Matrix stiffness regulates Notch signaling activity in endothelial cells

The Notch signaling pathway plays a crucial role in many tissues, regulating cell fate decisions, cell cycle progression and apoptosis during tissue development and maintenance (Liu et al., 2003; Totaro et al., 2018). Notch signaling is induced by a Notch ligand (Delta-like 1, Delta-like 4 or Jagged 1) interacting with one of four types of Notch receptors (Notch1–4) of an adjacent cell (trans interaction), which triggers a signaling cascade in the receiver cell (Boareto et al., 2015; Sprinzak et al., 2010; Totaro et al., 2018). Receptor–ligand binding leads to two receptor proteolytic cleavage events, resulting first in separation of the Notch extracellular domain (NECD) and subsequently in release of the Notch intracellular domain (NICD) (Hodkinson et al., 2007; Nandagopal et al., 2018; Nichols et al., 2007). The NECD is pulled into the ligand-presenting cell (sender cell) along with the ligand in the course of trans-endocytosis, while the NICD is translocated to the nucleus of the receiver cell, where it associates with the transcriptional regulator RBPJ, activating the Notch target genes such as those encoding hairy and enhancer of split (HES) transcription factors and HES-related proteins (Hodkinson et al., 2007; Liu et al., 2003; Meloty-Kapella et al., 2012).

During Notch signaling activation, mechanical forces generated by endocytosis in the sender cell are required for opening the negative regulatory region (NRR) in the Notch receptor, which leads to the first proteolytic cleavage event (Koon et al., 2018; Meloty-Kapella et al., 2012). However, the influence of mechanical cues from the surrounding matrix on this process (i.e. mechanosensitivity) is poorly studied.

The Notch signaling pathway is instrumental for angiogenesis, regulating the selection of tip and stalk cells (Antfolk et al., 2017; Merk et al., 2016). Of the four Notch receptor homologs and five Notch ligands that have been identified in mammals, the Notch receptor Notch1 and the ligands Dll4 and Jag1 are mainly involved in specification of tip and stalk cells (Antfolk et al., 2017; Stepanova et al., 2021; Totaro et al., 2018). Tip cells are characterized by increased Dll4 expression mediated by VEGFA–VEGFR2 (KDR) signaling (Antfolk et al., 2017; Stepanova et al., 2021). Stalk cells, on the other hand, exhibit increased Notch1 expression, and this is associated with the high Dll4 level in the tip cells. Activation of the Notch signaling pathway in stalk cells by Dll4 expression in tip cells induces cell proliferation of stalk cells, resulting in extension of sprouts and lumen formation (Pauty et al., 2018; Stepanova et al., 2021). Dll4–Notch1 signaling is antagonized by the ligand Jag1, which prevents the activation of Notch receptors in stalk cells and promotes tip cell selection and sprouting by Jag1–Notch signaling (Andrawes et al., 2013; Antfolk et al., 2017). Whereas the biochemical cues during angiogenesis are well studied, the mechanical forces during angiogenesis are only beginning to be understood (Mammoto et al., 2009; Santos-Oliveira et al., 2015). Other than the shear force due to blood flow, most of the mechanical cues during angiogenesis are exerted by the extracellular matrix (ECM) and are caused by matrix structural changes that can locally alter the ECM stiffness and density (Fischer et al., 2009; Perfahl et al., 2017; Shamloo et al., 2016). However, little is known about the mechanical influences via the matrix on the signaling pathways involved in angiogenesis, such as the Notch signaling pathway.

In this study, we investigate the influence of different physiologically relevant substrate stiffnesses on the Notch signaling pathway and related pathways in endothelial cells, such as Yes-associated protein (YAP, also known as YAP1) signaling and adhesion-dependent signaling. Using different stiffness-tuned silicone substrates and two different modes of Notch activation with Dll4 as a ligand (via cell–cell contact or a protein coat), we show that activation of the Notch signaling pathway is stiffness dependent. Furthermore, we investigate the influence of cell–cell contacts on Notch activation and the mutual influence of Notch and cell–cell or cell–matrix adhesion proteins, as well as the crosstalk between Notch and YAP signaling, a well-known mechanosensitive pathway (Totaro et al., 2018). Taken together, our results demonstrate a mechanosensitivity of the Notch signaling pathway that is likely associated with the process of trans-endocytosis upon Notch activation by juxtacrine signaling, suggesting a second mechanical aspect of the Notch signaling pathway in addition to the pulling force generated by the ligand-presenting cell.

