Upregulated flotillins and sphingosine kinase 2 derail AXL vesicular traffic to promote epithelial-mesenchymal transition

Aberrant reactivation of epithelial-to-mesenchymal (EMT) transition, an essential embryonic process that is also crucial for tumor initiation and metastatic cancer cell dissemination (reviewed by Dongre and Weinberg, 2018), is a key step in cancer progression. EMT induction is orchestrated by a family of EMT-inducing transcription factors (e.g. ZEB1, ZEB2, Snail1 and Slug) that are regulated by multiple signaling pathways [e.g. TGFβ receptors and receptor tyrosine kinases (RTKs)] and the downstream signaling Ras/MAPK and phosphoinositide 3-kinase (PI3K)/AKT pathways (Puisieux et al., 2014). Because of EMT pleiotropic roles in the acquisition of invasive properties and therapeutic resistance, it is important to identify new inducers of this cellular program.

For example, AXL, a member of the TAM (Tyro3, AXL and Mer) family of RTKs, is an attractive candidate target for anticancer therapies (reviewed by Zhu et al., 2019). AXL expression is low in normal adult tissues, but is overexpressed in several tumors. This correlates negatively with overall survival. In many tumors, aberrant AXL expression is associated with EMT, invasion, metastasis formation, cell proliferation, tumor angiogenesis, reduced anti-cancer immune response, stem cell maintenance and treatment resistance (Gjerdrum et al., 2010; Gay et al., 2017). AXL is activated through several mechanisms, including binding to its ligand GAS6, homodimerization when overexpressed and crosstalk with other RTKs (Gay et al., 2017). However, the rarity of AXL genetic mutations and amplification events, despite its upregulation in many tumor types, suggests that AXL expression is mainly regulated by post-transcriptional/translational modifications.

The endosomal system accumulates activated receptors and signaling molecules. Endocytosis was considered for many years to be only a means of limiting signal transduction; however, its crucial role in maintaining and even generating signaling cascades is now acknowledged (Dobrowolski and De Robertis, 2008; Murphy et al., 2009). Deregulation of vesicular trafficking, particularly endocytosis, has emerged as a major player in tumorigenesis (Mellman and Yarden, 2013; Schmid, 2017). Consistently, upregulation of flotillin 1 and/or 2, two proteins involved in the formation of membrane invaginations, is a common feature of many invasive tumors and a poor prognostic marker (Berger et al., 2013; Pust et al., 2013; Zhao et al., 2015; Cao et al., 2016; Gauthier-Rouvière et al., 2020). When co-upregulated, flotillins promote plasma membrane (PM) invaginations, and favor an endocytic pathway called upregulated flotillin-induced trafficking (UFIT) (Glebov et al., 2006; Frick et al., 2007; Babuke et al., 2009) that targets protein cargos to non-degradative late endosomes (LEs) where flotillins accumulate (Dermine et al., 2001; Neumann-Giesen et al., 2007; Langhorst et al., 2008; Planchon et al., 2018). Several studies have reported that, in cancer cells, flotillins influence the activation of signaling pathways that promote EMT and cellular invasion, and that they are scaffolding factors for several signaling molecules (Amaddii et al., 2012). Therefore, the UFIT pathway, which is at the crossroads between endocytosis and signaling, has emerged as a crucial mechanism in the regulation of oncogenic pathways.

Here, to study the contribution of flotillin-induced deregulation of the endosomal system to EMT, we have reproduced the flotillin 1 and 2 co-upregulation observed in tumors in non-tumoral mammary cell lines (MCF10A and NMuMG) with low endogenous flotillin levels. Indeed, previous studies on the role of flotillin in EMT and cell invasion (Zhao et al., 2015; Cao et al., 2016) did not allow the determination of the exact contribution of flotillin upregulation and of the molecular mechanisms because they were carried out using cancer cell lines harboring several mutations in signaling and structural genes that promote EMT and invasion. In non-tumoral cells, we found that flotillin upregulation was sufficient to promote EMT and to activate key oncogenic signaling pathways. We identified AXL as a major contributor to flotillin upregulation-induced EMT. We showed that the UFIT pathway promotes AXL endocytosis towards flotillin-positive LEs and AXL stabilization, leading to its overexpression. Finally, in non-tumoral cell lines, where flotillins are upregulated, and in invasive breast cancer cells, where flotillins are endogenously overexpressed, AXL stabilization by the UFIT pathway depends on the sphingosine kinase 2 (SPHK2) that produces the bioactive lipid sphingosine 1-phosphate (S1P), a key signaling component of the sphingolipid metabolic network during tumorigenesis.

To precisely determine the role of flotillin upregulation in EMT and the implicated molecular mechanisms, we used non-tumoral epithelial cell lines (human MCF10A and murine NMuMG) in which oncogenic pathways are not activated and endogenous flotillin expression levels are low. From these two cell lines, we generated stable cell lines (MFC10AF1F2 and NMuMGF1F2) that express flotillins at levels similar to those observed in MDA-MB-231 invasive breast cancer cells (Fig. S1A-C; Planchon et al., 2018).

First, to identify the gene expression changes and the downstream signaling pathways associated with flotillin upregulation, we compared, using RNA-sequencing (RNA-seq), the expression profiles of control MCF10AmCh cells (transfected with empty mCherry vector, mCh) and MCF10AF1F2 cells (transfected with flotillin 1-HA and flotillin 2-mCherry). Using two differential gene expression analysis pipelines (Tuxedo and DSeq2), we identified 232 upregulated and 570 downregulated genes (fold change 2.8, P<0.05) in MCF10AF1F2 cells compared with control cells (Table S1). Gene ontology (GO) cluster analysis (Fig. 1A) revealed that flotillin upregulation was associated with deregulation of genes involved in extracellular matrix, cell adhesion, migration, differentiation, apoptosis and signaling pathways, particularly MAP kinases. These processes are often deregulated during tumor development and particularly during EMT.

We compared the list of genes differentially expressed between MCF10AmCh and MCF10AF1F2 cells with the EMT transcriptional signature previously established by a meta-analysis that combined 18 independent gene expression studies on EMT induced by different treatments in different cell types (Gröger et al., 2012). In this EMT signature, 22/67 of the upregulated genes and 39/62 of the downregulated genes overlapped with the genes deregulated upon flotillin overexpression (Fig. 1B). We also confirmed the enrichment of the EMT signature in our ranked transcriptome profile by gene set enrichment analysis (GSEA) (Fig. 1C). The analysis of the top 20 up- and downregulated genes between MCF10AmCh and MCF10AF1F2 cells highlighted the decreased expression of E-cadherin and occludin, and the increased expression of ZEB1, fibronectin, vimentin and N-cadherin (Fig. 1D).

We validated these results in MCF10AF1F2 and NMuMGF1F2 cells by RT-qPCR (Fig. 1E and Fig. S1D), western blotting and immunocytochemistry analysis (Fig. 1F-H and Fig. S1E-G). We confirmed the downregulation (mRNA and protein) of the hallmark epithelial marker E-cadherin (Fig. 1E-G and Fig. S1D-F) and of the polarity marker ZO-1 (also known as TJP1) (Fig. 1H and Fig. S1G), and the upregulation of the mesenchymal markers N-cadherin and vimentin (Fig. 1E-G and Fig. S1D,G). Morphological analysis by scanning electron microscopy and F-actin cytoskeleton staining showed that cell polarity was lost in NMuMGF1F2 cells and in MCF10AF1F2 cells compared with NMuMGmCh cells and with MCF10AmCh cells, respectively (Fig. S1H-J).

Several transcription factors can inhibit E-cadherin expression (De Craene and Berx, 2013). Flotillin upregulation specifically induced ZEB1 and ZEB2 mRNA expression (Fig. 1I and Fig. S1K,L), and ZEB1 protein expression (Fig. 1J and Fig. S1M). ZEB1 localized in the nucleus (Fig. 1K and Fig. S1N). ZEB1 and ZEB2 knockdown in MCF10AF1F2 cells restored E-cadherin expression (Fig. 1J) and its accumulation at cell-cell contacts (Fig. 1L). This demonstrated the key role of ZEB1 and ZEB2 in E-cadherin downregulation upon flotillin upregulation. In conclusion, flotillin upregulation in non-tumoral mammary epithelial cells is sufficient to induce EMT.

The GO analysis of the genes differentially expressed between control and flotillin-upregulated MCF10A cells revealed a strong association with signal transduction pathways involving transmembrane RTKs and the MAPK pathway (Fig. 1A). Different bioinformatics analysis identified the PI3-K/AKT and Ras pathways: two major oncogenic signaling cascades (Table S2). In agreement, analysis using the human phospho-kinase array revealed the increased phosphorylation of at least four main tyrosine and serine/threonine kinases involved in oncogenic pathways [ERK1 and ERK2 (ERK1/2; also known as MAPK3 and MAPK1, respectively), FAK (also known as PTK2), AKT family proteins and Fyn] and of their substrates [Hsp27 (also known as HSPB1) and STAT] (Fig. 2A). We demonstrated Ras activation in MCF10AF1F2 cells (Fig. 2B), as well as ERK1/2, PI3K/AKT and STAT3 activation (Fig. 2C, Fig. S2A for NMuMG cells, Fig. 2D-F).

