Rab35 and its effectors promote formation of tunneling nanotubes in neuronal cells

Tunneling nanotubes (TNTs) are F-actin rich structures that connect distant cells, allowing the transport of many cellular components, including vesicles, organelles and molecules. Rab GTPases are the major regulators of vesicle trafficking and also participate in actin cytoskeleton remodelling, therefore, we examined their role in TNTs. Rab35 functions with several proteins that are involved in vesicle trafficking such as ACAP2, MICAL-L1, ARF6 and EHD1, which are known to be involved in neurite outgrowth. Here we show that Rab35 promotes TNT formation and TNT-mediated vesicle transfer in a neuronal cell line. Furthermore, our data indicates that Rab35-GTP, ACAP2, ARF6-GDP and EHD1 act in a cascade mechanism to promote TNT formation. Interestingly, MICAL-L1 overexpression, shown to be necessary for the action of Rab35 on neurite outgrowth, showed no effect on TNTs, indicating that TNT formation and neurite outgrowth may be processed through similar but not identical pathways, further supporting the unique identity of these cellular protrusions.


Scientific RepoRtS
| (2020) 10:16803 | https://doi.org/10.1038/s41598-020-74013-z www.nature.com/scientificreports/ tions induced by Rab35 identified by morphological criteria were functional TNTs, we next measured the transfer of DiD-labelled vesicles between two populations of cells in co-culture as previously described 30,60 . Briefly, donor cells (transfected with either GFP-Rab35-WT, active (Q67L) or inactive (S22N)) were loaded with a fluorescent membrane dye (Vybrant DiD) to label all internal vesicles, mixed at a 1:1 ratio with acceptor cells (transfected with mCherry-H2B, to distinguish them from donors) and co-cultured for 16 h. As controls, donor cells were transfected only with GFP empty vectors (see Fig. S2). The transfer was measured by flow cytometry and the corresponding gating strategy is shown in Fig. S3. Additionally, secretion-based transfer was quantified (Fig. S2, see "Materials and methods" for more details). To visualize the co-culture setup, confocal images were acquired and represented in Fig. S4. Overexpression of GFP-Rab35-WT and Q67L significantly increased the transfer of DiD-labelled vesicles, as measured by flow cytometry and visualized by microscopy (Fig. 1c, Figs. S3a, S4). In contrast, overexpression of the dominant negative mutant of GFP-Rab35-S22N showed a decrease in vesicle transfer between the cells as compared to the control (Fig. 1c, Fig. S3a). Secretion-based transfer was much lower than total transfer (Fig. S3a), suggesting that it had little to no impact on the transfer of the vesicles in our cells. Indeed, increase/decrease in cell-cell contact-based transfer is confirmed by the same trend in the amount of TNT-connected cells. Thus, this data shows that active Rab35 promotes the formation of TNTs and leads to an increase in cargo transfer between cells indicating that these TNTs are functional.
Scientific RepoRtS | (2020) 10:16803 | https://doi.org/10.1038/s41598-020-74013-z www.nature.com/scientificreports/ effector ACAP2/centaurin-β2 which was shown to be recruited by Rab35 and to act downstream of it in promoting neurite outgrowth 33 . By using the same methods described before, donor CAD cells were transfected with GFP (control) and GFP-ACAP2-WT for 24 h and re-plated for 16 h as single cell populations to analyse the percentage of TNT-connected cells or mixed in co-culture with mCherry-H2B acceptor cells to analyse vesicle transfer by flow cytometry. Overexpression of GFP-ACAP2-WT resulted in an increase in both the number of TNT-connected cells (Fig. 2a,b), where TNTs indeed contained actin and GFP-ACAP2 (Fig. S1), and vesicle transfer between the cells (Fig. 2c, Figs. S3b, S4). This data suggested that Rab35 may regulate TNT formation between the cells through ACAP2.
