We used CDRs as a read out for an RNAi-based phenotypic screening to identify new critical players promoting their formation and set out to test whether they are acting as oscillating waves steering cells during chemotaxis.
Given the involvement of CDRs in endocytic processes12,13, we specifically targeted each of the mammalian members of the RAB GTPase family, which includes more than 60 independent genes27. RAB GTPases, by controlling key steps of endocytosis and vesicular trafficking, are necessary for the execution of actin-based polarized functions that are in turn essential for cell migration and invasion27. We devised a multistep screening strategy. For the primary screening, we used mouse embryonic fibroblasts (MEFs) as model system because in response to PDGF stimulation the large majority of cells form easily detectable and prominent CDRs that appear in a highly synchronous, temporal fashion (with a peak after 10 min of stimulation). We employed a custom siRNA library, targeting all the members of the family (about 60 genes using three individual siRNAs for each gene, Supplementary Data 1), arrayed onto 96-wells imaging plates. We developed a fluorescence-based imaging pipeline to automatically monitor and quantitatively score the formation of CDR in PDGF-stimulated MEFs (for details see Methods and Supplementary Figure 1A–C). The accuracy of our pipeline was evaluated by manually/visually inspecting a subset of randomly chosen images in order to measure the ability to correctly recognize CDRs. The true-positive and false-positive rate of CDR recognition were 0.94 and 0.23, respectively. Raw data were quality controlled by removing images with a low number of nuclei (in all these cases the corresponding siRNA was considered as inconclusive/cytotoxic) and by discarding out-of focus images. In addition, we systematically evaluated plates based on the transfection efficiency of our cells (~80 %) that was measured by counting the percentage of polylobed nuclei upon Incenp downregulation (Supplementary Figure 1C). Finally, the efficiency in forming CDRs was normalized in each experimental condition with respect to the negative control (siEGFP) of the screening (Methods) to obtain a CDR score that was used to rank the various treatments (Fig. 1a). The top regulator of CDRs included, as expected, the β isoform of the PDGF receptor, and few RAB GTPases, among which RAB35, the silencing of which resulted in one of the most robust inhibition of CDRs formation (Fig. 1a and Supplementary Data 2). Next, we performed an independent secondary validation step by focusing on those genes for which at least two out of three siRNAs resulted in a CDR score <0.4 (Supplementary Data 2). Gene silencing in these cases was verified by quantitative RTPCR analysis (Fig. 1b). The silencing of RAB35, RAB8A and RAB8B resulted in a robust and reproducible decreased in CDR activity, thus corroborating the validity of the primary screen and providing evidence that RAB35 is the main regulator of CDR formation among the mammalian RAB protein family (Fig. 1b).
To prove that the CDR-phenotype associated with RAB35 silencing is the result of specific targeting of the gene and rule out off-target effects, we performed a rescue experiment. To this end, we generated a population of MEF cells expressing the HA-tagged human form of RAB35, which is resistant to the murine oligo used to induce the silencing of the endogenous gene product, in a doxycycline-inducible fashion. Cells interfered for endogenous RAB35 displayed reduced CDR formation, which was fully rescued by the expression of the human protein (Fig. 2a). Of note, ectopically expressed RAB35 was diffuse on the cytoplasm and present on the plasma membrane, but re-localized to CDRs following stimulation with PDGF (Supplementary Figure 1D).
In addition to be migratory, CDRs are also sites of macropinocytic internalization28. Real-time phase contrast microscopy, consistently, revealed the formation of large fluid-filled, vesicle-like structures that invariably form following CDR closure (Supplementary Movie 1). This was mirrored by PDGF-mediated increase in the internalization of large molecular weight, fluorescently labelled Dextran, a bona fide macropinocytosis cargo. Importantly, MEFs stably downregulated for RAB35 were severely impaired in the internalization of dextran (Fig. 2b), reinforcing the notion that this GTPase is critical for CDRs formation and their endocytic functional activity.
We further validated our finding in a different cellular context. We used HeLa cells, which are of epithelial origin and form CDRs in response to HGF stimulation29. Silencing of RAB35 in these cells severely decreased CDRs formation as compared to scrambled-transfected cells (Supplementary Figure 1E), indicating that RAB35 is a critical regulator of CDR formation in response to different growth factor-dependent signalling pathways.