In MCF10A cells, expression of a hyperactive H-Ras mutant or its downstream effector PI3-K leads to ZEB1 and ZEB2-dependent E-cadherin downregulation and EMT via the transcriptional repression of ΔNp63α, the predominant p63 (also known as TP63) isoform in MCF10A cells (Yoh et al., 2016). In these cells, ΔNp63α downregulation is an EMT marker (Yoh et al., 2016; Hu et al., 2017). Flotillin upregulation in MCF10A cells led to the reduction of ΔNp63α and ΔNp63β expression (Fig. 2G,H and Fig. S2B; Table S1). We excluded the involvement of ZEB1 and ZEB2 in ΔNp63α downregulation through a negative feed-back loop because ZEB1 and ZEB2 knockdown did not rescue ΔNp63α expression in MCF10AF1F2 cells (Fig. 2H). Moreover, data obtained from a previously described (Planchon et al., 2018) cohort of 527 individuals with breast cancer indicated that flotillin 1 and ΔNp63 mRNA expression levels were negatively correlated (Fig. 2I).

High TGFβ signaling pathway activity contributes to tumor development and to EMT induction (Dongre and Weinberg, 2018). Flotillins participate in its activation in nasopharyngeal carcinoma (Zhao et al., 2015; Cao et al., 2016). Here, we found that flotillin upregulation in non-tumoral cells activated TGFβ signaling, as indicated by the higher SMAD3 phosphorylation (Fig. 2J and Fig. S2C) and the nuclear localization of phosphorylated SMAD3 in MCF10AF1F2 (Fig. 2K) and NMuMGF1F2 (Fig. S2D) cells compared with control cells (mCh).

We never detected activation of the NF-κB and WNT canonical pathways (Fig. S2E-G), although several oncogenic pathways were activated upon flotillin upregulation. Altogether, these data demonstrate that flotillin upregulation in cells without oncogenic mutations is sufficient to activate major oncogenic pathways, such as the Ras/ERK MAP kinase, AKT and TGFβ signaling cascades.

To identify how flotillin upregulation in non-tumoral mammary cells activates these oncogenic signaling pathways, we used a phospho-RTK array to simultaneously screen 49 tyrosine-phosphorylated RTKs. The tyrosine phosphorylation of 11 RTKs was increased by >1.4-fold in MCF10AF1F2 cells compared with MCF10AmCh cells (Fig. S2H,I). To validate these results for the three top receptors, we wanted to compare their Y-phosphorylation status in the two cell lines by western blotting using antibodies against phosphorylation at specific Y residues. Unfortunately, the available antibodies allowed us to validate this result only for AXL, but not for EphA4 and ALK. AXL phosphorylation was increased (2.53±0.71-fold) in MCF10AF1F2 cells compared with control MCF10AmCh cells, as was the AXL protein level (2.57±0.61-fold versus control) (Fig. 3A). The identical phospho-Y702 AXL/AXL ratio in MCF10AmCh and MCF10AF1F2 cells indicated that flotillin upregulation increased AXL protein levels but did not stimulate AXL phosphorylation (Fig. 3A). The expression level of the transferrin receptor CD71 (also known as TFRC) was comparable in both cell lines (Fig. S3A). We detected increased AXL level in cells grown in the presence of serum (Fig. S3B), which might contain its ligand GAS6, and also in serum-starved cells, indicating that flotillin upregulation promotes an increase in the AXL level in MCF10A cells independently of GAS6. Moreover, AXL expression was barely detectable in normal breast tissues, but was increased in breast tumor samples and was correlated with that of flotillin 1, particularly in the luminal and HER2 (also known as ERBB2)-positive breast cancer subtypes (Fig. S4A,B).

Flotillin upregulation promoting an increase in AXL levels did not occur at the transcriptional level (Fig. 3B). We assessed AXL stability by comparing AXL expression in MCF10AmCh and MCF10AF1F2 cells incubated with cycloheximide (CHX) to block protein synthesis. AXL decay was significantly slower in MCF10AF1F2 cells than in MCF10AmCh cells (Fig. 3C). This indicates that flotillin upregulation promotes AXL stabilization. We analyzed the contribution of AXL to EMT and cell migration induced by flotillin upregulation in MCF10A cells. AXL knockdown by siRNAs in MCF10AF1F2 cells decreased ZEB1 expression (Fig. 3D,E), reversed the E- to N-cadherin switch (Fig. 3D,F) and reduced the migration of the cells (Fig. 3G).

Altogether, these results showed that flotillin upregulation in non-tumoral mammary epithelial cells promotes AXL stabilization, leading to the increase of its phosphorylated form, as observed in many invasive cancer types. These results also revealed that AXL plays a central role in flotillin upregulation-induced EMT and cell migration in mammary epithelial cells.

Deregulated endocytosis and vesicular trafficking alter the fate of signaling receptors, affecting their expression level and downstream signaling (Dobrowolski and De Robertis, 2008; Gonnord et al., 2012; Schmid, 2017); hence, we asked whether AXL is an UFIT pathway cargo. We developed an optogenetic approach based on the CRY2-CIBN system (Nguyen et al., 2016) to force flotillin oligomerization by light illumination combined with total internal reflection fluorescence (TIRF) video microscopy (see Fig. 4A for the description of the experimental set up and Fig. S5A for the validation of the tools). Using MCF10A cells co-expressing flotillin 2-CIBN-mCherry, CRY2-mCitrin and AXL-Halo, upon light illumination to induce CRY2 oligomerization and binding to CIBN, we demonstrated the formation at the PM of domains that contained both flotillin 1 and 2 (Fig. 4B) and AXL-Halo (Fig. 4B and Fig. S5B,D, Movie 1). Using time-lapse imaging, we detected the co-endocytosis of flotillin 2-mCherry and AXL-GFP (characterized in Fig. S5C) from PM sites in MCF10AF1F2 cells where both proteins initially co-accumulated (Fig. 4C, Movie 2). In agreement, we showed that, in MCF10AmCh cells, AXL was mainly at the PM. Conversely, in MCF10AF1F2 cells, AXL was still detectable at the PM in flotillin-rich regions, but was enriched in flotillin-positive vesicles (Fig. 4D), which were positive for the LE marker CD63 (Fig. 4J). Similarly, in MDA-MB-231 mammary tumor cells, in which endogenous flotillins are overexpressed (Planchon et al., 2018), AXL and flotillins 1 and 2 colocalized at the PM and in intracellular vesicles (Fig. S5E,F) that were CD63 positive (Fig. 4K). AXL-GFP co-trafficked with flotillin 2-mCherry in living MCF10AF1F2 (Fig. S5G, Movie 3) and MDA-MB-231 cells (Fig. S5H, Movie 4). About 40% of the flotillin 2-mCherry-positive vesicles were positive for AXL-GFP, and half of AXL-GFP-containing vesicles were positive for flotillin 2-mCherry (Fig. S3C). Conversely, CD71, the expression of which remained unchanged upon flotillin upregulation (Fig. S3A), poorly colocalized with flotillin-positive endosomes (Fig. S3C). These findings show that flotillins and AXL are present in the same PM domains, from where they are co-endocytosed towards LEs where they co-accumulate. Similarly, in breast tumor samples, flotillin 1 and AXL accumulate in intracellular vesicles in malignant cells (Fig. S4C). This co-distribution of flotillins and AXL is in agreement with the detection of flotillin 1 and AXL complexes by co-immunoprecipitation experiments in MCF10AF1F2 cells (Fig. 4E).

To monitor the impact of flotillin upregulation on AXL endocytosis, we performed an AXL uptake assay using the D4 antibody against the AXL extracellular domain coupled with immunocytochemical detection (Fig. S5I). Immediately after incubation with the D4 antibody, we detected cell-surface AXL signals in MCF10AmCh and MCF10AF1F2 cells (Fig. 4F). After 10 min of internalization, the AXL-D4 signal was more internalized in flotillin-positive endosomes in MCF10AF1F2 cells compared with MCF10AmCh cells (Fig. 4F). We obtained similar results by monitoring N-terminally Myc-tagged AXL surface expression with an anti-Myc antibody (Fig. S5J). Thus, AXL internalization occurs faster in cells in which flotillins are upregulated, which was confirmed using a cell surface biotinylation assay (Fig. 4G, upper panels). Reciprocally, to analyze the impact of flotillin downregulation on AXL internalization in MDA-MB-231 cells, we used the MDA-MB-231-shFlot2 cell line (Planchon et al., 2018). Flotillin downregulation slowed down AXL internalization (Fig. 4G, lower panels), and AXL accumulated at the PM in MDA-MB-231-shFlot2 cells (Fig. 4H,I). Conversely, CD71 internalization was not affected by flotillin upregulation (Fig.  S3D), in agreement with its low presence in flotillin-positive vesicles (Fig. S3C). This indicates that flotillin upregulation does not affect the endocytosis and trafficking of all membrane receptors.

Flotillin-positive vesicles are LAMP1-, Rab7a- and CD63-positive LEs, previously characterized as non-degradative LEs on the basis of their very low staining for fluorescent cleaved DQ-BSA (Planchon et al., 2018). In agreement with AXL stabilization in flotillin-upregulated cells, we found AXL in flotillin-, CD63- and LAMP1-positive vesicles in MCF10AF1F2 cells [endogenous AXL (Fig. 4J, and Fig. S5K) and Myc-tagged AXL (Fig. S5L)] and in MDA-MB-231 cells (Fig. 4K). Similarly, the presence of AXL in LAMP1-positive vesicles in mammary tumor cells was recently demonstrated (Zajac et al., 2020). In flotillin-positive LEs, we also detected other signaling transmembrane receptors and oncogenic molecules that participate in tumorigenesis, such as EphA4, TGFβ (Fig. S6A-F) and K-Ras4B (Fig. S6G). Specifically, flotillin upregulation modified the distribution of K-Ras4B but not of any other Ras isoforms. K-Ras4B was localized at the PM in MCF10AmCh cells, and in flotillin-positive vesicles in MCF10AF1F2 cells (Fig. S6G). This correlated with the co-immunoprecipitation of K-Ras4B, but not of K-Ras4A, with flotillin 1 (Fig. S6H).