Inactivation of ARF6 leads to an increase in TNTs. ACAP2 is a GAP for ARF6, and inactivation of ARF6 by ACAP2 is required for successful neurite outgrowth of PC12 cells 45 . We hypothesized that a similar mechanism could be involved in TNT formation. To test this idea, we first analysed the effect of ARF6 mutants on TNTs. We transfected cells with either GFP-ARF6-WT (wild type), GFP-ARF6-Q67L (active, GTP bound) and GFP-ARF6-T27N (dominant negative, GDP bound) mutants and measured the number of TNTs as described above. Overexpression of GFP-ARF6-WT and GFP-ARF6-Q67L had no impact on the number of TNT-connected cells (Fig. 3a,b), nor on vesicle transfer between the cells (Fig. 3c). Furthermore, the overexpression of inactive GFP-ARF6-T27N in the cells demonstrated an increase in the number of TNT-connected cells (Fig. 3a,b), whose TNTs were supported by actin and even contained DiD-stained vesicles; overexpression also increased vesicle transfer between cells, as quantified by flow cytometry and visualized by confocal microscopy (Fig. 3c, Figs. S3c, S4). Overall, this data showed that the inactivation of ARF6 (by overexpressing its dominant negative mutant) increases the formation of functional TNTs, indicating that ARF6 in its inactive state promotes TNT formation. These results are consistent with the observation that the overexpression of the ARF6 GAP ACAP2 (Fig. 2) increases TNT formation. EHD1 is required for TNT formation. Considering that active ARF6 converts PI4P to PI(4,5)P2 by activation of PIP5 kinase 61 , it was speculated that inactivation of ARF6 might be required to maintain PI4P which in turn can recruit EHD1 that has an essential role in promoting neurite outgrowth following Rab35 activation 33 . Following this line of thought, we next tested the role of EHD1 in TNTs. Overexpression of the wild type form of EHD1 showed an increase in the number of TNT-connected cells (Fig. 4a,b); these TNTs contained throughout their length not only actin, but the EHD1 protein, as well as DiD vesicles, suggesting the importance of EHD1 in TNT formation and the ability of TNTs to transfer cargo (Fig. S1). The amount of vesicles transferred between All graphs from at least three independent experiments and show mean ± SEM. (ns, not significant; *P < 0.05, **P < 0.01, ***P < 0.001; by unpaired Student's t-test). All bar graphs were analysed and represented using Graph Pad Prism version 7. Scale bars 10 µm.
Scientific RepoRtS | (2020) 10:16803 | https://doi.org/10.1038/s41598-020-74013-z www.nature.com/scientificreports/ the cells was also increased, as quantified by flow cytometry and represented by confocal microscopy (Fig. 4c, Figs. S3d, S4). By using shRNA against EHD1 we observed a decrease in the expression of endogenous EHD1 (Fig. 4e,  Fig. S5). Of importance, these cells showed a decrease in the number of TNT-connected cells (Fig. 4d,f) and in vesicle transfer between the cells (Fig. 4g, Fig. S3e). In addition, by re-expressing EHD1 in cells depleted of endogenous EHD1, we could rescue the number of TNT-connected cells (Fig. 4d,f) and the transfer rate (measured by flow cytometry as % of acceptor cells containing transferred vesicles) (Fig. 4g, Fig. S3e). Altogether, this data indicates that EHD1 is required for the formation of functional TNTs.
One important control for the specificity of this pathway in TNT formation, was to check whether Rab35 and its effectors were able to impact filopodia formation. Indeed, it has been clearly shown in CAD and other neuronal cells, that TNTs and filopodia are distinct structures 4 for which the mechanism of formation is different 30 . Interestingly, overexpressing either GFP-ARF6-T27N, GFP-ACAP2, GFP-Rab35 or GFP-EHD1 showed no effect on the number of attached filopodia (Fig. S6a,b), strengthening the specificity of this pathway for TNTs.
ARF6-GDP, ACAP2 and Rab35 act upstream of EHD1 to regulate TNTs. The data presented above suggest that Rab35, ACAP2, ARF6-GDP, and EHD1 act along the same pathway in the formation of TNTs. To check whether this was indeed the case, we started by overexpressing GFP-ARF6-T27N in cells in which endogenous EHD1 was depleted by shRNA and measured both the number of TNT-connected cells and the vesicle transfer by flow cytometry. We found no change in the percentage of TNT-connected cells compared to the control (Fig. 5a,b), as well as in vesicle transfer between the cells (Fig. 5c, Fig. S3f). Thus, the increase of TNTs observed upon GFP-ARF6-T27N overexpression requires the presence of EHD1, indicating that EHD1 acts downstream of ARF6-T27N. Next, we performed a similar experiment by overexpressing GFP-ACAP2-WT  (shControl + GFP = 100%, shEHD1 + GFP = 51.5 ± 9.0%, shEHD1 + GFP-EHD1 = 107.9 ± 5.6%). All graphs from at least three independent experiments and shows mean ± SEM. (ns, not significant; *P < 0.05, **P < 0.01, ***P < 0.001; by one-way ANOVA with Tukey's multiple comparison test). All bar graphs were analysed and represented using Graph Pad Prism version 7. Scale bars 10 µm.
