Palmitoylation and localisation of RAS isoforms are modulated by the hypervariable linker domain

Members of the RAS family of monomeric GTPases are molecular switches functioning at an early step in signal cascades that regulate many cellular programmes involved in survival, growth and differentiation. Four closely related RAS isoforms, Harvey rat sarcoma viral oncogene homolog (HRAS, hereafter termed H-RAS), neuroblastoma RAS viral oncogene homolog (NRAS, hereafter termed N-RAS) and Kirsten rat sarcoma viral oncogene homologs 4A and 4B [KRAS4A and KRAS4B, hereafter termed K(A)-RAS and K(B)-RAS], share almost complete sequence identity or homology between residues 1-165 – regions involved in GDP-GTP binding and effector interactions. Despite the fact that H-, K(B)- and N-RAS are expressed ubiquitously and almost identical, several biological differences have been characterised (Omerovic et al., 2007). These include preferential isoform coupling to the Raf and phosphoinositide 3-kinase effector pathways and a unique requirement for K-RAS during mouse development (Potenza et al., 2005; Walsh and Bar-Sagi, 2001; Yan et al., 1998). These differences are proposed to be mediated by differential localisation, enabling isoforms to interact with different subcellular pools of effectors and activators (Daniels et al., 2006; Onken et al., 2006; Rocks et al., 2006).

The area that distinguishes RAS isoforms from each other is the C-terminal hypervariable region (HVR), which contains the major motifs for membrane attachment and targeting to distinct intracellular compartments (Hancock, 2003). The HVR is subdivided into two functionally defined descriptors: a linker domain (amino acids 166-178/179) with few known functions and the targeting domain (amino acids 179/180-189) that is modified posttranslationally to facilitate membrane binding and trafficking. All RAS isoforms are farnesylated on their extreme C-terminal cysteine residue, allowing weak membrane binding, and a second signal targeting motif within the HVR stabilises membrane binding (Hancock et al., 1991). For K(B)-RAS, this consists of a hexa-lysine polybasic stretch (residues 175-180) that interacts electrostatically with membranes and facilitates localisation to the plasma membrane enriched in acidic lipids (Hancock et al., 1990). For the other RAS isoforms, membrane binding is stabilised by single [N-RAS, K(A)-RAS] or double (H-RAS) palmitoylation while in the endoplasmic reticulum (ER) or Golgi complex (Hancock et al., 1989).

More recently, the dynamic nature of RAS palmitoylation has been identified as a crucial determinant of subcellular localisation. Activation of H- and N-RAS stimulates de-palmitoylation and redistribution to the ER/Golgi, where they can be re-palmitoylated (Baker et al., 2003; Goodwin et al., 2005; Rocks et al., 2005; Smotrys and Linder, 2004). The de-palmitoylation–re-palmitoylation cycle is important for preventing nonspecific accumulation of H- and N-RAS on all membranes (Goodwin et al., 2005; Rocks et al., 2005). Furthermore, analysis of H-RAS trafficking revealed that individual palmitoyl groups play different roles in H-RAS localisation. Palmitoylation of Cys181 stabilises H-RAS plasma membrane localisation, whereas Cys184 palmitoylation facilitates redistribution between plasma membrane microdomains (Roy et al., 2005). Residues 166-172 within the linker domain of the H-RAS HVR are required for this redistribution (Jaumot et al., 2001; Prior et al., 2001). Confining H-RAS in cholesterol-dependent microdomains by generating Cys184-mono-palmitoylated RAS or mutating residues within the linker domain impairs H-RAS signalling through the Raf mitogen-activated protein kinase (MAPK) pathway (Prior et al., 2001; Roy et al., 2005). Therefore, reversible palmitoylation and the precise positioning of the palmitoyl group within the RAS HVR regulate RAS localisation and signalling output.

Understanding how RAS isoforms control their dynamic association with subcellular compartments has clear functional implications. We have performed a detailed dissection of the role of the HVR linker domain in modulating palmitoylated RAS trafficking and micro-localisation. Our results reveal novel functions for linker domain residues in stabilising or facilitating palmitoylation, membrane binding, plasma membrane targeting and compartmentalisation.