To investigate the mechanosensitivity of the Notch signaling pathway, synthetic polydimethylsiloxane (PDMS) substrates with defined stiffness in a range from 0.5 to 70 kPa were applied. Endothelial cells [human umbilical vein endothelial cells (HUVECs) and mouse cardiac endothelial cells (MCECs)] were used as prototypic models for Notch signaling. The Notch signaling pathway was not additionally induced or was activated either via seeding of HUVECs onto a coat of recombinant humanDll4 (rhDll4) or by co-culture of HUVEC cells (receiver cells) with Dll4-overexpressing MCECs (MCEC-Dll4–mCherry; sender cells). Due to the dependence of the primary endothelial cells in in vitro culture on a protein coat, all substrates were coated with collagen type I, unless stated otherwise. The purified native collagen type I solution formed a 3D matrix on the substrates and thus enabled uniform adhesion of the endothelial cells. Analysis of a Notch dual luciferase reporter gene assay, which allowed for compensation of potentially varying transfection efficiencies in the primary HUVECs, showed a continuous increase in Notch transcriptional activity on softer substrates both without additional induction of the pathway and after activation either by cell–cell contact using the Dll4-overexpressing cell line or by an rhDll4 protein coating (Fig. 1A–C). However, the differences in Notch transcriptional activity on the various substrates without Notch induction were only minor, especially in contrast to the results after Notch activation. These findings were supported by intensity analysis of the nuclear localization of the NICD after immunofluorescence staining: the softer the substrate, the higher the NICD intensity in the Notch receiver cells upon stimulation (Fig. 1D–F). However, it must be kept in mind that the analysis of stained cells is limited by cell numbers (≥240 in our case) and is not a bulk measurement. Furthermore, results from RT-qPCR experiments showed that mRNA expression of the Notch target gene Hey1 in HUVEC cells was increased on very soft substrates after Notch activation by an rhDll4 coat (Fig. S1C). mRNA expression of the ligand Dll4 showed a stiffness-dependent effect, whereas mRNA expression of the ligand Jag1 remained unchanged in the endothelial cells throughout the examined range of substrate stiffness (Fig. S1C). Thus, Notch activation, both by plate-bound Dll4 and by co-cultured sender cells, exhibits mechanosensitivity. Experiments measuring activation of the Notch signaling pathway upon co-culture with MCEC-Dll4–mCherry cells in different ratios, using either HUVEC cells or wild-type MCECs (MCEC-WTs) as signal receiver cells, showed a sender:receiver cell ratio of 1:1 to be optimal for activation. With a higher amount of sender cells, the overall signal intensity decreased (Fig. S2A,B). Human microvascular endothelial cells (HMECs) were used as an additional endothelial cell model; after co-culture with MCEC-Dll4–mCherry cells they showed results comparable to those obtained using HUVECs and MCECs, including the same mechanosensitivity (Fig. S2C).

To test for general effects of the different substrates on cell viability or proliferative capacity, the proliferation of HUVEC cells on substrates of different stiffness was investigated over periods of 24 h and 48 h. No differences in proliferation rates were detected (Fig. S1A). The efficiency and reproducibility of coating of the PDMS substrates with rhDll4 were controlled in an availability assay by immunostaining, which showed that the rhDll4 coat was evenly distributed and displayed a comparable intensity on all substrates (Fig. S1B).

Since interaction with the ECM is known to be mediated by integrin signaling, we next checked whether Notch signaling affects or is affected by integrins (Hodkinson et al., 2007). Integrin β1 (also known as ITGB1) represents the largest subchain of integrins and is involved in several biological processes, such as adhesion, migration and cell cycle regulation (Lei et al., 2008). Due to the involvement of β1 subchains in cell–ECM interaction, integrin β1 plays a major role, especially in endothelial cells (Howe and Addison, 2012). To investigate whether mechanosensitivity of Notch lies upstream or downstream of integrin signaling, we quantified total and activated β1 integrin levels with and without pretreatment of cells with the Notch inhibitors DAPT and SAHM1 on the different substrates. No cytotoxic effect of the inhibitors on HUVEC cells was detectable at the concentrations used (Fig. S1D,E). Previous studies have shown that levels of activated integrins, but not overall integrin levels, are substrate dependent and are increased on softer substrates (Du et al., 2011). Accordingly, in our model the overall intensity of integrin β1 in HUVEC cells did not change on the different substrates and was not affected by inhibition of the Notch signaling pathway with DAPT or SAHM1 (Fig. 2A,B). In contrast, the softer the substrate, the more integrin β1 was activated (Fig. 2C,D). Upon blocking of basal Notch1 cleavage by DAPT treatment, and thus downstream of Notch signaling, integrin β1 activation decreased substantially, although the correlation between softer substrates and increased integrin β1 activation remained to a small degree. A comparable effect was observed after Notch complex assembly was prevented by treatment with the alternative Notch inhibitor SAHM1 (Fig. 2C). Thus, Notch signaling occurs upstream of integrin activation, as far as mechanosignaling is concerned. This result was confirmed in the second endothelial cell line, MCEC-WT (Fig. S3A).

The role of YAP as a mechanotransducer, and the related mechanosensitivity of the YAP–TAZ (WWTR1) signaling pathway are well known (Matsuo et al., 2021; Nukuda et al., 2015). We and others have shown that YAP activity decreases on soft substrates (Gegenfurtner et al., 2018). To assess a possible YAP–Notch crosstalk in endothelial cells, we performed immunofluorescence staining in HUVEC cells seeded on plastic and PDMS substrates, again with and without addition of DAPT (25 µM, 24 h) or SAHM1 (10 µM, 24 h). As expected, our results show progressively decreased nuclear YAP intensity in HUVEC cells on softer substrates, demonstrating the mechanosensitivity of the YAP–TAZ signaling pathway in our cells (Fig. 3A,C). However, inhibition of the Notch signaling pathway had no clear effect on nuclear YAP intensity; only on the softest substrate of 0.5 kPa did Notch inhibition rescue YAP activation to some degree (Fig. 3B,C). Consistent results were obtained with the control cell line MCEC-WT (Fig. S3B). Since YAP and Notch activity are inversely regulated by substrate stiffness, there seems to be no direct crosstalk between these two signaling pathways in this context.