Finally, to establish whether RAB35 is also sufficient to induce these structures, we monitored by time-lapse phase contrast microscopy MEF cells expressing HA-RAB35 in a doxycycline-inducible fashion cultured in growing media without any addition of growth factors. Importantly, up-regulation of RAB35 was sufficient to promote the formation of multiple CDRs that display a persistent and rapid dynamics, as confirmed by the quantification of the number of events per cell monitored in 1 h (Fig. 2c and Supplementary Movie 2). This was mirrored by an increased rate of fluorescently labelled dextran internalization (Fig. 2d)
Collectively, these findings identify RAB35 as a non-previously characterized RAB GTPase essential for the formation of CDRs in response to stimulation with various growth factors.
CDRs play a key role in cytoskeleton remodelling associated with the transition from sessile to motile states7. In addition they frequently, if not invariably, form in close proximity to the cell leading edge30, and are capable of initiating endo/exocytic cycles of plasma membranes and integrins that are subsequently delivered in a polarized fashion to the prospective lamellipodia11,12,31,32. These properties suggest that CDRs may function as bona fide cellular steering compasses to initiate forward and directional chemotactic migration. Given the relationship between CDRs, cell locomotion and cell guidance, we further hypothesize that RAB35 by controlling their formation may also be essential for regulating directional, chemotactic motility.
To act as steering devices CDRs must form in a polarized fashion in response to local gradient of chemotactic growth factors, and their dynamics formation should be spatiotemporally correlated with the extension of lamellipodia protrusions. In addition, structural components and biochemical wiring involved in cell migration guidance are often behaving as excitable oscillatory systems, which may become spatiotemporally biased following chemoattractant exposure33. The propagation of actin waves at the ventral surface of neutrophil is a typical case in point34,35,36,37. We verified whether CDRs display all these features. Firstly, we monitored by time-lapse microscopy their formation and recorded the subsequent extension of lamellipodia in response to local delivery of PDGF. CDRs formed in polarized directions and their appearance/disappearance was followed by the subsequent extension of flat lamellipodia-like protrusions following the local delivery of PDGF (Fig. 3a and Supplementary Movie 3). The temporal correlation between CDR and leading-edge protrusions was robust in RAB35-expressing cells. RAB35-expressing, but not control cells, form multiple CDRs in the absence of any added growth factor and these structures precede the extension of lamellipodia with a lag phase of about 105″ (Fig. 3b and Supplementary Movie 4). We further exploited the ability of RAB35 to induce a constitutive wave of recurrent multiple CDRs to characterize their overall dynamics and kinematics in more details and relate it to the extension of membrane protrusions. We observed travelling CDR waves, where CDR formed at the rear of an elongated, spatially restricted pseudopodia-like protrusion and move persistently along and in synchrony with the extended protrusion, with nearly identical speed (Fig. 3c and Supplementary Movie 5). We further detected multiple CDRs expanding centrifugally, behaving as iterative waves that form in diverse, but generally peripheral positions expanding centrifugally toward the cell edge (see Supplementary Movies 4). Finally, we detected recurrent and oscillating CDRs waves, which form repeatedly in the same location expanded and enclosed with a typical oscillatory frequency of about 20 min (Fig. 3d and Supplementary Movie 6). Collectively, these features support the notion that CDRs behave as an excitable system of propagating waves, which can be biased by exogenously added PDGF to promote cell steering and chemotactic motility33.
If CDRs are indeed bona fide, excitable, steering devices, their perturbation should impair directional motility, particularly toward growth factors known to induce robustly their formation right at the onset of chemotaxis. To this end, we first measured the ability of MEFs to migrate through a microporous membrane towards a PDGF gradient in a Boyden chamber assay. Scrambled and RAB35-silenced cells (we used three independent siRNAs) were seeded into Transwell chambers. After 20 h, cells on the top of the membrane were scraped away and the ones migrated at the bottom were stained with crystal violet. An additional time point was taken 3 h after cell plating to demonstrate that we seeded an equal number of cells, which display similar adhesion efficiency across all the different experimental conditions. The loss of RAB35 significantly and robustly reduced the number of cells crossing the porous filter by chemotaxis (Fig. 4a). To monitor MEF chemotaxis in real time and to explore the potential underlying cellular alterations, we employ a microfluidic commercial device that generate a stable linear gradient of PDGF. We used the ImageJ plugin Chemotaxis tool to extract migratory parameters, including forward migration index (chemotaxis) and mean velocity. Silencing of RAB35 significantly impaired chemotaxis and reduced mean cell velocity (Fig. 4b and Supplementary Movie 7).