The RAS HVR is so named owing to the low sequence conservation (of <30%) across the four main RAS isoforms. However, a comparison of only the palmitoylated RAS isoforms [H-, K(A)- and N-RAS] revealed that almost 70% of their HVR residues are homologous (Fig. 1). In addition to the well-characterised conserved cysteine residues modified by palmitoyl or farnesyl groups, both acidic and basic/hydrophobic patches are conserved across isoforms and species. To investigate the role of these conserved motifs in modulating RAS localisation, we created a series of GFP-tagged HVR constructs (Fig. 1C). Minimal HVR sequences fused to GFP have been used previously as tools for analysing the micro-localisation of isoforms (Prior et al., 2001; Prior et al., 2003) because these minimal sequences are biologically inert and consequently divorced from potentially confounding effector interactions or expression-induced effector activation.

Owing to the sequence conservation within the palmitoylated RAS linker domains, we wanted to analyse whether residues within this region contributed directly to membrane binding or trafficking. Confocal microscopy analysis of the steady-state distributions of GFP-tagged HVR or full-length RAS constructs expressed in HeLa cells revealed two clear results. First, when the linker domain was removed, both N-RAS and K(A)-RAS targeting motifs showed reduced cell-surface labelling and a more prominent perinuclear localisation (Fig. 2; tN, tKA versus HVR-N, HVR-KA). For N-RAS, this was especially dramatic, with only a perinuclear pool remaining that colocalised with the Golgi marker GM130 (data not shown). By contrast, the localisation of the H-RAS targeting domain was indistinguishable from that of intact HVR and full-length H-RAS (Fig. 2; tH, HVR-H, H-RAS). These data indicate that singly palmitoylated N- and K(A)-RAS isoforms are more sensitive than doubly palmitoylated H-RAS to HVR linker residues modulating their steady-state distributions. For N-RAS, an intact linker domain is essential for correct cell-surface localisation.

A second, more subtle, observation was also apparent when the distributions of full-length RAS proteins were compared with the localisation of their respective HVR constructs comprising both the linker and targeting domains. As observed previously, H-RAS and the H-RAS HVR (HVR-H) displayed identical distributions (Hancock et al., 1991), and we now observed the same result for K(A)-RAS [Fig. 2; K(A)-RAS, HVR-KA]. However, for N-RAS, the distribution of the full-length protein and its cognate HVR (HVR-N) did not match. While full-length N-RAS displayed a plasma membrane localisation and a prominent perinuclear localisation, the N-RAS HVR was restricted to the plasma membrane, with minimal perinuclear labelling apparent (Figs 2 and 4; N-RAS, HVR-N). Therefore, the N-RAS G domain (amino acids 1-165) regulates subcellular localisation by promoting association with endomembrane organelles. While the H-RAS main body (G-domain) has been shown previously to provide a repulsive force to membrane interactions, promoting redistribution within the plane of the plasma membrane (Rotblat et al., 2004), these data reveal a novel modulatory role for the N-RAS G-domain in regulating gross compartmentalisation.

Our data reveal a requirement for an intact HVR to promote cell-surface localisation of mono-palmitoylated N- and K(A)-RAS isoforms. In order to dissect the important additional sequence elements regulating localisation, we first examined the localisation of the minimal RAS targeting motifs. The targeting domains of N- and K(A)-RAS (tN and tKA) displayed different distributions despite their identical posttranslational modifications of mono-palmitoylation and farnesylation (Fig. 2). tN was restricted to the ER/Golgi, whereas tKA was present on endomembranes and the cell surface. The obvious sequence difference between these constructs is the presence of a basic/hydrophobic patch within tKA that might contribute to membrane binding (Fig. 1). The inner leaflet of the cell surface is highly enriched with acidic phospholipids that facilitate cell-surface targeting through electrostatic interactions (Okeley and Gelb, 2004). We therefore focussed on the potential for the basic/hydrophobic patch of the K(A)-RAS targeting region to stabilise localisation at the plasma membrane. Substitution of the tKA basic residues for uncharged glutamine residues confined the K(A)-RAS targeting region to the ER/Golgi (Fig. 3; tKA-Q). Conversely, the introduction of a tri-lysine stretch into the N-RAS targeting region (tN-3K) promoted plasma membrane localisation. Therefore, we conclude that a single palmitoyl group is insufficient to enable steady-state cell-surface localisation of RAS targeting domains, but the addition of an adjacent basic/hydrophobic patch stabilises cell-surface localisation.

