Caveola disassembly increases mechanical tension at AJ in response to hypoosmotic stimulation

We next asked whether the increase in junctional F-actin was accompanied by changes in junctional mechanics, focusing on AJ, which are often sites of mechanical tension. Junctional tension was evaluated first by measuring the initial recoil of vertices after laser-ablation of the bicellular AJ that connect those vertices (Ratheesh et al., 2012; Liang et al., 2016; Michael et al., 2016). Control monolayers showed a baseline-initial recoil, consistent with there being preexisting mechanical tension in the bicellular junctions. However, initial recoil was increased significantly after acute hypoosmotic stimulation and restored upon addition of isoosmotic media (Figure 2a). This suggested that hypoosmotic stimulation could reversibly increase AJ tension. However, hypoosmotic stress did not increase AJ tension in Cav-1 KD cells (Figure 2a), although these cells had an elevated baseline tension, as previously reported (Teo et al., 2020). This suggested that caveolae might be required for the rapid increase in AJ tension that occurred upon hypoosmotic stimulation.

Open in a separate window

FIGURE 2:

Mechanical activation of caveolae increases tension at AJ. a) Initial recoil velocity (IRV) of cell vertices following junctional laser ablation in confluent MCF-10A epithelial monolayers. In WT monolayers, hypoosmotic shock increased IRV by ~55% (p = 0.0006), which is returned to baseline levels following a 1 h recovery period in isoosmotic growth medium (p = 0.0306). In monolayers lacking caveolae (Cav-1 KD), baseline tension is increased by ~28% compared with WT monolayers. Importantly, this is unchanged following hypoosmotic shock (p = 0.7896). b) IF imaging of total α-catenin (middle) and α-catenin in its open conformation (α18; top). Lower row shows an image calculation of α18/α-catenin ratio, where the warmer colors indicate increased ratio. c) Quantification of (b) showing the percentage changes in the α18/α-catenin ratio following hypoosmotic shock in WT (~17%; p ≤ 0.0001), Cav-1 KD (~7%; p ≤ 0.0001), and Cav-1 KDR (~26%; p ≤ 0.0001) monolayers. Cav-1 KD monolayers also show a baseline increase of ~10%. d) In separate experiments, a 1 h pretreatment with 25 nM para-aminoblebbistatin (an inhibitor of actomyosin contractility) attenuated the IRV of cell vertices in WT monolayers by ~77% under control conditions (p ≤ 0.0001). In the absence of para-aminoblebbistatin, IRV was found to increase by ~63% following hypoosmotic shock (p ≤ 0.0001), but this was desensitized with the addition of para-aminoblebbistatin (p = 3920). e) Full dataset of α-18/α-catenin ratio (see Figure 2b/c) showing data points from individual cell–cell junctions. All data are means ± SD. IRV statistical analyses calculated from N = 3 independent experiments (20 junctions per experiment) using unpaired t tests. All other statistical analysis calculated from N = 3 independent experiments (60 junctions per experiment). Points on graphs represent individual cell junctions. ns, not significant; *p < 0.05; **p < 0.01, ***p < 0.001; ****p < 0.0001.

To corroborate the results of recoil experiments, we then assayed changes in the conformation of the E-cadherin-associated protein, α-catenin, as an index of molecular-level tension at AJ. Application of mechanical tension causes conformational unfolding of α-catenin, exposing cryptic epitopes including one in the central M-domain that is recognized by the α18 monoclonal antibody (mAb; Yonemura et al., 2010; Noordstra et al., 2023). We therefore measured α18 mAb fluorescence intensity normalized for total-junctional α-catenin levels (α18/pan-α-catenin ratio) as a proxy for molecular-level tension applied to the cadherin–catenin complex. Supporting our recoil measurements, α18/pan-α-catenin ratios increased at AJ within 5 min of applying hypoosmotic media, and this was significantly reduced in Cav-1 KD monolayers (Figure 2, b and c). Specificity for Cav-1 was confirmed as the mechanical response was restored by expressing an RNAi-resistant Cav-1 transgene in the KD monolayers. Together, these data indicate that hypoosmotic stimulation increases mechanical tension at AJ in a caveola-dependent manner. This supported the notion that the junctional cytoskeleton was being altered by mechanical stimulation of caveolae.

