Surface morphology and mechanical properties of MDCK monolayers by atomic force microscopy

The atomic force microscope (AFM; Binnig et al., 1986) generates image contrast by a mechanism completely different from other forms of microscopy, which will lead to new information about cellular and biological structure (Engel, 1991; Hoh and Hansma 1992; Radmacher et al., 1992; Hansma and Hoh, 1994). The direct interaction of the AFM tip with the sample initially prompted concern that soft biological material might be damaged during imaging. However, it has recently become evident that cells imaged with the AFM remain viable during imaging. This has allowed the direct imaging of intracellular structures, such as the nucleus and actin stress fibers, and of dynamic processes in living cells (Butt et al., 1991; Haeberle et al., 1991; Henderson et al., 1992; Hoerber et al., 1992; Chang et al., 1993; Fritz et al., 1993; Kasas et al., 1993; Parpura et al., 1993). In addition to imaging, it is also possible to measure the local elastic and viscous (i.e. mechanical) properties of a cell by monitoring the response of the cantilever as the tip is pushed against the plasma membrane (Fritz et al., 1994; Weisenhorn et al., 1993).

The utility of the AFM varies considerably depending on the cell type. Some cells are intrinsically difficult to image because of poor adhesion while others reveal very little structure. Here we describe the surface morphology and mechanical properties of Madin-Darby canine kidney (MDCK) cells and R5 cells, a rastransformant derived from MDCK (Schoenenberger et al., 1991), by AFM. MDCK cells are a widely used system for studying epithelial polarization (see Simons and Fuller, 1985; Matlin and Caplan, 1992). They form continuous monolayers possessing an apical free surface and basolateral membrane domains engaged in cell-cell and cell-substratum interactions. More recently, MDCK cells have been employed as a model system to study the breakdown of polarity associated with carcinogenesis (Schoenenberger and Matlin, 1991).

MDCK strain II epithelial cells (low transmonolayer resistance; Matlin and Simons, 1983) were subcloned by limiting dilution and a clone was chosen on the basis of its cuboidal morphology, growth rate, ability to form domes, and transepithelial resistance. The R5 transformed cell line was generated by retroviral infection of the MDCK subclone with the viral Kirsten rasoncogene (Schoenenberger et al., 1991). For routine culture on plastic substrata (Falcon), MDCK and R5 cell lines were grown in Eagle’s minimal essential medium supplemented with Earle’s salts (Gibco BRL, Basel), 5% fetal bovine serum, 2 mM L-glutamine, and 10 mM HEPES, pH 7.3, at 37°C in a 95% air/5% CO₂humidified incubator. Confluent cells were released from the plastic using trypsin-EDTA (Gibco BRL) and passaged at appropriate dilutions. Routine and experimental cultures were fed with fresh medium every other day.

To prepare samples for AFM imaging, round glass coverslips (12 mm) were glued onto magnetic steel stubs (needed for mounting samples in the AFM), autoclaved and placed in tissue culture dishes. Confluent MDCK cells grown on plastic were trypsinized, counted and adjusted to 6×10⁵cells/ml. A 100 μl sample of the cell suspension was plated on a mounted coverslip and spread to cover the entire surface. Cells were allowed to attach for several hours at 37°C in 95% air/5% CO₂before adding medium to the tissue culture dish. On day 4 after plating, monolayers of MDCK cells had nearly reached saturation density and were used for imaging. R5 cells were plated at low densities (<10⁴cells/coverslip) and imaged 2 days after plating. All imaging was performed with cells at passages <15.

For imaging MDCK cells on permeable substrata, cells were cultured on polycarbonate filter supports (Millipore PCF, 0.4 μm pore size, 12 mm diameter, Millipore Corp., Bedford, MA). To mount the filters on steel stubs, the walls of the culture plate inserts were trimmed to a maximum height of 1.5 mm. The inserts were glued upside down onto steel stubs and the wall was perforated to allow access of culture medium from the bottom. The mounted filters were sterilized by UV irradiation. To plate cells on these filters, the bottom of the filter chamber was filled with medium and 100 μl of cell suspension was spread out on the filter. After cell attachment, medium was added to the tissue culture dish and the cells were cultured as above before imaging.

