Imaging Principles. We excite the evanescent waves by directing the laser light with the center wavelength of 638 nm onto the surface of a cover glass via a prism at the incident angle of ~65o, which is beyond the critical angle at the glass-water surface for total internal reflection (TIR). Under this condition, the reflection light does not provide any useful information, where the large background resulting from strong reflection will block any signal from the cells. In contrast, the ESM can collect the evanescent lights scattered by the analytes via a 10x objective (NA 0.28) to clearly observe the distribution of cell adhesion sites (Fig. 1, and Supplementary Fig. 1 for details).
In ESM, the evanescent waves only play the role of illumination, so the parabolic tails resulting from the long decaying length of the evanescent waves along the surface will not be seen in the images, making it possible to provide higher image contrast and lateral spatial resolution than traditional evanescent imaging systems such as the SPRM and TIR imaging devices20-23. Besides, the cover glasses employed by the ESM can provide friendly surfaces for cell attachments and small surface roughness to create a negligible image background, thus allowing clear observations of the cell edges for recognizing individual cells for single cell analysis (Fig. 1B and 1C). In addition, the ESM provides high image quality with a prism as an illumination device and a low magnification objective for observation, thus providing a field of view of about 1.4 mm ×1.0 mm for simultaneously analyzing ~200 cells in a single measurement, making it possible to conduct high throughput detection.
Membrane protein binding kinetics in fixed cells Fig. 2 shows the kinetics analysis of wheat-germ agglutinin (WGA; molecular weight, 35 kDa) binding onto N-acetylglucosamine (GlcNAc) in fixed A431 cells, where the cells are fixed with 4% paraformaldehyde in PBS buffer to minimize the random cellular micromotions for suppressing the background noise. The bright field and ESM images are shown in Fig. 2A. First, PBS buffer was injected into the channel to flow over the cells with a flow rate of 350 μL min−1 to obtain a baseline. Then, 25 μg mL−1 WGA in PBS buffer was introduced to allow the association of WGA with the GlcNAc on the cells. Finally, the flow was switched back to the PBS buffer to allow the dissociation of WGA from the cells. The ESM image intensities of the cells increase along with the WGA binding (Supplementary Video 1). This is expected because the WGA molecules are biological macromolecules, and evanescent waves scattered by WGA molecules can interfere with those scattered by the cell adhesion sites, resulting in increasing image intensity along with the WGA binding17,18,24-30. After averaging the image intensities of all adhesion sites of the cells and plotting the mean values against the time, the ensemble binding curve can be achieved as shown in Fig. 2B. By fitting the binding curve with a first-order binding kinetics model, the ensemble association rate constant (kon) and dissociation rate constant (koff) were found to be 1.3 × 104 M−1 s−1 and 2.7 × 10−4 s−1, respectively, and the dissociation constant (KD) can be determined to be 20.8 nM using the relationship of koff / kon for the fixed A431 cells.
In addition to the ensemble measurements, the ESM also permits the membrane protein binding kinetics analysis at a single cell level. The differential images achieved by subtracting the ESM image after flowing the WGA solution with the initial ESM image clearly show the heterogeneous cell responses to WGA binding (Fig. 2C). After averaging the image intensities of all adhesion sites of one individual cell and plotting the mean values against the time, we can achieve the binding curves resulting from WGA binding onto the membrane proteins in those three individual cells that are shown in Fig. 2C (Fig. 2D). By fitting the binding curve with a first-order binding kinetics model, the kon, koff, and KD were found to be 2.8 × 104 M−1 s−1, 1.4 × 10−3 s−1, and 50.0 nM for the cell 1, 5.4 × 103 M−1 s−1, 6.1 × 10−4 s−1, and 112.9 nM for the cell 2, 3.1 × 104 M−1 s−1, 7.4 × 10−4 s−1, and 23.9 nM for the cell 3, respectively. The results clearly show that the kinetic properties are heterogeneous among the individual cells even they belong to the same cell line. After analyzing all individual cells, we find that the binding curves from ~65% of the individual cells can be fitted with the first-order binding kinetics model to achieve the kon, koff, and KD, where the binding curves show notable individual differences (Supplementary Fig. 2 and 3). The distribution of kon values for these individual cells can be directly fitted with Gaussian fitting, and the distributions of koff, and KD values can be fitted with Gaussian fitting in semi logarithmic coordinates (Fig. 2E-G). The most probable kon, koff, and KDvalues can be determined to be 2.7 × 104 M−1 s−1, 7.2 × 10−4 s−1, and 26.9 nM, where the KD value looks similar but the kon and koff values obviously deviate from the ensemble measurements shown in Fig. 2B. These deviations can be attributed to at least two reasons. First, even for the 65% of the cells with binding curves that can be fitted have heterogeneous responses to the WGA binding, which can be seen from their binding curves (Supplementary Fig. 2 and 3). These cells also have different maximum response values(Fig. 2H), indicating that the contents of membrane proteins differ with cells19, which may be one of reason leading to heterogeneous binding kinetics. Second, the binding curves from the other 35% of the individual cells cannot be fitted with the first-order binding kinetics model, which is unlikely caused by the measurement errors because the cells are randomly distributed within the field of view (Supplementary Fig. 4). Most of them show notable responses to WGA binding but no measurable dissociation curves (Supplementary Fig. 5) while few show no dynamic features (Supplementary Fig. 6). Considering the koff as zero, the most probable konfor the cells with notable association curves and without measurable dissociation curves can be estimated to be 2.7 × 104 M−1 s−1, which is the same as the values for the cells with the binding curves that can be fully fitted, indicating that the koff, which determines the drug-target residence time that plays a decisive role in drug efficacy in vivo8,9, may dominate the cellular heterogeneity in membrane protein binding kinetics between different individual fixed cells.