Since the Notch signaling pathway is contact dependent, we examined the influence of substrate stiffness on the major endothelial adhesion molecule VE-cadherin (also known as CDH5). No correlation between VE-cadherin intensity (i.e. overall VE-cadherin expression) and substrate stiffness was detected. The junction patterns, however, showed stiffness-related changes. The softer the substrate, the less often typically branched and interlinked junction patterns were observed at the cell–cell contacts, and a continuous VE-cadherin junction with a larger intensity area without branches or comb-like structures was visible instead, as shown in the representative images (Fig. 4A,B). We investigated the influence of substrate stiffness on VE-cadherin trafficking at cell–cell contacts using a fluorescence recovery after photobleaching (FRAP) assay. HUVEC cells were transiently transfected with a Citrine-coupled VE-cadherin plasmid and seeded on PDMS substrates with different stiffnesses. VE-cadherin recovered to the same extent and with the same kinetics on all substrates, and therefore, no significant differences in the half time of recovery were observed (Fig. 4C). Thus, the altered junction pattern on the different substrates does not depend on or affect VE-cadherin kinetics.

To investigate the importance of existing cell–cell contacts for basal Notch activity, we first acutely destabilized cell–cell contacts using EGTA and then inhibited them in a prolonged manner using a VE-cadherin-blocking antibody. Treated cells were stained for NICD, and its intensity was analyzed in the nuclei as readout for Notch activity. Because EGTA might affect the structural activity of Notch1, leading to receptor cleavage and activation (Rand et al., 2000; Tremblay et al., 2013), we checked the effect of EGTA on NICD levels in comparison to further blocking of VE-cadherin by an antibody. Basic activity of Notch was weak, and EGTA alone only marginally decreased NICD levels in the nucleus independently of stiffness. The VE-cadherin-blocking antibody enhanced this effect (Fig. 5A) independently of substrate stiffness. The representative images shown in Fig. 5A (0.5 kPa PDMS substrate) confirm the quantitative analyses and show that the treatment changed the cell morphology and caused the cells to drift apart, leading to diminished cell–cell contacts. This is, however, not surprising, since in this setting without further stimulation inherent Notch signaling completely depends on the physical contact between endothelial cells.

To also test for a possible effect of Notch on VE-cadherin, we treated cells with DAPT (25 µM, 24 h) or SAHM1 (10 µM, 24 h). The results show that there was a significant reduction in VE-cadherin intensity after Notch inhibition on all substrates, and the reduction was greater after treatment with SAHM1 compared to the effects of treatment with DAPT (Fig. 5B,D). The junction patterns, however, did not change following the addition of DAPT or SAHM1 (Fig. 5C,D). Thus, changes in VE-cadherin morphology due to substrate stiffness seem to be independent of Notch signaling, whereas overall VE-cadherin expression is not.

To investigate the role of Notch receptor–ligand binding in increased Notch signaling activity on softer substrates, we performed a trans-endocytosis assay (Shaya et al., 2017). Separate cell populations were transfected with either a Notch1–Citrine fusion plasmid or a doxycycline-controlled Dll4–mCherry fusion plasmid. Dll4 expression was induced by adding doxycycline to the co-culture of both transfected cell populations. Analysis was performed using ImageJ in the areas where the signals of Notch1 receptor and Dll4 ligand overlapped at the cell–cell contacts of a signal-sending and a signal-receiving cell. To quantify trans-endocytosis and examine the stiffness effects, the colocalization of Notch receptor and ligand was analyzed both in the form of intensity analysis of the overlay areas and as a correlation analysis using Pearson's r coefficient. Only the signal overlays of Notch1 extracellular domain and Dll4 that clearly occurred in the signal-sending cell were examined. Trans-endocytosis peaked after 6 h of incubation, as described previously by Shaya et al. (2017). Our results show that trans-endocytosis increased on soft substrates, which was reflected in an increased intensity of the interaction area as well as in a higher colocalization coefficient – the Pearson's r value (Fig. 6A). A transferrin endocytosis assay was performed on the substrates to exclude a stiffness effect on endocytosis in general. The results show that the softer substrates did not enhance general endocytosis, so the increased trans-endocytosis cannot be attributed to general increase in endocytosis (Fig. S4A). Thus, trans-endocytosis seems to selectively exhibit mechanosensitivity in cell–cell contact-dependent receptor binding. The same experiments were performed with the endothelial cell line MCEC-WT for comparison, showing similar results and the same effect: the softer the substrate, the more trans-endocytosis occurred upon cell–cell contact (Fig. S4B). Furthermore, trans-endocytosis was analyzed using a flow cytometry assay wherein the presence of double-positive cells (i.e. cells showing a signal in both the FITC and PI channels) indicates that trans-endocytosis has occurred. The results are in line with those obtained by confocal microscopy, with increased trans-endocytosis observed among cells cultured on softer substrates (Fig. 6B).

