Embryonic stem cells are devoid of macropinocytosis, a trafficking pathway for activin A in differentiated cells Free

In multicellular organisms, the constantly forming ligand–receptor complexes transduce signals to the cell interior, transmitting vital information concerning the extracellular environment and neighbouring cells. Such internalisation proceeds along the various endocytic pathways, including via clathrin-mediated endocytosis (CME), caveolae, macropinocytosis (MP), and the APPL pathway (Miaczynska et al., 2004), the non-clathrin and non-caveolar pathway (Sabharanjak et al., 2002), the FEME pathway (Boucrot et al., 2015) and others. The internalisation route ligand–receptor complexes follow determines the signalling output by regulating the selection of effectors, inhibitors or modifiers that may interact with the signalling molecules along the various endocytic pathways or by defining the metabolic environment in which the signalling will exert its effects. For instance, MP is a key adaptive mechanism allowing internalisation of the abundant extracellular substrates (proteins) when nutrients delivered by the blood are scarce (e.g. amino acids and glucose) (Commisso et al., 2013).

Activin A, a homodimeric growth factor encoded by the INHBA gene, is a secreted member of the TGF-β superfamily ligands that has pleiotropic biological functions (Wijayarathna and de Kretser, 2016) including stimulation of hormonal activity, cell type-dependent effects on proliferation and survival, and dose-dependent developmental effects, being a potent mesoderm inducer in Xenopus XTC cells (Smith et al., 1990). Activin A as a morphogen exhibits concentration dependency, acting in gradients. Thus, at low doses, activin A maintains hESCs in pluripotency, activating SMAD2 and SMAD3 (SMAD2/3) proteins, which bind directly the promoter of the NANOG gene, regulating its transcription (Vallier et al., 2009a; Xu et al., 2008), whereas at high doses, activin A induces differentiation of hESCs into mesendoderm (D'Amour et al., 2005; McLean et al., 2007; Vallier et al., 2009b).

The complexity of the different biological functions of activin A is in stark contrast with the rather simple core signalling machinery. Indeed, a mature, dimeric activin A triggers heteromeric complex formation between two specific transmembrane type II (ACTR-II and ACTR-IIB, also called ACVR2A and ACVR2B) and two type I (ALK4, also called ACVR1B) Ser/Thr kinase receptors in which ACTR-II–ACTR-IIB transphosphorylates ALK4. The activated ALK4 phosphorylates SMAD2/3 proteins, which oligomerise with SMAD4, forming complexes that accumulate in the nucleus where they exert transcriptional activity (Shi and Massague, 2003). It is not yet fully understood how the core SMAD2/3 proteins mediate the plethora of different activities of activin A following ALK4 activation. Indeed, activin A may result in altered SMAD2/3 activation in many ways, such as by inducing non-SMAD cross-talking pathways, exemplified by activation of phosphoinositide 3-kinase (PI3K), which inhibits activin-A induced phosphorylation of SMAD2/3 affecting hESCs differentiation (D'Amour et al., 2005; Yu et al., 2015), or by imposing a differential spatio-temporal regulation on the activin A–receptor complex due to the use of various endocytic pathways and thereby affecting the signalling outcome.

Knowledge regarding activin A–receptor complex trafficking routes and associated signalling regulation is limited and conflicting. Initial experiments showing that activin A does not require internalisation to signal (Zhou et al., 2004; Hagemann et al., 2009) were challenged by experiments demonstrating that activin A dynamin-dependent internalisation is indispensable for signalling (Jullien and Gurdon, 2005). Also, it has been shown that, in Xenopus, RAP2, which is a member of the Ras GTPases family, is required for the rapid recycling of activin receptors, in a Rab11-dependent manner, in order to maintain their levels on the cell surface in the absence of ligand (Choi et al., 2008). However, no studies have been conducted, to our knowledge, regarding internalisation of activin A and the trafficking routes of the ligand and receptors. By contrast, TGF-β receptor endocytosis has been well studied. According to experimental data, TGF-β receptors are internalised through both clathrin-coated pits and caveolin-1-positive vesicles with opposing effects on signal transduction (Siegert et al., 2018). Interestingly, recent data has suggested that the 4′-phosphatase INPP4B has a role in TGF-β receptor internalisation (Aki et al., 2020); this phosphatase along with MTMR6, is essential for MP (Maekawa et al., 2014). Similar studies are missing for activin A and in the present study we address this gap in the field. Thus, we have investigated the trafficking of activin A in differentiated endothelial cells (ECs) and human embryonic stem cells (hESCs).

To identify the internalisation pathways for activin A, Alexa Fluor-488-labelled activin A (activin A–Alexa488) was incubated with human umbilical vein endothelial cells (HUVECs) for 60 min. The labelling process for activin A–Alexa488 revealed four populations of different specific activities that exhibited comparative activity to unlabelled activin A for transcriptional activation of the reporter cell line Α431/(CAGA)₁₂-Luc/Ren (Sflomos et al., 2011) or SMAD2/3 phosphorylation, showing that the labelling process did not affect the activity of the ligand (Fig. S1A,B). Internalisation of activin A–Alexa488 was dependent on binding to ACTR-IIB receptors, as silencing of ACVR2B gene abolished its internalisation in HUVECs, whereas silencing of ACVR1B gene (encoding ALK4) allowed a notable amount to be internalised via ActR-IIB (Fig. S1C), which is consistent with the literature (Attisano et al., 1993). Likewise, internalisation of activin A–Alexa488 was inhibited at 4°C compared to 37°C (Fig. S1D), and, as expected, was fully competed for when there was an excess of unlabelled activin A (Fig. S1E).

