Direct visualization of the extracellular binding structure of E-cadherins in liquid

E-cadherin is a key Ca-dependent cell adhesion molecule, which is expressed on many cell surfaces and involved in cell morphogenesis, embryonic development, EMT, etc. The fusion protein E-cad-Fc consists of the extracellular domain of E-cadherin and the IgG Fc domain. On plates coated with this chimeric protein, ES/iPS cells are cultivated particularly well and induced to differentiate. The cells adhere to the plate via E-cad-Fc in the presence of Ca2+ and detach by a chelating agent. For the purpose of clarifying the structures of E-cad-Fc in the presence and absence of Ca2+, we analyzed the molecular structure of E-cad-Fc by AFM in liquid. Our AFM observations revealed a rod-like structure of the entire extracellular domain of E-cad-Fc in the presence of Ca2+ as well as trans-binding of E-cad-Fc with adjacent molecules, which may be the first, direct confirmation of trans-dimerization of E-cadherin. The observed structures were in good agreement with an X-ray crystallographic model. Furthermore, we succeeded in visualizing the changes in the rod-like structure of the EC domains with and without calcium. The biomatrix surface plays an important role in cell culture, so the analysis of its structure and function may help promote cell engineering based on cell recognition.


Scientific Reports
| (2020) 10:17044 | https://doi.org/10.1038/s41598-020-72517-2 www.nature.com/scientificreports/ need for sample staining. With AFM observation, a resolution of nm level is expected. For example, the periodic double-helix structure of B-DNA in the solution was observed at a spatial resolution of 1.2 nm with the PeakForce Tapping mode 27 . The short axis diameter of the EC domain is expected to be around 2 nm (calculated from PDB ID: 3Q2V 28 , Fig. 3b), so that AFM may be well suited for our study to capture E-cad-Fc structure in cell culture.
Since cadherins are membrane protein and tend to aggregate and are difficult to analyze as a whole molecule, previous analyses were limited to their hydrophilic EC domains. To the best of authors' knowledge, the first AFM observation of cadherins in liquid was performed by Baumgartner et al. 29 on VE cadherin which belongs to the type II classical cadherin family. Later, Brasch et al. 30 . obtained AFM images of molecules under ambient condition which are presumed to be monomers and dimers of VE cadherin. Harrison et al. 31 made an AFM observation in liquid for the entire extracellular domain of N-cadherin which belongs to the type I classical cadherin family. For our observations, E-cad-Fc is employed, which is soluble and easy to handle. E-cad-Fc molecules in solution may take two states: in one state, two EC domains within the same E-cad-Fc molecule interact laterally, while the EC domain trans-interacts with different E-cad-Fc molecules in the other state. In other words, there is a possibility that both cis and trans dimerization may occur. Although there had been much debate on which dimer, namely cis or trans-dimer [32][33][34][35][36][37] , forms initially, various studies over the past decades have established that trans dimerization precedes cis-dimer formation 28,[38][39][40][41][42][43] . The features observed here were trans-binding as reported.
In this paper, we report the results of direct AFM observations of E-cad-Fc under physiological conditions in vitro. Our AFM observations successfully visualize a structure of the entire extracellular domain of E-cad-Fc in the presence and absence of Ca 2+ , as well as molecules configuration which may be presumed as trans-binding of E-cad-Fc with adjacent molecules. The observed structures were in good agreement with an X-ray crystallographic model. This may be the first AFM observation of the entire extracellular domain and trans-binding of E-cadherin in the presence of Ca 2+ . Furthermore, dynamic shape changes of the EC domain with the addition of EDTA are also shown.