While K(A)-RAS uses adjacent basic/hydrophobic residues within the targeting domain to traffic correctly, N-RAS requires sequence elements within the linker domain to ensure cell-surface localisation (Fig. 2; HVR-N and tN). This was confirmed when the linker domain of the N-RAS HVR was substituted with alanine residues (HVR-N-ala), and retention within the ER/Golgi similar to that seen with GFP-tN was then observed (Fig. 4). In order to exclude the possibility that these changes in distribution were an artefact of expressing minimal targeting motifs fused to GFP, we generated full-length N-RAS constructs with mutations to the linker domain. Importantly, the linker similarly modulated full-length N-RAS distribution: substitution of the N-RAS linker domain with alanine residues produced an ER/Golgi localisation equivalent to that seen with the minimal HVR linker mutant (Fig. 4; N-RAS-ala). These data reveal that the linker domain has the capacity to modulate the localisation of N-RAS at the plasma membrane versus endomembranes.

N-RAS shares a number of conserved linker domain residues and motifs with the other palmitoylated RAS isoforms, including a basic/hydrophobic patch, a pair of acidic residues and a serine residue (Fig. 1). We systematically mutated these residues to identify those that promoted cell-surface localisation of the N-RAS HVR (Fig. 4). In these studies, we compared HVR linker mutants with the HVR-N control that showed strong cell-surface and minimal perinuclear Golgi labelling (50% of cells had undetectable perinuclear and only 10% showed strong Golgi-type staining). If a motif promotes cell-surface localisation, then mutagenesis of this motif should cause an increase in endomembrane labelling compared with HVR-N. This was seen with HVR-N-Q-Δ2ala, in which the second half of the linker domain was substituted with alanine residues and the lysine residues within the basic patch were exchanged for glutamine residues. In this case, mutation of the hydrophobic/basic, serine and acidic residues resulted in a substantial decrease in cell-surface labelling that was replaced by a reticular and perinuclear endomembranous localisation. A more subtle redistribution to endomembranes was seen when we deleted the entire first half of the linker domain (HVR-N-Δ1), substituted the basic residues for glutamine residues (HVR-N-Q) or the serine residues for alanine residues (HVR-N-SA). Semi-quantitative analysis revealed that HVR-N-Q reverted to a similar localisation to that seen with full-length N-RAS, while the HVR-N-SA mutant showed an intermediate increase in endomembrane labelling (Fig. 4). Therefore, while the basic/hydrophobic patch and the serine residues facilitate cell-surface localisation, none of these is solely responsible for promoting stable cell-surface binding. A much stronger effect of a mutation of a single targeting motif was seen when the acidic residues were replaced with alanine residues (HVR-N-DA). In this case, plasma membrane staining was weaker (30% of cells possessed no cell-surface labelling), while labelling of endomembranes was now very prominent. For HVR-N-DA, almost all of the cells examined had no perinuclear staining but instead possessed a strong punctate localisation at endomembranes. In summary, basic/hydrophobic and serine amino acids contribute to the correct trafficking and localisation of N-RAS, whereas acidic residues appear to be key determinants of localisation to the plasma membrane.

The apparent confinement of GFP-tN within the ER/Golgi might be due to an inability to sort correctly into transport vesicles destined for the cell surface. Alternatively, the loss of crucial linker domain residues might alter the half-life of palmitoylation or palmitoyl-membrane interactions, thereby resulting in weaker or short-lived membrane binding. In order to address these possibilities, we focussed on the palmitoylation status and membrane versus cytosol partitioning of our palmitoylated RAS constructs. Previous analysis indicated that the half-life of palmitoylation of N-RAS is 20 minutes and total cellular N-RAS [³H]-palmitate incorporation is effectively complete within 2 hours (Magee et al., 1987). We used a 4 hour pulse of [³H]-palmitate to generate steady-state labelling of RAS constructs and observed no significant effect of the linker region on the extent of N- or H-RAS palmitoylation or the rate of N-RAS de-palmitoylation (Fig. 5A,B). Specifically, both tH and tN possessed amounts of [³H]-palmitate equivalent to those of their cognate HVR and full-length RAS constructs (Fig. 5B). This means that the differences in localisation seen with tN versus HVR-N are not due to either more rapid de-palmitoylation or impaired palmitoylation resulting in short-lived membrane interactions.