Scientific RepoRtS | (2020) 10:16803 | https://doi.org/10.1038/s41598-020-74013-z www.nature.com/scientificreports/ in EHD1 depleted cells. Like for ARF6-T27N, the overexpression of ACAP2 showed no change in the percentage of TNT-connected cells (Fig. 5d,e), nor in vesicle transfer (Fig. 5f, Fig. S3g) when EHD1 was depleted. This is consistent with the hypothesis that EHD1 acts downstream of ACAP2 and ARF6. Finally, to test the role of Rab35 in this activation cascade we overexpressed its wild type form in EHD1 depleted cells. Again, we observed no significant change in the percentage of TNT-connected cells (Fig. 5g,h) and vesicle transfer (Fig. 5, Fig. S3h). This corroborates our hypothesis that Rab35 is upstream of EHD1 in the pathway of TNT regulation. This pathway is similar to what was previously shown in neurite outgrowth, where Rab35 recruits both ACAP2 and MICAL-L1 on endosomes, and where MICAL-L1 works along with ACAP2 to increase the formation of neurite protrusions by contributing to directly recruit EHD1 33 . We thus tested whether the overexpression of GFP-MICAL-L1-WT would increase the number of TNTs and transfer of vesicles by using the same assays as described above. Interestingly, no significant effect either on the number of TNT-connected cells (Fig. 6a,b) or vesicle transfer between the cells (Fig. 6c, Fig. S3i) was observed. Thus, this data suggests that Rab35 may regulate TNTs through ACAP2 and ARF6, while it assumes no MICAL-L1 involvement in this pathway. Therefore, our results differentiate the pathway of neurite outgrowth and TNT regulation and strengthen the hypothesis that TNTs are unique protrusions with a distinct formation mechanism.

Discussion
In this study we have established that Rab35-GTP, ACAP2, ARF6-GDP and EHD1 positively regulate TNT formation in neuronal CAD cells. We have also shown that EHD1 is required for TNT formation and acts downstream of Rab35, ACAP2 and ARF6-GDP. We propose that Rab35 and its effector ACAP2 promote TNT formation by inactivating ARF6. The consecutive EHD1 recruitment (see below) then favours the formation of TNTs (Fig. 7). Additionally, we have shown that MICAL-L1 does not participate in the formation of TNTs thus differentiating the mechanism of TNT formation from the one for neurite outgrowth where MICAL-L1, following activation of Rab35, plays an important role. These data uncover new players in TNTs and broaden the currently insufficient knowledge of the role of vesicle trafficking in the field of TNT formation. Furthermore, this study demonstrates that TNTs use a distinct mechanism of formation, reinforcing the hypothesis that they are indeed specific structures different from other cellular protrusions.
Vesicle trafficking was reported to be essential in the process of protrusion formation 32,33 . Though it is intuitive that the cytoskeleton may be one of the major influencers in the formation of neurites, it was shown that the formation of these protrusions is also controlled by vesicle trafficking 62 and by Rab GTPases which regulate proteins that are directed to the neurite outgrowth 63 . Similar to neurites, in addition to being composed of actin cytoskeleton, TNTs are membranous structures, which raises the question of how membrane is supplied for their growth and what are the key molecular players involved in this process. Previous studies indicated that in addition to actin remodelling, membrane recycling pathways play a role in TNT formation 59 . In support of this hypothesis, we recently reported that inhibition of membrane recycling from the endosomes to the plasma membrane has a negative impact on TNTs, thus implicating the role of vesicle recycling in the formation of TNTs 32 .