Altogether, these results showed that AXL is a cargo of the UFIT pathway. Flotillin upregulation accelerated AXL internalization and promoted AXL endocytosis toward flotillin-positive LEs.

To identify the molecular mechanisms linking AXL accelerated internalization and stabilization, and its sorting towards flotillin-positive LEs, we focused on sphingolipids, the metabolism of which is linked to endocytosis and flotillins. Indeed, flotillins bind to sphingosine, and sphingosine-1-phosphate (S1P) levels are reduced in cells from flotillin knockout mice (Riento et al., 2018). The bioactive lipid S1P is implicated in endocytosis and membrane remodeling (Kajimoto et al., 2013; Shen et al., 2014; Lima et al., 2017). S1P production, which is induced upon sphingosine phosphorylation by sphingosine kinase (SPHK) 1 or 2, has been linked to cancer progression (Pyne et al., 2018). SPHK1 and SPHK2 subcellular localization, and consequently S1P production compartmentalization, play an important role in their functions (Siow and Wattenberg, 2011). We hypothesized that flotillin-rich regions of membranes can concentrate sphingosine, where SPHKs could be recruited to locally generate S1P.

We examined sphingosine cell distribution after addition of fluorescent NBD-sphingosine in live cells. In MCF10AF1F2 cells, and also in MDA-MB-231 and Hs 578T breast cancer cells that express flotillin 2-mCherry, NBD-sphingosine was almost undetectable at the PM, but it accumulated strongly and rapidly in flotillin 2-mCherry-positive vesicles (Fig. 5A,B and Fig. S7A). Conversely, fluorescent Oregon Green-phosphatidylethanolamine poorly accumulated in flotillin 2-mCherry-positive vesicles (Fig. 5A).

Then, we investigated SPHK localization in MCF10AF1F2 and MDA-MB-231 cells expressing GFP-tagged SPHK1 or SPHK2 because no antibody is available for immunocytochemistry. We detected SPHK2-GFP in ∼70% of flotillin-positive vesicles (Fig. 5C-E, Movie 5), in contrast to SPHK1-GFP (Fig. 5D and Fig. S7B). In Hs 578T cells, SPHK2-GFP was also localized in flotillin-positive vesicle (Fig. S7E). SPHK2- and flotillin-positive vesicles also contained AXL (Fig. 5F, Movie 6), Rab7 (Fig. S7C) and CD63 (Fig. S7D). SPHK2 and flotillin 2 also colocalized at discrete sites of the PM, from which we could observe the fast co-endocytosis of both proteins (Fig. 5E, Movies 7 and 8). Flotillin upregulation increased SPHK2 recruitment to CD63-positive LEs (Fig. 5G), without modifying its protein expression level (Fig. S7F).

On the basis of these results and the finding that S1P level is reduced in cells derived from flotillin-KO mice (Riento et al., 2018), we asked whether flotillin upregulation increased S1P intracellular levels. Mass spectrometry quantification showed that S1P levels were 2.8-fold higher in MCF10AF1F2 than MCF10AmCh cells (Fig. 5H).

Owing to increased S1P levels in cells with upregulated flotillins and to SPHK2 enrichment in flotillin-positive membrane domains and flotillin-positive vesicles, we then asked whether S1P generation by SPHK2 was involved in AXL stabilization upon flotillin upregulation. Addition of the selective SPHK2 inhibitor (Riento et al., 2018) opaganib to cells expressing SPHK2-GFP led to its rapid dissociation from flotillin-positive endosomes that lasted for several hours (Fig. S7G, Movie 9). Incubation of MCF10AF1F2 or MDA-MB-231 cells with opaganib for only 4 h did not affect AXL expression level (Fig. 6A). However, in MCF10AF1F2 cells incubated with opaganib for 10 h, AXL levels were reduced to those observed in MCF10AmCh cells (Fig. 6A). AXL was similarly decreased in MCF10F1F2 and MDA-MB-231 cells, where SPHK2 was efficiently knocked down by siRNA (Fig. 6B,C). Opaganib treatment also led to a decrease in AXL in MDA-MB-231shLuci and in Hs 578T cells (Fig. 6D and Fig. S7H) that express high flotillin levels. In MCF10AF1F2 and MDA-MB-231shLuci cells, SPHK2 inhibition decreased also ZEB1 levels (Fig. 6A-D), in agreement with our finding that its expression is AXL dependent (Fig. 3D,E). Of note, SPHK2 inhibition did not alter flotillin levels in MCF10AmCh and MCF10AF1F2 cells (Fig. 6A), or AXL levels in MCF10AmCh cells (Fig. 6A) and in MDA-MB-231-shFlot2 cells (Planchon et al., 2018) (Fig. 6D). This indicates that SPHK2 activity is required for maintaining high AXL levels specifically in the context of flotillin upregulation. Moreover, in agreement with the limited SPHK1-GFP localization in flotillin-positive vesicles (Fig. 5D and Fig. S7B), SPHK1 inhibition using PF543 had no effect on AXL levels in MCF10AF1F2 and MDA-MB-231 cells (Fig. S7I).

To confirm that SPHK2 was involved in AXL stabilization downstream of upregulated flotillins, we examined AXL levels in MCF10AF1F2 and MDA-MB-231shLuci cells incubated with CHX to inhibit protein synthesis. CHX considerably decreased AXL level in both cell lines, and the effect was stronger in cells co-incubated with opaganib (Fig. 6E). In agreement, in cells exposed to CHX for up to 9 h, AXL decay was more pronounced in cells in which SPHK2 was knocked down than in controls (Fig. 6F). Conversely, CD71, which poorly localized in flotillin-positive endosomes (Fig. S3C) and the endocytosis of which was not influenced by flotillin level and SPHK2 activity (Fig. S3A,D,E), remained stable upon incubation with CHX, even in cells in which SPHK2 was silenced (Fig. S3F). This demonstrated that SPHK2 activity was involved in AXL stabilization, but not in stimulating its synthesis.

We next investigated how SPHK2 contributed to AXL stabilization. As SPHK2 co-accumulated with flotillins at the PM before endocytosis (Fig. 5E, Movies 7, 8), we analyzed its participation in flotillin-mediated AXL endocytosis. We compared AXL internalization rate and cell surface level in cells incubated or not (control) with opaganib for only 4 h. This short incubation time, when AXL decrease is not yet detectable (Fig. 6A,D), allows a more straightforward analysis of AXL internalization and level at the cell surface. AXL internalization was significantly decreased when SPHK2 was inhibited in both MCF10AF1F2 and MDA-MB-231shLuci cells (Fig. 6G,H). In agreement, incubation with opaganib increased AXL expression at the cell surface in both cell lines (Fig. 6I,J). This indicates that SPHK2 participates in flotillin-mediated AXL endocytosis. In accordance with the finding that CD71 internalization is not influenced by flotillin level (Fig. S3D), we found that it was not altered by SPHK2 inhibition (Fig. S3E). This confirmed that CD71 is not endocytosed through the UFIT pathway. In agreement with the finding that AXL knockdown inhibited MCF10AF1F2 cell migration (Fig. 3G), SPHK2 inhibition or knockdown also reduced their migration (Fig. S7J), in addition to decreasing AXL level (Fig. 6A,B). In conclusion, our data demonstrate that the lipid kinase SPHK2, which colocalizes with flotillins at the PM and in LEs, possibly by locally generating S1P, is required to divert flotillin-mediated AXL endocytosis towards the UFIT pathway in order to promote its stabilization, thus favoring cell migration.

Here, we demonstrate that flotillin upregulation in non-tumoral mammary cell lines devoid of oncogenic mutations is sufficient to activate oncogenic signaling pathways that lead to EMT induction, a key step towards cancer cell invasion. Moreover, flotillin upregulation led to the stabilization of AXL through the UFIT pathway that promotes endocytosis of several cargos from the PM toward flotillin-positive LEs (Fig. 7). This requires the lipid kinase SPHK2, which demonstrates the role of sphingolipid metabolism on this UFIT pathway. Our findings demonstrate the unprecedented role of flotillin-induced endosomal trafficking on AXL upregulation, a major feature of several invasive tumors.

Flotillins are upregulated in many tumor types, and their upregulation is predictive of aggressive tumor behavior (reviewed by Gauthier-Rouvière et al., 2020). Flotillin co-upregulation is necessary and sufficient to promote breast, nasopharyngeal and rhabdomyosarcoma cancer cell invasion (Liu et al., 2015; Cao et al., 2016; Planchon et al., 2018). Their specific role in metastasis formation was demonstrated in flotillin 2 knockout mice (Berger et al., 2013). EMT is a crucial step in tumor initiation and cancer cell dissemination, but the contribution of flotillins to this process remained to be addressed (Zhao et al., 2015). Previous studies were performed in cancer cell lines with mutations in both signaling and structural genes that promote EMT induction and invasion. To determine the exact contribution of flotillin upregulation to this crucial process and the molecular mechanisms that are specifically deregulated, we co-upregulated flotillins 1 and 2 in immortalized non-tumoral mammary cell lines that physiologically display low flotillin expression levels. Flotillin upregulation was sufficient to promote EMT and activation of key signaling pathways involved in EMT induction. We identified AXL as a key upstream actor. However, other receptors implicated in mammary tumor development might also be co-deregulated, according to the results of our phospho-RTK arrays (Fig. S2H,I) and the presence in flotillin positive-endosomes of EphA4 and TGFβ receptors (Fig. S6) (Pardali and Moustakas, 2007; Gjerdrum et al., 2010; Gregory et al., 2011; Hachim et al., 2017). Following flotillin upregulation, Ras and PI3K and the downstream signaling pathways, which are frequently activated in several cancers (Bailey et al., 2018), were induced in MCF10A cells that do not harbor mutations in these cancer driver genes (Bessette et al., 2015). Moreover, flotillin upregulation mimics oncogenic Ras activation that in MCF10A mammary cells has been correlated with ΔNp63α downregulation, PI3-K/AKT activation and ZEB1- and ZEB2-mediated EMT (Yoh et al., 2016). These findings show that flotillin upregulation mimics the effects of cancer driver gene mutations, acts as an EMT driver, and promotes tumor cell invasion (Planchon et al., 2018).

