Homotypic endothelial nanotubes induced by wheat germ agglutinin and thrombin

Endothelial barrier formation is maintained by intercellular communication through junctional proteins. The mechanisms involved in maintaining endothelial communication subsequent to barrier disruption remain unclear. It is known that low numbers of endothelial cells can be interconnected by homotypic actin-driven tunneling nanotubes (TNTs) which could be important for intercellular transfer of information in vascular physiology. Here we sought insight into the triggers for TNT formation. Wheat germ agglutinin, a C-type lectin and known label for TNTs, unexpectedly caused striking induction of TNTs. A succinylated derivative was by contrast inactive, suggesting mediation by a sialylated protein. Through siRNA-mediated knockdown we identified that this protein was likely to be CD31, an important sialylated membrane protein normally at endothelial cell junctions. We subsequently considered thrombin as a physiological inducer of endothelial TNTs because it reduces junctional contact. Thrombin reduced junctional contact, redistributed CD31 and induced TNTs, but its effect on TNTs was CD31-independent. Thrombin-induced TNTs nevertheless required PKCα, a known mediator of thrombin-dependent junctional remodelling, suggesting a necessity for junctional proteins in TNT formation. Indeed, TNT-inducing effects of wheat germ agglutinin and thrombin were both correlated with cortical actin rearrangement and similarly Ca2+-dependent, suggesting common underlying mechanisms. Once formed, Ca2+ signalling along TNTs was observed.

Lectin subtype specificity and correlation with redistributed of F-actin. Lectins bind to many carbohydrate molecules. To determine the specificity of the WGA effect we investigated other types of lectin, the binding properties of which are provided in Table S1. Only unmodified WGA induced TNT formation ( Fig. 2A,B). Of particular note was succinylated WGA (a modified WGA containing a bulky succinyl group that binds N-acetylglucosamine but not to sialic acid 26 ) which failed to induce TNTs. In marked contrast, WGA evoked F-actin distribution to a cortical arrangement whereas succinylated WGA did not (Fig. 2C). The data suggest that the induction of TNTs was specific to WGA and required binding to a sialylated protein.
The WGA effect depends on CD31. We speculated that WGA might act via CD31 because CD31 is an important transmembrane protein of endothelial cells which is highly sialylated and previously suggested to be directly modulated by WGA 27 . We first tested if WGA affects the localisation of CD31 because binding of divalent antibodies directed against CD31 has been shown to induce protein clustering [28][29][30] . WGA caused redistribution of CD31 from cell-cell junctions to clusters across the cell surface (white arrows) and small aggregates at the plasma membrane ( Fig. 3A: red arrows). Therefore, we depleted endothelial cells of CD31 using short interfering RNA (Fig. 3B). In CD31-depleted cells, WGA failed to induce significant TNT formation (Fig. 3C,D) or cortical redistribution of F-actin (Fig. 3E). The data suggest that the ability of WGA to induce TNTs depended on CD31. We postulated therefore that WGA evoked TNT formation by binding and clustering the sialic acid residues of CD31 thereby inducing cortical F-actin redistribution.
Thrombin induces TNTs as well as F-actin and CD31 redistribution. We next sought to determine if the mechanism could be relevant to a physiological agonist. We focused on thrombin because it increases endothelial permeability and thus potentially creates fertile ground for TNT formation. We confirmed that thrombin increased inter-endothelial gaps in cells that were seeded at high density and grown to confluence in order to mimic the endothelial barrier ( Supplementary Fig. S2). Application of thrombin to HUVECs led to accentuated cortical F-actin (Fig. 4A). Although CD31 did not cluster in response to thrombin, it changed distribution to follow the line of the cortical F-actin fibres (Fig. 4A, compare the area indicated by the red arrow in control as compared to the red arrow in the thrombin treated cells and the enhanced box area). Thrombin strikingly increased hovering TNT formation that lasted for about 1 hour and then declined (Fig. 4B-D) and the effect was concentration-dependent (Fig. 4E).

Induction of TNTs by thrombin is CD31-independent.