To understand whether the altered migratory capability observed upon RAB35 ablation is an intrinsic defect in the molecular machinery sustaining cell locomotion, we also monitored the migratory behaviour of sparsely seeded Ctrl and short hairpin-silenced RAB35 cells in the absence of any external guiding factor in a random migration assay. Under these conditions, we found that RAB35 loss caused a significant but marginal decrease in cell velocity and had no effect on the formation of lamellipodia protrusions, suggesting that the machinery generating locomotory forces was not strictly dependent on RAB35 (Fig. 4c and Supplementary Movie 8). On the contrary, the stable up-regulation of RAB35, which promoted that extension of multiple and subsequent waves of CDRs, significantly reduced persistent motility (indicated as directionality) and altered mean velocity (Fig. 4c). These latter observations are consistent with the possibility that multiple and short-lived CDRs are induced by the expression of the transgene, leading to cells that frequently change the direction of their protrusions and motion, and cannot persistently move in a biased direction. To further substantiate this notion, we analysed the migratory behaviour of control and RAB35-silenced MEFs moving along a PDGF gradient through an array of pillars where directional decision choices must be made. The array is composed of pillars from a photocurable hybrid polymer separated by a 4-μm space (Fig. 4d). Control cells navigate toward the PDGF gradient by extending persistent migratory protrusions, most of which were oriented coherently with the direction of the gradient. Conversely, loss of RAB35 reduces significantly chemotaxis and the number of cells with protrusions oriented along the gradient (Fig. 4d).
Finally, if our model were correct, we would expect that RAB35 loss might not be strictly required for migration toward stimuli that poorly or do not induce CDRs formation. Consistently, we showed that RAB35 loss inhibited chemotaxis of MEFs toward serum, which poorly induces CDR formation, much less drastically than toward PDGF (Fig. 5a, b). Additionally, RAB35 was dispensable for chemotactic migration toward EGF, which is unable to induce the formation of CDRs in MEFs (Fig. 5c, d). Similarly, RAB35 loss had no effect on MEFs migration in scratch wound assays. Under these latter conditions, cells migrate by kenotaxis without forming CDRs, while extending lamellipodia and filopodia protrusions like control cells (Fig. 5e and Supplementary Movie 9).
Collectively, our findings argue that deficiency in CDR formation leads to a slight impairment in cell locomotion but a more dramatic inhibition of directional sensing and chemotactic-guided migration further impacting on cell persistence, at least during crawling locomotion on two-dimensional (2D) surfaces typically measured by all these assays. In vivo, however, cell migration occurs within complex three-dimensional (3D) matrices with different structural organization, fibres composition and physical properties. Under these conditions, cells frequently move along single ECM fibres or narrow channels that impose a defined, physical confinement. To this end, we performed migratory assays on one-dimensional (1D) micro-printed as well as suspended ECM lines. These assays mimic 1D tumour interstitial migration and allow, through the live monitoring of cell motion, a precise control of migratory parameters. Using suspended fibres, we showed that RAB35 ablation impeded the crawling mode of locomotion. Cells devoid of RAB35 were no longer able to extend trains of expanding wave and were stuck on the fibres (Supplementary Movie 10). Next, to provide a more reliable quantitation of this phenotype, we turned to 1D micro-printed fibronectin tracks. We seeded MEF shCtrl and shRAB35 cells on lines of 10 μm in width and monitored their 1D locomotion by time-lapse phase contrast microscopy for 24 h. Cell trajectories were automatically reconstructed by a build in-house ImageJ macro and a number of migratory parameters were extrapolated. Results showed that the absence of RAB35 affected cell velocity, the total length covered and, more relevantly, the persistence of cell motion (Fig. 6a and Supplementary Movie 11).
