Skip to main content
. 2021 Mar 30;10:e66151. doi: 10.7554/eLife.66151

Figure 1. Experimental/computational pipeline to analyze heterogeneous adhesion dynamics in ChoK1 cells.

(a–c) High-resolution traction maps co-imaged with mGFP-tagged adhesion protein, talin (d), vinculin (e), and paxillin (f). 5 kPa silicone gel coated with high density beads was used as a TFM substrate. (g) Trajectories of individual nascent and focal adhesions overlaid on a region of interest cropped from (e). Tracking is based on all detected point sources, (red circles). Big segmented focal contacts/adhesions (orange, closed freeform overlays) were used as additional information for feature selection. (h) Some of the key features used for supervised classification, tabulated in Supplementary file 1A. (i) Classification of adhesion trajectories into nine different groups, overlaid on the adhesion image. Five different NA groups, three FA groups and one noise group were distinguished by the support vector machine classifier. (j–m) Comparison of feature values among the five NA groups, G1, G2, G3, G4, and G5: edge protrusion speed (j), adhesion movement speed, positive when sliding toward protruding edge (k), mean intensity (l), and lifetime (m), extracted from six vinculin-tagged cells. All features show a significant shift in value for at least one subgroup. (n) Average traction magnitude, read from traction map, at individual NA trajectories per each group. The number of samples per each group is shown in the lower right corner of the figure.

Figure 1—source data 1. Source data for Figure 1j.
Figure 1—source data 2. Source data for Figure 1k.
Figure 1—source data 3. Source data for Figure 1l.
Figure 1—source data 4. Source data for Figure 1m.
Figure 1—source data 5. Source data for Figure 1n.

Figure 1.

Figure 1—figure supplement 1. Simultaneous TFM-adhesion experimental approach.

Figure 1—figure supplement 1.

(a) A schematic of the TFM-adhesion imaging using TIRF. Here, cells adhere to a fibronectin-coated, high-refractive index, mechanically defined substrate that is compatible with TIRF. 40 nm beads on top of the substrate allow cell-mediated traction forces to be visualized. (b) Bead spatial resolution analysis. (c) Beads on the silicone gel surface were detected using Gaussian-mixture model, and the inter-bead distance was calculated using KD-Tree algorithm (0.42 ± 0.17 μm, mean ± standard deviation). (d) The bead density calculated from 20 bead images was 1.54 beads/μm on average. (e–h) Analysis procedure for simultaneous TFM-adhesion imaging. (e) Overlay of adhesion channel (green) and bead channel (red). (f) Adhesions are analyzed for detection, tracking, and classification as described in the main manuscript and in Figure 1. (g) Bead displacement fields are calculated by comparing the substrate state before and after removing the cells from the substrate. Bead channel with a reference image taken after the cells were removed from the substrate is analyzed for bead displacement. (h) Traction fields are obtained from the displacements solving the inverse problem as in Han et al., 2015.
Figure 1—figure supplement 2. Overall average traction per cell and cell spreading area did not change with expression of talin-GFP, vinculin-GFP, or paxillin-GFP.

Figure 1—figure supplement 2.

(a) Bar plot of average traction quantified per cell. (b) Bar plot of cell spread area.
Figure 1—figure supplement 3. Validation of SVM-based machine learning.

Figure 1—figure supplement 3.

(a) confusion matrix among nine different adhesion groups. (b) Feature space. (c) Similarity among features. (d) Similarity among trained data. White lined boxes represent similarity within each group.
Figure 1—figure supplement 4. Boxplot of the fraction of G2 NAs that mature into FCs, FAs, or either FCs and FAs.

Figure 1—figure supplement 4.

N is the number of movies, which includes data obtained from cells expressing tagged variants of talin, vinculin, or paxillin. In total 10,028 G2 NAs were analyzed. Importantly, owing to the limited duration of imaging (10 min), this quantification is an underestimation since e.g. some G2 NAs do not progress to FC or FA because the movie concludes before this can occur.
Figure 1—figure supplement 5. Representative time series of fluorescence intensity, traction magnitude, edge protrusion speed, adhesion sliding speed, and distance to closest edge, for the five NA groups (G1–G5).

Figure 1—figure supplement 5.

The names for all of the features are listed in Supplementary file 1A.
Figure 1—figure supplement 6. Differences in the feature values for NA subgroups in paxillin and talin time-lapse images.

Figure 1—figure supplement 6.

NAs groups G1, G2, G3, G4, and G5, were classified with paxillin-mGFP and talin-mGFP, from which feature values were collected. The traction magnitude, which is a non-feature outcome (e.g. is not part of the SVM training set), shows significant differences among classifications groups, similar to differences found from vinculin-mGFP experiments in Figure 1j–n. The sample number per each group is summarized at the right side of each row. Adhesion numbers are extracted from four cells for paxillin and six cells for talin.
Figure 1—figure supplement 7. Classification shifts due to substrate stiffness.

Figure 1—figure supplement 7.

ChoK1 cells, transfected with vinculin-GFP, were cultured and imaged on 18 kPa and 5 kPa silicone gel substrates with TIRF. (a) Color-coded classes of kinematically and kinetically different adhesions, tracked and classified using a filter-based classifier (Supplementary file 1C) for a cell on an 18 kPa and (b) a 5 kPa gel. (c) Mean adhesion density for each adhesion class (G1–G9) as a function of substrate stiffness. N represents the number of frames during an active cell protrusion. Total nine movies for 18 kPa gel and six movies for 6 kPa gel were captured. Note that all classes exhibit significant differences between the two stiffness conditions, particularly in newly assembling NAs, G1 and G2 and large FAs in G8.