## Introduction

Epithelial cells form a dense, stable cell layer lining the outer surface of tissue and organs. Mechanical strength and communication between the cells within a layer is provided by different cell adhesion sites including tight junctions (TJs)1, adherens junctions (AJs)2, desmosomes3, and gap junctions4. Establishment of these intercellular junctions divides polarised cells in apical and basolateral divisions. Owing to their motility and dynamic conditions epithelia are capable of monitoring development5, tissue healing6, and cancer invasion7. To fulfil their role in force transmission between cells lateral cell-cell junctions are connected to the actin cytoskeleton.

Tight junctions are the uppermost cell-cell connection at the lateral cell membrane of polarised epithelia. They form a physical barrier to control the lateral flux of ions, macromolecules, pathogens, and other solutes within the paracellular pathway8. In addition, tight junctions are responsible for the separation of apical and basolateral membrane lipids and proteins9,10. TJs consist of transmembrane proteins such as occludin, claudins, or JAMs and peripheral membrane proteins from the zonula occludens (ZO) or cingulin family8. The actin cytoskeleton is connected to this junctional complex via these zonula occludens proteins, including ZO-1, ZO-2, and ZO-3. As a consequence, loss of ZO proteins influences the actomyosin cortex structure at the tight junctions, for instance, ZO-1/-2 double knock down led to accumulating actin structures as well as an enhanced myosin IIB level at the adherens junctions11. Actin becomes more clustered at the apical cell side11. It was also recently found that epithelial tension and effective viscosity are increased in ZO-1/-2 lacking cells12.

While the molecular details of cell-cell junctions are well-known, their role for cellular mechanics on the mesoscale is largely unexplored. Trepat and co-workers recently demonstrated that the formation of epithelial cracks during cell sheet dilatation is caused by tissue stretching independent of epithelial tension resulting from a transient pressure build-up in the substrate during stretching and compression17. Essentially, the authors found that the cracks do not originate from detachment of the cortex from the membrane but are due to dissolution of adherens junctions. Even tight junctions remain mostly intact in these cracks.

Espinosa and co-workers used micropillar arrays and AFM indentation experiments on single cells and cell pairs to investigate the impact of the desmoplakin–intermediate filament linkage on cell mechanics18. They found that desmosome–intermediate filament networks are important contributors to cell stiffness in cell pairs and sheets underlining the importance of cell-cell contacts for cellular mechanics. Therefore, depending on the cell type different mechanisms might exist to regulate tension.

In the present study, we examined how epithelial cells regulate their mechanical behaviour when cell-cell junctions are impaired. We employed atomic force microscopy (AFM) which is proven to be a versatile tool not only for imaging but also to assess local mechanical properties of adherent cells19,20. Local mechanical properties are an important readout how cells respond to environmental changes21,22,23,24. We stopped tight junction formation by ZO-1 silencing and interfered with the integrity of adherens junctions by a chemical stimulus. Mechanical measurements were analysed using a minimal mechanical model based on the assumption that the restoring force to deformation originates mainly from cortical tension. Our data indicate that the impact of tight junctions is of minor importance for the mechanics of epithelial cells from a monolayer, while their morphology is changed considerably. Loss of adherens junctions, however, changes tension and area compressibility significantly, while the morphology of the cell layer is only slightly affected. Therefore, our study suggests that TJs and AJs impact cellular mechanics in different, opposite ways highlighting their role in establishing a tight but also flexible barrier tissue.

## Results

### Tight junctions are responsible for a complete, dense cell monolayer

Tight junctions form the uppermost cell-cell connections in epithelial cells. This junctional complex consists of the transmembrane proteins occludin and various claudins. On the inner cellular site, it binds to zonula occludens, which establishes the contact to the F-actin cytoskeleton8. We used short interference RNA (siRNA) against zonula occudens-1 (ZO-1) in order to examine the role of tight junctions for the mechanical integrity of epithelial cells. We chose to knock out ZO-1 since it is known to bind to nearly all relevant proteins of the tight junctional complex, especially the perijunctional actomyosin network. It was also shown by Anderson and co-workers that depletion of ZO-1 preserves the transepithelial resistance but generates appreciable changes in cell morphology and organisation of the actomyosin cortex9.

