaining with rabbit polyclonal affinity-purified antibodies to p65/RelA (C-20), RelB (C-19) and p52 (K-27) (Santa Cruz Biotechnology) used at 2 μg/ml and Alexa Fluor® (AF) 633-conjugated F(ab′)2 goat anti-rabbit IgG (Invitrogen) used at 4 μg/ml. 10% normal goat serum (Sigma Aldrich) was used to block nonspecific binding of the secondary antibody, and all reagents were diluted with Tris-buffered saline. Nuclei were counterstained with 2 μg/ml of the nuclear stain DAPI (Sigma Aldrich) and coverslips were mounted onto glass slides (VWR) using Vectashield hard-set mounting media (Vector).
Fluorescence images were captured on a Leica SP2 confocal microscope. DAPI, GFP and Alexa Fluor AF633 fluorescence was captured using sequential acquisition to give separate image files for each stain. A pin hole of 1 Airy (114.5 μm), scan speed of 400 Hz and four-frame averaging were used. Photomultiplier tube gain and offset were adjusted to give a subsaturating fluorescence intensity with an optimal signal:noise ratio. Image analysis was performed using ImageJ software (http://rsb.info.nih.gov/ij). For each high-power field, image masks were created of GFP-, AF633- and DAPI-positive staining. This was achieved by applying a median filter (3×3 pixel radius) to remove noise and automatic thresholding (using the IsoData algorithm) to remove background fluorescence and for conversion to a binary image. The DAPI staining mask was used to define nuclear regions of interest (ROIs). Subtraction of the DAPI mask from the AF633 was performed to define cytoplasmic ROIs and the GFP staining mask was used to identify cells with and without lentiviral-encoded gene expression. Each of these staining masks was then applied to the original AF633 images to separate NF-κB subunit staining in the nuclei and cytoplasm of lentiviral vector-transduced and untransduced cells within each high-power field. Quantitative fluorescence data were then exported from ImageJ-generated histograms into Graphpad Prism 5 software for further analysis and presentation. Nuclear:cytoplasmic ratios of NF-κB subunit staining were then calculated by comparison of median values from histogram data of GFP-negative and GFP-positive cells.
SDS-PAGE and immunoblot analysis
Proteins were separated by SDS-PAGE and transferred to hybond ECL nitrocellulose membranes (Amersham) for immunoblot analysis. Blots were incubated for 1 hour at room temperature in blocking solution (PBS containing 2.5% low-fat milk and 0.1% Tween 20), followed by overnight incubation with primary antibody in blocking solution at 4°C: p100/p52 mouse mAb 1:1000 (Upstate Biotech 05-361), β-actin mouse mAb 1:5000 (AbCam ab6276-100), vFLIP 6/14 (Low et al., 2001) rat mAb 1:100. Bound antibodies were detected with appropriate peroxidase-conjugated secondary antibodies (1:2000 dilution) and visualised by ECL chemiluminescence reagents (Amersham).
This work was supported by a Programme Grant from Cancer Research UK and a Wellcome Trust Fellowship to M.N.