To determine if CD31 is relevant to thrombininduced TNTs we again depleted endothelial cells of CD31. In contrast to WGA, the thrombin effect was unaffected by CD31 depletion (Fig. 5A). Therefore, although there were similarities in the effects of WGA and thrombin, only the action of WGA depended on CD31.  Fig. S3A) had no effect on WGA induced TNT formation (Fig. 5B) and WGA failed to mobilise intracellular Ca 2+ (Fig. 5C). We therefore propose that in endothelial cells WGA clamps cell-cell adhesion complexes via sialylated CD31, evoking TNT formation as the cells move apart. We therefore sought to understand the signalling mechanism by which the physiological activator, thrombin, induces TNTs. Thrombin acts through its proteolytically activated receptor 1 (PAR1) to evoke signalling events that impact upon the molecular organisation of endothelial adhesion complexes 31 . Both Src and the threonine/serine kinase PKC have roles in adherence junction remodelling [32][33][34] . Therefore, we explored the role of Src and PKC in thrombin stimulated TNT formation. HUVECs transfected with siRNA specifically targeted to Src kinase responded to thrombin (Fig. 5D) to a similar extent to cells transfected with control siRNA. These data suggest Src-mediated tyrosine phosphorylation does not play a role in TNT formation evoked by thrombin. To investigate the role of PKC we first used bisindolylmaleimide 1 (BIM), a selective total PKC inhibitor 35 . BIM treatment had no effect on its own but blocked thrombin-induced TNT formation (BIM; Fig. 5E). We investigated the role of PKC isoform specificity using rottlerin, a PKC δ inhibitor and Go6976, a PKCα inhibitor (rottlerin, Go6976; Fig. 5E). Pre-incubation with rottlerin had no effect on thrombin-induced TNTs but treatment of HUVECs with Go6976 significantly inhibited the effect of thrombin. To further test the role of PKCα we transfected HUVECs with PKCα siRNA (Supplementary Fig. S3B). Compared with a scrambled control siRNA, knockdown of PKCα significantly attenuated thrombin-induced TNT formation (Fig. 5F). The data indicates a significant role for PKCα on TNT formation upon thrombin exposure.
WGA and thrombin induced TNT formation is Ca 2+ -dependent. Because of the similarities in the actions of WGA and thrombin on TNTs, we speculated that there might be a common downstream mechanism. We hypothesised that this might be a Ca 2+ -dependent process. Consistent with this suggestion, chelation of extracellular Ca 2+ with EGTA mostly prevented the induction of TNT formation by WGA (Fig. 6A,B) or thrombin (Fig. 6C,D). The data suggest common requirement for extracellular Ca 2+ in WGA and thrombin induction of TNTs. Homotypic endothelial TNTs support Ca 2+ signalling. We speculated that homotypic TNTs would mediate the transfer of ions in order to maintain barrier coordination. To test this suggestion we have observed the propagation of Ca 2+ from a stimulated cell ( Fig. 7: fine arrow at time 0: Movies 1 and 2) along a TNT (arrow at time 60 secs) to a connected cell (shown at 106 secs after stimulation). As observed in other studies 36,37 Ca 2+ propagated between a network of endothelial cells that were connected by TNTs or direct cell-cell contact (Supplementary Videos 1 and 2) but unconnected cells failed to respond (Fig. 7) unless there was cell damage (Supplementary Video 3). These data suggest that homotypic TNTs could maintain endothelial communication by the propagation of Ca 2+ signals.