We demonstrate that flotillin upregulation deregulates the trafficking of PM receptors, and identify them as new cargos of the UFIT pathway. This endocytic pathway emerges in the context of flotillin upregulation, and the proteins located in the flotillin-containing PM microdomains are endocytosed and delivered to flotillin-positive non-degradative LEs (Planchon et al., 2018). Besides their weak degradative activity (Planchon et al., 2018), these vesicles have characteristics of signaling endosomes because they contain activated RTKs (Fig. 4 and Fig. S6), and also the cancer driver K-Ras4B, which is associated with increased Ras activity upon flotillin upregulation (Fig. S6G and Fig. 2B). Interestingly, K-Ras and activated ERK1/2 are recruited to Rab7-positive endosomes (Lu et al., 2009). Endosomes act as signaling platforms to facilitate the multiplexing of signaling molecules that, in turn, affect the endosomal system by modifying the number, size and position of endosomes, and ultimately the trafficking of signaling molecules (Murphy et al., 2009; Villaseñor et al., 2016). The UFIT pathway is a key player in connecting deregulation of endocytosis and signaling in cancer cells.

This is illustrated here for AXL, a cargo of the UFIT pathway that is crucial to sustain EMT-promoting signaling upon flotillin upregulation. AXL is expressed in solid cancers, particularly breast carcinoma, where its level is associated with high risk of metastases and poor prognosis (Gjerdrum et al., 2010; Asiedu et al., 2014). AXL expression is correlated with the acquisition of a mesenchymal phenotype, invasive properties and treatment resistance. AXL is an attractive therapeutic target, and several AXL inhibitors are currently in preclinical and clinical development (Leconet et al., 2014, 2017; Gay et al., 2017). Although AXL overexpression in tumors is attributed to post-transcriptional regulation, the underlying mechanisms are unknown. When overexpressed, AXL activation is ligand independent and is the result of its oligomerization and/or association with other RTKs (Goyette et al., 2018). Using an optogenetic system to force flotillin oligomerization, we demonstrated the formation of flotillin/AXL clusters at the PM that are co-endocytosed. The UFIT pathway allows AXL delivery to flotillin-positive endosomes and AXL stabilization. Thus, flotillin upregulation emerges as a new control mechanism in the ligand-independent regulation of AXL post-transcriptional regulation. Previous studies identified ligand-independent AXL stabilization at the cell surface through its association with the HER2 receptor (Goyette et al., 2018), ligand-dependent AXL sequestration in cholesterol-enriched PM domains and internalization (Paolino et al., 2014; Antony et al., 2018). To our knowledge, this is the first report to describe AXL stabilization following its endocytosis. This flotillin/AXL link is also observed in invasive breast tumor cells. Indeed, AXL is very weakly expressed in non-tumoral breast lesions, whereas its expression is broadly detected in all malignant breast tumor subtypes (Goyette et al., 2018). AXL expression appears to be correlated with that of flotillin 1, especially in the luminal and HER2-positive subtypes (Fig. S4).

Our discoveries open a new interesting perspective for cancer therapy. Inhibition of flotillin functions might affect the efficacy of anti-AXL (which had promising results in preclinical models of pancreatic and breast tumors; Ye et al., 2010; Leconet et al., 2014, 2017) and anti-RTK antibodies by increasing the membrane receptor availability at the cell surface, particularly in flotillin-positive tumors. Endocytosis is dysregulated in human tumors. Its inhibition increases the availability of therapeutic monoclonal antibody targets and promotes antibody-dependent cell cytotoxicity (Joseph et al., 2019; Chew et al., 2020).

Flotillin upregulation-induced AXL stabilization requires the activity of SPHK2, a lipid enzyme that maintains the cellular sphingolipid homeostasis and is a target for cancer therapies (Neubauer et al., 2016; Britten et al., 2017). SPHK2 colocalizes with flotillin-positive domains at the PM and is enriched in flotillin-positive endosomes, in agreement with previous findings showing SPHK2 localization in multivesicular endosomes in K562 leukemia cells (Mohamed et al., 2017). Flotillin binding to sphingosine and ectopic flotillin overexpression allow the formation of sphingosine domains that could favor SPHK2 recruitment (Riento et al., 2018). We found that SPHK2 inhibition or knock-down in cells with upregulated flotillins reduced AXL endocytosis (Fig. 6E,F) and AXL expression levels (Fig. 6A-D), and decreased cell invasion (Fig. S7J). Accordingly, SPHK2 inhibition did not affect AXL level in control MCF10A and in MDA-MB-231 invasive breast cells where flotillins were knocked down. In these cells, AXL accumulated at the PM, suggesting that the UFIT pathway targets AXL to flotillin-positive endosomes to escape degradation. AXL can be degraded at the PM or in the cytoplasm, through several mechanisms involving metalloproteases or gamma secretases, or the proteasome and the autophagic pathway (Krishnamoorthy et al., 2013; Bae et al., 2015; Rao et al., 2020). It remains to be determined from which degradation mechanism(s) the UFIT pathway protects AXL.

SPHK catalyzes S1P production from sphingosine and there is a strong causal association between overactive SPHK, S1P generation and cancer (Hatoum et al., 2017). SPHK2 is involved in MDA-MB-453 cell migration toward EGF (Hait et al., 2005), and SPHK2 upregulation promotes oncogenesis. Moreover, results from a phase 1 clinical trial support the use of SPHK2 inhibitors in cancer (Neubauer et al., 2016; Britten et al., 2017). Showing that SPHK2 inhibition affects AXL might help to understand the anti-tumoral function of these inhibitors and the role of SPHK2 during tumor development, which is unknown. The UFIT pathway does not influence SPHK2 expression level (Fig. S7E), but affects its recruitment to cell membranes, at flotillin-rich endocytic sites (Fig. 5D), and to flotillin-positive endosomes (Fig. 5E). In agreement, modifications of SPHK2 localization are linked to tumorigenesis (Siow and Wattenberg, 2011; Neubauer et al., 2019). Additional studies are required to understand how sphingolipid modifications participate in endocytic events at the PM and how they influence endosome fusion, motion and/or function.

In conclusion, this study adds new evidences for a crucial gain-of-function role of upregulated flotillins during EMT. We found that the UFIT pathway, by perturbing sphingolipid metabolism, allows the endocytosis of several PM receptors, particularly AXL, a major RTK during oncogenesis. AXL targeting to flotillin-positive LEs promotes its stabilization and signaling during EMT.

All cell lines were authenticated, tested for contamination and cultured in a 37°C incubator with 5% CO₂. The MCF10A mammary epithelial cell line (American Type Culture Collection, ATCC CRL-10317) and the derived MCF10AmCh and MCF10AF1F2 cell lines were grown in DMEM/HAM F-12 medium supplemented with 5% horse serum (Gibco, Thermo), 10 µg/ml insulin, 20 ng/ml EGF, 0.5 µg/ml hydrocortisone and 100 ng/ml cholera toxin. The NMuMG mouse mammary epithelial cell line (ATCC CRL-1636) and the derived NMuMGmCh and NMuMG-F1F2 cell lines were grown in DMEM supplemented with 10% fetal bovine serum (FBS) (S1810-100; Biowest), 100 units/ml penicillin, 100 μg/ml streptomycin and 10 μg/ml insulin (I9278; Sigma-Aldrich). The human breast adenocarcinoma cell line MDA-MB-231 (ATCC HTB-26) and the MDA-MB-231shLuci- and MDA-MB-231-shFlot2-derived cell lines (described by Planchon et al., 2018), and the Hs 578T cell line were grown in DMEM with 10% FBS (Biowest, Eurobio). Retrovirus production in Phoenix cells (Garry Nolan, Center for Clinical Sciences Research, Stanford University, USA), infection and selection were performed as described previously (Fortier et al., 2008). Cells were grown continuously in 1 µg/ml puromycin and 200 µg/ml hygromycin.