Phase contrast images of confluent MDCK II cells show a tight, densely packed cell monolayer (Fig. 1A1). Fluorescence images of ZO-1 stained cells show that tight junctions built up a gapless mesh surrounding each cell (Fig. 1A2). After applying siRNA against ZO-1 the cell layer still forms but is less uniform. Cells appear to be separated from each other and cell borders are brighter in phase contrast images (Fig. 1B1). ZO-1 is blocked with high efficiency by siRNA as nearly all cells lose their fluorescence signal (Fig. 1B2). This spatial separation between cells is further confirmed by AFM-imaging of ZO-1 deficient cells (Fig. 2). A continuous connection appears in AFM images for a fully established cell monolayer of untreated cells (Fig. 2A–E). In contrast, for ZO-1 deficient cells we found that adjacent cells are no longer connected to each other and only a few finger-like connections are visible (Fig. 2F–J).

Since ZO-1 connects the tight junctional complex with the actin cytoskeleton, the actin distribution of cells after ZO-1 depletion was also analysed by fluorescence microscopy (Fig. 3). On the apical side small point-like structures are found for untreated cells (Fig. 3B1), reflecting the actin-filled microvilli of polarized MDCK II cells25. At the basal side stress fibres are visible (Fig. 3D1). For successfully ZO-1 silenced cells the actin distribution is more clustered on the apical side (Fig. 3B2), while on the basal side stress fibres disappear (Fig. 3D2). At the central focal plane actin fences untreated MDCK II cells, whereas a distinct cell-cell separation becomes visible for ZO-1 deficient cells and the actin signal appears blurry (Fig. 3C).

Mechanical properties are inferred from AFM-based indentation experiments that provide information about cortical strength, membrane tension and excess membrane area. We indented siRNA treated cells with an AFM tip in order to analyse the impact of ZO-1 on these properties. Interestingly, although the cells are visibly separated from each other, the mechanical parameters are only slightly altered by ZO-1 depletion (Fig. 4). The biggest effect is an increased cortical strength after weakening of tight junctions. The average overall tension of untreated cells is found to be $${T}_{0}=(0.275\pm 0.010)\,\text{mN}/{\rm{m}}$$ $$({\rm{median}}\pm {\rm{SEM}})$$. Cells with compromised tight junctions show a significantly increased overall tension of $${T}_{0}=(0.334\pm 0.009)\,\mathrm{mN}/{\rm{m}}$$ (Fig. 4A). The apparent area compressibility modulus is also higher for cells lacking ZO-1 ($${\hat{K}}_{{\rm{A}}}=(0.228\pm 0.007)\,{\mathrm{nN}/\mu {\rm{m}}}^{{\rm{3}}}$$, control: $${\hat{K}}_{{\rm{A}}}=(0.153\pm 0.008)\,{\mathrm{nN}/\mu {\rm{m}}}^{{\rm{3}}}$$) (Fig. 4B). The membrane tension Tt, which is a measure for the strength of membrane-cytoskeleton attachment sites, is found to be $$(0.161\pm 0.004)\,\mathrm{mN}/{\rm{m}}$$ for untreated cells. Nearly the same value is found for ZO-1 lacking cells ($${T}_{{\rm{t}}}=(0.164\pm 0.004)\,\mathrm{mN}/{\rm{m}}$$) (Fig. 4C).

In order to exclude mechanical alterations due to the transfection procedure control measurements using non-targeting siRNA are shown in the supplementary information (suppl. Fig. S1)26. Applying the transfection protocol shifts mechanical parameters to slightly lower values, i.e. in the opposite direction.

In summary, ZO-1 is extremely important for the formation of a dense cell layer, which shields underlying cells. The morphology of ZO-1 lacking cells is changed tremendously but the overall stiffness of the cells increases only slightly compared to well-connected cells.