Activin A–Alexa488 colocalised with the early endosomal markers EEA1 and Rab5a, albeit with different kinetics. Colocalisation with EEA1 reached a maximum at 20 min, whereas with Rab5, the colocalisation peaked at 30 min, decreased at 45 min and showed a secondary increase at 60 min (Fig. 1A). Moreover, activin A–Alexa488 colocalised extensively with rabankyrin-5 in vesicular structures, especially at later time points, suggesting internalisation via the macropinocytic route (Fig. 1A). Interestingly, labelled activin A colocalised with Rab4 family proteins, indicating that it follows the fast recycling pathway towards the plasma membrane (Fig. 1A), while colocalisation with Rab11 family proteins, a late recycling pathway marker, was very low (Fig. S2). The colocalisation of activin A–Alexa488 at 30 min with the late endosomal marker Rab7 family proteins suggests lysosomal targeting and degradation and, indeed, in cells overexpressing Rab7 at high levels, activin A–Alexa488 was completely degraded within 60 min (data not shown). Finally, labelled activin A colocalised minimally with caveolin-1, APPL2 and CD63 (Fig. S2).

Monitoring ligand–receptor routes by administering activin A following mild overexpression of ACTR-IIB and ALK4 receptors revealed that both receptors colocalised with the early endosomal marker EEA1, peaking at 20 min (Fig. 1B) and colocalised with Rabankyrin-5 and Rab11, throughout the 60 min time course (Fig. 1B). However, there were notable differences, ACTR-IIB colocalised with the late endosomal marker Rab7, whereas ALK4 colocalised with Rab4. This suggests that ACTR-IIB is targeted for degradation, while ALK4 is recycled through a Rab4 positive compartment. Both receptors traffic through a Rab11-positive pathway (Fig. 1B).

Next, we addressed the role of each of the main endocytic routes on activin A internalisation and SMAD2/3-dependent signalling. Depletion of clathrin heavy chain 1 protein (CHC, also known as CLTC) in HUVECs, monitored by inhibition of Transferrin–Alexa568 uptake, caused a significant reduction in activin A–Alexa488 internalisation (Fig. 2A) and a statistically significant decrease in the level of activin A-induced phosphorylation of SMAD2/3 (Fig. 2B). In agreement, decreased nuclear translocation of SMAD2/3 and increased cytoplasmic and juxtamembrane localisation in the CHC-depleted cells was observed (Fig. 2C). TGF-β1 receptor complexes internalise both via CME and the caveolar pathway (Di Guglielmo et al., 2003). Despite the fact that there was no significant colocalisation of activin A with caveolin-1, we investigated a potential role of caveolar endocytosis. Silencing of caveolin-1 had no effect on SMAD2/3 phosphorylation (Fig. S3A) or on activin A–Alexa488 internalisation (Fig. S3B). As a positive control for the effect of caveolin-1 knockdown, we examined the proliferation of ECs. The lower amount of caveolin-1 protein after siRNA transfection led to decreased VEGF-induced proliferation of ECs (Fig. S3C), in accordance with the literature (Tahir et al., 2009; Xu et al., 2017).

In conclusion, our results indicate that CME is an important internalisation route for the activin A–receptor complex, whereas the caveolar pathway is not involved. Furthermore, inhibition of CME decreased the internalisation of activin A and also resulted in decreased SMAD-dependent signalling. As the inhibition of internalisation of activin A was ∼50% when CME was inhibited, it was likely that a second endocytic pathway is also involved.

We considered MP as a possible trafficking route for activin A because (1) depletion of CHC did not completely inhibit internalisation of activin A–Alexa488 or activin A-induced SMAD2/3 phosphorylation and (2) there was extensive colocalisation of activin A–Alexa488 and receptors with rabankyrin-5 on vesicular structures, which appeared larger than those containing transferrin–Alexa568 (Fig. 1A). We obtained more direct evidence that MP is an internalisation route for activin A in ECs (1) by showing colocalisation of activin A in macropinosomal-like structures with a marker of macropinocytic structures (Araki et al., 1996) 70 kD Texas Red-labelled dextran (Dextran-70 kDa) at early time points (5 min) after its addition (Fig. 3A) and (2) by demonstrating that the size of activin A–Alexa488-positive vesicles was greater than 0.25 μm, and a higher number of larger vesicles were activin A–Alexa488-positive rather than transferrin-positive (Fig. 3B). Moreover, depletion of the CDC42 protein, a key player in formation of macropinosomes (Garrett et al., 2000; Koivusalo et al., 2010), resulted in lower phosphorylation of SMAD2/3 upon activin A treatment (Fig. 3C) and in significantly lower amounts of internalised activin A–Alexa488 and Dextran-70 kDa (Fig. 3D and Fig. S4).

Consistent with results in previous reports, where different growth factors stimulate MP (Basagiannis et al., 2016; Schmees et al., 2012), treatment of cells with activin A for 20 min increased the amount of internalised Dextran-70 kDa compared to that in unstimulated cells (Fig. 3E). Phorbol myristate acetate (PMA), a well-known inducer of MP was used as a positive control (Swanson, 1989). Furthermore, activin A induced actin reorganisation (Fig. 3F) and membrane ruffling (processes important in MP), as indicated by the arrows and shown in the accompanying movies (Movies 1 and 2).