Discussion
Our findings suggest that endothelial cells maintain Ca 2+ signals during an assault on barrier formation by forming intercellular TNTs. We provide important insight into the mechanisms that trigger endothelial-to-endothelial (homotypic) TNTs. We show striking ability of WGA to induce TNTs and reveal that this induction depends on the highly sialylated endothelial-to-endothelial contact protein CD31. CD31 dependence suggests that this type of induction is likely to be a relatively endothelial-specific phenomenon because CD31 expression is largely restricted to endothelial cells. As well as a potential approach to achieving relatively endothelial-specific TNTs, the CD31-dependent effect of WGA suggests an important relationship between TNT formation, mechanical force detection and the molecular mechanisms of cell-to-cell contact. CD31 is a lectin that displays sialic acid-dependent homophilic binding 29 , in addition to heterophilic lectin activity to other sialylated proteins 38 . In Fig. 8, we depict CD31 homotypically bound to sialic acid residues on the extracellular surface of juxtaposed cells. In addition, CD31 also interacts heterotypically with other sialylated junctional proteins such as VE-cadherin. Here, we postulate that WGA clusters sialic acid residues exposed on non-bound CD31 whilst aggregating an as yet unknown CD31-containing complex at the cellular junction. Thus, WGA anchors a sialic acid-dependent complex at cell-cell junctions permitting thin membrane extensions to form as cells move apart. The concept of regulated cell-to-cell contact being important in TNT induction is reinforced by our observation that thrombin-induced cell-separation readily led to the appearance of TNTs. Thrombin has previously been shown to act via proteolysis of PAR1 receptors on the endothelial membrane inducing an intracellular signalling cascade that impacts upon the cytoskeleton 31 . Although thrombin evoked TNT formation was CD31-independent, like CD31-dependent WGA-induced TNT formation this correlated with enhanced cortical F-actin, suggesting common downstream mechanisms. Indeed, Konstantoulaki et al. observed that thrombin induced aggregates of VE-cadherin and catenins (actin binding proteins) at the plasma membrane and, as cells retracted, thin membrane extensions were formed between the cells 34 . These remodelling events required PKC modifications of VE-Cadherin and β-catenin although the group had not characterised the intercellular connections as TNTs 39 . Here we extend these findings to show that thrombin signalling via PKCα leads to TNT formation. Future studies would explore this signalling pathway in order to understand the cell-cell junctional remodelling required for thrombin-induced TNT formation. The findings support the hypothesis that homotypic TNTs are relevant to endothelial biology and suggest contexts in which the TNTs might be induced in physiology and patho-physiology. The mechanism of the Ca 2+ -dependence of TNT formation is unclear but a possibility is that it reflects Ca 2+ -dependence of VE-cadherin, a key Ca 2+ -dependent junctional protein of endothelial cells and modulator of actin organisation 40 . Junctional proteins are disassembled and endocytosed during extracellular Ca 2+ depletion 41,42 therefore the inhibition of TNT formation in Ca 2+ free medium suggests they are necessary for TNT formation. WGA, but not succinylated WGA, failed to induce intracellular Ca 2+ mobilization and inhibited VEGF-evoked Ca 2+ elevation ( Supplementary Fig. S4A,B), suggesting that the WGA effect on TNTs was not related to elevation of intracellular Ca 2+ . Similarly, although thrombin induces a dose-dependent Ca 2+ response ( Supplementary Fig. S4C), thrombin-mediated actin rearrangements have been suggested to be Ca 2+ -independent 43,44 .
CD31 is abundant in endothelial cells, a commonly used endothelial cell marker and a constituent of intercellular endothelial junctions where it tightens junctions and regulates leukocyte extravasation and other aspects of endothelial biology 8,45 . Desialylation by NEU1 sialidase plays a role in counteracting CD31's positive impact on endothelial cell migration and angiogenesis 46 . In addition, evidence suggests CD31 plays a pivotal role in the rapid restoration of the vascular permeability barrier following inflammatory or thrombotic challenge 47 (Supplementary Figure S2). As sialylation also plays important roles in the formation of junctional proteins further studies should elucidate the role of sialic acid in the induction of endothelial TNTs and their role in barrier restoration.
There have been few studies describing TNTs in vivo or ex vivo 48,49,12,50 , presumably because the technical challenges are substantial. Detailed studies of this biology, at least in mammalian vasculature, are likely to remain challenging until specific TNT tools are identified and developed for live in vivo imaging at high spatial resolution. Nevertheless, our findings suggest a framework for studies in which TNT formation is provoked by thrombin and modulated CD31. A potentially relevant modulator of CD31 is shear stress. CD31 has been suggested to act as a shear stress sensor in a triad or larger complex with VE-cadherin and VEGF receptor 2 51,52 . Dependence of the sensing of shear stress on the glycocalyx has been suggested and glycocalyx is a sialylated structure 53,54 . Perhaps when there is separation of endothelial cells there is desialylation that triggers TNT formation via CD31. Although mitochondrial exchange between TNTs has been observed in cultured endothelial cells, this relatively slow propagation, which has been shown to have important functions in the rescue of distressed cells, is unlikely to be necessary for the preservation of intercellular communication during barrier disruption. Here we have shown that homotypic TNTs permit the propagation of Ca 2+ to connecting cells. Whether this is dependent on Cx43-containing gap junctions and IP 3 propagation as described in other studies 55 requires further analysis. However, these data suggest that TNTs could maintain the Ca 2+ oscillations and waves that are required for the spatiotemporal control of endothelial permeability.