### Adherens junctions influence topography and mechanics of cells

Phase contrast imaging shows no changes in the cellular structure during the entire observation period (Fig. 5 row A). In contrast to ZO-1 silenced cells, the cell monolayer stays intact after incubation with DTT-containing medium, also tight junctions are not affected by DTT treatment (Fig. 5 row C). The distribution of actin on the apical cell side remains unchanged (Fig. 5 row D), small point-like structures remain. Interestingly, in the central focal plane (Fig. 5 row E) (at the same z-position, where E-cadherin imaging is carried out) the actin fence, found as a solid ring in control cells (Fig. 5E1), is perturbed after 3 h of DTT treatment (Fig. 5E2). 3 h after removal of the drug the cells do not restore the original actin distribution (Fig. 5E3). Only after keeping the cells free from DTT for altogether 6 h, the actin ring reshapes and appears as homogeneous as in untreated cells (Fig. 5E4). In contrast, stress fibre formation at the basal cell side is not significantly affected by disruption of the adherens junctions (Fig. 5 row F).

The intracellular E-cadherin is known to be connected to the actin cytoskeleton via catenins and we found that the actin distribution at the lateral cell side is severely perturbed by DTT treatment (Fig. 5E1, E2). Therefore, we also investigated the distribution of β-catenin (Fig. 6). Control cells display a ring of β-catenin in the same focal plane, where E-cadherin is located (Fig. 6B1). After dissolution of the E-cadherin-E-cadherin bonds the β-catenin ring is also destroyed (Fig. 6A2, B2). The recovery of the β-catenin structure occurs within the same time as the E-cadherin reformation. 3 h after DTT removal the distributions of both, E-cadherin and β-catenin are still irregular (Fig. 6A3, B3). However, after keeping the cells free from DTT for 6 h, they finally manage to restore the distribution of E-cadherin (Fig. 6A4), β-catenin (Fig. 6B4) and F-actin (Fig. 5E4).

AFM imaging confirms the presence of an intact cell monolayer during the whole experimental procedure (Fig. 7). However, a detailed inspection of the topography revealed that the morphology of the connected cells is altered by DTT incubation. Control cells display a typical height of the apical cap of $$(2.0\pm 0.1)\,\mu {\rm{m}}$$ (Fig. 7A). After exposure of the cells to 10 mm DTT for 3 h, a substantially elevated cap of $$(2.8\pm 0.2)\,\mu {\rm{m}}$$ is found (Fig. 7B). After removal of DTT and a recovery time of 3 h cells start to restore their initial topography (cap height: $$(2.5\pm 0.2)\,\mu {\rm{m}}$$, Fig. 7C). 6 h after DTT removal the height of the apical cap is completely restored to $$(2.0\pm 0.2)\,\mu {\rm{m}}$$ (Fig. 7D). Considering that the adherens junctions ‘outstretch’ the cells by pulling at adjacent cells, bulging of the cells after compromising the junctions and therefore removing the tension between them is expected and a strong hint to an altered cellular mechanics.

Next, we asked for the impact of DTT treatment and recovery from the treatment on the mechanical properties of the cells. Fundamental changes of the mechanical parameters are observed during the exposure to DTT and recovery from the drug treatment. Destroyed adherens junctions lead to a substantial drop in the overall tension from $${T}_{0}=(0.275\pm 0.010)\,\text{mN}/{\rm{m}}$$ down to $${T}_{0}=(0.164\pm 0.005)\,\mathrm{mN}/{\rm{m}}$$ after 3 h of DTT treatment. 3 h after DTT removal T0 is found to be $$(0.315\pm 0.010)\,\mathrm{mN}/{\rm{m}}$$, even higher than the initial value. Later, after a recovery time of 6 h the initial value is eventually restored ($${T}_{0}=(0.265\pm 0.013)\,\mathrm{mN}/{\rm{m}}$$) (Fig. 4A). DTT treatment also causes a halving of the apparent area compressibility modulus ($${\hat{K}}_{{\rm{A}},{\rm{c}}{\rm{o}}{\rm{n}}{\rm{t}}{\rm{r}}{\rm{o}}{\rm{l}}}=(0.153\pm 0.008)\,{{\rm{n}}{\rm{N}}/\mu {\rm{m}}}^{3}$$, $${\hat{K}}_{{\rm{A}},3{\rm{h}}{\rm{D}}{\rm{T}}{\rm{T}}}=(0.083\pm 0.006)$$ $${\mathrm{nN}/\mu {\rm{m}}}^{{\rm{3}}}$$. When the drug is removed, cells manage to restore this value within 3 h also with a slight overshoot. It is found to be $${\hat{K}}_{{\rm{A}}}=(0.194\pm 0.024)\,{\mathrm{nN}/\mu {\rm{m}}}^{{\rm{3}}}$$. After 6 h of recovery from the DTT treatment, an apparent area compressibility modulus comparable to untreated cells is detected $$({\hat{K}}_{{\rm{A}}}=(0.187\pm 0.078)\,{\mathrm{nN}/\mu {\rm{m}}}^{{\rm{3}}})$$ (Fig. 4B).