A Nanoscope III atomic force microscope (Digital Instruments Inc., Santa Barbara, CA) was used for all AFM imaging. This microscope was equipped with a large-area scanner (J type) with a maximum xyscan range of 160 μm × 160 μm and a zrange of 5 μm. Cantilevers were standard microfabricated (Albrecht et al., 1990) V-shaped, 200 μm long with 40 μm wide legs, silicon nitride Nanoprobes (Digital Instruments Inc.). A standard fluid cell was used without the o-ring, similar to the ‘trapped drop’ approach used in the first AFM imaging in liquids (Drake et al., 1989). Imaging of cells was performed in PBS containing 1 mM CaCl₂and 0.5 mM MgCl₂, in most instances at room temperature. Alternatively, imaging was performed in a 37°C environment (warm room). To compensate for evaporation from the unsealed fluid cell, buffer was exchanged at regular intervals. Microscope parameters were similar to those previously described (Hoh et al., 1993). The most critical imaging parameters were imaging force and scan rate. Scan rates were kept at 50-500 μm/s, with a typical image acquisition time of 50-100 seconds, and the imaging force was maintained just high enough to achieve good contrast. Error signal images were collected by monitoring the deflection signal with the gains set as high as possible without the feedback loop oscillating (Putman et al., 1992c). This produces an effect similar to a high-pass filter in which quantitative height information is lost. Constant force images with the true height (zpiezo position) were collected simultaneously and used for all height measurements. All images shown are raw data from the AFM with no processing.

Force curves were collected by monitoring cantilever deflection while ramping the piezo scanner (i.e. sample) in zwith the xyscanning disabled, resulting in a plot of force versus sample position (Weisenhorn et al., 1989; Burnham and Colton, 1989). The deflection sensitivity of a cantilever was calibrated on a bare glass coverslip prior to imaging cells. Force curves were used to measure the elastic properties of the cells and determine imaging forces. The imaging forces reported here represent the force normal to the surface. Spring constants were calibrated from the measured resonant frequency of the 200× 40 μm silicon nitride cantilevers. Calibration curves of resonant frequency versus spring constant were used to derive a value of 0.06 N/m, as described by Cleveland et al. (1993).

Stiffness curves, in which the stiffness Sis defined as S=dF/dD, where Fis force and Dis indentation depth, were generated by first making an indentation curve as described by Weisenhorn et al. (1993), in which the deflection of the cantilever is plotted versus the indentation depth into the cell. The first derivative of this indentation curve, which shows the slope of the indentation curve at each point, was then calculated to produce a stiffness curve.

Cells were grown on unmounted glass coverslips as described for AFM samples. For fixation, coverslips were rinsed twice with phosphate buffered saline (PBS, Dulbecco formulation), followed by 20 mM HEPES, 100 mM KCl, 2 mM MgCl₂, pH 7.0 (buffer A), and then incubated for 4 hours with 2.5% glutaraldehyde in buffer A at 4°C. After extensive rinsing with buffer A, coverslips were incubated with 1% OsO₄in buffer A for 1 hour at 4°C, rinsed with buffer, and dehydrated through increasing concentrations of ethanol (50,70,90,100%), ethanol/acetone (50/50, v/v), and acetone. Samples were critical point dried and sputter coated with ∼15 nm gold. The field emission scanning electron microscope (Hitachi S-800) was operated at 20 keV. Samples shown are tilted at 30°.