Nevertheless, Notch recipient cells also showed increased Notch signaling activity after activation by an rhDll4 coat, where no trans-endocytosis takes place. In order to investigate whether the stiffness effect of the Notch signaling pathway was also detectable in the absence of cell–cell contacts – and thus independently of trans-endocytosis and solely by Notch activation with rhDll4 – a dual luciferase reporter gene assay was performed in nonconfluent HUVECs, in contrast to the confluent cells used in the experiment shown in Fig. 1A. Again, the data were normalized to a transfection control to overcome the effects of both transfection and low cell density on the results. We observed a significant increase in reporter gene activity after Notch activation by rhDll4 on softer substrates (Fig. 6C), where the rhDll4 coat is evenly distributed on all substrates (Fig. S1). In contrast, no change in reporter gene activity was evaluated on the substrates without Notch activation (collagen G coat). Thus, cell–cell contacts contribute to Notch activity but are not necessary for the mechanomodulation of Notch signaling in this setting.

The Notch signaling pathway plays multiple and crucial roles in developmental and pathological processes (Artavanis-Tsakonas et al., 1999; Aster et al., 2017; Bray, 2016). Many of these scenarios are related to changes in the composition and the biomechanical features of the ECM. During angiogenesis, the Notch signaling pathway regulates cell fate decisions such as migration, proliferation and differentiation, which are essential for vascular development and angiogenesis (Merk et al., 2016; Takeshita et al., 2007). Endothelial Notch signaling can be activated by VEGFA, and at the same time VEGFA controls matrix composition, causing local ECM softening or stiffening (Estrach et al., 2011; Stepanova et al., 2021). Previous studies on biophysical aspects of Notch signaling have focused on a pulling force exerted by the bound receptor (Sprinzak and Blacklow, 2021) that is a prerequisite for Notch receptor cleavage. To date, however, little is known about whether the Notch signaling pathway is modulated by substrate stiffness. To address this question, we used an endothelial cell model based on synthetic PDMS substrates with tunable stiffness. The substrates we used cover the physiological stiffness ranges of soft organ tissues such as breast, brain or liver (0.5 kPa); of endothelial tissue or pathological hepatoma tissue (1.5 kPa); of firmer tissue such as skin, but also pathological cervical cancer tissue (70 kPa); and of very firm tissue such as bone structures (plastic) (Cui et al., 2017; Zanotelli and Reinhart-King, 2018).

Interestingly, nuclear NICD localization, transcriptional activity of Notch, and mRNA expression of Notch receptor and target genes were found to increase with decreasing substrate stiffness. This phenomenon occurs irrespective of whether Notch activation was achieved by surface-bound ligand (rhDll4) or Dll4-overexpressing sender cells (Fig. 1).

Endothelial cells sense changes of matrix properties via cell–matrix adhesion proteins, such as integrin β1 (Estrach et al., 2011; Fischer et al., 2009), which also plays a crucial role in regulating cellular adhesion to enable migration and proliferation (Estrach et al., 2011; Matsuo et al., 2021). Since integrins have been previously located both upstream (Rallis et al., 2010) and downstream of Notch in signaling pathways (Hodkinson et al., 2007), we first investigated this issue. For integrin β1, it has already been shown that the overall levels do not change with different substrate stiffnesses, but that the active state of integrin β1 increases on softer substrates (Du et al., 2011). We confirm these observations in our model. Furthermore, our data show that cell–matrix adhesion is directly influenced by the Notch signaling pathway. Notch inhibition by DAPT or SAHM1 treatment significantly reduced the integrin β1 intensity of the activated form, although substrate-dependent activation was further observed (Fig. 2). These results corroborate the findings of Hodkinson et al., which show that Notch1, through NICD cleavage and subsequent R-Ras binding, can induce a conformational change in the surface-bound integrin from a low-affinity state to a high-affinity state, mobilizing the active form (Hodkinson et al., 2007). This clearly positions integrins downstream of Notch. Despite Notch inhibition, a continuous increase in integrin β1 levels on softer substrates can be observed; however, this occurs on a much lower level (Fig. 2). Thus, either pharmacological Notch inhibition was not complete or integrin activity is additionally influenced by substrate stiffness via an alternative pathway.