The MCF10AF1F2 and NMuMGF1F2 cell lines that stably overexpress flotillin 1 and 2 were generated by double retroviral infection of two vectors encoding the chimeric proteins flotillin 2-mCherry and flotillin 1-HA (described by Planchon et al., 2018), followed by selection for resistance against puromycin and hygromycin. Cells were then sorted by flow cytometry according to the mCherry fluorescence signal.

peGFP-N1-AXL was generated by subcloning the human AXL sequence (NM_021913.4) amplified by PCR using the primers 5′-AAGCTTATGGCGTGGCGGTGCCCCAGG-3′ and 5′-GAATTCCGGCACCATCCTCCTGCCCTG 3′ between the HindIII and EcoRI sites in peGFP-N1. pHalo-N1-AXL was obtained by replacing GFP by the Halo-sequence in peGFP-N1-AXL between HindIII and EcoRI. To generate the pBabe-Flot2-CIBN-mCherry construct, the sequence encoding the CIBN domain was amplified by PCR using the primers 5′-GAATTCTAATGAATGGAGCTATAGGAGG-3′ and 5′-CGGATTATATTCATGTACCGGTCACACA-3′, and inserted, using the EcoRI and AgeI sites, between flotillin 2 and mCherry in the pBabe-Flot2-mCherry plasmid (Planchon et al., 2018). The peGFP-ERK2 construct was a gift from Dr H. Farhan (University of Oslo, Norway). The CD63-GFP plasmid was a gift from Dr A. Weaver (Vanderbilt University, Nashville, TN, USA) (Sung et al., 2015). The Rab7a-GFP plasmid has been described previously (Planchon et al., 2018). The pCMV3-N-Myc human AXL (HG10279-NM) plasmid was purchased from Sino Biological. The peGFP-C1-KRas4A, peGFP-C3-KRas4B, peGFP-C3-HRas and peGFP-C3-NRas plasmids were gifts from Dr M. Philips (NYU Cancer Institute, New York, USA). The phCMV3-Flot1-HA plasmid was a gift from Dr V. Niggli (University of Bern, Switzerland) and pmCherry-N1-Flot2 has been previously described (Planchon et al., 2018). The plasmids encoding the different TAp63 α,β,γ and ΔNp63 α,β,γ isoforms were gifts from Dr C. Caron de Fromentel (INSERM UMR590, Lyon, France) (Petitjean et al., 2008), the PCS2+chickenEphA4-GFP plasmid was a gift from Prof F. Fagotto (CRBM, Montpellier, France) (Rohani et al., 2014). The TGFβRI-GFP and TGFβRII-Flag constructs were gifts from Dr R. Derynck (UCSF, San Francisco, CA, USA) (Budi et al., 2015). The SPHK1-GFP and SPHK2-GFP constructs were gifts from Prof. P. de Camilli (Yale School of Medicine, New Haven, CT, USA) (Shen et al., 2014). The SPHK2-Halo plasmid was obtained by replacing GFP with the Halo-sequence between EcoRI and NotI. The CRY2-mCitrin plasmid was a gift from Prof. W. Do Heo (KAIST, Dajeon, South Korea), previously described by Nguyen et al. (2016). The TfR(CD71)-GFP vector was purchased from Addgene (#45060). Cells were transfected with the indicated constructs using the JetPEI or JetPrime Transfection Reagents (Polyplus Transfection).

Mouse antibodies used were against: actin (1:5000, A5441, Sigma), flotillin 1 (1:1000, 610820, BD Biosciences), flotillin 2 (1:100, 610383, BD Biosciences), E-cadherin (used for MCF10A cells, 1:1000, 13-1700, Life Technologies), E-cadherin (used for NMuMG cells, 1:500 immunofluorescence, 1:2000 western blot, 610181, BD Biosciences), N-cadherin (1:200 immunofluorescence, 1:2000 western blot, 61920, BD Biosciences), LAMP1 (used for MCF10A cells, 1:1000, 555798, BD Biosciences), CD63 (1:300, 857.770.000, Diaclone), SMAD3 (1:500, MA5-15663, ThermoFisher), vimentin (1:100, V6630, Sigma), α-tubulin (1:200, Hybridoma Mab356), phosphorylated ERK1/2 at T202/Y204 [1:1000, 9106, Cell Signaling Technology (CST)], AKT (1:1000, 2920, CST), pan-Ras (1:500, OP40, Millipore), phosphorylated Y (1:50, Hybridoma G10) and STAT3 (1:1000, 9139, CST). Rabbit antibodies were against flotillin 1 (1:50, F1180, Sigma), flotillin 2 (1:100, 3436, CST), AXL (1:1000, 8661, CST), phosphorylated AXL at Y702 (1:250, D12B2, CST), ERK1/2 (1:1000, 9102, CST), phosphorylated AKT at S473 (1:1000, 4060, CST), ZEB1 (1:200, sc-25388, Santa Cruz), ZO-1 (1:200, 61-7300, Invitrogen), phosphorylated SMAD3 at S423/425 (1:2000, ab52903, Abcam), p63 (1:2000, ab124762, Abcam), phosphorylated EphA2,3,4 at Y588/596 (1:200, ab62256, Abcam), EphA4 (1:200, a gift from Dr Greenberg, Harvard University, Boston, MA, USA) (Rohani et al., 2014), phosphorylated STAT3 at Y705 (1:1000, 9131S, CST), p65-NFκB (1:1000, sc372, Santa Cruz), β-catenin (1:2000, C2206, Sigma, 2206), HA-tag (1:250, 71-5500, Invitrogen), Flag-tag (1:1000, F7425, Sigma), Myc-tag (1:100, 06549, Millipore) and SPHK2 (1:1000, 32346, CST). The humanized anti-AXL antibody D4 (https://patents.google.com/patent/EP3229836A1/en?oq=EP3229836A1) and the mouse anti-AXL antibody D9 (Leconet et al., 2014) were gifts from Dr B. Robert (ICRM, Montpellier, France). Alexa-488-, -546- and -633-conjugated secondary antibodies were from Thermo Scientific (Alexa-488-anti-mouse antibody, A-11017, 1:2000; Alexa-546-anti-mouse antibody, A-11018, 1:2000; Alexa-633 anti-mouse antibody, A21050, 1:1000; Alexa-488-anti-rabbit antibody A-11034, 1:2000; Alexa-546-anti-rabbit antibody, A-11035 dilution, 1:2000; Alexa-633 anti-rabbit antibody, A21070, 1:1000). Alexa-Fluor-488-phalloidin was from Invitrogen and Hoechst 33342 (0.1 mg/ml; #B2261) was from Sigma-Aldrich. Lipopolysaccharide (LPS) (Sigma, L5668) was used at 1 µg/ml for 1 h. Halo-tagged proteins were labeled by incubating cells with 200 nM HaloTag Ligand conjugated to Janelia Fluor 646 (Promega) for 30 min before imaging. Opaganib (ABC294640) (used at 50 µM up to 10 h) and PF543 (used at 10 µM up to 24 h) were from Selleckchem. Cycloheximide (used at 100 µg/ml for 6 h) was from Sigma-Aldrich.

The pSIREN-RetroQ-Flot2-shRNA (targeting both human and murine Flot2) and pSIREN-RetroQ-Luciferase shRNA vectors have been previously described (Guillaume et al., 2013). Luciferase shRNA (shLuci) was used as a control (Fortier et al., 2008). For RNA-mediated interference experiments, cells were transfected using interferin (Polyplus Transfection) with siRNA oligonucleotide duplexes (Eurogentec). The sequences of the siRNA duplexes were: for human ZEB1, 5′ GGUAGAUGGUAAUGUAAUA 3′ matching the human ZEB1 sequence (NM_001128128.2) (Lehmann et al., 2016); for human ZEB2, 5′ GCAUGUAUGCAUGUGACUU 3′ matching the human ZEB2 sequence (NM_014795.3) (Xia et al., 2010); for human AXL, 5′ CGAAAUCCUCUAUGUCAACAU 3′ matching the human AXL sequence (NM_021913.4) (Antony et al., 2016); control siRNA (Eurogentec, SR-CL000-005); for human SphK2, 5′ GCUGGGCUGUCCUUCAACCU 3′, matching the human SphK2 sequence (NM_020126.4) (Hait et al., 2005).

Protein concentrations in whole-cell lysates were determined with the BCA protein assay kit (Pierce). Proteins from whole-cell lysates (20-60 µg) were separated on 8-10-12% polyacrylamide gels and transferred to Immobilon-FL membranes (Millipore). Primary and secondary antibodies were diluted in blocking buffer (Millipore) containing 0.1% Tween 20. Detection and analysis were performed using the Odyssey Infrared imaging system (LI-COR Biosciences). Immunoblots were quantified by densitometry using Odyssey V3.0 and Fiji.

The rabbit polyclonal anti-AXL (CST, 8661) antibody (1 µg) was incubated with Dynabeads Protein G (25 µl) at room temperature for 1 h. Antibody coated-beads were then incubated with 1 mg of proteins from post-nuclear supernatants of the indicated cells at 4°C for 2 h, washed and loaded on SDS-PAGE. Immunoprecipitates were analyzed by western blotting.

The Proteome Profiler Human Phospho-RTK array kit and human phospho-kinase antibody array kit (R&D Systems) were used according to the manufacturer's instructions. Both array types were incubated with 200 µg of proteins from post-nuclear supernatants isolated from each tested cell line maintained in complete culture medium for 48 h after seeding.

Cells were lysed in lysis buffer A [50 mM Tris (pH 7.5), 10 mM MgCl₂, 0.3 M NaCl, 2% (v/v) IGEPAL]. Lysates were then incubated with 30 µg of Ras-GTP Binding Domain (RBD of human c-Raf kinase) fused with GST associated with beads (RF02, Cytoskeleton) at 4°C for 1 h. Prior to incubation with beads, 1/20 of each cell lysate was collected to be used as ‘input’. Precipitates and inputs were revealed by western blotting using an anti-pan-Ras antibody.

Cells were seeded in 12 mm sterile poly-lysine-coated round glass coverslips and grown to 60-70% confluence. After 15 min incubation at 4°C in serum-free medium, cells were incubated at 4°C in serum-free medium containing 250 μg/ml of D4 anti-AXL humanized antibody for 45 min. Alternatively, cells transfected with the plasmid encoding N-terminally Myc-tagged AXL (N-Myc-AXL) were incubated with the anti-Myc (9E10) antibody. Cells were extensively washed with cold serum-free medium to remove unbound antibodies and surface-bound antibodies were internalized by incubating cells at 37°C in serum-free medium for the indicated time points. Cells were then fixed and permeabilized, and the internalized primary antibody molecules were detected using the appropriate fluorescent secondary antibody (FITC-conjugated anti-human IgG for cells incubated with the D4 antibody, and Alexa488-conjugated anti-mouse IgG for cells that express N-Myc-AXL). Samples were then processed as described in the ‘Immunofluorescence, image acquisition and analysis’ section.