AFM images of living MDCK cells cultured on glass reveal the apical surface of the monolayer. These images were acquired in ∼60 seconds (similar for all images presented). As shown in Fig. 1A, height variations across the surface of the monolayer are generally 1-3 μm, though occasionally areas that extend beyond the zrange (5 μm) of our scanner are encountered. Some images reveal a large structure in the center of many cells that appears to be the nucleus. The nucleus is particularly prominent in thin areas of the monolayer, where the total thickness of individual cells is often <2 μm. The corresponding error signal image, which is the cantilever deflection while imaging with the feedback loop on, in effect acting as a high-pass filter, shows the cell boundaries that clearly delineate neighboring cells (Fig. 1B). This boundary appears as a ridge about 10-100 nm high. The thickness of the monolayer was assessed by increasing the imaging force and thus pressing through the cell to the glass surface with the tip, in an area adjacent to the nucleus. The overall organization of the monolayer was extremely stable and showed only small lateral deformations in the position or boundaries of the cells. This is determined by comparing images that were recorded as the tip was moving left to right (trace) with images from the right to left movement of the tip (retrace). Cells were stable over time and could be imaged for several hours with no loss of image quality. MDCK cells grown on permeable filters had a surface morphology comparable to those grown on glass (data not shown). However, on filters the feedback loop was extremely prone to oscillating. Gains of a factor of 10 lower than normal imaging conditions had to be used, and scan rates were reduced to 10-50 μm/s, resulting in image acquisition times of up to 50 minutes.

Evidence of cell viability is provided by the dynamic behavior of the cells over time. Cellular extensions move over the time course of hours and the healing of an intentional wound in the plasma membrane made by the tip still occurs after imaging cells for several hours (unpublished observation). The viability of cells was also confirmed by returning individual coverslips after imaging to culture conditions for another 12-24 hours, and staining with trypan blue. Only scattered dead cells could be seen in the monolayer (data not shown).

Cells are soft objects that are deformed by the imaging forces used. We examined the amount of deformation in the monolayer by collecting force curves on the cells. These curves show the deflection of the cantilever as the AFM tip is approaching a surface. On a hard surface the movement of tip and sample become coupled (one to one) after the tip makes contact. However, on a soft surface the sample deforms, resulting in less than a one to one movement of tip and sample after contact. Fig. 2Ashows a typical force curve from an MDCK monolayer, in which the sample is advanced, after contact, ∼1 μm in zto produce a cantilever deflection of ∼35 nm (2.2 nN). Therefore, the apical plasma membrane has an average spring constant of 0.002±0.0006 N/m (n=11) over this deflection range, less than 10% of that of the cantilever (kcant.=0.06 N/m). These data can also be displayed as a stiffness curve (Fig. 2B). This curve shows that the measured stiffness varies with indentation depth, as the contact area with the tip increases. For MDCK cells, the stiffness is ∼0.0035 N/m at 1 μm depth.

As illustrated in Fig. 3, varying the speed at which the sample is moved in zwhile force curves are recorded did not significantly alter the approaching curve. However, at scan rates above 4.6 μm/s (Fig. 3B-D) the withdrawing curve indicated an adhesive interaction, which is strongly scan ratedependent. This adhesion is not the typical kind of adhesion seen in AFM force curves, in which the tip is stuck to the surface and at sufficient force suddenly disengages (for example, see Hoh et al., 1992). Rather, the rounded shape of the adhesive part of the curve suggests that the tip slowly disengages from the surface. This could result from a weak interaction between the tip and plasma membrane over a relatively large contact area that does not release fast enough as the tip is withdrawn at high scan rates. Therefore, the cell deforms upward as the tip is withdrawn (Fig. 3D, inset).

We also examined the effect of imaging force on image quality by imaging the same area at increasing forces (Fig. 4). The images shown in Fig. 4Aand Bwere collected at 2.0 and 4.4 nN, respectively, corresponding to the positions shown on the force curve in Fig. 4C. These images show that imaging forces above 2 nN, which correspond to zdeformations in the membrane of >1000 nm (from Fig. 2A), were required to achieve good image contrast. However, this is an upper limit for the amount of deformation during imaging, since the viscous response of the cell will reduce the amount of deformation at the relatively high scan rates used when imaging. The large forces needed suggest that sample deformation is an important component of the contrast mechanism in imaging living cells.