DTT treated cells show a lowered membrane tension ($${T}_{{\rm{t}}}=(0.121\pm 0.036)\,\mathrm{mN}/{\rm{m}}$$) compared to untreated cells $$({{\rm{T}}}_{{\rm{t}},{\rm{c}}{\rm{o}}{\rm{n}}{\rm{t}}{\rm{r}}{\rm{o}}{\rm{l}}}=(0.161\pm 0.003)\,{\rm{m}}{\rm{N}}/{\rm{m}})$$. When the drug is removed for 3 h, Tt is found to be $$(0.227\pm 0.010)\,\mathrm{mN}/{\rm{m}}$$. After 6 h of recovery, a homeostasis occurs. The membrane tension drops again down to $${T}_{{\rm{t}}}=(0.188\pm 0.006)\,\mathrm{mN}/{\rm{m}}$$ (Fig. 4C) close to the initial value.

In summary, we found that opening of adherens junctions by DTT leads to a considerable softening of the individual cells. Both, cortical and membrane tension drop and either the bending modulus of the composite shell is reduced or the excess membrane area is enlarged depending on the interpretation of the apparent area compressibility modulus. Interestingly, the cells entirely recover from this treatment after keeping them for 6 h in the absence of DTT. This holds true both for structure and mechanics. Cells reach their initial values by going through a state of higher stiffness as initially found for control cells (overshoot).

## Discussion

In the present study we examined how different cell-cell junctions contribute to the mechanical properties of individual epithelial cells and how the cells maintain tension homeostasis after selectively destroying the junctions.

It is clear that cell-cell junctions are important for providing efficient and selective barriers against the environment. This also includes a stable mechanical connection between adjacent cells that permits the cell layer to respond to external mechanical stress without losing its integrity. There are seminal studies addressing questions like, which forces are necessary within cells to establish strong cell-cell binding complexes29,30, how different types of junctions influence each other31,32, and how forces are transmitted within a cell layer33,34,35,36. However, it is still not entirely clear if and to what extent cell-cell junctions impact the mechanical behaviour of individual cells of a confluent monolayer.

When one of the major players of tight junctions, ZO-1, is depleted, we found that cells are visibly separated from each other. AFM imaging reveals that only a few finger-like connections between the cells remain (Fig. 2). Surprisingly, however, these significant alterations of the cell morphology have only a small effect on the mechanical behaviour of the individual cells that are still connected via adherens junctions. Slightly elevated overall tension and area compressibility moduli (Fig. 4) can be explained by an altered actin cytoskeleton (Fig. 3B2) with coarser structures on the apical side. This is in good agreement with results from Fanning et al. They also found larger aggregates of actin and accumulating myosin at the cell apex in ZO-1/-2 depleted MDCK II cells11. An enhanced myosin II level and higher actomyosin contractility may also contribute to higher tension values. Contrary to Fanning et al., however, we did not find a stronger actin belt at the lateral apical junctional complex region in response to ZO-1 depletion (Fig. 3C2). This may be due to the knock down of only ZO-1 and thereby explain the only moderately enhanced tension in our case. Compared to their work, we also found less pronounced stress fibre formation, when the cells are separated from each other (Fig. 3D2). Ikenouchi et al. demonstrated that polarisation of EpH4 epithelial cells is slowed down when ZO-1/-2 is lacking. Actin on the lateral cell side is separated between individual cells37. We found the same actin distribution in our experiments with MDCK II cells.