Changes in junctional tension can reflect contributions from multiple elements of the AJ, potentially including the membrane itself as well as the cytoskeleton that physically couples to the cadherin–catenin adhesion complex. To dissect these possible factors, we inhibited Myosin II, which is responsible for the contractile forces that the cytoskeleton can directly and indirectly apply to AJ (Smutny et al., 2010; Ratheesh et al., 2012). Monolayers were treated with the Myosin II inhibitor, para-aminoblebbistatin (blebbistatin, 25 nM), before hypoosmotic shock. Blebbistatin reduced baseline-junctional tension as measured by recoil assays, consistent with a dominant contribution of cellular contractility to AJ tension. Importantly, hypoosmotic stimulation did not increase junctional tension in the presence of blebbistatin (Figure 2d). This suggested that cellular contractility might be principally responsible for the tensional response of AJ to hypoosmotic stimulation. This was confirmed by measuring α18/pan-α-catenin ratios, which increased when control monolayers were stimulated with hypoosmotic media, but not in the presence of blebbistatin (Figure 2e). Although we cannot exclude a role for a change in membrane tension, these findings imply that cytoskeletal change may be the major process responsible for the caveola-dependent increase in junctional tension that occurs upon hypoosmotic stimulation. Furthermore, because Cav-1 KD did not affect the response of NMII to hypoosmotic stimulation, we surmise that the increase in tension was principally due to the change in F-actin. This is consistent with earlier evidence that regulation of F-actin in the actomyosin system can increase mechanical tension, without concomitant change in NMII (Kovacs et al., 2011; Wu et al., 2014; Reymann et al., 2016; Teo et al., 2020).

Caveolae mechano-activation engages a PtdIns(4, 5)P ₂-FMNL2 signaling pathway

We next sought to characterize the molecular pathway that reinforced the junctional-actin cytoskeleton when caveolae were stimulated with hypoosmotic media. Recently, we reported that the chronic depletion of caveolae activated a lipid-based signaling pathway that regulated the actin cytoskeleton (Teo et al., 2020). Specifically, Cav-1 RNAi increased phosphoinositide-4, 5-bisphosphate (PtdIns(4, 5)P₂) at AJ membranes, which directly recruited the FMNL2 formin to promote actin assembly at the junctional cortex. We therefore examined whether this pathway might have a physiological role when preexisting caveolae were acutely disassembled by mechano-stimulation.

PtdIns(4, 5)P₂ was visualized by transiently expressing a location sensor bearing the pleckstrin homology (PH) domain of phospholipase Cδ (mCherry-PH). In control monolayers, mCherry-PH localised primarily to the PM, being clearly evident at cell–cell contacts. Junctional PtdIns(4, 5)P₂ levels increased by ~20% within 5 min of adding hypoosmotic media (Figure 3, a and b; Supplemental Figure S3c). This response did not occur, however, in Cav-1 KD cells, although baseline levels were higher, as previously reported (Teo et al., 2020). Reconstitution of KD cells with an RNAi-resistant transgene confirmed a specific effect of Cav-1 (Figure 3, a and b and Supplemental Figure S3c). This change was relatively selective for PtdIns(4, 5)P₂, as hypoosmotic stress did not affect the membrane levels of PtdIns(3, 4, 5)P₃, detected by expression of a sensor bearing the PH domain of Akt (PH-Akt-GFP; Supplemental Figure S3, a and b). Similarly, PtdIns(4, 5)P₂ was increased by monolayer stretch (~17%, Figure 3, c and d). Together, these findings indicated that mechanoactivation of caveolae increased the identifiable levels of junctional PtdIns(4, 5)P₂ at AJ.

Open in a separate window

FIGURE 3:

Mechanical stimulation of caveolae activates the PtdIns(4, 5)P₂-FMNL2 signaling pathway. a) Fluorescence imaging of transiently expressed mCherry-PH, a biosensor for PtdIns(4, 5)P₂ in WT (top), Cav-1 KD (middle), and Cav-1 KDR (bottom) MCF-10A monolayers before and after hypoosmotic shock (left and right columns, respectively; scale bar = 10 μm). b) Quantification of (a) reveals that WT monolayers are sensitive to hypoosmotic shock, with mCherry-PH increasing by ~20% (p ≤ 0.0001). Cav-1 KD monolayers were relatively insensitive to hypoosmotic shock, increasing by just ~4% (p = 0.0823). Cav-1 KDR monolayers were resensitized to hypoosmotic shock, with mCherry-PH increasing by ~18% (p ≤ 0.0001). c) Fluorescence imaging of mCherry-PH expressed in control monolayers (left) and those following 5 mins of 10% stretch. d) Quantification of (c) shows that mechanical stretching also increases mCherry-PH by ~17% (p ≤ 0.0001). e) Fluorescent imaging of transiently expressed FMNL2-EGFP in WT and Cav-1 KD monolayers before and after hypoosmotic shock. f) Quantification of (e) shows that FMNL2-EGFP is increased by ~30% following hypoosmotic shock (p ≤ 0.0001). In Cav-1 KD monolayers, baseline FMNL2-EGFP is elevated by ~21% compared with WTs (P ≤ 0.0001), and this is insensitive to further change with hypoosmotic shock (p = 0.4319). g) Fluorescence imaging of FMNL2-EGFP in control WT MCF-10A monolayers (top) or those pretreated for 1 h using 4 mM neomycin, a PtdIns(4, 5)P₂ blocking agent (bottom). h) In the absence of neomycin, FMNL2-EGFP levels were increased by ~18% following hypoosmotic shock (p ≤ 0.0001). In the presence of neomycin, baseline FMNL2-EGFP levels were reduced by 14% (p ≤ 0.0001) and were not remained unchanged by hypoosmotic shock (p = 0.5540). All data are means ± SD. All statistical analyses calculated from N = 3 independent experiments (60 junctions per experiment) using unpaired t tests. Points on graphs represent individual cell junctions. ns, not significant; *p < 0.05; **p < 0.01, ***p < 0.001; ****p < 0.0001.

We then expressed FMNL2-EGFP in MCF-10A cells to assess whether it was engaged by caveola mechanoactivation. Consistent with the observed change in PtdIns(4, 5)P₂, junctional levels of FMNL2-EGFP also increased acutely when control monolayers were stimulated with hypoosmotic media and this response was abolished in Cav-1 KD cells (Figure 3, e and f). We then used the PtdIns(4, 5)P₂ antagonist neomycin to test whether recruitment of FMNL2 required PtdIns(4, 5)P₂. Pretreatment with neomycin (4 mM, 1 h) reduced baseline levels of FMNL2 and abolished its recruitment to junctions upon hypoosmotic stimulation (Figure 3, g and h). This supports the notion that the PtdIns(4, 5)P₂-FMNL2 mechanism is activated when caveolae dissociate upon application of acute mechanical stress.

Molecular stabilization of caveolae antagonizes cortical reinforcement

We next sought to test whether this signaling pathway was engaged as a response to the acute disassembly of caveolae. To address this notion, we required a molecular strategy that would allow us to preserve caveolae in cells but render them more resistant to disassembly by mechanical stimulation.

The stability of caveolae is enhanced by interactions between undecad cavin1 (UC1) repeat domains and phosphatidylserine (PtdSer) and the number of UC1 repeats influences the ability of cavin1 to stabilize caveolae (Tillu et al., 2018). Mammalian cavin1 contains two UC1 domains, whereas zebrafish cavin1b (DrCavin1b) has five UC1 domains, and expression of DrCavin1b in mammalian cells increased the resistance of their caveolae to disassembly upon hypoosmotic stimulation (Tillu et al., 2018; Figure S1i). Therefore, we used lentiviral transduction to reconstitute cavin1 KD MCF-10A cells with either mouse cavin1 (MmCavin1-EGFP) or zebrafish cavin1 (DrCavin1b-EGFP), predicting that the greater number of UC1 domains would allow DrCavin1b-EGFP to stabilize caveolae against mechanical stress, compared with MmCavin1-EGFP (Figure 4a). We then measured the change in cavin1 fluorescence at AJ before and after mechanical stimulation as an index of cavin dissociation, where negative change indicates cavin dissociation. Indeed, cavin1 dissociated from the membrane upon hypoosmotic stress to a significantly lesser degree in DrCavin1b-EGFP cells compared with MmCavin1-EGFP cells (Figure 4, b and c; Supplemental Figure S4a). As a further control, we expressed a zebrafish cavin1 mutant construct in which four UC1 repeat domains were deleted (4UC1-EGFP), bringing its UC1 complement closer to that of mouse cavin1. Stress-induced cavin dissociation in 4UC1-EGFP cells was similar to that in MmCavin1-EGFP cells (Figure 4, b and c; Supplemental Figure S4a). This supported the strategy that DrCavin1b, bearing a higher number of UC1 repeats, could stabilize caveolae against mechanical stress.