WGA is a plant lectin widely used for staining plasma membranes and thus TNTs. Although CD31 is most abundant in endothelial cells, our observations suggest caution when using WGA as a label for TNTs even in non-endothelial cell types. There are reports of effects of WGA on cytoskeleton of enterocytes 56 and granulocytes 57 . To further this research, WGA would serve as a useful tool to identify the exact molecular composition of the junctional complex required for TNT formation between endothelial cells. WGA is also a Ca 2+ -dependent (C-type) lectin. The human genome encodes for over 100 proteins that are classified as containing C-type lectin like domains (CLEC superfamily of lectins) 58 . Interestingly, there are members of this superfamily that play significant roles in the integrity or the function of the vasculature 59,60 . For instance, selectins are C-type lectins expressed on both immune cells and endothelial cells that are important for leukocyte migration across the endothelium 61 . It would be interesting to see whether these lectin-containing proteins induce TNT formation in order to preserve cell-cell signalling during endothelial barrier disruption.
In conclusion, this study strengthens the hypothesis that homotypic TNTs have under-appreciated importance in endothelial biology. Importantly it has identified molecular mechanisms by which the nanotubes can be regulated -specifically through modulation of sialylated CD31 and the action of thrombin via PKCα. Based on these findings we suggest that homotypic endothelial TNTs are likely to be important in vascular regions exposed to disturbed mechanical force, inflammation or thrombosis, in which contexts the TNTs may serve to facilitate endothelial cell coordination when gaps between endothelial cells emerge.  Methods HUVEC culture. Human umbilical vein endothelial cells (HUVECs: Lonza) were maintained in endothelial growth medium (EGM-2) supplemented with 2% foetal calf serum (FCS) and growth factor bullet kit (Lonza). Experiments were performed on passage 2-5 cells. Cells were maintained at 37 °C in a 5% CO 2 incubator.

EPCs.
Whole peripheral blood was drawn from subjects into EDTA coated tubes. All participants provided written informed consent, according to the declaration of Helsinki; ethical approval was provided by the Harrogate and Leeds (Central) research ethics committees. All procedures were carried out in accordance with these regulations. For EPC extraction an equal volume of Dulbecco Phosphate buffered saline (DPBS) was added to the EDTA tubes, and the resultant mixture was layered onto Ficoll paque plus. As per manufacturer's instructions, density gradient centrifugation was carried out, and peripheral blood mononuclear cells (PBMCs) were aspirated from the buffy coat layer. These were washed and re-suspended with DPBS and subject to re-centrifugation at 300 × g for 10 mins at 18 °C. The cell pellet obtained was suspended in EGM-2 growth medium with EGM-2 single quot bullet kits and 10% foetal calf serum (FCS) (v/v), with an equivalent of 5 million cells per 2 ml of medium placed per well of a 6-well fibronectin coated plate. Cells were then incubated at 37 °C with 5% CO 2 . After 3 days, cells were washed with DPBS and fresh medium replaced. The cells firmly adherent between 4-7 days were predominantly early outgrowth EPCs. The phenotype of early EPCs was confirmed by staining with anti-CD31 using fluorescence microscopy. Immunofluorescence. Cells were detached (48-hour post transfection for siRNA experiments) and transferred to glass-bottomed 35 mm dishes, 30,000/dish (WGA) or 40,000/dish (thrombin), with fresh culture medium and allowed to spread for 24-48 hours. Cells were treated in serum-free M199 plus 10 mM HEPES medium for 30 mins then incubated with the stated concentration of lectin (Vector labs). Thrombin (Sigma) was used at 1-2 U/ml unless otherwise stated. One NIH U/ml of human thrombin activity is equivalent to 0.324 μg/ml (i.e. 9 nM). 5 mM EGTA was used in some experiments to chelate extracellular Ca 2+ . Lectin solution was replaced with fresh medium for live cell imaging or cells were fixed in situ with 1% glutaraldehyde/2% PFA for 10 mins to preserve the TNT structure. For pharmacological studies cells were serum starved for 30 mins prior to incubation for 30 mins with 5 µM bisindolylmaleimide 1 (BIM) or 5 µM rottlerin or 100 nM Go6976. Immunostained cells were permeabilised in 0.1% TritonX-100 for 10 mins at room temperature then incubated with anti-tubulin 1:500 (Covance) or anti-CD31 1:100 (Abcam) for 1 hr at room temperature, then washed and incubated with the appropriate secondary IgG (Stratech Scientific, Jacksons Immuno Research). Cells were stained with either Cell Mask (1 μl in 10 mls PBS), rhodamine phalloidin (Cytoskeleton Inc.), MitoTracker (Thermofisher) or WGA-FITC (Vector Labs) along with DAPI as per manufacturer's instructions.