Cells were seeded and after 24 h were washed with ice-cold PBS, incubated with biotin NHS [0.2 mg/ml in PBS (pH 7.9)] at 4°C for 45 min, and shifted to 37°C for the indicated times. Cells were then incubated with Mesna buffer at 4°C for 15 min to remove the remaining cell-surface biotin, followed by a 10 min incubation at 4°C with iodoacetamide to stop the action of Mesna. Cells were then scraped in lysis buffer [10 mM Pipes (pH 7), 100 mM NaCl, 300 mM sucrose, 3 mM MgCl₂, 0.5% (v/v) IGEPAL CA-640, 1 mM EDTA, 1 mM Na₃VO₄ and protease inhibitor cocktail] and the post-nuclear supernatants were harvested after centrifugation (800 g, 4°C for 10 min). One cell plate was lysed immediately after biotinylation and before incubation with Mesna buffer, and was used to estimate the amount of total labeled surface proteins. An equal protein amount for each time point was incubated with streptavidin-coated agarose beads (at 4°C for 45 min). After two washes in PBS, beads were precipitated and resuspended in Laemmli buffer. Samples were separated by SDS-PAGE followed by western blotting with antibodies against AXL and CD71.

Results from the experiments performed using MCF10AmCh and MCF10AF1F2 cells were not expressed as the percentage of total surface-labeled AXL at t=0 min because AXL expression level in MCF10AF1F2 cells was much stronger than in MCF10AmCh cells (see Figs 3A and 4G). In these conditions, expressing the data classically as the percentage of the maximum level of surface-labeled AXL at t=0 min would have led to a strong underestimation of any difference in the internalization rate. Thus, to estimate correctly the internalization rate, results at each time point were expressed as the percentage of the maximum level of internalized protein (reached at 8 min for MCF10AF1F2 cells and at 12 min for MCF10A cells). Because the AXL expression level remained unchanged between untreated MDA-MB-231shLuci cells, MDA-MB-231shLuci cells incubated with opaganib for 4 h and MDA-MB-231shFlot2 cells, and between untreated MCF10AF1F2 cells and MCF10AF1F2 cells incubated with opaganib for 4 h, results at each time point were expressed as the percentage of the maximum level of surface-labeled AXL or CD71 at t=0 min.

Cells were fixed in 3.2% paraformaldehyde for 10 min, followed by a 2 min permeabilization with 0.1% Triton X-100 or by a 10 min permeabilization with 0.05% saponin, and a 15 min saturation in the presence of 2% BSA. Cells were then incubated with primary and secondary antibodies. For CD63, LAMP1 and flotillin immunostaining, cells were fixed and permeabilized with acetone/methanol (v/v, 1/1) at −20°C for 2 min. Images were taken with a confocal SP5-SMD microscope (Leica) with 40×/1.3 or 63×/1.4 oil HCX PL APO CS objectives (Leica), and captured with a hybrid detector (Leica HyD) controlled using the C software. In some cases, stacks of confocal images with a step size of 0.3 μm were processed with Imaris (Bitplane, Zurich, Switzerland) for visualization and volume rendering. Images were processed using Adobe Photoshop and assembled using Adobe Illustrator. For the colocalization measurements of flotillin1/2-mCherry with AXL-GFP, CD71-GFP and SPHK1/2-GFP in vesicles, images were analyzed with the Fiji software and using the Icy software (http://icy.bioimageanalysis.org); a detailed protocol will be provided upon request.

Fixed non-permeabilized cells were immunostained using the monoclonal D4 antibody against AXL extracellular domain and an Alexa Fluor 488-conjugated anti-human secondary antibody. Images were acquired by confocal microscopy under identical illuminated conditions. Using the Fiji software, the outline of each cell was defined and the mean fluorescence intensity (taking into account the cell surface) was quantified in the area defined by the outline.

MCF10A cells were co-transfected with vectors encoding CRY2-mCitrin, flotillin 2-CIBN-mCherry and AXL-Halo labeled with the Halo-Tag-Ligand-Janelia Fluor 646. TIRF images were acquired using an inverted Nikon microscope and a 100×/1.49 oil objective controlled using the Metamorph software (1 image/s). First, Flot2-CIBN-mCherry and AXL-Halo were imaged for 10 s before 488 nm illumination. Cells were then globally illuminated by 22 pulses (1 pulse/s, 100 ms time length for each pulse) of 488 nm light, and CRY2-mCitrin, Flot2-CIBN-mCherry and AXL-Halo signals were simultaneously imaged for 3 min.

Live cells were incubated with NBD-sphingosine or Oregon Green 488-phosphatidylethanolamine (Avanti-Polar lipids) (0.5 µM) added to the culture medium, and cells were imaged by spinning disk confocal microscopy.

Sphingolipid extracts, fortified with internal standards (N-dodecanoylsphingosine, N-dodecanoylglucosylsphingosine, N-dodecanoylsphingosylphosphorylcholine, C17-sphinganine and C17-sphinganine 1-phosphate; 0.2 nmol each), were prepared and analyzed by UPLC-TOF MS as described previously (Cingolani et al., 2014), except for SoP, which was analyzed as follows. Analysis of the extracts was performed with a system consisting of Acquity ultraperformance liquid chromatography (UPLC) (Waters) connected to a triple-quadrupole mass spectrometer (Xevo TQ-S, Waters) and controlled by Waters/Micromass MassLynx software. Detection was performed with an electrospray interface operating in the positive ion mode, capillary voltage set to 3.1 kV, source temperature at 150°C and desolvation temperature at 500°C. The following selected reaction monitoring transitions were acquired: C17 D-erythro-dihydrosphingosine-1-phosphate, 368-252 Da, cone voltage 30 V, collision energy 20 eV; and S1P, 380–264 Da, cone voltage 30 V, collision energy 20 eV.

MCF10AF1F2 cells that express SPHK2-GFP were imaged by spinning disk confocal microscopy (1 image/15 min). The Fiji software and images acquired at 0 and 90 min after opaganib (50 µM) addition in the culture medium were used to define same-size individual regions of interest (ROIs) surrounding one flotillin 2-mCherry-positive vesicle. For each ROI, the mean intensity corresponding to the mCherry and the GFP signals were quantified. To take into account the decrease in signal intensity due to photobleaching, quantifications were also carried out on three ROIs of the same size, but outside flotillin 2-mCherry-positive vesicles. The mean of the obtained values was considered as background and subtracted from the GFP and mCherry signals measured in the ROIs surrounding vesicles. The GFP/mCherry signal ratio was then individually calculated for each ROI surrounding a vesicle on images acquired at 0 and 90 min after opaganib addition.

Time-lapse imaging experiments were performed using an inverted Spinning disk Nikon Ti Andor CSU-X1 microscope equipped with a focus Fluor 100× objective (NA 0.55). Time-series images were acquired with a EMCCD iXon897 Andor camera controlled by the Andor iQ3 software. Time lapse was started 24/48 h after transfection. Time series of the captured images were saved as tiff files that were compiled into avi movies using Fiji software.

PBS-washed cells were fixed with 2.5% glutaraldehyde in PHEM buffer (pH 7.2) at room temperature for 1 h, and then washed in PHEM buffer. Fixed cells were dehydrated using a graded ethanol series (30-100%), followed by 10 min in graded ethanol-hexamethyldisilazane and 10 min in hexamethyldisilazane. Samples were then sputter coated with a 10 nm gold film and examined using a scanning electron microscope (Hitachi S4000, at CoMET, MRI-RIO Imaging, Biocampus, INM Montpellier France) and a lens detector with an acceleration voltage of 10 kV at calibrated magnifications.

Total RNA was prepared from MCF10AmCh and MCF10AF1F2 cells using the RNeasy Mini Kit (Qiagen) according to the manufacturer's instructions. RNA quantity and quality were determined with an Agilent RNA 600 Nano Bioanalyser (Agilent Technologies), and total RNA was concentrated to 100 ng/µl. Pure RNA preparations without DNA or protein contamination were used for sequencing. Libraries, processed by Fasteris (Geneva, Switzerland), were prepared with the TruSeq Stranded mRNA Library Prep Kit from Illumina. Fragments were sequenced on an Illumina Hi-Seq 4000 sequencer according to the manufacturer's instructions.

The differential gene expression analysis was carried out using three replicates per condition to compare the transcript levels between MCF10AmCh and MCF10AF1F2 cells. Two analysis methods were used: DESeq (http://bioconductor.org/packages/release/bioc/html/DESeq2.html) and Tuxedo (http://cufflinks.cbcb.umd.edu/).

Genes with a P_adj<0.05 were assigned as differentially expressed. The Deseq2 and Tuxedo results were compared. Only genes with an expression fold change >2.8 were considered. Using this criterion, 802 up- and downregulated genes were identified.

The PANTHER (protein annotation through evolutionary relationship, version 13.1) classification system (http://www.pantherdb.org/) was used to perform a gene ontology (GO) enrichment. Enriched terms were considered statistically significant when P_adj<0.05, and a minimum of five genes were grouped for each significant term. GO terms were categorized in three major functional groups: ‘biological process’, ‘molecular function’ and ‘cellular component’. A gene set enrichment analysis (GSEA) was performed with the GSEA software v4.1.0. The pre-ranked method was used and the number of permutations was set to 1000. Gene sets with adjusted q-values (FDR) <0.25 were considered significant (Subramanian et al., 2005).