Membrane protein binding kinetics in live cells The fixation processing varies the native environments of the membrane proteins, which may change the molecular interaction properties. Therefore, determining the membrane protein binding kinetics in single live cells makes more sense in biochemical applications. Herein, we show that the ESM can extract the signals of ligand binding onto membrane proteins in single live cells. The bright field and ESM images of live A431 cells are shown in Fig. 3A. First, the live cell imaging solution was injected into the channel to flow over the cells with a flow rate of 350 μL min−1 to obtain a baseline. Then, 25 μg mL−1 WGA in live cell imaging solution was introduced to allow the association of WGA with the GlcNAc on the cells. Finally, the flow was switched back to the live cell imaging solution to allow the dissociation of WGA from the cells. The ESM image intensities of the cells increase along with the WGA binding (Supplementary Video 2). After averaging the image intensities of all adhesion sites of the cells and plotting the mean values against the time, the ensemble binding curve can be achieved as shown in Fig. 3B. By fitting the binding curve with a first-order binding kinetics model, the ensemble kon, koff, and KD were found to be 1.3 × 104 M−1 s−1, 3.9 × 10−3 s−1, and 300.0 nM for the live A431 cells. The koff and KD vlaues obviously differ from those achieved from fixed cells, which should be caused by different cell conditions.
We further show that the ESM can analyze the membrane protein binding kinetics at a single live cell level. The differential images achieved by subtracting the ESM image after flowing the WGA solution with the initial ESM image clearly show the heterogeneous cell responses to WGA binding under live conditions (Fig. 3C). After averaging the image intensities of all adhesion sites of one individual live cell and plotting the mean values against the time, we can achieve the binding curves resulting from WGA binding onto targets in those three individual live cells that are shown in Fig. 3C (Fig. 3D). By fitting the binding curve with a first-order binding kinetics model, the kon, koff, and KD were found to be 3.7 × 104 M−1 s−1, 1.3 × 10−3 s−1, and 35.1 nM for the cell 1, 2.9 × 104 M−1 s−1, 5.7 × 10−3 s−1, and 196.6 nM for the cell 2, 2.2 × 104 M−1 s−1, 6.6 × 10−4 s−1, and 30.0 nM for the cell 3, respectively. The results clearly show that the kinetic properties are heterogeneous among different single cells. After analyzing all individual live cells, we find that the binding curves from ~61% of the cells can be fitted with the first-order binding kinetics model to achieve the kon, koff, and KD, where the binding curves show notable individual differences (Supplementary Fig. 7 and Fig. 8). The distribution of kon values for these individual cells can be directly fitted with Gaussian fitting, and the distributions of koff, and KD values can be fitted with Gaussian fitting in semi logarithmic coordinates (Fig. 3E-G). The most probable kon, koff, and KDvalues can be determined to be 1.9 × 104 M−1 s−1, 1.8 × 10−3 s−1, and 63.1 nM, where the kon value looks similar but the koff and KD values obviously deviate from the ensemble measurements shown in Fig. 3B. Considering that the KD is achieved by koff / kon, the koff should be the more fundamental factor determining the cellular heterogeneity on kinetics of WGA biding onto membrane proteins. Similar to fixed cells, the deviations can be also attributed to two reasons. First, the contents of membrane proteins differ with individual cells, which can be seen from the heterogeneous maximum response values among different single cells (Fig. 3H). Second, the binding curves from the other 39% of individual cells cannot be fitted with the first-order binding kinetics model, which should not be caused by the measurement errors because they are randomly distributed within the field of view (Supplementary Fig. 9). Most of them present notable association curves but no measurable dissociation curves (Supplementary Fig. 10 and Fig. 11) while few show no dynamic features (Supplementary Fig. 12). Considering the koff as zero, the most probable konfor the cells with notable association curves and without measurable dissociation curves is estimated to be 1.9 × 104 M−1 s−1 (Supplementary Fig. 11), which is the same as the values for those individual cells with the binding curves that can be fully fitted, indicating that the koff may play a more important role in determining the heterogeneity on the responses of individual live cells to WGA as the fixed cells. In addition, an interesting phenomenon is that the most probable maximum response value is almost three times higher in live cells (Fig. 3H) than in fixed cells (Fig. 2H), indicating that the membrane protein may be more active in live than in fixed cells.