In Fig. 5, the comparison of a scanning electron micrograph (Fig. 5A) with AFM images (Fig. 5B-D) shows that the numerous microvilli and the apical surface of MDCK monolayers are not visible in the AFM images. It is likely that the AFM tip sweeps them in the scan direction, thereby obscuring individual microvilli. Consistent with this notion, the high-magnification AFM images show that much of the fine surface structure is swept in the direction of the tip movement.

In addition to the cell boundaries and nucleus, there are two distinct structures in the plasma membrane of these cells that we cannot identify. They appear as two types of protrusions from the membrane. One is a smooth bulge in the apical membrane that is typically 2-10 μm wide and 0.1-1 μm tall (Fig. 5B). These bulges occupy a volume of ∼10-100 μm³above the surrounding plasma membrane, and appear to represent a relatively rigid subapical structure that is transient. The other type of protrusion is a sharp spike that extends above the membrane (Fig. 5C) and is characterized by the fact that it is swept along the scan direction by the tip, as seen by a comparison of trace and retrace images (Fig. 6). Spikes are more frequent than bulges, often with two or more appearing in a single cell. They are generally 0.1-1 μm tall and have a triangular shape, which most likely reflects the AFM tip shape (in the scan direction). At present, we envisage three possible origins of the spikes. First, they could represent an evagination of the plasma membrane. However, the deformation measurements above do not favor this explanation. Second, they could result from cilia, which have been reported in MDCK cells by some investigators. However, we have not seen any cilia in scanning electron microscopy of the strain of MDCK cells used here. Also, there is only one cilium per cell, while there are often many spikes in each cell. Third, and more likely, the spikes also result from the presence of a mechanically rigid subapical structure. The morphology suggests that this structure could be filamentous, such as a microtubule, with one end abutting the plasma membrane (Fig. 6). In addition, fibrous structures that run parallel to the apical surface were also seen (Fig. 5D).

Fixation of the MDCK monolayer with glutaraldehyde results in a rapid stiffening of the monolayer and a dramatic change in the surface morphology. Comparison of the cell surface before (Fig. 7A) and after (Fig. 7B) the glutaraldheyde addition reveals that the morphology changes in response to fixation. After glutaraldehyde treatment, the membrane surface becomes highly corrugated and there is a small (∼100 nm) thinning of the monolayer. The origin of these surface corrugations is not known; however, it could result from a mechanical stabilization of the microvilli, making them visible to the probe. The cell boundaries and overall organization of the monolayer remained unchanged, although the nucleus is not seen in fixed cells. As illustrated in Fig. 7C, the mechanical stiffening of the membrane can also be monitored in real time by collecting force curves during the glutaraldehyde fixation reaction. These curves demonstrate that the elastic response changes to a rigid response to the cantilever in ∼1 minute.

In contrast to normal MDCK cells, ras-transformed R5 cells do not form homogeneous, polarized monolayers (Schoenen-berger et al., 1991). At low density, R5 cells on glass substrata do not grow as discrete colonies but spread out and frequently extend processes. Fig. 8Aillustrates that R5 cells grow very flat in the periphery, although the center of the cell with the nucleus is often micrometers tall and sometimes out of the zrange of our piezo scanner. Cross-sections through the AFM images show that the peripheries of these cells are often <100 nm thick. The corresponding error signal image (Fig. 8B) reveals cell borders and nuclei as the most prominent structures. In thin areas of the periphery, details of intracellular structures underlying the plasma membrane are visualized by AFM, but not by scanning electron microscopy (Fig. 9). Cytoskeletal elements in various configurations were clearly seen (Fig. 9B,C). Many of the bumps seen are likely to represent intracellular vesicles, as suggested by their movement along cytoskeletal elements (unpublished observation). Like MDCK cells, R5 cells also have spikes protruding from the membrane, as well as a number of unidentifiable structures, some of which move over time.