While GFP-tN displayed palmitoylation equivalent to that seen with HVR-N and N-RAS, other N-RAS HVR linker mutants showed significantly reduced palmitoylation (Fig. 5C). Substitution of the linker region with alanine residues almost completely ablated palmitoylation (Fig. 5C; HVR-N-ala), consistent with the complete redistribution to the ER/Golgi (Fig. 4). Dissection of the HVR linker residues responsible for maintaining palmitoylation revealed that the basic/hydrophobic patch was not involved (HVR-N-Δ1-ala). Instead, we noted that, in the context of an intact HVR, whenever the acidic residues were mutated, reduced palmitoylation is observed; and specific mutation of the acidic linker residues resulted in an 85% reduction of HVR palmitoylation (Fig. 5C; HVR-N-DA, HVR-N-ala, HVR-N-Q-Δ2-ala). Therefore, the acidic amino acids within the N-RAS HVR promote stable palmitoylation.

Although palmitoylation was necessary for localisation to the plasma membrane (Fig. 4; HVR-N-DA), it was not sufficient as tN was strongly palmitoylated but not present at the cell surface. Interestingly, all of the HVR mutants displayed similar levels of membrane binding regardless of the extent of palmitoylation (Fig. 5D). For example, tN and tKA exhibited dramatically different levels of palmitoylation (Fig. 5B) but were equivalently membrane bound. However, this might also be partly due to the adjacent basic residues within tKA contributing to stabilising membrane binding. Equally significant is the observation that N-RAS and the minimal N-RAS CAAX motif, which is not palmitoylated, displayed equivalent levels of membrane binding (Fig. 5D), while showing dramatically different subcellular localisation (Fig. 4). These data reveal a complex interplay between posttranslational modifications and HVR sequence elements in ensuring strong membrane binding and appropriate subcellular localisation of mono-palmitoylated RAS isoforms. These include acidic residues stabilising palmitoylation, which, together with adjacent hydrophobic and basic residues, enhances membrane binding and promotes targeting to the cell surface.

In recent years, compartmentalisation of RAS isoforms has emerged as a crucial regulator of RAS function, and RAS interactions with membranes and compartments have been shown to be highly dynamic. Early studies characterised farnesylation and a second signal of palmitoylation or a polybasic region as being sufficient for H- or K(B)-RAS membrane binding and trafficking to the plasma membrane (Hancock et al., 1991). Subsequent studies have focussed on how these motifs are regulated through de-acylation or phosphorylation, respectively, to weaken membrane interactions and promote redistribution to endomembranes (Rocks et al., 2006). The role of the linker has been less clear, but H-RAS linker domain residues have been characterised as functioning as regulators of localisation to cell-surface signalling microdomains (Rotblat et al., 2004). Loss of the regions containing the basic and hydrophobic residues (amino acids 166-171) prevents H-RAS from moving out of lipid rafts and prevents efficient activation of the Raf-MAPK cascade (Jaumot et al., 2001; Prior et al., 2001).

Our data reveal that a minimal combination of mono-palmitoylation and farnesylation are insufficient for stable cell-surface localisation (Fig. 2; tN). Instead cooperating sequence elements within the HVR drive correct localisation by stabilising palmitoylation and providing specific targeting information. First, acidic residues within the N-RAS HVR linker domain facilitate stable palmitoylation (Fig. 5C; HVR-N-DA). Palmitoylation is known to be necessary for plasma membrane localisation because H-RAS targeting domain mutants lacking palmitoylated cysteine residues are not observed at the cell surface (Hancock et al., 1991). At this stage, it is unclear from our data whether acidic residues regulate the on- or the off-rate of palmitoylation. No consensus sequence has been found directing palmitoylation (Linder and Deschenes, 2007), but it is tempting to speculate that the acidic residues contribute to promoting interactions with the RAS protein acyltransferase. Changes in the extent or rate of cycling of palmitoylation represent an obvious mechanism to explain the role of the linker domain in promoting cell-surface labelling. However, we see that cellular tN and N-RAS are equivalently palmitoylated and have similar de-acylation rates despite being resident in different locations (Fig. 5), which argues against a simple model in which changes in the dynamics of palmitoylation restrict tN to the Golgi.