Furthermore, a screen of 41 Rab proteins previously conducted in our lab led to the discovery that Rab11a and Rab8a act in a cascade mechanism to stimulate TNT formation through the action of VAMP3 and membrane recycling 32 . Among the others, Rab35, was another strong positive candidate for TNT formation in our screen. As Rab35 is an important regulator of membrane recycling 47,48 , and was shown to be a master regulator of neurite outgrowth 33 , we hypothesize that neurite outgrowth and TNT formation may be related or they might form by employing a similar mechanism.
During neurite outgrowth, Rab35 acts along with ACAP2 to inactivate ARF6 and facilitate the recruitment of EHD1 33 . Since in our study we have found Rab35 to be a positive regulator of TNT formation, and knowing the Rab35 involvement in protrusion formation through vesicle trafficking, we propose that when Rab35 is overexpressed as a wild type or active form in our cells, it operates along the same vesicle recycling pathway to positively regulate TNTs.
Remarkably, Rab35 effectors ACAP2, GDP-bound (inactive) ARF6 and EHD1 were individually shown to induce functional TNTs. In addition, we found that their effect was dependent on the presence of EHD1 which acts as a downstream effector molecule in the pathway (Fig. 7). ACAP2, being a GAP of ARF6, inactivates ARF6, which has been shown to regulate the level of the lipids in the membrane. Specifically, the active form of ARF6 promotes the conversion of PI4P to PI(4,5)P2 61 . Therefore, when ACAP2 inactivates ARF6 we speculate that there should be more PI4P in the membrane of recycling endosomes. EHD1 has an affinity for PI4P over PI(4,5) P2 64 , which might consequently lead to EHD1 recruitment to the membrane compartments rich in PI4P. In our neuronal cells inactive GDP-ARF6 positively influenced the formation of TNTs. Furthermore, activated ARF6 was previously found to indirectly impact actin cytoskeleton by recycling the proteins such as CDC42 and Rac to the cell leading edge to promote cell migration 65 . Considering CDC42 negatively regulates TNTs in CAD neuronal cells 30 , we postulate that in our case ARF6-GDP positively regulated TNTs by recruiting EHD1, but we do not exclude an additional role of ARF6 on CDC42 and actin cytoskeleton that would also affect formation of TNTs.
Interestingly, during neurite outgrowth, EHD1 was found to be recruited to recycling endosomes by two different pathways, on one side through a direct interaction with MICAL-L1, and on the other indirectly through the action of ACAP2 on AFR6 33 . In our hands, MICAL-L1 overexpression had no effect on TNTs, suggesting that it does not have a role in TNT formation. Therefore, we can hypothesize that Rab35 recruits ACAP2, which in turn inactivates ARF6 leading to an increase in PI4P levels, thus enabling the membrane recruitment of EHD1 which then positively regulates TNT formation (Fig. 7). Overall this data demonstrates that even though they use similar pathways through the activation of Rab35 and the recruitment of EHD1, neurite outgrowth and TNT formation are regulated in a different manner. We also show here that this pathway does not affect the number of How would EHD1 promote TNT formation? Endocytic recycling has been proposed to be crucial to supply membranes and/or proteins to neurite tips and growing protrusions like cilia, filopodia and also TNTs to enable their extension 33,36,54,66,67 . In a previous study we demonstrated that Rab11a and Rab8a act in a cascade upstream of VAMP3 to induce TNTs 32 . In the same study, we showed that by blocking membrane recycling from the endosomes to the plasma membrane using a drug primaquine, there is a subsequent reduction of TNT formation, reinforcing the hypothesis of the involvement of this process also in TNTs like in other protrusions 32 . The mechanism by which EHD1 works for both neurite outgrowth and TNT formation could be the same, through promoting endocytic recycling, and in particular the trafficking from recycling endosomes to the plasma membrane 33,53 . Mechanistically, being a dynamin-like protein which is found on the endocytic recycling compartment, EHD1 enables the fission of PI4P-rich vesicles 68 which eventually may provide membrane material for TNTs. In accordance with what has been previously published and with the information we obtained in this study, we propose that the mechanism through which EHD1 may facilitate TNT growth in response to active Rab35 is by supplying membrane material by vesicles targeted to the budding TNT from recycling endosomes. Interestingly, EHD1 was found to localize also at the preciliary membranes where, by its involvement in ciliary vesicle tubulation, was shown to initiate a cilium 36 . Albeit our work clearly emphasizes that Rab GTPases and their respective interacting partners, i.e. EHD1, have a crucial role in the process of TNT formation, it would be interesting to elucidate how the vesicles positive for the aforementioned proteins favour the formation of TNTs and whether these proteins are found in its base and/or along its length.