Likewise, depletion and overexpression of rabankyrin-5 in HUVECs, a protein that has been previously shown to be required for MP in A431 and NIH3T3 cells (Schnatwinkel et al., 2004), resulted in a reduction of SMAD2/3 phosphorylation (Fig. 4A) and uptake of activin A–Alexa488 by ECs upon depletion (Fig. 4B), and an enhancement of SMAD2/3 phosphorylation by activin A upon overexpression (Fig. 4C). The efficiency of gene silencing and the level of overexpression are shown in Fig. S5A,B. Importantly, activin A–Alexa488 and its type I receptor ALK4 exhibited an extensive degree of colocalisation with rabankyrin-5 suggesting that activin A may follow a macropinocytic pathway in ECs (Figs 4D and 1A).

In agreement with the above hypotheses that activin A undergoes MP, chemical inhibitors of MP, such as 5-(N-ethyl-N-isopropyl)amiloride (EIPA), which inhibits the Na⁺/H⁺ exchanger (NHE) (West et al., 1989), cytochalasin D, which alters actin dynamics, abolishing the membrane ruffling required for MP, and LY294002, a PI3K inhibitor, which inhibits ruffle formation and cup closure (Amyere et al., 2000; Hoeller et al., 2013), reduced phosphorylation of SMAD2/3 post induction with activin A (Fig. 5A–C) and, almost, abolished the internalisation of labelled activin A (Fig. 5D,E). Finally, treatment of cells with SU6656, an inhibitor of Src kinase, a crucial mediator of MP (Mettlen et al., 2006), reduced activin A-induced phosphorylation levels of SMAD2/3 in a statistically significant manner (Fig. S6A). Collectively, these results strongly suggest that MP is one of the key trafficking routes of activin A in HUVECs. Furthermore, addition of both EIPA and Dynasore abolished activin A-induced phosphorylation of SMAD2/3 (Fig. S6B), suggesting that CME and MP are prominent endocytic routes of activin A in HUVECs. However, we cannot rule out the contribution of additional endocytic pathways that might be involved in the uptake of activin A, nor indeed can we dismiss the possibility of signalling of activin A from the plasma membrane.

Treatment of hESCs with EIPA, an inhibitor of MP, did not affect activin A-induced phosphorylation of SMAD2/3, whereas administering Dynasore to the cells (Macia et al., 2006), an inhibitor of dynamin-dependent endocytosis (including CME), reduced activin A-induced SMAD2/3 phosphorylation (Fig. 6A). Furthermore, cell reprogramming rendered activin A-induced human foreskin fibroblasts (HFFs), which could be inhibited by both EIPA and Dynasore (Fig. 6D), insensitive to EIPA, suggesting that MP is not utilised when HFFs were reprogrammed to human induced pluripotent stem cell (hiPSCs) (Fig. 6E). Moreover, both Dynasore and EIPA exhibited a statistically significant inhibition of activin A-induced SMAD2/3 phosphorylation when administered to mature ECs generated by differentiating H1 hESCs via CD34⁺ vascular progenitor cells (VPCs) (Tsolis et al., 2016) (Fig. 6B), to a comparable level to that observed in HUVECs (Fig. 6C).

Importantly, previous microarray analysis of human fibroblasts before and after reprogramming to hiPSCs, as cells reverted from the differentiated to the stem cell phenotype, identified several molecules that were downregulated in the reprogrammed cells including myotubularin-related protein 6 (MTMR6) and inositol polyphosphate-4-phosphatase type IIB (INPP4B) (Kyrkou et al., 2016), both of which have been shown to be implicated in MP (Maekawa et al., 2014). Initially, we verified the essential role of both phosphatases in MP in differentiated cells. As shown in Fig. S7A,B, knocking down either MTMR6 or INPP4B strongly inhibited Dextran-70 kDa uptake. Knockdown efficiency is shown in Fig. S7D. Interestingly, silencing MTMR6 or INPP4B also decreased activin A internalisation in a highly statistically significant manner (Fig. S7A,B). In addition, activin A-induced SMAD2 phosphorylation was also decreased, showing that functional MP mediated by MTMR6 and INPP4B is essential for activin A signal transduction (Fig. S7C).

Quantitative RT-PCR analysis of both the MTMR6 and INPP4B genes revealed that their expression was low in pluripotent H1 and HUES 1 cells; however, levels were high in ECs, such as HUVECs and in ECs (CD34⁺) derived from the differentiation of H1 hESCs. Likewise, their concentration was high in HFF cells and was reduced to baseline upon their reprogramming to hiPSCs (Fig. 6F). In addition, in hESCs, the expression of CDC42, a regulator of MP (Dharmawardhane et al., 2000), was elevated during differentiation (Fig. 6G).

In pluripotent hESCs, activin A–Alexa488 and transferrin exhibited weak vesicular localisation upon internalisation (Fig. 7A, upper panel), whereas Dextran-70 kDa was not internalised (Fig. 7B, upper panel), as expected from the lack of an inhibitory effect of EIPA on SMAD2/3 phosphorylation. Subsequent differentiation of pluripotent hESCs towards CD34⁺ VPCs caused accumulation of activin A in macropinosome-like vesicles of increased size (Fig. 7B), unlike what was observed for transferrin, which was internalised in vesicles of relatively smaller size (Fig. 7A). To ensure that these results were not just a result of 2D culturing, the same experiment was carried out on gastruloids generated by culturing H1 hESCs on patterned surfaces (Warmflash et al., 2014; Martyn et al., 2018). Again, in these biomimetic 3D culturing conditions, internalisation of activin A and transferrin were minimal, and Dextran-70 kDa was not internalised at all (Fig. 7C). These data show that activin A is not internalised in hESCs by MP and apparently utilises CME as a trafficking route in these cells.