The role of YAP as a mechanotransducer and the related mechanosensitivity of the YAP–TAZ signaling pathway are well known and studied. Upon YAP activation – for example, by higher ECM rigidly or substrate stiffness – YAP and TAZ are not phosphorylated and can thus be relocated from the cytoplasm to the nucleus, where binding to a transcription factor regulates genes necessary for cell migration and proliferation (Das et al., 2016; Nukuda et al., 2015). Accordingly, we investigated, how the YAP–TAZ pathway relates to stiffness-modulated Notch signaling. As expected, YAP activity decreased with lower stiffness of the substrate (Fig. 3). It has been previously shown that activation of YAP reduces Notch signaling by repression of the ligand Dll4, whereas knockdown of YAP increases Dll4 (Yasuda et al., 2019). However, this mechanism is not sufficient to explain the correlation between substrate stiffness and Notch activity in our model because we examined Notch activation after co-culture with a stably overexpressing cell line. Additionally, increased activity of Notch with lower stiffness was also observed after activation by rhDll4, which is independent of YAP activity. There also seems to be no feedback loop between Notch and YAP translocation, since neither DAPT nor SAHM1 treatment altered stiffness dependence of YAP translocation to the nucleus (Fig. 3). However, we cannot exclude concomitant activation of YAP and Notch on a transcriptional level, which has been described previously (Totaro et al., 2018).

In addition to the effect on cell–matrix adhesions, a stiffness effect has already been established for cell–cell contacts. On stiffer substrates, VE-cadherin junctions are wider and discontinuous, whereas on soft substrates they are narrower but continuous (Bordeleau et al., 2017). We have corroborated this finding by analyzing VE-cadherin patterns through mean branch length evaluation. The VE-cadherin junctions on plastic show a branched, comb-like structure with longer branches, in contrast to the narrower junctions observed on softer substrates, which have shorter branches and appear more distinct and continuous (Fig. 4). A role for cadherins in the regulation of the Notch signaling pathway has previously been described by Kwak et al., who demonstrated that cadherin-based adherens junctions control Notch–γ-secretase interactions (Kwak et al., 2022). In this context, the changes in VE-cadherin junctions on the soft substrates could be related to the increased Notch activity. However, it should be noted that the VE-cadherin patterns on the changing substrate stiffnesses do not affect the intensity level or VE-cadherin trafficking (Fig. 4). Furthermore, Notch signaling is regulated by the cell–cell contact area (Shaya et al., 2017; Totaro et al., 2018), as signaling is directly linked to the contact area of two interacting cells (Shaya et al., 2017). By destabilizing and blocking the cell–cell contacts, we have confirmed that, as expected, Notch activity is reduced. This effect is, however, independent of substrate stiffness (Fig. 5). In this context, it should be noted that the Notch receptor has a Ca²⁺ dependency, which means that the use of EGTA to destabilize the cell–cell contacts could also destabilize the receptor, leading to Notch cleavage and activation (Rand et al., 2000; Tremblay et al., 2013). Thus, this result should be interpreted with care. Blocking the Notch signaling pathway leads to a uniform decline of VE-cadherin on all substrates. The change in junctional patterns due to variation of stiffness is not influenced by DAPT treatment (Fig. 5). These findings argue against an upstream role of VE-cadherin in stiffness regulation of Notch signaling.

Endocytosis of the Notch ligand into the signal-sending cell after binding to the receptor can, on one hand, counteract the accumulation of Notch receptor on the surface of the signal-receiving cell and, on the other hand, serve primarily for receptor activation (Antfolk et al., 2017; Shergill et al., 2012). During trans-endocytosis, the extracellular part of the Notch receptor bound to the Notch ligand is pulled into the ligand-presenting cell, thus releasing the intracellular part of the receptor (the activated form), which can be detached from the membrane in a further cleavage event (Nichols et al., 2007). Our results from the trans-endocytosis assay on the different substrates show that trans-endocytosis is significantly enhanced on soft substrates, which is consistent with the enhanced Notch activity on softer substrates. A stiffness effect on the general endocytosis can be excluded. Taken together, these observations demonstrate a mechanosensitivity of the Notch signaling pathway that may in part be due to trans-endocytosis. Interestingly, however, we find that activation of the Notch signaling pathway in a cell–cell contact-independent manner – by using rhDll4 as a stimulus for subconfluent single cells – still responds to substrate stiffness. Whereas a mechanical aspect of Notch pathway activation following interaction with a ligand-presenting cell – namely the pulling movement of the ligand into the sender cells – has previously been established, the mechanism of mechanosensitivity of the Notch pathway after rhDll4 activation, and thus without trans-endocytosis and pulling movement, is to date unknown and is based on a different mode of action or mechanical force.

These findings might be of importance to better understand the effects of matrix remodeling on endothelial cell behavior and adjustment of their gene expression (Shamloo et al., 2016; Stepanova et al., 2021). Matrix remodeling increases the elasticity of the matrix and thus decreases the stiffness of the ECM (Perfahl et al., 2017; Shamloo et al., 2016), thus triggering a feedback loop. With Notch being one of the main regulators of endothelial cell fate decisions as well as being the key mediator of angiogenesis by controlling tip and stalk cell differentiation (Antfolk et al., 2017; Fernandez-Chacon et al., 2021), our results provide further insights into the consequences of normal and pathological matrix changes for endothelial behavior during angiogenic sprouting in development and maintenance.