Results and discussion
The molecular model shown in Fig. 1d was created by combining two molecular structures obtained by crystal structure analysis. Figure 1b  The adhesive capability of E-cad-Fc was reported by Nagaoka et al. [8][9][10] . EB3 mouse embryonic stem (ES) cells adhered to the E-cad-Fc coated surface with the same efficiency as the conventional gelatin coated surface, and this adhesion was Ca 2+ dependent. Many applications using the recombinant E-cad-Fc were also reported [16][17][18][19][20]45 .  www.nature.com/scientificreports/ We also examined the adhesive activity of cadherin using E-cad-Fc coated polystyrene beads. The results showed that the beads adhere to each other depending on Ca 2+ concentration (Supplementary Data S1). Cells are cultured on E-cad-Fc coated polystyrene dishes, whereas AFM observation was performed on atomically flat mica. Mica was used after being made hydrophobic by treatment with poly-l-ornithine coat. The series of figures in Fig. 2 show AFM images of E-cad-Fc molecules in PBS buffer solution containing 0.9 mM Ca 2+ . Various conformational E-cad-Fc molecules are observed in the wide area image of Fig. 2a, with EC domains and Fc domains clearly distinguishable. Figure 2b,d are magnified views from Fig. 2a, which may be presumed an E-cad-Fc monomer, and Fig. 2c,e are molecular models constructed from these AFM images respectively. As will be described in detail below, the observed curvature and height of the entire EC domain are in good agreement with the x-ray structural model. Figure 2f is a magnified view from Fig. 2a, which may be presumed three E-cad-Fc molecules, and Fig. 2g is a molecular model constructed from these AFM images. The comparison with the molecular model by crystal structure analysis Fig. 3b suggests that three E-cad-Fc molecules are transbonded in the EC1 domain, where the following three points are of particular interest.
(1) The curvature of the rod-like structure of EC domain (Fig. 2f, the blue dotted circle).
The curvature of the EC domains observed with AFM are similar to that found in the crystal structure (Fig. 3b). The crystal structure analysis 46 and the electron micrographs observation 47,48 reported that the five EC domains are curved such that the long axis of EC1 is approximately perpendicular to the long axis of EC5 (Fig. 2h).
(2) The angle of the EC1-binding part (Fig. 2f, the blue arrow). www.nature.com/scientificreports/ The angle between EC1-binding part was measured to be about 90° in our AFM image, which coincides very well with the previous results of ~ 88° by Harrison et al. 28 and Häussinger et al. 49 .
(3) The measured size of the rod-like structure of EC domain.
When the crystal structure data are examined with the structural analysis PyMOL software (version 2.0.4) 50 (Schrödinger, Inc., USA), the short axis diameter of EC is about 2 nm, with the length in the major axis direction being about 18 nm (Fig. 3b). According to the analysis of Harrison et al. 28 , the distance between the C-termini of the trans-dimer was 37.3 nm. The AFM image of Fig. 2f gives the results that the diameter measured by height is 2 nm, and the distance between C-terminals is about 35 nm. When measuring a cylindrical diameter by AFM, the height (Z) information was employed instead of the X-Y plane information, because the X-Y plane measurement includes a convolution effect with the cantilever tip 51,52 . Since the C-terminus overlaps with the Fc domain and the position cannot be determined accurately, it is likely that the value is slightly smaller than that obtained by the crystal structure analysis. The height of the overlapping part of the EC1 (Fig. 2f blue arrow) of the dimer was measured to be about 4 nm. In Fig. 2a, ten locations were identified where the overlapping of the EC1 could be clearly recognized, and the average height of those locations was found to be 4.0 (0.27) nm (mean (SD)). These are in good agreement with the X-ray crystallographic model examined by PyMOL. www.nature.com/scientificreports/ The height of the EC domain was calculated as the height difference between the two peaks observed in the height distribution (Fig. 3a) of the AFM image (Fig. 2a). The calculation by using the software bundled to the AFM instrument. The height distribution was obtained from the height data at the 262,144 (= 512 × 512) locations in the image, where the peak location of the largest frequency is considered as the substrate portion and the other peak location is deemed as the EC domain. The calculated EC domain height was 2.02 nm, which coincides very well with that obtained from the crystal structure analysis.