To elucidate between activin A being unable to utilise the MP machinery of hESCs and the genuine lack of this machinery in hESCs, first, we tested the effects of PMA (Swanson, 1989) and EGF (Bryant et al., 2007), two known inducers of MP on differentiated cells. Neither compound induced internalisation of Dextran-70 kDa in hESCs (Fig. S8A) and hence we investigated the possibility that MP was refractory to activin A action in hESCs. Indeed, it is known that when activin A is apically applied to polarised compact hESC colonies, it cannot easily access its receptors, which are localised laterally (Etoc et al., 2016). Therefore, we investigated internalisation in small colonies and Versene-generated single cell cultures of H1 hESCs. Whereas activin A was active in small colonies of H1 hESCs as evidenced by the nuclear localisation of SMAD2/3 upon ligand administration (Fig. S8B), the uptake of activin A–Alexa488 and transferrin–Alexa568 remained marginal, and Dextran-70 kDa did not internalise in small colonies of H1 hESCs or single cell cultures (Fig. S8C and D,E, respectively). These results strongly support the conclusion that the molecular machinery of MP is not utilised in hESCs.

Since MP is a property shared by most differentiated cells, we addressed whether MP is established early or late during the differentiation process. Towards this purpose, using a method for the differentiation of hiPSCs to ECs via VPC progenitors (Tsolis et al., 2016), we observed that during days 3–5 of the differentiation process, Activin–Alexa488 and Dextran-70 kDa internalisation increased and internalised activin A–Alexa488 colocalised with Dextran-70 kDa and rabankyrin-5 (Fig. 8A). Strikingly, rabankyrin-5 exhibited a diffuse and vesicular staining in hESCs, however on the third and fifth day of differentiation rabankyrin-5 localised to larger vesicles, reminiscent of macropinosomes (Fig. 8B). It is noteworthy that activin A and Dextran-70 kDa were internalised by differentiated cells at the hESC gastruloid colony border (Fig. S8F), suggesting that as cells differentiate the MP pathway becomes activated.

In the present study, we demonstrate that activin A internalisation via both the CME and MP pathways is required for downstream signalling via SMAD2/3 phosphorylation in ECs. Inhibiting CME, by silencing CHC, reduced activin A internalisation and downstream signalling, in agreement with previous reports for activin A (Jullien and Gurdon, 2005) and TGF-β1 (Hayes et al., 2002; Penheiter et al., 2002). By contrast, the identification of MP as an internalisation route for activin A was unexpected as there are no reports of TGF-β family members trafficking through this pathway, even though recently MP has been implicated in the BMP pathway, as a BMPR homeostatic mechanism to regulate BMP-mediated synaptic development (Kim et al., 2019). Here, we show the role of MP in activin A internalisation and signal transduction. First, we show colocalisation of internalised activin A with a marker of macropinocytic structures (Dextran-70 kDa) in macropinosomal-like vesicles greater than 0.25 μm. Second, we show inhibition of activin A internalisation and SMAD2/3 phosphorylation through inhibition of MP by a number of different mechanisms, including chemical inhibition of Na⁺/H⁺ exchanger NHE, PI3K, actin dynamics and Src kinase. Finally, depletion of CDC42 and rabankyrin-5, proteins required for MP, resulted in reduced activin A-induced SMAD2/3 phosphorylation, whereas overexpression of rabankyrin-5 enhanced activin A-induced SMAD2/3 phosphorylation.

Apparently, activin A utilises both internalisation pathways, CME and MP, in parallel in ECs. CME is characterised by high selectivity of the internalised cargo due to clathrin and adaptor proteins, whereas MP is considered to constitute a non-specific uptake of fluid into large cytoplasmic vesicles exhibiting little selectivity in the surface proteins internalised as macropinosomes, even though some plasma membrane proteins are excluded from macropinosomes (Mercanti et al., 2006). Following internalisation, ActivinA–Alexa488 is mainly localised in Rab5- and rabankyrin-5-positive vesicles in a time-dependent manner and displays a moderate colocalisation with the early endosomal marker EEA1. Interestingly, the increased colocalisation of activin A with Rab5 and its effector rabankyrin-5 is suggestive of the formation of macropinosomes, since the recruitment of Rab5 along with phosphatidylinositol 3-phosphate (PI3P) generation are essential for macropinosome maturation (Yoshida et al., 2009). Moreover, it has been suggested that the amount of Rab5 on macropinosomes continuously increases in order to regulate macropinosomal trafficking (Welliver and Swanson, 2012).