While palmitoylation is essential for localisation to the plasma membrane, it is not sufficient. The second motif present in the palmitoylated RAS HVR contains basic residues that have a well-established role in plasma membrane targeting of K(B)-RAS (Hancock et al., 1991). Polybasic motifs interact electrostatically with negatively charged lipids that are present within cellular membranes at their highest concentration within the cytosolic leaflet of the plasma membrane (Okeley and Gelb, 2004; Yeung et al., 2006). Indeed, these motifs have been used as sensors for acidic lipids, revealing their key role in plasma-membrane-specific functions such as phagocytosis (Yeung et al., 2006). Introduction of basic residues allows mono-palmitoylated tN to interact stably with the cell surface, and mutation of equivalent intrinsic basic residues within tKA prevents plasma membrane localisation (Fig. 3). Importantly, a stretch of three basic residues is not sufficient to ensure association of K(B)-RAS with the cell surface (Hancock et al., 1990). This suggests that the basic patch and mono-palmitate cooperate to facilitate localisation of RAS to the cell surface, whereas neither is sufficient in isolation. We propose that the basic/hydrophobic patch in the first half of the linker domain of palmitoylated RAS isoforms performs a similar, albeit less effective, function to that of the tKA basic motif. While mutation of this first basic patch within the N-RAS linker domain increases endomembrane labelling (Fig. 4; HVR-N-Q), plasma membrane labelling is still observed. Therefore, the first basic patch contributes to localisation but is not essential, revealing that other HVR linker sequences are also crucial for correct targeting. These include the conserved serine residue that, when mutated, also increases endomembrane labelling (Fig. 4; HVR-N-SA).

Our studies have been conducted predominantly with minimal HVR sequences fused to an inert reporter protein. Importantly, similar mutations in the context of the full-length protein mirrored the localisation, confirming the importance of these HVR residues in directing RAS targeting (Fig. 4; N-RAS-ala and HVR-N-ala). Chemically synthesised, lipid-modified tN and full-length H- and N-RAS have been analysed spectroscopically and modelled using Molecular Dynamics software to determine how the HVR orientates when inserted into membranes. The majority of the HVR is embedded at the lipid-water interface, whereas the lipid modifications insert into the hydrophobic core of the membrane (Gorfe et al., 2004; Huster et al., 2003; Reuther et al., 2006). These studies revealed that the linker domain is highly flexible and enables basic residues within the HVR and the G-domain of H-RAS to interact with membranes (Gorfe et al., 2007). A separate observation from the spectroscopy and modelling studies was that, in the context of the N-RAS targeting domain, the farnesyl group forms a stable anchor, whereas the palmitoyl group can flip in and out of the membrane (Vogel et al., 2007). Both our data and previous studies of H-RAS reveal that the N- and H-RAS G-domains partially oppose the membrane interaction forces supplied by the HVR (Fig. 5D) (Rotblat et al., 2004). For N-RAS, this is especially dramatic, with full-length N-RAS displaying membrane partitioning equivalent to a minimal CAAX motif. At this stage, it is unclear whether this is a direct effect or occurs through modulation of HVR interactions with the membrane.

Based on these studies and our data, two compatible models can be envisaged to describe the role of the linker domain in promoting surface localisation of mono-palmitoylated RAS isoforms. Most simply, the linker domain might stabilise insertion of the palmitoyl group into membranes, allowing N-RAS to be sorted more efficiently into plasma-membrane-directed vesicles (so-called `rasosomes') (Rotblat et al., 2006). This would be analogous to the effect of the second H-RAS palmitoyl group, which is sufficient to allow plasma membrane targeting in the absence of a linker domain. Alternatively, residues within the linker domain, such as the basic/hydrophobic patch, might more actively target RAS to the cell surface by ensuring retention on this membrane by means of electrostatic interactions.

In summary, we have characterised new targeting motifs within the HVR of both N- and K(A)-RAS that facilitate stable cell-surface localisations. This is achieved by acidic residues stabilising N-RAS palmitoylation; however, this is insufficient for promoting localisation to the plasma membrane. Additional amino acids, including hydrophobic/basic residues, are required to provide specific targeting information or to stabilise insertion of the palmitoyl group into the membrane. The intrinsic ability of RAS isoforms to modulate their localisation through these dynamic HVR interactions with the membrane might provide the basis for ensuring compartmentalised RAS signalling.