These results imply that observed structures are trans-bonded E-cad-Fc molecules in the EC1 domain. Some of the obvious trans-bonded E-cad-Fc molecules in Fig. 2a are shown as magnified views in Fig. 2i-k where their EC1-binding parts are pointed with blue arrows.
Regarding the height of the Fc domain, two peaks were found in its height distribution as shown in Fig. 3e. The distribution was constructed by using the height data obtained from locations which are considered to correspond to the Fc domain of the E-cad-Fc in Fig. 2a. The followings were excluded from the height distribution construction: the overlapping portion of the EC1, the portion where no connection with the EC1 was observed, the aggregate, and the portion where the measurement was corrupted with noise. The heights of the two peaks were obtained by Gaussian fitting using IGOR Pro software (version 6.3.6). The median and variance for the first peak were 4.1 nm ± 0.66, while those for the second peak were 5.5 nm ± 0.42 (n = 148). Examination with PyMOL revealed that the size of the Fc domain was 6.7 nm in length, 6.3 nm in width, and 3.6 nm in height (Fig. 3c).
It is unknown which part of the Fc domain is in contact with the mica on substrate. Judging from the shape of the molecular model, it is likely that the widest side of the Fc is "lying down" in contact with the mica. If this is the case, the height of the Fc domain measured by the AFM should be around 3.6 nm (Fig. 3c), but this value does not match the measurement result. The Fc domain consists of two CH2CH3 chains, which are the constant regions of an immunoglobulin heavy chain. The two CH2CH3 chains are bent and in a skewed position, rather than on the same plane (Supplementary Video S2). Hinge regions are not included in this crystallography structural model. The hinge region follows the N-terminal, residue 237 53 , in this model. The two EC domains are connected via linkers to Fc domains comprising CH2, CH3 and hinge regions. Because residue 237 is located on the inner periphery of the skewed U-shaped Fc domain, it is presumed the hinge and linker region between EC5 domain and CH2 is located in this neighborhood (Fig. 3d). Therefore, the EC5 domains may tilt Fc domains which are 'lying down' . If Fc domains are tilted, the height will be more than 3.6 nm. Our AFM measurements revealed two peaks at 4.1 nm and 5.5 nm in the height distribution for the Fc domain (Fig. 3e), which are larger than 3.6 nm. These results support the assumption that Fc domains are tilted. Further if EC5 domain wedges in between the CH2CH3 chain and the mica substrate, the height of Fc will be 5.6 nm at maximum (Fig. 3f).
These measurements have been achieved with a very weak load of 60 pN. This has been made possible by the excellent performance of the device and the carefully prepared cantilever. The PeakForce Tapping mode controls the load on the sample by the movement of the piezo without using the resonance frequency of the cantilever, which improves the operability in liquids where the measurement is usually more challenging [54][55][56][57] . In the case of a soft sample, the cantilever tip may deform the sample if the load is too large. In the DNA measurement example, it was reported that the difference in loading force causes a difference in DNA height 27 . To keep the load as small as possible, we used a cantilever with an extremely low spring constant of 0.12 Nm −1 . Since our measured EC height of E-cad-Fc measured by AFM was around 2 nm, which is consistent with the crystal structure analysis, the loading force of 60 pN was small enough not to give a significant deformation to the sample. Since the measurement was performed in solution, there is a possibility that hydration water around the sample and the electric double layer may affect the measurement 58 . To evaluate the absolute value of the height measurements, it is necessary to consider these influences.