It is an open issue for how long the activin A–ACTR-IIB–ALK4 complex is intact and active in transducing downstream signals at the various topologies of the CME and MP routes and in what context. For instance, activin A–receptor complexes internalised via CME are transported to early endosomes, from where they signal or are recycled to the cell surface or directed to degradative compartments. In the presence of ligand, both ACTR-IIB and ALK4 were detected on early endosomes (colocalisation with Rab5 and EEA1), macropinosomes (colocalisation with Rab5, rabankyrin-5 and Dextran-70 kDa) and the slow recycling compartment (Rab11), whereas ALK4 was preferentially localised on the fast recycling compartment (colocalisation with Rab4) and ACTR-IIB on late endosomes (colocalisation with Rab7). It seems that although the ligand–receptor complex is intact and active transducing downstream signals at the level of the early endocytic compartment (Hayes et al., 2002), receptor dissociation occurs as ALK4 is recycled back to plasma membrane and ACTR-IIB is degraded in lysosomes. Information about the processing of the ligand–receptor complex during its passage through the macropinosomal route is scarce. Macropinosome maturation and the interactions between macropinosomes and other endocytic pathways is cell type specific. In macrophages, macropinosomes fuse with clathrin-derived tubular endosomes (Racoosin and Swanson, 1993), however in A431 and NIH3T3 cells macropinosomes rarely fuse with the clathrin-mediated endocytic compartments (Schnatwinkel et al., 2004; Hewlett et al., 1994). In ECs, there is little information on macropinosome maturation and it is unclear whether the ligand–receptor complex is active as the macropinosomes mature and whether the processing of mature macropinosomes that leads to their lysosomal degradation occurs in a similar manner to that seen in the clathrin-mediated endocytic pathway. Recycling of surface proteins from macropinosomes is reported to occur very early through the action of WASP and SCAR homologue (WASH) and retromer sorting complex (Buckley et al., 2016) to prevent rapid turning over of the entire cell surface given that MP lacks cargo-sorting coats resulting in non-specific internalisation of entire segments of the plasma membrane. In this respect, recent data suggest that type I BMP and activin receptors are regulated by retromer (Gleason et al., 2014; Steinberg et al., 2013).

The question then arises about the specific contribution of CME and MP internalisation of activin A–receptor complexes in activin A signalling output and the context thereof. In CME, clustered ligand–receptor complexes are internalised in early endosomes where SARA presents SMAD2/3 proteins to ALK4 for phosphorylation (Tsukazaki et al., 1998). Although the internalised and active activin A–receptor complex is an indispensable component of activin signal transduction (Jullien and Gurdon, 2005), other studies demonstrate that activin A signals from the plasma membrane (Zhou et al., 2004). In ECs, we found little contribution of activin A signalling from the plasma membrane, as inhibition of both CME and MP completely inhibited ligand internalisation and signalling. In hESCs internalisation of activin A is marginal and inhibition of dynamin-dependent internalisation in hESCs, including CME, while resulting in a significant reduction in the amount of phosphorylated SMAD2/3, did not completely block signalling, thereby suggesting that activin A may also signal from the plasma membrane in this cell type. In Xenopus, activin A signals emanating both from the plasma membrane (Hagemann et al., 2009) and intracellular endocytic structures (Jullien and Gurdon, 2005) determine mesodermal fate.

Activin A is a known morphogen, determining cell fates during development in a concentration and time dependent manner (Gurdon et al., 1994; Kutejova et al., 2009). This is compatible with the recent finding that activin signalling is integrated over a time frame in which ACTR-IIB and ALK4 are not depleted from the plasma membrane, because internalisation and degradation are matched by renewal (Miller et al., 2019), rendering cells responsive to ligand over a large time frame. It is also compatible with our results showing that activin A is minimally internalised in hESCs, suggesting signalling from the membrane. For TGF-β, on the other hand, which has not been shown to exhibit morphogenic activity during development, upon ligand addition, receptors are internalised and the cell becomes refractory to further addition of the ligand (Vizan et al., 2013). Therefore, the trafficking of TGF-β receptors render it incompatible with the rapid responses to ligand gradients required by cells during development (discussed in Miller et al., 2019). In addition to ligand concentration, a recent study led to the conclusion that, during development, cells respond to rapid concentration changes in activin A and NODAL (Heemskerk et al., 2019). We postulate that rapid internalisation of ligand–receptor complexes may be incompatible with the requirements of the cell during embryonic patterning, and this may explain the basis for differences in endocytic trafficking of activin A between differentiated and stem cells. Unless the cell carefully balances receptor internalisation/degradation and recycling it could remain unresponsive to morphogen gradients. This is especially relevant in the case of MP, where large segments of plasma membrane are internalised.

Interestingly, we found in the present study that MP is not utilised in hESCs and is associated only with their differentiated phenotype. Indeed, whereas H1 cells internalised activin A–receptor complexes via CME, the mature ECs derived by differentiation of H1 cells (Tsolis et al., 2016) internalised the ligand–receptor complexes by both CME and MP. Moreover, HFF cells that were capable of internalising activin–receptor complexes by both CME and MP, upon reprogramming to hiPSCs (Kyrkou et al., 2016) retained only the former. Importantly, previously performed microarray analysis from our group, before and after reprogramming of HFFs to hiPSCs revealed that the MTMR6 and INPP4B phosphatases were found dramatically downregulated in HFF-derived hiPSCs (Kyrkou et al., 2016). Likewise, in the present study, we show that their expression was very low in hESCs, but high in HUVECs and in ECs derived from the differentiation of H1 cells (Fig. 6F). In fact, MP was already functioning in mesodermal progenitors, as evidenced by the increased internalisation of activin A–Alexa488 and Dextran-70 kDa at day 3 and 5 and the colocalisation of activin A–Alexa488 with Dextran-70 kDa and rabankyrin-5, using a method for the differentiation of hiPSCs to ECs via VPC progenitors (Tsolis et al., 2016).