Next, we wondered whether this protein might also play an active role in tumour cell dissemination in 3D matrix. To address this point, we tested MCF10.DCIS.com cells stably downregulated for RAB35 in a set of in vitro migratory/invasive assays. This cell line is an oncogenic variant of MCF-10A that is widely used to recapitulate the transition from an in situ ductal to an invasive breast carcinoma38,39. We first measured the ability of RAB35-depleted cells in migrating through a microporous membrane towards an HGF gradient in a Boyden chamber assay. Ablation of RAB35 profoundly affected the chemotactic migratory ability of MCF10.DCIS.com cells (Fig. 6b). In addition, we also observed that control cells displayed a higher ability to invade and migrate through a thin layer of Matrigel in comparison to RAB35-downregulated cells, as shown in the Matrigel invasion assay (Fig. 6c). To further validate this findings, we monitored by time-lapse phase contrast microscopy the invasive migration of control and RAB35-depleted MCF10.DCIS.com cells into 3D gels of native type I collagen enriched with the motogenic factors HGF40,41,42. Control cells readily invaded the 3D matrix. Conversely the chemoinvasive potential of RAB35-deficient cells was impaired as demonstrated by the reduced invasive forward index and cell velocity (Fig. 6d and Supplementary Movie 12). Finally, we exploited the ability of MCF10.DCIS.com cells to generate invasive outgrowths in 3D basement membrane43. Control and RAB35-silenced MCF10.DCIS.com cells were seeded as single cells onto a gel composed of Matrigel and type I collagen and allowed to form spheroids. The addition of HGF in the presence of collagen type I is known to trigger an invasive programme, characterized by the outgrowths of multicellular structures that expand from the regular contour of the spheroids44. This transition recapitulates what seen in vivo when in situ ductal carcinoma, confined into the lumen of a duct, convert into invasive ductal carcinoma through the extension of local multicellular protrusions44. The percentage of acinar structures that form invasive outgrowths was significantly reduced in RAB35-depleted cells (Fig. 6e), reinforcing the notion that RAB35 is necessary to promote a mesenchymal programme of chemotactic cell invasion in vitro.
What are the cellular processes and molecular pathways RAB35 uses to promote CDR and steer cells in response to chemotactic stimuli?
To address this question, we initially turned to the well-established functional role of RAB35 in controlling clathrin-dependent endocytic internalization and membrane trafficking. More specifically, we tested the possibility that manipulation of RAB35 levels would impact on the trafficking of growth factor receptors, focusing on PDGFRB. However, no difference in either total PDGFRB or surface amounts could be found following silencing (Supplementary Figure 2A) or ectopic up-regulation (Supplementary Figure 2B) of RAB35.
We next utilized a molecular epistasis approach to position RAB35 action on known pathways controlling CDRs7. We have previously shown that the formation of these structures requires an active PDGFR, which function as a first-line sensor and a transducer of PDGF signalling, and a functional endosomal trafficking route, in turn, necessary to spatially restrict RAC1-activity and RAC1-dependent actin remodelling into CDR8, defining an epistatic relationship between these components (Fig. 7a). To position RAB35 along this pathway, we exploited the finding that elevation of the levels of this protein is sufficient to promote multiple CDRs in the absence of GF. Specifically, we monitored by time-lapse phase contrast microscopy the dynamics of CDRs in MEF cells silenced for Pdgfrb, or Rab5a, band c or Rac1 in control (−DOX) and doxycycline-inducible, RAB35-expressing populations (+DOX). The number and dynamics of CDR formation, as expected, was robustly increased upon elevation of RAB35 levels. The silencing of PDGF receptor had no effects on CDR, consistently with the notion that elevation of RAB35 is sufficient to bypass the need of ectopic addition of growth factor to induce these protrusions. On the contrary, silencing of Rab5 or Rac1 genes nearly completely abrogated RAB35-induced CDRs (Fig. 7b and Supplementary Movie 13). Thus, RAB35 appears to act downstream of PGDF receptor and either upstream or in a parallel RAB5/RAC1 pathway.