In Fig. 4, a series of images for E-cad-Fc are shown when the concentration of Ca 2+ is changed by adding chelating agent. Figure 4a is an AFM image in a PBS buffer which contains 0.9 mM Ca 2+ before adding chelating agent. It shows an EC domain with a rod-like structure, a trans-binding portion of the EC1 (pointed by white arrow), and Fc domains. EDTA was added to the PBS buffer solution so that its final concentration is 5 mM. Here we focus on the molecular structures at two locations in Fig. 4 which have shown clear morphology changes. The first one is the molecule within the green circle, which appears to be an Fc domain with a rod-like structure. In Fig. 4b,c, the rod-like structure can be clearly discerned, but in Fig. 4d which is the image for 6 min after the addition of EDTA, the rod-like structure seems to have disappeared. Our second focus is on the molecule within the blue circle. In Fig. 4b, two rod-like structures can be observed. But in Fig. 4c, the connections seem to have been severed. Figure 4f is an image obtained 18 min after the EDTA addition, where rod-like structures are almost invisible. In other words, the chelation effect was successfully visualized at the molecular level. Figure 4g shows an AFM image of E-cad-Fc in a PBS solution without Ca 2+ . It has the same shape as the image of E-cad-Fc in a PBS solution without Ca 2+ . Baumgartner et al. 29 reported that AFM-observed VE-cad-Fc in a buffer lacking Ca 2+ was globular or V-shaped, but we observed only globular shapes. In this study, it took 3 min to obtain one frame. With the high-speed AFM technology [59][60][61][62][63] , which is capable of imaging at 50-100 ms per frame under physiological conditions, more detailed dynamic observation could be possible.
In the AFM observations of VE-Cadherin-Fc in a buffer by Baumgartner et al. 29 , the spatial resolution was about 10 nm, while it was about 2-4 nm in our study. One of the key factors for our success in acquiring highresolution images in liquid is the use of cantilevers whose tip is extremely sharp with the nominal radius being 1 nm. When the same E-cad-Fc sample was measured with another cantilever, whose nominal tip radius is 2 nm, Scientific Reports | (2020) 10:17044 | https://doi.org/10.1038/s41598-020-72517-2 www.nature.com/scientificreports/ www.nature.com/scientificreports/ the Fc and the EC domains could not be distinguished, with only a rod-like structure being observed (Supplementary Fig. S3). The resolution of an AFM image on the XY-plane depends on the radius of curvature at a cantilever tip. Specifically, an apparent width (W) becomes W = 4(Rr) 1/2 , where R is the radius of curvature of the cantilever tip and r is the radius of a sample which is assumed to have a circular cross section 51,52 ( Supplementary  Fig. S4). If we assume the EC domain is cylindrical with the radius of its cross section being 1 nm, the apparent width is calculated as 4 nm for a cantilever tip of R = 1 nm, and 5.64 nm for a R = 2 nm tip. The half-width value measured from the cross-sectional images of the EC domains in Fig. 2a is 4.73 nm ± 0.83 (n = 10) with a 1 nm cantilever tip. For the data measured with a cantilever whose nominal tip radius is 2 nm, the half-width value becomes 6.39 nm ± 0.88 (n = 10). Of course, there should be a variation in each cantilever tip, but we can certainly obtain higher resolution images by using a cantilever which has a sharp tip with a smaller radius. Further, force regulating capability as small as 60 pN contributes greatly to the successful imaging. Mica is a substrate commonly used for the sample fixation in AFM measurement because it is easy to obtain a cleavage plane that is clean and atomically flat. The mica surface as prepared is generally hydrophilic, which makes sample fixation in an aqueous solution difficult. However, the coating with poly-l-ornithine turns the surface positively charged and the immobilization of E-cad-Fc via the Fc domain on the hydrophobic surface in an aqueous solution is made possible through the hydrophobic interaction, but the details of interaction forces contributing to the fixation are unknown.

Methods
Sample preparation and substrate. E-cad-Fc's were genetically engineered by using the previously reported method 10  For the time laps measurement, 12.5 mM EDTA was prepared with EDTA 2Na. For the control experiment, a 30 µl droplet of 20 μg/mL E-cad-Fc in the PBS solution was deposited on poly-l-ornithine coated mica. Thirty minutes later, the substrate was rinsed with the PBS solution and was measured in the PBS solution.