Human umbilical vein endothelial cells (HUVECs) were obtained from Promocell (Heidelberg, Germany), Human microvascular endothelial cells (HMECs) were a gift from Dr Guido Jürgenliemk (University of Regensburg, Germany). Cells were cultured in ECGM medium (PELO Biotech, Planegg, Germany) including the supplement kit, 10% heat-inactivated fetal calf serum (FCS; PAA Laboratories GmbH, Pasching, Austria), 250 µg/ml amphotericin B and 10,000 U/ml penicillin-streptomycin (PAN Biotech, Aidenbach, Germany) at 37°C and 5% CO₂ under constant humidity. Cells were maintained until passage six for the experiments.

Mouse cardiac endothelial cells (MCECs) were a gift from Andreas Fischer (DKFZ, Heidelberg, Germany) and were maintained in DMEM medium supplemented with 10% FCS (PAA Laboratories GmbH) and 200 mM L-glutamine (Thermo Fisher Scientific, Waltham, MA, USA). Two MCEC cell lines were used: a wild-type cell line (MCEC-WT) as well as a cell line overexpressing Dll4 coupled to an mCherry reporter (MCEC-Dll4–mCherry). At confluence, cells were split and cultivated with continuous passage.

Unless stated otherwise, all experiments were performed using confluent cultures. HUVEC–MCEC co-cultures were suspended in ECGM medium; MCEC–MCEC co-cultures were suspended in DMEM medium.

The generation of MCEC TetR-TO-Dll4-mCherry (MCEC-Dll4–mCherry) was performed in two steps. First, MCEC-TetR cells were generated by stable transfection of the pcDNA6-TR-Blast plasmid (gifted from the Bjorkman lab, Division of Biology and Biological Engineering, California Institute of Technology, Pasadena, CA, USA) and selection of single clones using 10 µg/ml blasticidin (H-1012-PBS; A.G. Scientific Inc., San Diego, CA, USA). Then, MCEC-TetR cells were transfected with the pcDNA5-TO-hDll4-mCherry plasmid (Sprinzak et al., 2010) and were placed under another selection with 250 µg/ml hygromycin (B-1247-SOL; A.G. Scientific Inc.). Single colonies were picked and tested for fluorescence under conditions with and without 100 µg/ml doxycycline (Sigma-Aldrich, St Louis, MO, USA).

To ensure uniform adhesion, all applied surfaces and substrates were coated with collagen G from bovine skin (10 µg/ml in PBS; 50104, MATRIX BioScience GmbH, Mörlenbach, Germany) before cell seeding. The coating solution was incubated at 37°C, 5% CO₂ for 30 min and then aspirated for cell seeding.

For coating with rhDll4, the rhDll4 (R&D Systems, Minneapolis, MN, USA) was reconstituted at 200 µg/ml with pure PBS and used at working concentrations of 1 µg/ml. The rhDll4 solution was incubated overnight at 4°C and, for temperature equilibration, for 1 h at room temperature on the different substrates before it was aspirated for cell seeding.

PDMS substrates were prepared using the Sylgard™ 184 silicone elastomer kit (Dow Corning, Midland, MI, USA). PDMS base was mixed with a ratio of 10%, 2% or 1.3% curing agent and was transferred to 8-well or 2-well µ-slides from ibidi (Martinsried, Germany). Gels were degassed in a desiccator for 10–15 min and polymerized in a compartment drier at 60°C for 20 h, resulting in substrate stiffnesses of 70 kPa, 1.5 kPa and 0.5 kPa. PDMS substrates were hydrophilized with oxygen plasma at 0.3 mbar for 3 min using the Zepto plasma system (Diener electronic GmbH & Co. KG, Ebhausen, Germany). Stiffnesses were determined after hydrophilization by rheometric measurements using the Modular Compact Rheometer MCR 100 (Physica, Stuttgart, Germany).

Cells were seeded at 250,000 cells/well in 6-well plates on plastic and PDMS substrates coated with collagen G in PBS. After incubation of 24 h and 48 h, cells were stained with Crystal Violet and lysed in sodium citrate buffer. For detection, 100 µl/well of the solution was transferred to a 96-well plate, and the absorbance was measured at 595 nm. All cell proliferation assays were carried out in triplicate.

Cytotoxicity of the Notch inhibitors SAHM1 and DAPT on endothelial cells was assessed using the CellTiter-Blue cell viability assay (Promega, Madison, WI, USA). Cells were seeded at 20,000 cells/well, incubated for 24 h and treated with the inhibitors as indicated for another 24 h. CellTiter-Blue Reagent was added in a ratio of 1:5. Fluorescence intensity was measured using the SpectraFluor Plus plate reader (Tecan, Männedorf, Switzerland) at 550/595 nm. Fluorescence intensity is proportional to the cell number. Cells without treatment were used as negative controls. All experiments were conducted in triplicate.

The TP1-Luc construct was first published by Minoguchi et al. (Minoguchi et al., 1997). pRL-SV40P Renilla was obtained from Addgene (Addgene plasmid 27163; deposited by Ron Prywes). The pcDNA3-hN1-citrine plasmid was constructed from pcdna3-hNECD-G4esn-cit, described in previous work (Sprinzak et al., 2010). The G4esn was replaced with the full length of human Notch1, leaving only the extracellular domain. pcDNA5-TO-hDll4-mCherry has been described previously (Sprinzak et al., 2010). mCitrine-VE-Cadherin-N-10 was obtained from Addgene (Addgene plasmid 56319; deposited by Michael Davidson; RRID: Addgene_56319).