The normal MDCK cells and rastransformants described here have a number of features that make them suitable for imaging by atomic force microscopy under physiological conditions.

The MDCK cells form continuous monolayers that are laterally supported by the close packing of cells, making them stable to lateral movement by the AFM tip. The relatively small height variations along the apical surface are well within the range of current AFM scanners. In contrast, the ras-transformed R5 cells grow extremely flat and reveal extensive intracellular structure. The cells remained viable during AFM imaging, as evidenced by dynamic changes in the plasma membrane (unpublished), and further confirmed by Trypan Blue exclusion.

By actually touching the cells with a sharp tip, the AFM uses a unique sensing system to generate images. Interpretation of these images is in its infancy, and considerable efforts are being made to understand the nature of the structures seen. Henderson and colleagues (1992)have proposed two possible contrast mechanisms for detecting intracellular structure, deformation of the plasma membrane around intracellular structures and penetration of the tip through the membrane. The data presented here show that the plasma membrane is highly deformable, and suggest that the AFM primarily detects cellular structures that are rigid relative to their surroundings. The imaging conditions we have used on MDCK cells require forces >2 nN to achieve good image contrast. This corresponds to ∼1000 nanometers of deformation in the force curve. However, the deformation measured by force curves is an upper limit for the actual deformation during imaging, since the viscous response of the cell will prevent the full deformation at the scan speeds used. Further, none of the structures we could visualize with the AFM, such as nuclei, stress fibers, cell boundaries or intracellular vesicles, could be seen in SEM images. The ability to visualize structures based on their rigidity therefore allows certain structures within cells to be selectively imaged. The effect is to ‘stain’ structures on the basis of their mechanical properties, which allows the direct examination of their dynamic behavior. This notion is in agreement with the results of Chang et al. (1993), who conclude that the ability to image the cytoskeleton in living cells stems from variations in the ‘surface stiffness’ of the specimen.

The physical interaction of the AFM tip with the cell surface raises the question of what types of perturbations to cellular structure and behavior occur during imaging. That imaging perturbs the cellular structure is most clearly seen by comparing trace and retrace images, in which structures such as the spikes on MDCK cells are swept along the scan direction (Fig. 6). This movement of structure is produced by a sheer force that is generated by the tip moving across the cell surface. We have also seen large-scale displacement of structures, with the tip in effect placing structures within a cell (data not shown). A more subtle type of response to the AFM tip could occur at the physiological level, where, for example, calcium waves can be induced in cell monolayers by mechanical stimulation (Charles et al., 1991).

Measuring the dimensions of structures from AFM images is often difficult because of sample deformation, as shown in Fig. 2, and contributions of the tip geometry to the image. In addition, living material that may change shape on the time scale of imaging presents a new type of problem. An AFM image is composed of data that are acquired on several different time scales. The height signal (z) is limited by the response of the cantilever or feedback loop, both being on the order of milliseconds in our instrument. In the lateral dimensions, a line in the fast scan direction (x) is generally acquired in about 100 ms, while lines in the slow scan direction (y) are acquired on the order of tens of seconds. Therefore, a structure changing shape in a few seconds would be accurately represented in any single line trace in x, but would be distorted in y. It is in principle possible to compensate for movement of structures, if the direction and rate of movement at each point are known or can be predicted (Hillner et al., 1992). However, this information is usually not available. Therefore, the most reliable measurements are obtained from features that are stable over time periods that are much longer than the imaging acquisition time. Such structures are most easily identified in sequential images.

In this report we have not addressed the identification of various structures seen. Some, such as the cell boundaries and nuclei are identified on the basis of similarities with morphological features of these cells revealed by light and electron microscopy. Methods for the definite identification of structures in AFM images are not yet well developed. Putman and co-workers (1992a) showed that cell surface markers could be identified on air-dried lymphocytes labeled with colloidal gold. Although, no labeling techniques for living cells in the AFM have been devised, Henderson and co-workers (1992) identified actin filaments by performing immunofluorescence on cells after AFM imaging. New AFMs that are integrated with optical microscopes will allow for simultaneous identification of structures by classical staining approaches used in light microscopy, including immunolabeling (Henderson and Sakaguchi, 1993; Putman et al., 1992b, 1993; Schabert et al., 1994).