In our experiments cells undergo substantial morphological changes upon ZO-1 depletion (vide supra). The membrane-cytoskeleton attachment describing membrane tension Tt, however, remains largely at the same level that we found for control cells. This supports the work from Fanning et al. who found that cell polarity and ezrin distribution is unaffected in ZO-1/-2 double knockdown cells.

In contrast to ZO-1 depletion, E-cadherin interference has dramatic consequences for the mechanical properties of MDCK II cells, while the impact on the morphology of the cell monolayer is small. Capaldo and Macara reported a swelling of the apical cap in E-cadherin silenced MDCK T23 cells38. Our AFM studies confirm this result, the removal of adherens junctions results in a substantial rounding of the cells indicative of an interfacial tension release. Considering that contractile cortical tension is balanced by the presence of cell-cell contacts it is expected that cells bulge into a more roundish shape after losing the cell-cell counter force. Interestingly, Capaldo and Macara report that β-catenin remains at the cell-cell-boundaries, whereas we found that the protein spreads within the whole cell after DTT treatment (Fig. 6B2). Changes in this protein distribution provoke a reduction of the overall tension and the area compressibility modulus since actin is no longer connected to the lateral cell membrane. A tension generating actin belt is partially degraded, as shown by confocal fluorescence microscopy at the central focal plane (Fig. 5E2). Studies on Xenopus revealed that cadherins influence the actin architecture39. Thus, the mechanical stability provided by the actin belt cannot be maintained after E-cadherin disruption. Additionally, we were able to show that recovery of the original E-cadherin distribution after DTT removal is accompanied by a recovery of the initial cell topography (Fig. 7). Interestingly, after only 3 h of recovery both mechanical parameters, $${T}_{0}$$ and $${\hat{K}}_{{\rm{A}}}$$, return to even higher values as found for untreated cells, which might be due to an altered geometry. At the same time β-catenin is still distributed within the whole cell interior. This is a strong hint that other regulatory mechanisms can rescue the mechanical integrity without participation of β-catenin. At this time point E-cadherin is already partly located at the cell-cell junctions. Other studies put the cadherin-catenin-actin complex as the only cadherin-cytoskeleton connection way into question40. However, there is evidence in the literature that force generation within the cell is necessary to establish cadherin mediated cell-cell contacts41,42,43. In our experiments cells start to rebuild adherens junctions after DTT removal through a temporary increase (overshoot) of the measured mechanical parameters, essentially forming stiffer cells than initially present. During the next 3 h keeping the cells in the absence of DTT, this effect is balanced out through a tug-of-war mechanism. Eventually the cells restore E-cadherin mediated cell-cell junctions (Figs 5B4 and 6A4). β-catenin is again concentrated at the E-cadherin enriched adherens junctions in accordance with junction formation described in textbooks40.

After AJ degradation we found a small decrease in membrane tension followed by an enhanced Tt level during E-cadherin reassembly. It is indisputable that membrane tension is largely governed by the attachment of the cytoskeleton to the plasma membrane44. In studies with C. elegans it was shown that the apical membrane-cytoskeleton linker ezrin homologue ERM-1 is necessary for apical junction formation45. In order to re-establish adherens junctions after DTT removal MDCK II cells might therefore enhance their ezrin level. This linker protein is then available for enhancement of the apical membrane-cytoskeleton connection leading to higher Tt values26. More pinning points to the membrane also increase the level of cross-linking of the actomyosin network that also allows the cells to adopt a higher cortical tension through motor activity. After 6 h, the cells eventually manage to return to a normal AJ formation reflected in a recovery of all three mechanical parameters (Fig. 4).