RNA was extracted from the indicated cell lines as described above. mRNA samples were then reverse transcribed to cDNA using SuperScript II Reverse Transcriptase (Thermo Fisher Scientific) and qPCR was performed using an LC480 apparatus (Roche) with the SYBR Green PCR Master Mix (Applied Biosystems). The primer sequences were: for human CDH1, forward 5′-TCATGAGTGTCCCCCGGTAT-3′ and reverse 5′-CAGCCGCTTTCAGATTTTCAT-3′; for human CDH2, forward 5′-GGCAGAAGAGAGACTGGGTC-3′ and reverse 5′-GAGGCTGGTCAGCTCCTGGC-3′; for human vimentin, Duplex QT00095795 (Qiagen); for human ZEB1, forward 5′-GCACCTGAAGAGGACCAGAG-3′ and reverse 5′-TGCATCTGGTGTTCCATTTT-3′; for human ZEB2, forward 5′-GGGAGAATTGCTTGATGGAGC-3′ and reverse 5′-TCTCGCCCGAGTGAAGCCTT-3′; for human MRPL19 (internal control), forward 5′-GGGATTTGCATTCAGAGATCAG-3′ and reverse 5′-GGAAGGGCATCTCGTAAG-3′; for murine Cdh1, forward 5′-GGTCTCTTGTCCCCACA-3′ and reverse 5′-CCTGACCCACACCAAAGTCT-3′; for murine Cdh2, forward 5′-AGGGTGGACGTCATTGTAGC-3′ and reverse 5′-CTGTTGGGGTCTGTCAGGAT-3′; for murine vimentin, forward 5′-AATGCTTCTCTGGCACGTCT-3′ and reverse 5′-GCTCCTGGATCTCTTCATCG-3′; for murine Zeb1, forward 5′-GCTCAGCCAGGAACCCGCAG-3′ and reverse 5′-TGGGCACCCTCTGCCACACA-3′; for murine Zeb2, forward 5′-ATGGCAACACATGGGTTTAGTGGC-3′ and reverse 5′-ATTGGACTCTGAGCAGATGGGTGT-3′; for murine Mrpl32 (internal control), forward 5′-TTAAGCGAAACTGGCGGAAAC-3′ and reverse 5′-TTGTTGCTCCCATAACCGATG-3′; and for human AXL, forward 5′-GTGGGCAACCCAGGGAATATC-3′ and reverse 5′-GTACTGTCCCGTGTCGGAAAG-3′.

The mRNA levels of flotillin 1 and ΔNp63 in tumor samples from the cohort of patients with breast cancer were analyzed as described previously (Rossé et al., 2014).

Cells (5×10⁴) were seeded on the upper chamber of the Boyden chamber inserts (Corning FluoroBlok cell culture insert, 8 µm pores) in 250 µl of DMEM/F12 with 1% horse serum. In the lower chamber, complete medium (750 µl) was added, as a chemoattractant. Opaganib (50 µM) was added in the upper and lower chambers.

After 14 h, cells at the bottom of the filter were fixed with 3.2% paraformaldehyde and stained with Hoechst. The nucleus of cells that had migrated through the pores was imaged with an EVOS microscope (Life Technologies) using a 10× objective. Semi-automatic quantification was performed with Fiji. For each experiment, each condition was performed in triplicate.

Patients (n=43) with invasive breast carcinoma were treated at the Cancer Research Institute, Tomsk NRMC (Tomsk, Russia). Frozen and formalin-fixed paraffin-embedded (FFPE) tumor tissue specimens obtained during surgery were used for immunofluorescence analyses.

The histological type was defined according to the World Health Organization recommendations (Lakhani et al., 2012). Tumor grade was determined using the Bloom and Richardson grading system (Bloom and Richardson, 1957). The expression of estrogen receptor (ER) and progesterone receptor (PR) was scored using the HSCORE method (Kinsel et al., 1989). HER2 expression was analyzed by immunohistochemistry and calculated on a scale 0-3+, according to the ASCO/CAP guidelines (Wolff et al., 2007). Ki-67 expression was calculated as the percentage of Ki-67-positive tumor cells relative to all tumor cells. Molecular subtypes were categorized on the basis of the primary tumor ER, PR, HER2 and Ki-67 status, according to the St Gallen recommendations (Wolff et al., 2007): luminal A (ER⁺ and/or PR⁺, HER2⁻ and Ki-67<20%), luminal B (ER⁺ and/or PR⁺, HER2^+/− and Ki-67≥20%), HER2⁺ (ER⁻ and PR⁻, HER2⁺) and triple negative (ER⁻, PR⁻ and HER2⁻). The study procedures were in accordance with the Helsinki Declaration (1964, amended in 1975 and 1983). This study was approved by the institutional review board of the Cancer Research Institute, Tomsk National Research Medical Center, Tomsk, Russia; all patients signed an informed consent for voluntary participation.

Clinicopathological characteristics were noted for the patients who gave the breast cancer samples used for the analysis (Fig. S4).

Patient 1: 65-year-old woman, invasive carcinoma (left side), T2N1M0, grade 3, ER⁻, PR⁻, HER2⁻, Ki-67-positive cells (50%), triple negative. Lymph node positive (one metastasis from six lymph nodes). Five courses of neoadjuvant chemotherapy.

Patient 2: 55-year-old woman, invasive carcinoma (left side), T1N0M1, multifocal, grade 2, ER⁻, PR⁻, HER2⁻, Ki-67-positive cells (31%), triple negative. Lymph node negative (0 metastases in seven lymph nodes). Six courses of neoadjuvant chemotherapy.

Patient 3: 55-year-old woman, invasive carcinoma (left side), T2N1M0, grade 2, ER⁻, PR⁻, HER2⁻, Ki-67-positive cells (31%), triple negative. Lymph node negative (0 metastases in seven lymph nodes). Six courses of neoadjuvant chemotherapy.

Patient 4: 55-year-old woman, invasive carcinoma (left side), T1N0M0, grade 3, ER⁻, PR⁻, HER2⁻, Ki-67-positive cells (52%), triple negative. Lymph node-negative (0 metastases from 11 lymph nodes). No neoadjuvant chemotherapy.

Patient 5: 48-year-old woman, invasive carcinoma (right side), T2N1M0, grade 3, ER⁻, PR⁻, HER2⁻, Ki-67-positive cells (37%), triple negative. Lymph node positive (two metastases from eight lymph nodes). Five courses of neoadjuvant chemotherapy.

Patient 6: 48-year-old woman, invasive carcinoma (left side), T2N1M0, grade 3, ER⁻, PR⁻, HER2⁻, Ki-67-positive cells (80%), triple negative. Lymph node negative (0 metastases from 10 lymph nodes). Six courses of neoadjuvant chemotherapy.

Patient 7: 24-year-old woman, invasive carcinoma (left side), T4N0M0, grade 2, ER⁻, PR⁻, HER2⁻, Ki-67-positive cells (60%), triple negative. Lymph node negative (0 metastases from 10 lymph nodes). Six courses of neoadjuvant chemotherapy.

Patient 8: 59-year-old woman, invasive carcinoma (right side), T1N1M0, grade 2, ER⁻, PR⁻, HER2⁻, Ki-67-positive cells (61%), triple negative. Lymph node positive (one metastasis from eight lymph nodes). No neoadjuvant chemotherapy.

Patient 9: 51-year-old woman, invasive carcinoma (right side), T2N1M0, grade 2, ER⁻, PR⁻, HER2⁻, Ki-67-positive cells (48%), triple negative. Lymph node positive (one metastasis from 10 lymph nodes). Five courses of neoadjuvant chemotherapy.

Patient 10: 65-year-old woman, invasive carcinoma (left side), T1N1M0, grade 2, ER⁻, PR⁻, HER2⁻, Ki-67-positive cells (50%), triple negative. Lymph node positive (one metastasis from 10 lymph nodes). No neoadjuvant chemotherapy.

Patient 11: 60-year-old woman, invasive carcinoma (left side), T2N0M0, grade 2, ER⁻, PR⁻, HER2⁻, Ki-67-positive cells (45%), triple negative. Lymph node negative (0 metastases from eight lymph nodes). No neoadjuvant chemotherapy.

Patient 12: 43-year-old woman, invasive carcinoma (left side), T2N0M0, grade 2, ER⁻, PR⁻, HER2⁻, Ki-67-positive cells (60%), triple negative. Lymph node negative (0 metastases from eight lymph nodes). No neoadjuvant chemotherapy.

Patient 13: 54-year-old woman, invasive carcinoma (right side), T2N1M0, grade 3, ER⁻, PR⁻, HER2⁻, Ki-67-positive cells (60%), triple negative. Lymph node positive (one metastasis from 11 lymph nodes). Six courses of neoadjuvant chemotherapy.

Patient 14: 65-year-old woman, invasive carcinoma (left side), T3N2M0, grade 2, ER⁻, PR⁻, HER2⁻, Ki-67-positive cells (80%), triple negative. Lymph node positive (four metastasis from 12 lymph nodes). Six courses of neoadjuvant chemotherapy.

Patient 15: 59-year-old woman, invasive carcinoma (left side), T1N1M0, grade 2, ER⁻, PR⁻, HER2⁻, Ki-67-positive cells (74%), triple negative. Lymph node positive (one metastasis from nine lymph nodes). Six courses of neoadjuvant chemotherapy.

Patient 1: 49-year-old woman, invasive carcinoma (right side), T2N0M0, grade 2, ER+, PR+, HER2⁻, Ki-67-positive cells (10%), luminal. Lymph node negative (0 metastases in three lymph nodes). No neoadjuvant chemotherapy.

Patient 2: 66-year-old woman, invasive carcinoma (right side), T2N1M0, grade 2, ER⁺, PR⁺, Her2⁻, Ki-67-positive cells (35%), luminal. Lymph node positive (3/10 lymph nodes with metastases). No neoadjuvant chemotherapy.