Force curves collected on the MDCK monolayers show that these are very soft relative to the AFM cantilever. A 35 nm deflection of the cantilever corresponds to the apical plasma membrane being pushed in ∼1000 nm. The apical cell surface can therefore be thought of as having an average spring constant of less than one tenth of that of the cantilever (kcell=0.002 N/m versus kcant=0.06 N/m). This places inherent limits on the imaging of cells with the AFM, since membrane deflections will always be several times larger than the cantilever deflection with currently used cantilevers. The deformation seen in MDCK cells is in close agreement with that seen by Weisenhorn et al. (1993), who reported that 1-10 pN results in a 1 nm deformation of the plasma membrane of cells from a lung carcinoma cell line. In addition, they calculated a Young’s modulus of 0.013-0.15 MPa for these cells. However, because of concerns about the highly heterogeneous composition of a cell, we have not attempted to calculate an elastic modulus. Instead we report an effective spring constant as a measure of the elastic properties of the plasma membrane. This value is measured from the initial cantilever deflection through 1 μm of membrane deflection, and therefore represents an average value over this deformation range. The effective spring constant at any indentation depth can be seen in the stiffness curve (Fig. 2B). It should be noted that both these values depend on the shape of the indentor (the tip), and are therefore only valid for the pyramid-shaped tips used.

The adhesion manifest in force curves at high scan rates was unexpected, since the scans at low rates did not indicate an attractive interaction between the tip and the sample. A likely explanation for this adhesion is that liquid cannot flow into the gap between the tip and the membrane fast enough as the tip is withdrawn. Therefore, an underpressure develops, which holds the membrane onto the tip. The adhesion seen would then depend on the fluid dynamics at the tip-membrane interface and the viscoelastic properties of the cell. Such an adhesion mechanism is consistent with the measurements of Fritz et al. (1994), who concluded that the interaction force between a silicon nitride AFM tip and the plasma membrane of throm-bocytes is repulsive. While approaching force curves have previously been used to measure the viscoelastic properties of cells (Weisenhorn et al., 1993), the adhesion in the retracting curve described here is also influenced by these properties and may provide information about the viscoelasticity not seen in the approaching curve.

Mechanical stability is important in several cellular systems (McNeil, 1993). For example, cells in intestinal epithelia are often disrupted by external stress. Moreover, one current model for the protein dystrophin suggests that it is involved in the stabilization of the sarcolemma, and that the cells of dystrophic muscle are mechanically more fragile than those of normal muscle cells (Menke and Jockusch, 1991; Petrof et al., 1993). Present methods for examining the mechanical properties of cells, such as pipette aspiration, osmotic stress or dye loading, have poor spatial resolution. The ability of the AFM to measure deformation and apply deforming forces locally will soon allow the mechanical properties of cells to be mapped at a resolution of <100 nm by collecting arrays of force curves over the cells. Using the force modulation technique, similar images have recently been described for thrombocytes (Fritz et al., 1994). Further, as demonstrated here by the chemical fixation of MDCK cells with glutaraldehyde, changes in the mechanical properties of a cell in response to external treatments can be monitored in real time.

We thank Andreas Hefti for his assistance with the SEM and Kenneth Goldie for preliminary AFM images of MDCK cells. We thank Ueli Aebi, Andreas Engel and Paul Hansma for their generous support, and Jason Cleveland, Monica Fritz, Helmut Knapp, Manfred Radmacher and Frank Schabert for helpful discussions. This research was supported by a postdoctoral fellowship from the Human Frontier Science Program Organization (to J.H.H., LT-438/92), a junior scientist award from the Paul Blümel Foundation (to C.A.S.), and the M. E. Müller Foundation.