Tension homeostasis of epithelial cells depends on many complex mechanisms in single cells and the whole tissue. Our study shows that intact lateral cell junctions are not solely responsible for control of mechanical integrity. Tight junctions enable a sealed epithelial monolayer with a controlled perijunctional flux. The impact on mechanical stability of single cells of cell sheets is, however, small. The presence of intact tight junctions is synonymous with softer cells, maybe by decoupling the apical from the basolateral part of the polarised cell. In contrast, loss of functional adherens junctions results in a severe softening of cells that is, however, fully reversible.

## Material and Methods

### Cell culture

Madin-Darby canine kidney cells (strain II, MDCK II; Health Protection Agency, Salisbury, UK) were maintained in minimum essential medium (MEM) with Earle’s salts and 2.2 g/L NaHCO3 supplemented with 4 mm l-glutamine and 10% foetal calf serum at 37 °C in a 5% CO2 humidified incubator. Cells were grown to 70–90% confluency, released from culture flasks using trypsin/EDTA (0.05%/0.02%) and subcultured three times a week. Medium additionally contained penicillin (0.2 mg/mL), streptomycin (0.2 mg/mL), Amphotericin B (0.5 mg/mL) and HEPES (15 μm) during experimentation.

Cells were seeded on Petri dishes (μ-Dish 35 mm, low Grid-500; ibidi, Martinsried, Germany) and grown to confluency. 1,4-dithiothreitol was dissolved in antibiotics and fungicide containing cell culture medium with a drug concentration of 10 mm. Cells were incubated with the drug containing medium for the indicated time at 37 °C.

### ZO-1 silencing

MDCK II cells were seeded on Petri dishes (μ-Dish 35 mm, low Grid-500; ibidi, Martinsried, Germany) and grown to 50% confluence. Pooled siRNA targeting zonula occludens-1 sequences GCAAAGACAUUGAUAGAAA, GAGAAGAAGUGACCAUAUU, CAAAAGAUCUGCAUCCUUA, and CCUGAACCAUGGUCGGAAA (siGENOME SMARTpool human TJP1 siRNA; Thermo Fisher Scientific, Lafayette, USA) were transfected using Lipofectamine® RNAiMAX transfection reagent (Life Technologies, Carlsbad, USA) according to the manufacturer’s instructions. Experiments were performed 72 h after incubation with siRNA.

For control measurements non-targeting siRNA sequences UAAGGCUAUGAAGAGAUAC, AUGUAUUGGCCUGUAUUAG, AUGAACGUGAAUUGCUCAA, UGGUUUACAUGUCGACUAA (siGENOME non-targeting siRNA Pool #2, GE Healthcare, Lafayette, USA) were applied.

### Cell labelling

Cells were grown on Petri dishes (μ-Dish 35 mm, low Grid-500; ibidi, Martinsried, Germany), treated as desired, and fixed with 4% paraformaldehyde in phosphate buffered saline without calcium and magnesium (PBS) for 20 min. To block unspecific binding sites and to permeabilise the plasma membrane, cells were treated with blocking buffer (5% (w/v) bovine serum albumin (BSA), 0.3% (v/v) Triton X-100 in PBS) for 30 min. For ZO-1 labelling the fluorescently labelled primary antibody (ZO-1 Monoclonal Antibody (ZO1-1A12), Alexa Fluor 488; Thermo Fisher Scientific, Schwerte, Germany) was diluted with dilution buffer (1% (w/v) BSA, 0.3% (v/v) Triton X-100 in PBS) to a concentration of 5 μg/mL, and cells were incubated for 1 h at room temperature. E-cadherin was labelled by incubation of the permeabilised cells with a fluorescently labelled antibody (Alexa Fluor 488 or Alexa Fluor 555 Mouse anti-E-Cadherin (Clone 36), c = 5 µg/mL; BD Biosciences, Heidelberg, Germany). β-catenin staining was performed incubating permeabilised cells with mouse anti-β-catenin antibody (c = 5 µg/mL in dilution buffer; BD Biosciences, Heidelberg, Germany) for 1 h. The secondary antibody (Alexa Fluor 546-conjugated goat anti-mouse IgG; Life Technologies, Carlsbad, USA) was diluted with dilution buffer down to a concentration of 5 μg/mL and incubated with the cells for 45 min. F-actin was labelled by incubating cells with 165 nm Alexa Fluor 546-phalloidin or Alexa Fluor 647-phalloidin (Life Technologies, Carlsbad, USA) for 45 min. Between every labelling step, cells were rinsed three times with PBS for 5 min each on a vibratory plate (75 rpm).