Small molecule binding to membrane proteins Small molecules comprise over 90% of FDA-approved drugs, so it is vital to characterize the kinetics of small molecules binding to membrane proteins but this task remains challenging because of the tiny mass of small molecules. Erlotinib (429.9 Da) is an FDA-approved tyrosine kinase inhibitor targeting the epidermal growth factor receptor (EGFR). Here we employ the ESM to analyze the kinetics of erlotinib binding to EGFR in live A431 cells, where the previous experiments have shown that the cell response to ligand binding is greater in live than fixed conditions. The bright field and ESM image are shown in Fig. 4A. First, live cell imaging solution was injected into the channel to flow over the cells with a flow rate of 350 μL min−1 to obtain a baseline. Then, 1 μM erlotinib in live cell imaging solution was introduced to allow the association of erlotinib with the EGFR. Finally, the flow was switched back to the live cell imaging solution to allow the dissociation of erlotinib from the cells. Due to the tiny ligand mass, we can observe the intensity variation during the binding process, but the signals are pretty weak (Supplementary Video 3), and the binding curve achieved by tracking the ESM image intensity present response to some extent but is hard to fit (Fig. 4B). Fig. 4C shows that the heterogeneity can be seen in the differential images achieved by subtracting the ESM image after flowing the erlotinib solution with the initial ESM image for the individual cells. Meanwhile, the binding curves achieved by tracking the image intensity variations of the cells shown in Fig. 4C present heterogeneity but are still hard to fit (Fig. 4D). After analyzing the binding curves achieved by tracking the image intensity variations of all individual cells, we find that only 19% of the binding curves can be fitted with a first-order binding kinetics model (Supplementary Fig. 13), 12% of the binding curves can be fitted with assuming the koff as zero (Supplementary Fig. 14), and the other 69% of the binding curves have no dynamic features (Supplementary Fig. 15 and Fig. 16). Different kinds of the cells are randomly distributed within the field of view (Supplementary Fig. 17), indicating that the measurement errors are not the main reason for difficulty in fitting.
Although the binding curves achieved by tracking the image intensity variations are hard to fit, they indicate the cell response to the erlotinib to some extent. The ESM has demonstrated that the capability to analyze the small molecule binding behaviors by analyzing their effect on the conformation changes of receptors via quantifying the spring constants25,26,31, where the conformation changes are more intrinsic features of molecular interactions than molecular weight variations. The adhesion spring constants of cell adhesion sites were calculated as shown in Fig. 5A. The dynamic process shows that the adhesion spring constant maps can clearly show the cell variations during erlotinib binding (Supplementary Video 4). This is expected because the small molecule binding events can vary the membrane protein conformations contained in the adhesion sites. We first achieve the ensemble binding curve by averaging the spring constants of all adhesion sites of the cells and plotting the mean values against the time as shown in Fig. 5B, and the ensemble binding curve does not provide any meaningful information. However, the individual cells shown in Fig. 5C characterized with spring constant, where the cells are the same as those in Fig. 4C, present heterogeneity and their binding curves created by tracking the spring constant variations can be fitted with a first-order binding kinetics model (Fig. 5D). After analyzing the binding curves of all individual cells achieved by tracking the spring constant variations, 60% of the binding curves can be fitted with a first-order binding kinetics model (Supplementary Fig. 18 and Fig. 19), 4% of binding curves can be fitted with assuming the koff as zero (Supplementary Fig. 20), and the other 36% of binding curves have no dynamic features (Supplementary Fig. 21). Different kinds of the cells are randomly distributed within the field of view (Supplementary Fig. 22), indicating that the cell-to-cell heterogeneity, rather than the measurement errors, determines the cellular heterogeneity. The ratio of binding curves with no dynamic features is higher for erlotinib binding onto the cells than the WGA binding onto the cells, which may be one reason leading to the difficulty on fitting the ensemble binding curve. The distributions of the kon, koff, KD, and maximum response values from the fitted binding curves are shown in Fig. 5E, 5F, 5G, and 5H, respectively, which clearly show the cellular heterogeneity in kinetics of erlotinib binding onto EGFR. The response of RBL-2H3 cells expressing no EGFR to the erlotinib is employed as a negative control, where the image intensity or spring constant of RBL-2H3 cells presents no measurable variations during erlotinib binding (Supplementary Fig. 23). This experiment demonstrates the ESM can analyze the kinetics of small molecules binding onto membrane proteins at a single live cell level. Also, the comparison between ensemble and single cell analysis shows that the single cell analysis can uncover the fact that cells have response to the ligand stimulations, while the ensemble measurements may average out the individual differences and provide misleading information.