MTMR6 belongs to the myotubularin family of phosphoinositide 3-phosphatases (Robinson and Dixon, 2006; Hnia et al., 2012) degrading PI3P to phosphoinositide (PI), whereas INPP4B is a specific 4-phosphatase for phosphatidylinositol (3,4)bisphosphate [PI(3,4)P2] (Norris et al., 1997; Ivetac et al., 2005; Gewinner et al., 2009) generating PI3P. Formation of macropinosomes requires the sequential localised synthesis of (1) phosphoinositides, such as PI(4,5)P2 and phosphatidylinositol (3,4,5)-trisphosphate [PI(3,4,5)P3], to co-ordinate actin polymerisation via actin-associated proteins and (2) small GTPases which are known regulators of actin cytoskeleton. However, phosphoinositide breakdown and Rho GTPase deactivation are also required for complete macropinosome formation. Cells depleted of MTMR6 and INPP4B still exhibit membrane ruffling following EGF induction, suggesting that both function in the steps that follow the formation of membrane ruffles (Maekawa et al., 2014). Indeed, MP is initiated by the formation of such ruffles, which are formed by actin-driven plasma membrane protrusions. Whereas most ruffles rapidly recede, some become circular (macropinocytic cups), which upon closure of the ‘cup’ they become the intracellular macropinosomes vesicles and undergo further maturation for degradation or recycling (Egami et al., 2014). Although it is not clear at which step MTMR6- and INPP4B-induced breakdown of PIs is required, apparently, the consequences of the low expression of INPP4B and MTMR6 and subsequent reduction in the breakdown of PI(3,4)P2 to PI(3)P and subsequently PI(3)P to PI are sufficient to inhibit the MP process. In agreement with the above, we verified that both MTMR6 and INPP4B participate in the regulation of MP in ECs, as demonstrated by inhibition of Dextran-70 kDa uptake. Moreover, we provide experimental data that MTMR6 and INPP4B also contribute to activin A internalisation and signalling in ECs. Although recent experimental evidence implicates INPP4B in TGF-β receptor internalisation (Aki et al., 2020), and INPP5B, a member of the family of inositol polyphosphate-5-phosphatases, in macropinosome formation (Maxson et al., 2021), the contribution of endocytic pathways in the uptake of activin A or signalling of activin A from plasma membrane cannot be ruled out.

In conclusion, we demonstrate that, in differentiated cells, such as ECs, CME and MP constitute the main modes of internalisation of activin A–receptor complexes. Particular attention was paid to the validation of MP internalisation as there are no reports of TGF-β family members trafficking through this pathway. Embryonic stem cells are devoid of MP, a hallmark associated with low expression of CDC42, MTMR6 and INPP4B phosphatases, all key players in MP. Furthermore, during differentiation, rabankyrin-5 becomes localised to enlarged macropinosomal-like structures that are absent from stem cells. Our results reveal, for the first time, that MP is an internalisation route for activin A in differentiated cells, and that MP is not active in ESCs and is induced as cells differentiate.

HUVECs, p2-3, pooled from 30 individuals (VEC Technology), were plated on collagen and cultured in Medium 199 (Gibco) containing 20% heat-inactivated fetal calf serum (FBS, Gibco), endothelial cell growth supplement (ECGS, 0.045 μg/ml), heparin (10 units/μl, Sigma-Aldrich), 100 U/ml penicillin and 100 μg/ml streptomycin (Bellou et al., 2009). The Matrigel-adapted hESC line H1 (WiCell WB0113; Madison, WI, USA), HUES-1 cells (HUES Cell Facility/Melton Laboratory, Harvard University) and hiPSCs (Kyrkou et al., 2016) were cultured in serum free chemically defined medium mTeSR^™ 1 (STEMCELL Technologies, Vancouver, Canada). All cell lines were routinely checked for contamination.

ECs differentiated from hESCs and hiPSCs (Tsolis et al., 2016) were cultured in full HUVEC medium. For generating pluripotent single cells, H1 were incubated with Versene–EDTA 0.02% (Lonza, BE17-711E) for 7 min at room temperature (RT) and were subsequently dissociated into small clumps (3–5 cells) or single cells. The cell aggregates were transferred into ibidi dishes and allowed to attach and proliferate for 24 h in fresh medium supplemented with 5 μM fasudil, a ROCK inhibitor (LC Laboratories, F4660). 10⁶ H1 cells treated with Versene were also plated on each 3D micro-patterned coverslips of CYTOO (CYTOOchips^™ Arena A, 10-020-00-18). Before seeding, coverslips were coated with 50 μg/ml poly-L-lysine (Sigma-Aldrich, P1524) at RT for 2 h, washed with H₂O 6 times by serial dilutions (1:4) and then incubated with Matrigel overnight at 4°C. Matrigel was removed by serial dilutions (1:4) with ice cold PBS and 2 washes. Reprogramming of HFFs to hiPSCs and differentiation of H1 cells to ECs have been previously described (Tsolis et al., 2016; Kyrkou et al., 2016). The accession number for the microarray data discussed here is available from the Gene Expression Omnibus under accession #GSE58932.