Microscopy. Cells were visualized on an Olympus IX-70 inverted microscope using ×20, ×40 or ×60 UPLAN objectives supported by a DeltaVision deconvolution system (Applied Precision LLC) with SoftWorx image acquisition and analysis software. Images were captured on a Roper CoolSNAP HQ CCD camera at 0.05 secs exposure. Epifluorescence was recorded using filter sets for FITC/TRITC. To analyse hovering TNTs 35-40 0.5 μm z focal planes were captured and analysed with ImageJ orthogonal views, dynamic resplice or 3D project function.
For the analysis of WGA-induced TNTs random fields across the dish were captured in a blinded manner using the SoftWorx stage viewer. As thrombin-induced TNTs were more prevalent during gaps, to prevent bias imaging of gaps, a control dish was used to capture random fields of view where the points of image capture was marked onto the stage viewer and used as markers for all subsequent images. ImageJ was used post acquisition to process and prepare micrographs.
Live cell Ca 2+ imaging. A Zeiss Axiovet 200 microscope coupled to an Eppendorf FemtoJet microinjection system was used to capture intracellular Ca 2+ changes in Fluo4-loaded HUVECs. A polished glass pipette mounted onto the microinjection system was used to stimulate the cells and both epifluorescence (FITC) and bright field images were simultaneously captured every 2 secs using OpenLab software. Images were further compiled using ImageJ software. 10; Osmolality adjusted to 290 mOsm with NaCl; pH titrated to 7.4 with 4 M NaOH. Ca 2+ free solution was SBS without added Ca 2+ . Cells were washed three times with SBS before measurements were made at room temperature (21 ± 2 °C) on a 96-well fluorescence plate reader (FlexStation II 384 , Molecular Devices). Fura-2 was excited at 340 and 380 nm and emission collected at 510 nm. Readings were made every 10 secs. The change (Δ) in intracellular Ca 2+ concentration was indicated as the ratio of Fura-2 emission intensities for 340 nm and 380 nm excitation (ΔF ratio). Data analysis. Subsequent to confirming that TNTs hovered over the substrate, for consistency, TNTs were discerned if they connected 2 endothelial cells, were more than 5 μm in length and less than 1 μm in width. TNT length was analysed using the line and measurement function of ImageJ and TNTs counted using the cell counter. TNT counts are displayed as the percentage of TNTs to cells in a field of view. For thrombin-induced intercellular gap analysis images were segmented using the ImageJ threshold function, inverted and gaps analysed as particles measured as the percentage area of the field of view. Each time point was normalized to its own control but simplified to one control bar on the graph.

Intracellular Ca
Data were analysed and figures prepared using Origin 8.0 software (OriginLab Corporation). 5-10 fields of view ×3 replicates were analysed for each condition and presented as n/N, which indicates the number of replicates (n) and the number of individual fields of view per point (N). Averaged data are shown as mean ± SD. Data were produced in pairs and analysed statistically with two-sample t tests or one-way ANOVA using OriginPro 8.6 software (OriginLab). In all cases statistically significant difference is indicated by *P < 0.05. All measurements performed blind to verify non-bias. FlexStation experiments data are presented as n/N, which indicates the number of independent experiments (n) and the number of individual wells (replicates) in the 96-well plate. Paired data sets were compared using two-tailed Student's t-tests and expressed as mean ± standard deviation (SD) unless stated. Data Availability. All data generated or analysed during this study are included in this published article (and its Supplementary Information files).