HeLa cells grown in DMEM supplemented with 10% heat-inactivated FBS, at 37°C, were transfected at 50% confluence with GeneJuice (Novagen) according to the manufacturer's instructions. After overnight incubation, cells were processed for confocal or biochemical analysis, as indicated. GFP-tagged tH, HVR-H, H-RAS and N-RAS have been described previously (Prior et al., 2001). N-terminally GFP-tagged K(A)-RAS was generated using Gateway cloning of human K(A)-RAS ORF clone IOH44611 (Evoquest) and verified by sequencing. RAS HVR constructs were prepared from PAGE-purified sense and antisense oligos (Invitrogen) designed with pre-digested Xho1/BamH1 sites at the N- and C-termini, respectively (sequences available upon request). Human RAS HVR sequences correspond to HVR fragments of loci NM_002524 (N-RAS) and NM_033360 [K(A)-RAS]; in some cases, basic residues or portions of the linker domain (residues 166-179) were mutated (Fig. 1B). Following annealing, constructs were ligated into pEGFP-C1, screened for membrane targeting of GFP by immunofluorescence microscopy and sequenced. To generate full-length RAS constructs with mutated HVR motifs, RAS acceptors were made from a wild-type N-RAS template by PCR. BspE1 and Xho1 restriction sites were added to the N- and C-termini, respectively, to allow in-frame cloning of RAS constructs corresponding to residues 1-165 of RAS in between the GFP and HVR sequences. The resultant GFP-tagged RAS-HVR constructs have a linker of three amino acids (ARA) between the RAS and HVR sections.

Membrane and cytosol (P100/S100) fractions were prepared as described previously (Prior et al., 2001). In brief, cells were harvested in PBS, homogenised and pelleted to remove nuclei, followed by ultracentrifugation of the supernatant at 100,000 g. Membrane pellets were resuspended in 100 μl of HB (0.25 M sucrose, 0.03 M imidazole, pH 7.2); 5 μg of membrane pellet (P100) and an equivalent proportion of cytosolic supernatant (S100) were analysed by western blotting for GFP-tagged constructs. Bands corresponding to GFP-labelled proteins were digitally imaged (UViChemi, Uvitec, Cambridge, UK) and quantified using Image J. Averages (±s.e.m.) are the result of at least four independent fractionations. For microscopy, cells grown on cover slips were fixed 24 hours post-transfection with 3% paraformaldehyde in PBS. Representative examples of confocal images are shown. Semi-quantitation of perinuclear and cell-surface labelling on blind samples was scored by two independent observers; 120-200 cells from two separate experiments were counted per condition.

Before use, [³H]-palmitate (PerkinElmer Life Sciences) was concentrated by vacuum centrifugation and stored under nitrogen. Palmitate incorporation (0.2-0.7 mCi/ml [³H]-palmitate in DMEM) into transfected HeLa cells was terminated after 4 hours by rinsing cells twice with ice-cold PBS and lysing into NP40 lysis buffer (25 mM Tris-Cl pH 7.5, 100 mM NaCl, 0.5% Nonidet P-40). For de-palmitoylation measurements, cells were labelled for 4 hours with [³H]-palmitate-containing DMEM (0.5-1 mCi/ml), washed twice with unlabelled media and incubated with DMEM containing 200 μM palmitic acid (Sigma) for the times indicated before lysing. Following lysis, a post-nuclear lysate was prepared and incubated at 4°C with affinity-purified sheep antibody against GFP for 30 minutes before precipitation with protein-G-agarose beads (Sigma) for 2 hours at 4°C. The immunoprecipitates were washed (0.1% NP40, 25 mM Tris pH 7.5, 150 mM NaCl) before resuspension into sample buffer (2% SDS, 10 mM Na₂PO₄ pH 7, 50 mM DTT, 0.02% bromophenol blue). Samples were heated to 98°C for 2 minutes before SDS PAGE and transfer to nitrocellulose. Membranes were exposed to film (Kodak Biomax LE) for between 9 hours and 11 days at –80°C with the aid of a Kodak Biomax Transcreen LE intensifier screen. Subsequently, they were probed for GFP using a monoclonal antibody against GFP (Roche) to enable normalisation of palmitate incorporation to GFP expression. The resulting images of GFP and palmitoylated RAS were quantified using densitometry (Image J/LI-COR Odyssey software). Averages are the result of at least three independent labelling experiments.

This work was supported by funding from the Northwest Cancer Research Fund and the Royal Society.