Open in a separate window

FIGURE 4:

Molecular stabilization of caveolae disrupts the cytoskeletal response to mechanical stimulation. a) Schematic of experimental design for molecular stabilization of caveolae through the expression of zebrafish cavin-1b (DrCavin1b) in human MCF-10A epithelial monolayers. Caveola stability is conferred through interactions between cavin-1 UC1 domains (shown in red) and phosphatidylserine (PtdSer) at the PM. In contrast to the two UC1 domains of mammalian cavin-1 (top), zebrafish cavin-1b (middle) contains five UC1 domains. This confers greater stability to caveolae, resisting their disassembly following mechanical stimuli. A further cavin-1 construct containing a single UC1 domain (4UC1) was adapted from zebrafish cavin-1b (bottom). This has characteristics closer to that of mammalian cavin-1 and is an additional control. b) IF imaging of various cavin-1 constructs in MCF-10A monolayers which are depleted of endogenous cavin-1 by shRNA-mediated KD (scale bar = 10 μm). c) Quantification of (b) showing that both mouse cavin-1 (MmCavin1; left) or a mutated zebrafish cavin-1b variant (4UC1; right) (both homologous to human cavin-1) are dissociated from the PM following hypoosmotic shock by ~18% (p ≤ 0.0001) and 17% (p ≤ 0.0001), respectively. In monolayers expressing DrCavin1b (middle), cavin-1 is resistant to dissociation following hypoosmotic shock, decreasing by just ~2% (p = 0.7277). d) Fluorescence imaging of F-actin in cavin-1 depleted MCF-10A monolayers expressing MmCavin1, DrCavin1b, or 4UC1 cavin-1 constructs (scale bar = 10 μm). e) Quantification of changes to cortical F-actin levels following hypoosmotic shock in native cavin-1-depleted MCF-10A monolayers expressing the aforementioned cavin-1 variants. In monolayers expressing MmCavin1 and 4UC1 (homologous to native cavin-1) cortical F-actin increased by ~40% (p ≤ 0.0001) and ~27% (p ≤ 0.0001), respectively, as analyzed by a change in fluorescence intensity. Monolayers expressing DrCavin1b did not show an increase in cortical F-actin (p = 0.0012). All data are means ± SD. All statistical analyses calculated from N = 3 independent experiments (60 junctions per experiment) using unpaired t tests. Points on graphs represent individual cell junctions. ns, not significant; *p < 0.05; **p < 0.01, ***p < 0.001; ****p < 0.0001.

We then used this molecular stabilization strategy to test how acute disassembly of caveolae contributed to reinforcing the AJ cortex against mechanical stress. For this, we reconstituted cavin1 KD monolayers with our panel of cavin1 constructs. Cavin1 KD monolayers showed mechanosensitive responses in the cortex that were altered in a manner similar to those seen in Cav-1 KD cells. Namely, levels of junctional F-actin were increased at baseline in cavin1 KD cells, and they did not increase upon addition of hypoosmotic media. This similarity in effect of cavin1 and Cav-1 KD supports a role for caveolae in mediating the response to hypoosmotic stress, rather than either of these proteins independent of caveolae. Reconstitution of cavin1 KD with MmCavin1-EGFP restored their ability to reinforce the cortex, as did 4UC1-EGFP cells (Figure 4, d and e; Supplemental Figure S4b). Strikingly, however, cortical reinforcement was effectively abolished by the DrCavin1b-EGFP, which stabilized caveolae. This implied that the process of caveola disassembly was necessary for caveolae to reinforce the cortex in response to mechanical stress.

Cell culture, drugs, and transfection

MCF-10A cells were cultured at 37°C with 5% CO₂ in a medium consisting of DMEM/F-12 with GlutaMAX, supplemented to a final concentration with 5% horse serum, 20 ng mL^–1 human epidermal growth factor (hEGF), 0.5 mg mL^–1, 0.5 mg mL^–1 hydrocortisone, 100 ng mL^–1 cholera toxin, 10 μg mL^–1 human insulin, 50 U mL^–1 penicillin, and 50 μg mL^–1 streptomycin. Cells were transfected using Lipofectamine 3000 (Thermo Fisher Scientific #L300015) for the expression of plasmid constructs according to the manufacturer’s guidelines. Cells were analyzed 24 h posttransfection. Where blebbistatin was used, monolayers were exposed to 25 nM para-aminoblebbistatin for 1 h before mechanical stimulation/analysis. For neomycin experiments, monolayers were exposed to 4 mM neomycin for 1 h before mechanical stimulation/analysis.