The anti-rabankyrin-5 rabbit polyclonal antibody was kindly provided by Marino Zerial (MPI-CBG, Dresden, Germany). Rab5a 4F11 mouse monoclonal antibody was purified from the hybridoma. Rabbit monoclonal antibodies recognizing phosphorylated (p-)SMAD2 and SMAD2/3 were purchased from Cell Signaling (Cell Signaling Technology) and used at dilution 1:1000. Antibodies against p-SMAD3 (Rockland, 1:1000), APPL2 (Abnova, 1:500), caveolin-1 (Santa Cruz Biotechnology, 1:500), CD63 (DHSB, 1:500), CDC42 (Santa Cruz Biotechnology, 1:500), EEA1 (BD, 1:100), BrdU (DSHB, 1:250), biotin (Sigma-Aldrich, 1:100) and actin (Chemicon International Inc., 1:2000), were also used. All secondary antibodies were purchased from Jackson-ImmunoResearch and used at dilution 1:200.

Chemical inhibitors 5-(N-ethyl-N-isopropyl) amiloride (EIPA), Dynasore, Cytochalasin D and LY294002 were purchased from Sigma (Sigma, St Louis, MO) and dissolved in DMSO and used at a 0.5% final concentration. 70 kD-Dextran–Texas Red (Dextran-70 kDa) and Transferrin–AlexaFluor 568 were from Thermo Fisher Scientific (MA, USA). SiR-actin was purchased from Spirochrome. Endotoxin levels were estimated using the QCL1000 kit (BioWhittaker, Inc.) and in the case of activin A were 0.4 pg/ng.

Mature, purified activin A was labelled with NHS–Alexa Fluor 488 (ThermoFisher Scientific) in 100 mM Hepes pH 7.0 to drive the labelling to the N-terminus. 20% acetonitrile was used in the reaction to keep activin A in solution. The labelled protein was acidified with 0.1% trifluoroacetic acid and labelled protein separated using reversed-phased chromatography on Vydac C4 column (4.6×250 mm, 5 µm beads, 300 Å pore size) and eluted with acetonitrile gradient. Labelled protein eluted in a trailing peak, with protein with higher degree of labelling eluting later. Activin-A–Alexa488 was divided into four pools and dried under vacuum for later use. Biotinylated activin A (Biot-ActA) was prepared similar to Activin-A–Alexa488, using Sulfo-NHS-LC-Biotin (ThermoFisher cat. no 21335) as the label and purified by reversed phase chromatography to remove excess biotin. For Biot-ActA detection by immunofluorescence, goat anti-biotin was used as primary antibody and Alexa488-conjugated anti-goat-IgG as secondary antibody.

Cultured cells were lysed with lysis buffer (1% SDS and PMSF). Whole-cell protein lysates were quantified using the Pierce BCA assay kit (Thermo Fisher Scientific). Proteins were fractionated by SDS-PAGE and transferred to nitrocellulose membranes. After incubation with 5% non-fat milk in 10 mM Tris-HCl pH 8.0, 150 mM NaCl, 0.5% Tween 20 (TBST) for 30 min, membranes were incubated with antibodies against p-SMAD2, p-SMAD3, SMAD2/3 and β-actin, overnight at 4°C. Then were washed three times for 10 min and incubated with anti-rabbit-IgG or -mouse-IgG antibody conjugated to HRP for 2 h at RT. Blots were washed three times with TBST and developed with the ECL system (Amersham BioSciences) according to the manufacturer's protocol.

HUVECs were transfected with Endotoxin-free DNAs 24 h after trypsinisation using Metafectene-pro reagent (Biontex Laboratories GmbH). siRNAs (20 or 50 nM) were transfected using Lipofectamine RNAiMAX (Invitrogen, Life Technologies Corporation), according to the manufacturer's specifications.

pcDNA3-ALK4-HA wt was kindly provided by Masahiro Kawabata (Division of Biochemistry, the Cancer Institute of the Japanese Foundation for Cancer Research (JFCR, Tokyo, Japan). pRK5-ActR-IIB-Flag was generated by PCR using a vector harbouring ACTR-IIB kindly supplied by Jeff Wrana (Lunenfeld-Tanenbaum Research Institute, Mount Sinai Hospital, Toronto). The plasmids pCMV-Rab5a, pcDNA3-HA-Rab4a, pcDNA3-HA-Rab7a, pcDNA3-HA-Rab11a, pEGFP-C3-Rab4a, pEGFP-C3-Rab11, pEGFP-C3-hRab5a, pEGFP-C1-Rab7, pEYFP-C1-Rabankyrin-5 were kindly provided by Marino Zerial (MPI-CBG, Dresden, Germany). siRNAs were purchased from Thermo Fisher Scientific (CLTC, ANKFY1, ACVR1b, ACVR2b, CDC42), Biospring (Caveolin-1) or ORIGENE (MTMR6, INPP4B) and are listed in Table S1. For all RNA interference experiments, control siRNA (Silencer Negative Control 5, Ambion or Trilencer-27 Universal scrambled negative control siRNA duplex, ORIGENE) was used.

Rabankyrin-5 and LacZ adenoviruses were kindly provided by Marino Zerial. Alk4–HA adenovirus was generated using the pADEASY system by cloning the HA-tagged open-reading frame into the pCMV shuttle vector (Luo et al., 2007). To detect β-galactosidase expression in the infected cells, cells were fixed in 1% formamide, 1% glutaraldehyde, and 0.02% NP-40 in PBS prior to application of chromogenic substrate/buffer [0.1% X-gal, 2 mM MgCl₂, 5 mM K₄Fe(CN)₆, 5 mM K₃Fe(CN)₆ in PBS, pH 7.3]. Cells were incubated at 37°C until the blue colour was distinct in the LacZ-infected cells and images were acquired of the control cells (non-infected) and LacZ-infected cells at the same time point using Leica DM IRBE Microscope.

