Fig 1.
A: Example of single HeLa cell expressing mCherry::histone 2B, grown on an L-shaped micropattern coated with fibronectin Alexa fluor 555, and monitored using time-lapse microscopy. Top: raw data from time-lapse microscopy experiment; time is indicated in hours; bottom: corresponding schematic representations. Note that the contours of the cell are shown in the schematics solely for illustration purposes and are not monitored experimentally. After seeding, the interphase cell becomes oriented on the micropattern such that the nucleus is positioned centrally between the two arms of the L. Upon cell rounding in mitosis, the metaphase plate is positioned with an angle of ~45° with respect to the arms of the L-shaped micropattern. B: Example of visual field of cells grown in a 96-well plate coated with L-shaped micropatterns, monitored using time-lapse microscopy; time is indicated in hours. The regions highlighted with the blue insets in the lower magnification views are enlarged below to illustrate the different classes of cells that must be discriminated by TRACMIT: green (analyzable): cell correctly positioned on the micropattern and dividing during the recording; red [1], empty: no cell on the micropattern; red [2], crowded: more than one cell on the micropattern; red [3], bad position: cell on one arm of the L; red [4], not dividing: cell not dividing during the recording. Scale bar in insets = 10μm.
Fig 2.
Five blocks that characterize TRACMIT (preprocessing, pattern detection & filtering, anaphase detection, prometaphase backtracking, export results) are shown, together with the steps that they contain (numbered 1 through 18), as well as their computer implementation (left) and an illustration thereof (right). See text for details.
Fig 3.
Assessment of key steps implemented in TRACMIT.
A: Comparison of frames aligned without SIFT (left) or using SIFT (right). Note tighter area with the latter. In this panel, as well as in panels B and C, the insets correspond to those shown in Fig 2, step 1. B: Illustration of Fig 2, steps 6, 7 and 8, exclusion of micropatterns that are empty or contain more than one cell. Shown are z-projections of 180 frames. StDev: standard deviation of pixel intensities, used to determine cell movement, thereby reflecting the number of cells on the pattern. Average pixel intensity (middle panel): used as denominator in order to transform absolute StDev values into relative values (right panel). In this way a common bandpass threshold can be applied on ROIs with highly variable fluorescence intensities, and thus StDevs. C: Comparison of parameters used for calculation of LoG. Upper 4 images: smoothing values of 1, 2.5, 4 or 10. Lower panels: variable thresholds (Fig 2, step 9). Values boxed in green show optima that enable best recognition of anaphase figures. Red pixels in bottom row are above the applied threshold. D: Comparison of two local threshold methods to spot neighboring cells after anaphase detection (Fig 2, step 13). This step is challenging because the mCherry::histone 2B signal of neighboring cells varies significantly. Shown is a receiver-operator (ROC) curve comparing default and intermodes threshold calculations methods for 56 micropatterns with detected anaphases amongst ctrl and LGN siRNA conditions (95% confidence interval). An ideal ROC curve should show 100% true positives and 0% false positives (green dashed line). Note that the default threshold method efficiently excludes micropatterns with more than one cell, but also excludes a significant fraction of micropatterns that are properly occupied, thereby decreasing the detection of true positives (Y axis, true positives). Therefore, we used instead the less stringent intermodes threshold method, which increases sensitivity of detection at the cost of including more false positives with more than one cell per micropattern.
Fig 4.
TRACMIT validation and performance.
A: Correlation of “analyzable” cells detected per well (n), comparing manual analysis with TRACMIT; 209 siRNA conditions (wells), >9000 cells in total. Shown are high magnification views from plates with "less filled" and "very filled" conditions, extracted from the graph below, as indicated. Pearson correlation coefficient, R2 = 0.52, p<0.0001. B: Correlation of angle measurements between manual analysis and TRACMIT. Pearson correlation coefficient, R2 = 0.99, p<0.0001. C: Proof of principle: spindle positioning phenotype (= % of cells with mispositioned spindle per well, i.e. a metaphase plate with an angle >40° off from the normal 45° position) in wells treated with control (ctrl) or LGN siRNA, comparing manual analysis and automated analysis with TRACMIT. N = 8 wells per condition (15–40 cells per well). Two tailed Student’s t-test *** p<0.001.