Image Analysis

Images were analyzed using ImageJ/Fiji (NIH, Bethesda, MD) image analysis software. Stress fiber formation was quantified through manual counting of cells containing lateral stress fibers. A stress fiber was defined as an actin filament across the lateral width of the cell: If the cell expressed one or more stress fibers, we mark it as positive, and if not we define it as negative (Supplementary Figure 1 gives examples of each). For Fig. 1c, each HeLa GFP-LKB1 cell line was compared with the respective empty GFP control lines and also to its farnesylation mutant partner (WT vs. C430S, K78I vs. K78I-C430S, CTD vs. CTD-C430S) using a 2-tailed Chi-squared analysis with a p-value of 0.05. ****p ≤ 0.0001. All analysis was conducted over 3 separate experiments. In Fig. 1, Empty GFP: n = 103 cells; LKB1 WT: n = 150 cells; LKB1 C430S: n = 92 cells; LKB1 K78I: n = 198 cells; LKB1 K78I-C430S: n = 121 cells; LKB1 CTD: n = 142 cells; LKB1 CTD-C430S: n = 145 cells. Additionally (Supplementary Figure 1), the percent of stress fibers per field was quantified. These data passed the assumptions of parametric tests (homogeneity of variance), and this percentage was then analyzed using an ANOVA test, which showed statistical significance at a p-value of 0.05. Sidak’s multiple comparisons test was then performed to compare each construct to empty GFP control and its farnesylation mutant partner with a p-value of 0.05. ****p ≤ 0.0001. n = 3 fields of view/construct. In Fig. 2, Part B was broken into parts i and ii for simplicity, and each construct was compared to Empty GFP control, as well as within each family (ie: Empty GFP vs Empty GFP + RhoA Q63L and Empty GFP + cdc42 Q61L), using a 2-tailed Chi-squared analysis with a p-value of 0.05. In Fig. 2d, each construct was compared to empty GFP control as well as the respective treated vs untreated control (ie: empty GFP vs empty GFP + ROCKin) using a 2-tailed Chi-squared analysis with a p-value of 0.05. ****p ≤ 0.0001. In Fig. 2b(i), Empty GFP: n = 29 cells; Empty GFP + RhoA Q63L: n = 62 cells; Empty GFP + cdc42 Q61L: n = 40 cells; LKB1 Wildtype: n = 40 cells; LKB1 Wildtype + RhoA Q63L: n = 60 cells; LKB1 Wildtype + cdc42 Q61L: n = 82 cells; LKB1 C430S: n = 40 cells; LKB1 C430S + RhoA Q63L: n = 88 cells; LKB1 C430S + cdc42 Q61L: 103 cells. In Fig. 2b(ii), the Empty GFP, Empty GFP + RhoA Q63L, and Empty GFP + cdc42 Q61L values were repeated from Fig. 2b(i), LKB1 CTD: n = 36 cells; LKB1 CTD + RhoA Q63L: n = 86 cells; LKB1 CTD + cdc42 Q61L: 76 cells; LKB1 CTD-C430S: n = 45 cells; LKB1 CTD-C430S + RhoA Q63L: n = 52 cells; LKB1 CTD-C430S + cdc42 Q61L: n = 62 cells. In Fig. 2D, Empty GFP: n = 103 cells; Empty GFP + Y-27632: n = 110 cells; LKB1 Wildtype: n = 110 cells; LKB1 Wildtype + Y-27632: n = 92 cells; LKB1 CTD: n = 128 cells; LKB1 CTD + Y-27632: n = 101 cells.

In order to study dynamic protein colocalization and the relationships between membrane dynamics and colocalization we developed several enhancements to Machacek and Danuser’s CellEdgeTracker algorithm^35,45. CellEdgeTracker software provides boundary tracking and discretization. We used CellEdgeTracker as a platform to develop Cellular Analysis of Dynamic Events (CADE) which uses tracked discretized boundaries to measure protein colocalization dynamics and associations between colocalization membrane motion dynamics. Time-lapse fluorescent images were first segmented using 3D graph cuts algorithm developed by ref. ⁴⁶ and analyzed with CellEdgeTracker to obtain membrane positions {(x(t), y(t), t)}. The boundary was then divided into a sequence of sectors approximately 10 μm in length. The velocity V of each sector is calculated by equation (1), where d is the displacement vector and n is the normal vector for sector i. Normalized V is plotted in a heat map as a function of time (x-axis) and the position around the cell edge (left y-axis). Heatmap color is used to encode boundary velocity with red, blue and green denoting protrusion (V = 1), retraction (V = −1) and quiescence (V = 0) of the cell edge velocity respectively.