To investigate further this latter possibility and gain a molecular understanding of the mechanisms of action of RAB35, we systematically silenced all, the so far-identified, molecular effectors (Ocrl, Rusc2, Acap2, Mical1, Micall1, Fscn1 and p85) as well as guanine nucleotide exchange factors (Dennd1a, Dennd1b, Dennd6b and Flc) of this GTPase45,46. MEF cells were systematically downregulated for the indicated genes [only in the case of p85 we used MEFs double KO of the two isoforms p8547 instead of performing the transient downregulation of the genes], serum starved for 2 h, stimulated with PDGF and scored for the ability to form CDRs. The resulting CDR score showed that the genetic ablation of p85 isoforms was the sole manipulation able to phenocopy the loss of RAB35 (Fig. 7c). The lack of effects of critical downstream effectors mediating RAB35 known role as modulator of endocytosis further strengthened the notion that this function may not be the one used to control CDR formation.
To provide evidence of a direct causal link between p85/PI3K axis and RAB35, we performed three sets of experiments. Firstly, we inhibited PI3K activity, which strictly depends on its association with the regulatory subunit p85α and p85β, using a pharmacological inhibitor. Treatment of PDGF-stimulated with LY294002 or the more specific GDC-0941 PI3K inhibitors, the efficacy of which was tested by measuring the phosphorylation levels of the PI3K target and effector protein AKT, severely abrogated the formation of CDRs (Supplementary Figure 3A–B). Inhibition of AKT with MK-2206 had, instead, no effect on PDGF-induced CDRs formation nor on macropinocytosis (Supplementary Figure 3C). More importantly, pharmacological inhibition of PI3K also effectively abrogated CDRs induced by the sole expression of RAB35 (i.e. in the absence of PDGF addition) (Fig. 8a and Supplementary Movie 14). We corroborated the latter findings using MEF-KO for p85α and β. These cells and the related control cells were engineered to express RAB35 in a doxycycline-inducible fashion and monitored by time-lapse phase contrast microscopy. Removal of p85α and β completely prevented the formation of highly dynamic CDRs induced by RAB35 expression (Fig. 8b and Supplementary Movie 15). Finally, we tested biochemically whether RAB35 acts by directly modulating p85/PI3K activity. Indeed, while we were working on this project, RAB35 was identified as a critical and direct activator of the p85/PI3K–AKT pathway, shown to directly interact with the regulatory p85α subunit and to mediate, through this pathway, cell transformation48. In agreement with this latter finding, we found that silencing of RAB35 reduced significantly the phosphorylation of AKT in response to PDGF stimulation, without affecting PDGFR phosphorylation status, and PDGFR-dependent MAPK activity (Fig. 8c). We also showed that the ectopic expression of RAB35, in doxycycline-stimulated MEF pSLIK-HA-RAB35 cells cultured in growing conditions without addition of growth factors, caused the hyper-activation of AKT signalling without affecting the phosphorylation levels of other transducers (Fig. 8d). Finally, RAB35 and p85α co-immunoprecipitated. This interaction was enhanced upon growth factors stimulation. Importantly, wild-type RAB35, but not an inactive dominant-negative RAB35S22N mutant, associated with endogenous p85α, whereas dominant active RAB35 interacted with p85α in a constitutive growth factors-independent fashion (Fig. 8e, f). These latter findings imply that stimulation with growth factors might increase RAB35-GTP levels, leading to the activation of p85/PI3K pathway. Consistent, with this notion a dominant-negative RAB35S22N mutant abrogated PDGF-induced CDRs formation and directional migration (Supplementary Figure 3D), whereas two recently identified activated, tumour-associated RAB35 mutants, RAB35A151T and F161L48, promoted CDRs formation and elevated AKT phosphorylation in the absence of growth factor stimulation (Supplementary Figure 3E and Movie 16). In addition, there is a correlation between the levels of RAB35 and of phosphoAKT in prostate cancer cell lines (Supplementary Figure 4A-B). Of note, the analysis of the TCGA data set indicated that RAB35 display a significant elevated copy number variation in about 16% of human prostate cancers (Supplementary Figure 4C). The variable expression of RAB35 was also observed across a panel of human prostate cancer in a Tissue Microarray (TMA) with 12/56 (21.4%) adenocarcinoma displaying elevated levels of RAB35 (score ≥ 1.5), whereas only 2/32 (6%) of normal prostate tissues displayed RAB35 scores = 1.5 (Supplementary Figure 4D). These latter results indicate that deregulation of the levels of this GTPase may be positively selected in a subset of prostate tumours.