Quantitative reverse transcription-PCR (qRT-PCR) experiments were performed using QuantiTect SYBR Green RT-PCR Kit (QIAGEN, GmbH, Hilden, Germany) and LightCycler^® 2.0 (Roche Diagnostics GmbH, Mannheim, Germany). Total RNA was isolated using the RNeasy Kit (QIAGEN). The concentration and the quality of RNA were measured using Agilent RNA 6000 Nano Kit in Agilent 2100 Bioanalyzer (Agilent Technologies Inc, USA). Primers used for the RT-PCR are listed in Table S1.

Relative expression of genes in different samples was calculated by qRT-PCR using the Relative Standard Curve Method. In more detail, serial dilutions of RNA, extracted from HFFs, were used for the generation of standard curve for the gene of interest (i.e. MTMR6 or INPP4B). For each experimental sample, the target quantity is determined by interpolating from the standard curve. For the normalization, the same procedure was followed using GAPDH as the endogenous control. Graphs represent the target quantity of gene of interest (i.e. MTMR6 or INPP4B), divided by the target quantity of GAPDH in each sample. Each sample was run in duplicates and only standard curves with R²>0.99 were used. At least three biological replicates (n=3) were used for statistical analysis.

For western blot analysis, the density of each band was calculated using area×optical density units (ODU) with Quantity One Quantification software (Bio-Rad Laboratories, Hercules, CA). Statistical significance (*P<0.05, **P<0.01, ***P<0.001) was determined by a one-tailed unpaired Student's t-test.

Colocalization analysis, vesicular size measurements and fluorescence intensities were performed using a custom designed, multiparametric image analysis software, MotionTracking, which is freely available at http://motiontracking.mpi-cbg.de (http://pluk.mpi-cbg.de/media/f/docs/manual_3.pdf), developed at MPI-CBG, Dresden, as previously described (Collinet et al., 2010; Rink et al., 2005). This software allows the user to set the scale of an image (μm/pixel) allowing the measurement of the physical size of the detected objects as well as other parameters, such as a background window size, to discriminate fluctuation of background fluorescence from endosomes and other compartments. The threshold of object detection is defined in units of standard deviation of Poisson noise, which is determined for each pixel of the image. The software allows one to discriminate objects in very close proximity to each other, track objects, etc. Additionally, the software includes a wide set of functions for statistical analysis of the data obtained, such as colocalization analysis (Kalaidzidis et al., 2015) and object characteristics distributions (size, area, elongation, intensity and more).

We used object detection threshold corresponding to 4 standard deviations of background noise and minimum object detection window equalling 8 pixels. Colocalization is assessed on basis of cross-sectional overlap, scoring >40% overlap as colocalized, using 12–15 z stacks for each cell. Mean colocalization was calculated for each stack and then for each cell. For one experimental condition, at least 20 cells were analysed. No maximum projections were used. The percentages of colocalization presented in Fig. 1 mean that, for example, 20% of activin A–Alexa488-positive vesicles colocalize with EEA1-positive vesicles at 20 min. Comparison between percentages is meaningful between different time points for one vesicular marker (changes over time), but not between different markers, since the percentages depend on the number of vesicles that are positive for each marker.

For internalisation assays, fluorescence integral intensities were calculated as the sum of integral fluorescence intensities of all vesicles. The ‘integral intensity’ corresponds to the integral of fluorescent marker intensity per vesicle. The ‘total integral intensity’ is defined as the sum of integral intensities of all vesicles in an image normalized by the area covered by the cells. Since MotionTracking approximates real image intensity by using a sum of analytical functions (Rink et al., 2005), the resulting area and intensity have no pixel granularity.

Based on the scale of a confocal microscopy image (μm/pixel) the software can measure the physical size of all the tracked vesicles in micrometers and export the data on number of vesicles and their size, as indicated in Fig. 3B.

Cells were grown on μ-dishes (ibidi, GmbH) for live-cell imaging or 11 mm coverslips at 4×10⁴ cells/well for indirect immunofluorescence and then treated as indicated in figure legends. Indirect immunofluorescence was performed as previously described (Panopoulou et al., 2002). Coverslips were mounted in ProLong Gold Antifade (Thermo Fisher Scientific). Images were obtained using a Leica TCS-SP5 scanning confocal microscope, equipped with Argon/SS-561/HeNe lasers, APO CS 40.0×1.25 OIL UV and APO CS 63.0×1.40 OIL UV objective lenses and Leica Las AF Lite software.

HUVECs were plated on ibidi dishes until 50% confluence and incubated with 1 μM SiR-actin for 1 h. Cells were then induced with activin A (50 ng/ml, Movie 1) or not (Movie 2), placed in a 37°C, 5% CO₂ chamber and monitored using a Leica TCS-SP5 scanning confocal microscope equipped with an Argon/SS-561/HeNe lasers and Leica Las AF Lite software. Frames were taken every 30 s for 30 min with an APO CS 40.0×1.25 OIL UV objective. Images were converted into a video using ImageJ and exported in avi format (Movies 1, 2; at 3 fps). The video is accelerated for visualization (fast-forward ×90 times).

We thank the confocal laser microscope facility of the University of Ioannina for the use of the Leica TCS-SP scanning confocal microscope.