HUVECs were transfected transiently using either the Targefect-HUVEC™ transfection kit (Targeting Systems, El Cajon, CA, USA) or the Cell Line Nucleofector™ Kit V with the program A-034 (Lonza Group AG, Basel, Switzerland) according to manufacturers' protocols. MCECs were transiently transfected with the FuGENE^® Transfection Regent (Promega) following the manufacturer's instructions. Transfected cells were incubated for 24 h before further assays or evaluations were applied. With HUVECs, we routinely reached transfection efficiencies of ∼30%.

Notch-responsive dual luciferase reporter assays were performed 24 h subsequent to co-transfection of endothelial cells with the CSL-binding plasmid TP1-Luc and Renilla luciferase. Firefly luciferase and Renilla luciferase vector levels were applied in a ratio of 10:1. Using the Dual-Luciferase^® Reporter Assay System (Promega) and the Orion II microplate luminometer equipped with Simplicity analysis software (Berthold Technologies, Bad Wildbad, Germany), luciferase levels of triplicates per condition were determined. Firefly luciferase relative light units (RLUs) were normalized to the Renilla luciferase control. This approach compensates for changes of transfection efficiencies between experiments or potential cell death.

Cells grown to 80% confluency (HUVECs or MCEC-WTs) were transiently transfected separately with the plasmids pcDNA3-hN1-citrine and pcDNA5-TO-hDll4-mCherry and incubated on plastic for 24 h at 37°C under 5% CO₂. Cells were then washed, detached and reseeded together in a co-culture ratio of 1:1. Doxycycline was added during reseeding at a concentration of 100 ng/ml for activation of Dll4–mCherry expression and incubated for 6 h at 37°C under 5% CO₂. For visualization of the trans-endocytosis by confocal microscopy, co-cultures were fixed with 4% methanol-free formaldehyde (Thermo Fisher Scientific) in PBS for 10 min. Samples were washed twice with PBS and sealed with FluorSave mounting medium. Quantification was performed by analysis of colocalized areas of Notch extracellular domain and Dll4 in the signal-sending cell (Dll4-expressing cell) by defining regions of interest (ROIs) and measuring their intensity using ImageJ. For the detection of trans-endocytosis by flow cytometry, the cells were detached, collected by centrifugation, and washed and resuspended in ice-cold PBS. Flow cytometry was performed on a BD FACS Canto II (BD Biosciences, Franklin Lakes, NJ, USA). Fluorescence intensity of Notch1–Citrine (plasmid) was analyzed using the FITC channel, and fluorescence intensity of Dll4–mCherry (plasmid) was analyzed using the PI channel. 10,000 cells were analyzed per run. Data were evaluated using the FlowJo 7.6 software.

Cells grown to 100% confluence were washed once with PBS including Ca²⁺ and Mg²⁺ (PBS⁺; 0.25 mM MgCl₂ and 0.5 mM CaCl₂) and incubated for 10 min with 5 µg/ml Transferrin, Alexa Fluor 488 conjugate (Life Technologies, Carlsbad, CA) at 37°C under 5% CO₂. Cells were washed once with room temperature acid-wash medium (150 mM NaCl, 1 mM MgCl₂, 0.125 mM CaCl₂, 0.1 M glycine, pH 2.5) and then fixed with 4% methanol-free formaldehyde (Thermo Fisher Scientific) in PBS for 10 min. Samples were washed twice with PBS and sealed with FluorSave mounting medium. Transferrin uptake was visualized by confocal microscopy.

The following primary antibodies were used in this study: anti-cleaved Notch1 (Val1744; rabbit mAb IgG, 4147; Cell Signaling Technology, Danvers, MA, USA), anti-integrin β1 (rabbit pAb, 4706; Cell Signaling Technology), anti-integrin beta 1 (12G10; mouse mAb IgG₁; ab30394, Abcam, Cambridge, UK), anti-YAP (D8H1X; XP^® rabbit mAb IgG; 14074, Cell Signaling Technology) and anti-VE-Cadherin (rabbit pAb; 2158, Cell Signaling Technology). All primary antibodies were diluted 1:200 for application. The following secondary antibodies were applied in this study: Alexa Fluor 488-conjugated goat anti-mouse IgG (H+L), A-11001; Alexa Fluor 488-conjugated goat anti-rabbit IgG (H+L), A-11008; and Alexa Fluor 647-conjugated chicken anti-rabbit IgG (H+L), A-21443 (all from Thermo Fisher Scientific). All secondary antibodies were diluted 1:400 for application.