Patient 3: 40-year-old woman, invasive carcinoma (right side), T2N1M0, grade 2, ER⁺, PR⁺, HER2⁻, Ki-67-positive cells (33%), luminal. Lymph node positive (2/10 lymph nodes with metastases). No neoadjuvant chemotherapy.

Patient 4: 46-year-old woman, invasive carcinoma (right side), T2N1M0, grade 2, ER⁺, PR⁻, HER2⁻, Ki-67-positive cells (12%), luminal. Lymph node positive (5/6 lymph nodes with metastases). No neoadjuvant chemotherapy.

Patient 5: 57-year-old woman, invasive carcinoma (right side), T4N3M0, grade 2, ER⁺, PR⁺, HER2⁻, Ki-67-positive cells (17%), luminal. Lymph node positive (2/6 lymph nodes with metastases). Six courses of neoadjuvant chemotherapy.

Patient 6: 32-year-old woman, invasive carcinoma (right side), T2N1M0, grade 2, ER⁺, PR⁺, HER2⁻, Ki-67-positive cells (10%), luminal. Lymph node positive (6/10 lymph nodes with metastases). No neoadjuvant chemotherapy.

Patient 7: 60-year-old woman, invasive carcinoma (left side), T2N1M0, grade 2, ER⁺, PR⁺, HER2⁻, Ki-67-positive cells (30%), luminal. Lymph node positive (2/11 lymph nodes with metastases). Six courses of neoadjuvant chemotherapy.

Patient 8: 45-year-old woman, invasive carcinoma (right side), T2N1M0, grade 2, ER⁺, PR⁻, HER2⁻, Ki-67-positive cells (21%), luminal. Lymph node negative (0 metastases from 8 lymph nodes). No neoadjuvant chemotherapy.

Patient 9: 36-year-old woman, invasive carcinoma (left side), T2N0M0, grade 2, ER⁺, PR⁺, HER2⁻, Ki-67-positive cells (28%), luminal. Lymph node negative (0 metastases from six lymph nodes). No neoadjuvant chemotherapy.

Patient 10: 60-year-old woman, invasive carcinoma (left side), T2N0M0, grade 3, ER⁺, PR⁺, HER2⁻, Ki-67-positive cells (10%), luminal. Lymph node negative (0 metastases from 10 lymph nodes). No neoadjuvant chemotherapy.

Patient 11: 55-year-old woman, invasive carcinoma (left side), T2N1M0, grade 3, ER⁺, PR⁺, HER2⁻, Ki-67-positive cells (25%), luminal. Lymph node positive (1/10 lymph nodes with metastases). No neoadjuvant chemotherapy.

Patient 12: 56-year-old woman, invasive carcinoma (left side), T2N1M0, grade 2, ER⁺, PR⁺, HER2⁻, Ki-67-positive cells (10%), luminal. Lymph node positive (2/7 lymph nodes with metastases). No neoadjuvant chemotherapy.

Patient 13: 62-year-old woman, invasive carcinoma (right side), T1N0M0, grade 2, ER⁺, PR⁻, HER2⁻, Ki-67-positive cells (20%), luminal. Lymph node negative (0 metastases from 10 lymph nodes). Six courses of neoadjuvant chemotherapy.

Patient 14: 48-year-old woman, invasive carcinoma (left side), T2N0M0, grade 2, ER⁺, PR⁻, HER2⁻, Ki-67-positive cells (10%), luminal. Lymph node negative (0 metastasis in 9 lymph nodes). No neoadjuvant chemotherapy.

Patient 15: 56-year-old woman, invasive carcinoma (left side), T2N1M0, grade 2, ER⁺, PR⁺, HER2⁺, Ki-67-positive cells (25%), luminal. Lymph node positive (1/10 lymph nodes with metastases). Six courses of neoadjuvant chemotherapy.

Patient 16: 52-year-old woman, invasive carcinoma (right side), T2N2M0, grade 3, ER⁺, PR⁺, HER2⁻, Ki-67-positive cells (33%), luminal. Lymph node positive (6/18 lymph nodes with metastases). Six courses of neoadjuvant chemotherapy.

Patient 17: 53-year-old woman, invasive carcinoma (right side), T1N1M0, grade 2, ER⁺, PR⁻, HER2⁻, Ki-67-positive cells (12%), luminal. Lymph node negative (0 metastases in 8 lymph nodes). No neoadjuvant chemotherapy.

Patient 1: 47-year-old woman, invasive carcinoma (right side), multifocal, T4N2аM0, grade 3, ER⁻, PR⁻, HER2⁺; Ki-67-positive cells (65%). Lymph node positive (7/16 lymph nodes with metastases). Six courses of neoadjuvant chemotherapy.

Patient 2: 52-year-old woman, invasive carcinoma (left side), T2N1M0, grade 2, ER⁻, PR⁻, HER2⁺; Ki-67-positive cells (42%). Lymph node positive (6/18 lymph nodes with metastases). No neoadjuvant chemotherapy. Distant metastasis in liver and lung.

Patient 3: 58-year-old woman, invasive carcinoma (right side), T1N0M0, grade 2, ER⁻, PR⁻, HER2⁺; Ki-67-positive cells (35%). Lymph node positive (6/10 lymph nodes with metastases). Six courses of neoadjuvant chemotherapy.

Patient 4: 37-year-old woman, invasive carcinoma (right side), T2N0M0, grade 2, ER⁻, PR⁻, HER2⁺; Ki-67-positive cells (50%). Lymph node negative (0/11 lymph nodes with metastases). Six courses of neoadjuvant chemotherapy.

Patient 5: 56-year-old woman, invasive carcinoma (left side), T3N1M0, grade 3, ER⁻, PR⁻, HER2⁺; Ki-67-positive cells (34%). Lymph node positive (1/15 lymph nodes with metastases). Six courses of neoadjuvant chemotherapy.

Patient 6: 64-year-old woman, invasive carcinoma (left side), T2N0M0, grade 2, ER⁻, PR⁻, HER2⁺; Ki-67-positive cells (70%). Lymph node negative (0 lymph nodes with metastases from 6 lymph nodes). Six courses of neoadjuvant chemotherapy.

Patient 7: 41-year-old woman, invasive carcinoma (left side), T2N1M0, grade 2, ER⁻, PR⁻, HER2⁺; Ki-67-positive cells (76%). Lymph node negative (0 lymph nodes with metastases from 8 lymph nodes). Six courses of neoadjuvant chemotherapy.

Patient 8: 51-year-old woman, invasive carcinoma (right side), T2N1M0, grade 2, ER⁻, PR⁻, HER2⁺; Ki-67-positive cells (60%). Lymph node positive (2/6 lymph nodes with metastases). Six courses of neoadjuvant chemotherapy.

Patient 9: 50-year-old woman, invasive carcinoma (left side), T2N1M0, grade 2, ER⁻, PR⁻, HER2⁺; Ki-67-positive cells (54%). Lymph node positive (2/8 lymph nodes with metastases). Six courses of neoadjuvant chemotherapy.

Patient 10: 53-year-old woman, invasive carcinoma (right side), multifocal, T3N1M0, grade 2, ER⁻, PR⁻, HER2⁺; Ki-67-positive cells (55%). Lymph node positive (1/15 lymph nodes with metastases). Six courses of neoadjuvant chemotherapy.

Patient 11: 56-year-old woman, invasive carcinoma (right side), T3N2M0, grade 3, ER⁻, PR⁻, HER2⁺; Ki-67-positive cells (45%). Lymph node positive (8 metastases in 10 lymph nodes). Six courses of neoadjuvant chemotherapy.

The clinicopathological characteristics of the patients without cancer who gave the breast samples used for this expression analysis were:

Patient 1: 32-year-old woman, fibrocystic breast disease (left side).

Patient 2: 54-year-old woman, fibrocystic breast disease (left side).

Patient 3: 45-year-old woman, fibrocystic breast disease (left side).

Patient 4: 55-year-old woman, fibrocystic breast disease (right side).

Patient 5: 37-year-old woman, fibrocystic breast disease (right side).

Patient 6: 28-year-old woman, fibrocystic breast disease (left side).

FFPE tumor sections (7 μm) were deparaffinized, rehydrated, processed for heat-induced epitope retrieval with EDTA buffer (pH 8.0) using PT Link (Dako, Denmark), and blocked with 3% bovine serum albumin (Amresco, USA) in PBS. Expression of flotillin 1 (F1180, 1:50, Sigma) and AXL (C89E7, CST) was assessed using the BondRXm immunostainer (Leica). TSA visualization was performed with Opal 520 and Opal 690 (Opal seven-color IHC Kit; NEL797B001KT; PerkinElmer), nuclei were stained with DAPI, and samples were mounted in Fluorescence Mounting Medium (Dako, Denmark). Samples were analyzed using the Vectra 3.0 Automated Quantitative Pathology Imaging System (PerkinElmer).

All statistical analyses and graphs were performed with GraphPad Prism version 9. All data were first tested for normality using the Kolmogorov-Smirnov normality test. For experiments with n>30, an unpaired, two-tailed Student's t-test was used to identify significant differences between experimental conditions. For experiments with n<30, the non-parametric Mann-Whitney U-test was used. The n value and number of independent experiments are listed in the figure legends. At least four independent experiments were performed. In the figures, only significant differences are indicated. For experiments measuring AXL and CD71 decay over time upon incubation with CHX, non-linear regression with least squares fitting was used.

We thank Peggy Raynaud, Dominique Helmlinger, Dylane Detilleux, Christelle Dantec, Dimitris Xirodimas and Nathalie Abadie-Andrieu. We acknowledge the imaging facility MRI (supported by the French National Research Agency ANR-10-INBS-04).

The peer review history is available online at https://journals.biologists.com/jcs/article-lookup/doi/10.1242/jcs.259178.