Generation of stable-cell lines

Stable cell lines were produced using lentiviral transduction. To generate lentiviruses, HEK-293T cells were transfected with the expression vector, PLL5.0, and lentiviral packaging constructs, pMDLg/pRRE, RSV-Rev, and pMD.G using Lipofectamine 2000 according to the manufacturer’s guidelines. Viruses were concentrated from the supernatant after 48 h using Amicon Ultra-15 Centrifugal Filters (Millipore UFC-9100). MCF-10A cells were transduced with the concentrated viruses, and positive cells isolated using a FACS Aria Cell Sorter (Queensland Brain Institute, University of Queensland).

Immunostaining

MCF-10A cells were cultured to confluence and fixed with 4% (wt/vol) paraformaldehyde (PFA) in cytoskeleton stabilization buffer (10 mM piperazine-N,N’-bis(2-ethanesulfonic acid [PIPES] at pH 6.8, 100 mM KCl, 200 mM sucrose, 2 mM ethylene glycol-bis(b-aminoethyl ether)-N,N,N’,N’-tetraacetic acid [EGTA], 2 mM MgCl₂) on ice for 20 min. Alternatively, cells were fixed for 10 min in –20°C MeOH at -20°C. PFA-fixed cells were permeabilized for 5 min in 0.25% Triton X-100 in phosphate-buffered saline (PBS). Cells were then blocked for 1 h on ice using 3% bovine serum albumin (BSA) in PBS, followed by incubation with primary antibodies diluted in 3% BSA in PBS overnight at 4°C. Cells were washed in PBS and then incubated with similarly diluted secondary antibodies for 90 min at room temperature before final washing in PBS. Coverslips were mounted onto slides using ProLong Gold containing DAPI (Cell Signaling Technologies; catalogue#8961). Immunostaining for phosphorylated proteins was performed using tris-buffered saline (TBS) instead of PBS.

Junctional laser ablation

For junctional laser ablation, MCF-10A monolayers expressing E-cadherin-mCherry were used to visualize cell–cell junctions. To sever these junctions, a Zeiss 710 Inverted LSM equipped with Mai Tai two-photon capabilities was used. Ablation was performed using a 790 nm laser with 30 iterations and 22% transmission. Images were taken at 5 s intervals until the recoil of cell vertices was complete.

Microscopy

Cells were imaged using several microscopes. Fixed samples on glass coverslips were imaged using either a Zeiss Axio two epifluorescence microscope using a 40× 0.95 NA Plan Neofluar or 63× 1.4 NA Plan Apochromat objective, or a Zeiss 880 inverted confocal microscope with AiryScan using a 40× 1.3 NA, or 63× 1.4 NA Plan Apochromat objective. For imaging monolayers cultured on BioFlex plates, imaging was performed using a Zeiss 710 upright confocal microscope using a 40× 1.0 NA water dipping N-Achroplan objective. For TFM, imaging was performed using a Zeiss AxioObserver microscope using a 40× 0.75 NA EC-Plan Neofluar objective. AiryScan scans of F-actin are presented as conventional contrast images; other confocal-imaging modalities are presented as inverted contrast.