DAPT was purchased from Sigma-Aldrich (D5942), dissolved in DMSO to a stock concentration of 10 mM and used at a working concentration of 25 µM. DMSO controls were performed with the appropriate DMSO concentrations. SAHM1 was obtained from R&D Systems (6477/1), dissolved in H₂O to a stock concentration of 1 mg/ml and used at a working concentration of 10 µM. EGTA was purchased from Sigma-Aldrich (E4378) and applied at 5 mM. The VE-cadherin-blocking antibody clone BV6 (MABT134) was obtained from Merck Millipore (Darmstadt, Germany) and was diluted 1:10 for use (50 µg/ml). Hoechst 33342 was purchased from Sigma-Aldrich (B2261), dissolved in PBS and applied at a final concentration of 10 µg/ml. FluorSave™ Reagent was purchased from Merck Millipore (345789; Darmstadt, Germany).

Cells were washed once with PBS⁺ and were fixed with 4% methanol-free formaldehyde (Thermo Fisher Scientific) in PBS for 10 min. Fixation was followed by a brief washing with PBS and cell permeabilization with 0.1% Triton X-100 in PBS for 10 min. After another brief washing with PBS, nonspecific binding sites were blocked with 5% BSA (Roth, Karlsruhe, Germany) in PBS for 60 min. Cells were then incubated with the primary antibody diluted in PBS with 1% BSA (1:200) overnight at 4°C. Next, samples were washed 3×10 min with 1% BSA in PBS, then incubated with the secondary antibody (1:400) and Hoechst 33342 (1:100) for nuclear counterstain, again diluted in PBS with 1% BSA for 1 h at room temperature. Cells were washed again 2×10 min with 1% BSA in PBS and once for 10 min with PBS, and were sealed with FluorSave mounting medium.

Confocal images were taken with a Leica TCS SP8 microscope equipped with an HC PL APO CS2 63×/1.4 oil objective and photomultiplier (PMT) or HyD detectors, using the LAS X core software. In sequential scanning mode, two frames were acquired for every channel with a scanning speed of 400 Hz, and the pinhole size was set to 1.0 airy units. The following excitation laser lines were applied: 405 nm, 488 nm and 647 nm.

HUVECs were transiently transfected with mCitrine-VE-Cadherin-N-10, seeded directly on different substrates (plastic and PDMS) and incubated for 24 h at 37°C under 5% CO₂. The FRAP assay was conducted with the Leica TCS SP8 SMD microscope with the HC PL APO CS2 63×/1.4 NA oil objective and the heating and gas incubation system from OCO Lab (Naples, Italy) ensuring constant 37°C under 5% CO₂ and 80% humidity. Using the LAS X Core Software, the FRAP settings were adjusted to one pre-bleach iteration, 20 bleach iterations, five post-bleach iterations with 30 s intervals and seven with 60 s intervals. Images were taken with a pinhole size adjusted to 1.0 airy units and a scanning speed of 400 Hz. The 488 nm (argon) laser line and the PMT detector were applied.

RNA was isolated from cell samples using the RNeasy Mini Kit (Qiagen, Venlo, Germany) according to the manufacturer's instructions. After determining the yield with a NanoDrop spectrophotometer (Thermo Fisher Scientific), 1250 ng of the mRNA was reverse-transcribed using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Waltham, MA, USA) as described by the manufacturer. RT-qPCR experiments were performed using an Applied Biosystems^® QuantStudio™ 3 Real-Time PCR System (Thermo Fisher Scientific) with standard procedures. SYBR™ Green Master Mix (Thermo Fisher Scientific) was used for cDNA amplification and detection. GAPDH served as a housekeeping gene. The following primers were purchased from Metabion (Planegg, Germany): Dll4 forward, 5′-CTGCGAGAAGAAGTGGACAGG-3′; Dll4 reverse, 5′-ACAGTCGCTGACGTGGAGTTCA-3′; Jag1 forward, 5′-TGCTACAACCGTGCCAGTGACT-3′; Jag1 reverse, 5′-TCAGGTGTGTCGTTGGAAGCCA-3′; Hey1 forward, 5′-TGTCTGAGCTGAGAAGGCTGGT-3′; Hey1 reverse, 5′-TTCAGGTGATCCACGGTCATCTG-3′. Data were evaluated using the ΔΔCT method, as described previously (Fleige et al., 2006).

All confocal images were analyzed using ImageJ version 1.53c (NIH, Bethesda, MD, USA). Only two-dimensional images were evaluated. Pearson's r coefficients were determined using the Coloc2 plugin for ImageJ. Nuclear:cytoplasmic intensity ratios were evaluated with the Intensity Ratio Nuclei Cytoplasm Tool plugin (RRID: SCR_018573) for ImageJ. Expression patterns were determined by segmentation and skeletonizing of the images, as described previously (Arganda-Carreras et al., 2010, 2017).

All data were derived from three independent experiments and are represented as the mean±s.d. In the case of confluent cell layers, entire image sections were analyzed unless the separate analysis of individual cells is described. An image section of confluent cells contained ≥20 cells. Statistical analysis were performed as ordinary one-way or two-way ANOVA with Tukey's multiple comparison tests using GraphPad Prism 9.2.0 (sample size and variance were homogenous).

The peer review history is available online at https://journals.biologists.com/jcs/lookup/doi/10.1242/jcs.260442.reviewer-comments.pdf