AFM measurement. AFM images were measured by a MultiMode 8 with NanoScope V controller (Bruker Corp., Billerica, MA, USA). PeakForce Tapping (PFT), which we used for the imaging and measurement of the adhesion force, enables us to obtain quantitative nanomechanical properties at a relatively high speed. In this study, we also used PeakForce QNM (PFQNM) 64 . PFQNM uses PFT mode. PFQNM provides us information on nanomechanical properties such as modulus and adhesion. This PFQNM imaging mode enables us to identify variations of material properties at high resolutions simultaneously with the imaging of sample topography. The samples were imaged in the PBS solution with Ca 2+ at room temperature and captured in 512 × 512 data points for 500 nm scan area with a scan rate of 1 Hz. For the E-cad-Fc imaging (Fig. 2), the set point was chosen as 60 pN, and force-distance curves were recorded 20 nm (PF amplitude of 10 nm), at a frequency of 2 kHz, using sharp probes (PEAKFORCE-HIRS-F-B, Bruker Nano Inc., Goleta, CA, USA) with a spring constant of 0.12 Nm -1 , a nominal resonant frequency of 100 kHz in air, a nominal tip radius of 1 nm and reflective gold coating cantilever. For the time laps imaging (Fig. 4), the samples were captured in 256 × 256 data points for 250 nm scan area with 1.2 ~ 1.5 Hz, the set point was about 200 pN, and force-distance curves were recorded 100 nm (PF amplitude of 50 nm) at a frequency of 2 kHz, using the same kind of the probe. For the control imaging (Fig. 4g), the set point was set to 700 pN, force-distance curve were recorded 60 nm (PF amplitude of 30 nm) at a frequency of 2 kHz, using the standard probe for PFT measurements (SCANASYST-EFLUID + , Bruker Nano Inc., Goleta, CA, USA) with a spring constant of 0.7 Nm -1 , a nominal resonant frequency of 150 kHz in air, a nominal tip radius of 2 nm and reflective gold coating cantilever.
The images were mainly analyzed with the bundled software and the height distribution analysis was done by Gaussian fitting using standard function of IGOR Pro software (version 6.3.6) (HULINKS Inc., Japan).
In the time-lapse observation, the chelating agent solution was added using a syringe from the solution inlet of the holder for measuring in liquid. The solution was provided through an appropriately sized silicone tube which was placed between the syringe and its inlet, rather than directly from the syringe, so that the solution injection may not disturb the measurement.

Molecular structure analyses.
We used the free available PyMOL software (version 2.2) 50 for the molecular structure analyses. Standard functions such as image rotation, spatial distance measurement and movie editing were utilized for the analyses.

Conclusion
In this paper, E-cad-Fc molecules were visualized in liquid by AFM. Particularly, the AFM images we have provided may be the first, direct confirmation of trans-dimerization for E-cadherin occurring independent of cis-dimerization. The EC binding structure was similar to the one obtained in the crystal structure analysis. Moreover, the shape changes during the addition of chelating agent were clearly captured.
The key to the successful observation may be attributed to the use of Fc bounded to the cadherin EC domain, which facilitates the fixation of the EC domain to the substrate and helps distinguish its orientation identifying its N-terminus. Also, the AFM's capability of high resolution imaging of biological samples in liquid has been crucial.
The cells adhere to the plate via E-cad-Fc in the presence of Ca 2+ and detach when Ca 2+ is removed by a chelating agent. The corresponding morphological changes of E-cad-Fc as well as trans-binding of cadherins with adjacent molecules were successfully observed at molecular level. It should be noted these observations were made in liquid Ca 2+ concentration of 0.9 mM which is the same as that for cell cultivation. Clarification of the structure and function of biomatrix molecules play crucial role in future application development in cell engineering.