Fluorescence imaging was carried out using a confocal laser scanning microscope (CLSM; FluoView1200; Olympus, Tokyo, Japan) using an 100x oil immersion objective (UPLFLN100xO2PH, NA = 1.3; Olympus, Tokyo, Japan). Fluorescence images are shown in false colours.

### AFM imaging

Cells were prepared as desired and fixed by incubation with 2.5% (v/v) glutaraldehyde in PBS for 20 min. Cell imaging was carried out using a Nanowizard® II, III or 4 atomic force microscope (JPK Instruments, Berlin, Germany) equipped with silicon nitride cantilevers (MLCT, k = 10 mN/m, Bruker AFM Probes, Camarillo, USA). The AFM was mounted on an inverse optical microscope (IX 81 or IX83, Olympus, Tokyo, Japan) to enable phase contrast microscopy during imaging. Cells were imaged in contact mode in PBS at room temperature with a scan rate of 0.5 Hz. Image processing was performed using software provided by the AFM manufacturer.

### AFM force indentation experiments

Force indentation cycles were recorded on the above-mentioned setup. Cantilevers (MLCT, k = 10 mN/m, Bruker AFM Probes, Camarillo, USA) were plasma cleaned (Argon, 30 s) and incubated with FITC-conjugated Concanavalin A (c = 2.5 mg/mL in PBS; Sigma-Aldrich, Steinheim, Germany) for 1.5 h. This enables strong interaction between the cantilever and the plasma membrane during indentation and results in pulling out membrane tethers upon retraction of the indenter. The exact spring constant of each cantilever was determined using the thermal noise method46. Petri dishes containing samples of living cells were mounted on an inverted microscope and kept at 37 °C during experimentation. Cells were indented up to a force of 1 nN. After a dwell time of 0.5 s the indenter was retracted from the plasma membrane and membrane nanotubes were frequently pulled out of the plasma membrane. The indentation and retraction speed was set to 2 µm/s. These force indentation experiments were performed while scanning laterally across the sample, referred to as force mapping25.

### Data analysis of force indentation and retraction experiments

Force indentation curves were analysed as described in our previous studies34. Briefly, Force (F)-indentation (δ) curves were fitted by a polynomial starting at the contact point, where the indenter touches the cell membrane:

$$F(\delta )=a\cdot \delta +b\cdot {\delta }^{3}$$

Prefactors a and b were used to calculate the overall tension T0 and apparent area compressibility modulus $${\hat{K}}_{{\rm{A}}}$$ taking the geometrical properties of the indenter with the half-opening angle $$\varphi$$ into account:

$${T}_{0}=\frac{a}{2\pi \,\cos ({\rm{\Theta }})},$$

and

$${\hat{K}}_{{\rm{A}}}=\frac{b}{2{\pi }^{2}(1-\,\cos ({\rm{\Theta }}))}\,\tan ({\rm{\Theta }}),$$

where $${\rm{\Theta }}=\frac{\pi }{2}-\varphi$$.

The advantage of this simplified model over more sophisticated ones, where the geometry of the cell is considered47, is its analytical tractability and that it only considers the area increase due to the presence of the indenter. It has been shown by Discher and co-workers that a polynomial fit (linear and cubic term) serves well to describe the experimental data48.

Membrane tension Tt was calculated from force steps (Ftether) according to:

$${T}_{{\rm{t}}}=\frac{{F}_{{\rm{tether}}}^{2}}{8{\pi }^{2}\kappa }$$

$$\kappa =2.7\cdot {10}^{-27}{\rm{J}}$$ was chosen25,49,50.

An example of a force indentation and retraction cycle and the fitting procedure is shown in the supplementary information (suppl. Fig. S2).