This study provides novel insights into the role of Rab GTPases and recycling endosomes and thus deepens the knowledge of TNT formation but also assigns a new and appealing role to Rab proteins in their already broad pool of functions. This will provide us with a novel direction for the study of molecular effectors involved in this currently poorly understood but highly important process of TNT formation.
The two cells which were connected with at least one continuous connection were marked as TNT-connected cells. TNT-connected cells were assessed and quantified only in the middle and upper stack; first 5 slices were excluded from the analysis to avoid counting substrate-attached protrusions (Movie S1). At least 50 transfected green cells were counted for each sample per each individual experiment, having in total at least 150 cells analysed. To quantify TNTs following transfection, only TNTs found between two transfected cells, and between one transfected and one non-transfected were counted; TNTs formed between two non-transfected cells were not taken into consideration. Image analyses and displays of raw data, such as Z-projections, were obtained using ICY software and ImageJ 60 . Z-stack animation (Movie S1) was processed in ImageJ and Adobe Premiere Pro.
Attached filopodia detection and quantification. As described before 20 , confluent CAD cells were mechanically detached and counted, and 300,000 cells were plated for 6 h per well in a 6 well plate. Cells were transfected as described above, using the abovementioned plasmids. At 24 h post-transfection, cells were detached and counted, and 140,000 cells were plated for 16 h on ibidi μ-dishes (ibidi GmbH). Cells were then fixed with 4% PFA for 20 min at RT and washed three times with PBS. Cells were then incubated for permeabilization and blocking with 2% BSA including 0.0075% saponin at RT for 1 h. Primary monoclonal antibody of vinculin (Sigma, 1:500) was prepared in PBS having 2% BSA and 0.01% saponin and incubated at RT for 1 h. After several washes with PBS cells were incubated with secondary antibody goat anti-mouse Alexa Fluor 546 (Invitrogen, Thermo Fisher Scientific) in the same solution at RT for 1 h. Cells were then stained with HCS Cell Mask Blue Stain (Invitrogen, Thermo Fisher Scientific, 1:5000) in PBS for 20 min, washed several times and mounted. Quantification was performed as described before 20,30,32 by (1) creating the ROI restricted to the outer Fluorescence microscopy to image the transfer of DiD-labelled vesicles. As  www.nature.com/scientificreports/ The labelled donor cells were mixed in a 1:1 ratio with H2B-mCherry transfected acceptor cells and plated at subconfluence (200,000 cells per well) on 24-well plates for 16 h at 37 °C. Cells were then washed with PBS, mechanically detached from the dish by pipetting up and down with 500 μl PBS and passed through sterile 40 μm nylon cell strainers (BD Falcon) in order to obtain single-cell suspensions. Cell suspensions were fixed with 500 μl of 4% PFA (2% final solution), as previously described 30,32 . A 'supernatant' control was performed to verify that the transfer of vesicles between cells is mainly cell-cell dependent and not secretion-based. After overnight culturing, conditioned medium from donor cells was centrifuged at 1000 rpm for 5 min to remove cell debris and preserve the vesicles, and was then transferred to the acceptor cells for an additional 16 h of culture (Fig. S2). An additional 'mixture' control was added by mixing samples of separate donor and acceptor cells after fixation to remove any possible false positive results due to sample preservation in the fixative. Both of these controls were subtracted from the total transfer in coculture to obtain cell-cell dependent transfer represented in the final figures. Flow cytometry data were acquired using an LSR Fortessa flow cytometer (BD Biosciences). GFP fluorescence was analysed at 488 nm excitation wavelength, mCherry fluorescence was analysed at a 561 nm excitation wavelength and DiD fluorescence was analysed at a 640 nm excitation wavelength. Samples were analysed at a high flow rate, corresponding to 200-400 events per second, and at least 10,000 events were acquired for each condition, as previously stated 30,32 . The data was analysed using FlowJo analysis software.