Based on the cell membrane algorithm, we developed our colocalization algorithm to quantify the spatiotemporal colocalization of two proteins. We first erode the binary cell mask to obtain a 2 μm * 10 μm band at the cell edge, and then partition this band into N contiguous regions where protein expression G, R and colocalization signals between green and red channels coloc(G,R) can be quantified as in equations (2),(3),(4), where and denote the average background signals from green and red channels, respectively. μ_G and μ_R are the mean value of green and red signals. Cov and Var are the covariance and the variance of the mean subtracted green and red signals, respectively. The normalized colocalization is finally plotted as a heat map both in time and spatial domain. Red indicates high colocalization (coloc = 1) and blue indicates low colocalization (coloc = 0).

To investigate the spatiotemporal relationship C between membrane dynamics and protein colocalization, we calculated the dot product between V and coloc using equation (5), where N and t are the number of rows and columns in velocity heat maps. C is also normalized according to the maximum and minimum C values among all cells. Similarly as velocity and colocalization heat maps, normalized C is plotted into heat maps. Red indicates highly coupled protrusion and high colocalization (C = 1); Orange indicates highly coupled protrusion and low colocalization (C = 0.5); Green indicates coupling between quiescence and low colocalization (C = 0); Cyan indicates coupling between retraction and low colocalization (C = −0.5) and blue indicates highly coupled retraction and high colocalization (C = −1).

In Fig. 3, significance was initially measured using a Kruskal-Wallis test with a p-value of 0.05, which showed statistical significance. A Dunn’s multiple comparisons test was then performed to compare each construct to empty GFP control and its farnesylation mutant partner with a p-value of 0.05, where **p ≤ 0.01; ***p ≤ 0.001; ****p ≤ 0.0001. In Fig. 3b, Empty GFP: n = 16 cells; LKB1 Wildtype: n = 17 cells; LKB1 C430S: n = 16 cells; LKB1 K78I: n = 15 cells; LKB1 K78I-C430S: n = 14 cells; LKB1 CTD: n = 14 cells; LKB1 CTD-C430S: n = 15 cells. In Fig. 3e–f, LKB1 Wildtype: n = 9 cells; LKB1 K78I: n = 7 cells; LKB1 CTD: n = 5 cells.

We developed a semi-automated graphical user interface to facilitate measuring protrusion/retraction regions on the membrane velocity heatmap. Several regions of interest (ROIs) were manually selected on the heatmap to represent the protrusion/retraction regions using CROIEditor algorithm written by Jonas Reber (http://www.mathworks.com/matlabcentral/fileexchange/31388-multi-roi-mask-editor-class/content/CROIEditor/CROIEditor.m). Then these ROIs were segmented into binary masks. Several morphology features for each ROI was calculated using regionprops function in Matlab. The features for each ROI are the number of ROIs, the area of the ROI, angle of the ROI from major axis to the x-axis, and the duration time, which is the length in x-axis converted to time unit. Since there is no standard definition of a protrusion/retraction region, we asked multiple users to select the ROIs for the same heatmap, similar to refs ^35,45,47 and ⁴⁸. We then performed a one-way ANOVA test amongst all users to check reproducibility and found the p-value larger than 0.05, which indicated the selection among different users was similar. The ROIs of the multiple users was then averaged for statistical analysis. Colocalization duration and events/cell/minute (Fig. 3e–f) was compared between empty GFP control, LKB1 K78I, and LKB1 CTD using a Kruskal-Wallis test with a p-value of 0.05, which showed statistical significance. A Dunn’s multiple comparison test was then performed to compare each construct to each other with a p-value of 0.05. In Figs 4 and ​and5,5, membrane event duration, angle, and number of events/cell/minute was measured using a Kruskal-Wallis test with a p-value of 0.05, which showed statistical significance. A Dunn’s multiple comparisons test was then performed to compare each construct to empty GFP control and its farnesylation mutant partner with a p-value of 0.05. In Figs 4f and ​and5d,5d, a 2-tailed Chi-squared analysis was performed with a p-value of 0.05. *p ≤ 0.05; **p ≤ 0.01, ***p ≤ 0.001, ****p ≤ 0.0001. Values were determined over three separate experiments. In Fig. 4, Empty GFP: n = 7 cells; LKB1 Wildtype: n = 9 cells; LKB1 C430S: n = 8 cells; LKB1 K78I: n = 8 cells; LKB1 CTD: n = 5 cells; LKB1 CTD-C430S: n = 7 cells. In Fig. 5, Empty GFP: n = 7 cells; Empty GFP + FAKin: n = 5 cells; LKB1 K78I: n = 8 cells; LKB1 K78I + FAKin: n = 5 cells; LKB1 CTD: n = 5 cells; LKB1 CTD + FAKin: n = 5 cells.