Immunoblotting

Immunoblotting was performed using sodium dodecyl sulphate-polyacrylamide gel electrophoresis (SDS–PAGE). Protein was harvested by lysing confluent monolayers in a solution containing 0.1% bromophenol blue, 2% sodium dodecyl sulphate (SDS), 50 mM Tris-Cl (pH 6.8), 10% glycerol, 200 mM dithiothreitol (DTT), and PhosSTOP (Sigma PHOSS-RO). Note that PhosSTOP was only used for detecting nonphosphorylated proteins. Cells were scraped from the dish and transferred to 1.5 ml tubes and heated to 100°C for 10 min. Proteins were loaded into 1.5-mm thick homemade polyacrylamide gels consisting of a lower-running gel containing between 8 and 15% acrylamide and an upper-stacking gel consisting of 4% acrylamide. Protein separation was achieved by electrophoresis at 180V (400 mA) in a running buffer consisting of 25 mM Tris Base, 192 mM glycine, and 3.47 mM sodium dodecyl sulphate. Proteins were transferred to nitrocellulose filters and blocked using either 3% skim milk powder in TBS or 3% BSA in TBS. Primary antibodies were diluted in blocking buffer and incubated with the nitrocellulose filters overnight at 4°C. Filters were then washed in TBS containing 0.1% Tween (TBS-T) and the secondary (horseradish peroxidase [HRP]) antibodies, similarly diluted, were incubated for 90 min at room temperature. For visualization, membranes were exposed to a solution which allows the HRP-conjugated antibodies to emit chemiluminescence (Thermo Fisher ScientificTM 34578).

Application of mechanical tension in monolayers

Hypoosmotic shock was used to induce rapid-cell swelling. This consisted of regular growth media, diluted in a 1:1 ratio with dH2O, giving a concentration of approximately 150 mOsm L^–1. Cells were exposed to hypoosmotic medium for 5 min (unless otherwise stated) under usual culture conditions. If applicable, cells were recovered for 1 h in usual growth medium under usual-culture conditions following hypoosmotic shock. Mechanical stretching was performed using the Flexcell FX-5000 Tension System. Cells were cultured on BioFlex plates which have a silicone base coated in type I collagen. Unless otherwise stated, stretching consisted of cyclic flexing at a frequency of 1 Hz, to a strain of 10%, for 5 min. Stretching was performed at room temperature. To stimulate hypercontraction of actomyosin, growth medium was supplemented with calyculin A to a concentration of 50 nM for 5 min under regular-growth conditions (unless otherwise stated).

Knockdown of human cavin1 and expression of MmCavin1_EGFP, DrCavin1b_EGFP, or 4UC1_EGFP

Endogenous cavin1 was KD from MCF-10A cells via lentivirus-mediated RNAi using the following shRNA sequence 5′-CACCTTCCACGTCAAGAAGATCCGCGAGG-3′ inserted downstream of a U6 promoter within the PLL5.0_EGFP expression vector. This shRNA targets a translated region within human cavin1. Cavin1 variants (MmCavin1, DrCavin1b, or 4UC1) were inserted upstream of an EGFP sequence, with expression driven by an upstream LTR viral promoter. Plasmids were synthesized by Addgene.

Image analysis

Image analysis was performed using FIJI (ImageJ) for fluorescent microscopy, including MATLAB for TFM microscopy. For the analysis of F-actin, a line of standardized length was drawn perpendicular to the cell–cell junction and the fluorescence of all pixels along this line quantified using FIJI. The minimum intensity of this line was subtracted from the maximum intensity as background using Microsoft Excel. For all other junctional/cortical proteins, a freehand line was drawn over the entire junction, and a duplicate of this line transferred to the underlying cytosol. A ratio of the junctional to cytoplasmic ratio was calculated to determine the level of these proteins. For junctional laser ablation, cell–cell junctions were severed, and the length of the junction measured at 5 s intervals for 60 s, normalized to the length of the junction at time 0. For monolayer fracturing, monolayers were imaged as a time sequence at 30 s intervals following exposure to 50 nM calyculin A. Monolayer rupture was quantified according to the first frame in which cells appeared to pull apart. Prism (GraphPad) was used to generate graphs and perform statistical analysis.

We thank our lab colleagues for their support throughout this project, notably Ivar Noordstra for many helpful suggestions and Julia Eckert for help with the biophysical analysis. This work was supported by funds from the National Health and Medical Research Council (1140090 to A.S.Y. and R.G.P., 1136592 to A.S.Y., APP2016410 and APP1181135 to B.M.C.) and Australian Research Council (DP19010287 and DP220103951 to A.S.Y.). R.G.P. was supported by an National Health and Medical Research Council Fellowship APP1156489 and is now an Australian Research Council (ARC) Laureate Fellow. J.W.B. was supported by a scholarship from the UQ Research Training Program. Additionally, this work was supported by a research training scholarship from the University of Queensland (UQRTP). Microscopy was performed at the ACRF/IMB Cancer Research Imaging Facility created with the generous support of the Australian Cancer Research Foundation.