Volocity (PerkinElmer) image analysis software was used to quantify fluorescent paxillin sites. A ROI was drawn around a single lamellipodia and the image was cropped around the ROI. The images were then thresholded and the smoothing filter was used. Automatic detection of objects was employed and objects less than a volume of 0.1 μm³ were excluded from the analysis. We manually identified the cell boundary for each image and used this for the analysis. Fluorescent site area, elongation, and distance from the membrane were then analyzed and recorded. The site area, elongation, and distance to membrane of each HeLa GFP-LKB1 cell line was measured using a Kruskal-Wallis test with a p-value of 0.05, which showed statistical significance. A Dunn’s multiple comparison test was then performed to compare each construct to empty GFP control and LKB1 wildtype with a p-value of 0.05, *p ≤ 0.05; **p ≤ 0.01, ***p ≤ 0.001, ****p ≤ 0.0001. N = 10 cells/condition over 3 separate experiments.

Volocity (Perkin Elmer, Waltham, MA) image analysis software and manual tracking was used to quantify meandering index and percent area invaded. An ROI was manually drawn around the scratch wound at 0, 3, 6, 9, 12, 15, 18, and 21 hours, and the percent area closed was quantified over time. Meandering index (displacement/distance) and area closed was measured using a Kruskal-Wallis test with a p-value of 0.05, which showed statistical significance. A Dunn’s multiple comparisons test was then performed to compare each construct to empty GFP control and LKB1 wildtype with a p-value of 0.05. Wound closure passed the assumptions of parametric tests (homogeneity of variance) and was thus measured using ANOVA, which showed statistical significance. We then performed a Sidak’s multiple comparisons test to compare each construct to empty GFP control and LKB1 wildtype at 3 and 6 hours with a p-value of 0.05. * ≤ 0.05; **≤0.01; ***≤0.001; ****≤0.0001. n = 30 cells, 3 scratches/condition, over 3 separate experiments.

Non-confluent cells were imaged, and image analysis using Imaris Cell (Bitplane, South Windsor, CT) was performed using the cell tracking function. Cytoplasmic LKB1 was identified using phalloidin as a cytoplasmic marker, while nuclear LKB1 was identified using DAPI. Mean intensity of LKB1 was quantified in both the cytoplasmic and nuclear regions, and the cytoplasmic:nuclear ratio was determined for each cell. C:N ratio was compared between wildtype LKB1 and the various constructs, as well as the isogenic parental and farnesyl-mutant condition (WT vs C430S, K78I vs K78I-C430S, CTD vs CTD-C430S), using the Dunn’s multiple comparisons test with a p-value of 0.05. *≤0.05; ***≤0.001; ****≤0.0001. Empty GFP: n = 15 cells; LKB1 Wildtype: n = 21 cells; LKB1 C430S: n = 20 cells; LKB1 K78I: n = 17 cells; LKB1 K78I-C430S: n = 16 cells; LKB1 CTD: n = 21 cells; LKB1 CTD-C430S: n = 20 cells.