Automated analysis of cell migration and nuclear envelope rupture in confined environments

Abstract

Recent in vitro and in vivo studies have highlighted the importance of the cell nucleus in governing migration through confined environments. Microfluidic devices that mimic the narrow interstitial spaces of tissues have emerged as important tools to study cellular dynamics during confined migration, including the consequences of nuclear deformation and nuclear envelope rupture. However, while image acquisition can be automated on motorized microscopes, the analysis of the corresponding time-lapse sequences for nuclear transit through the pores and events such as nuclear envelope rupture currently requires manual analysis. In addition to being highly time-consuming, such manual analysis is susceptible to person-to-person variability. Studies that compare large numbers of cell types and conditions therefore require automated image analysis to achieve sufficiently high throughput. Here, we present an automated image analysis program to register microfluidic constrictions and perform image segmentation to detect individual cell nuclei. The MATLAB program tracks nuclear migration over time and records constriction-transit events, transit times, transit success rates, and nuclear envelope rupture. Such automation reduces the time required to analyze migration experiments from weeks to hours, and removes the variability that arises from different human analysts. Comparison with manual analysis confirmed that both constriction transit and nuclear envelope rupture were detected correctly and reliably, and the automated analysis results closely matched a manual analysis gold standard. Applying the program to specific biological examples, we demonstrate its ability to detect differences in nuclear transit time between cells with different levels of the nuclear envelope proteins lamin A/C, which govern nuclear deformability, and to detect an increase in nuclear envelope rupture duration in cells in which CHMP7, a protein involved in nuclear envelope repair, had been depleted. The program thus presents a versatile tool for the study of confined migration and its effect on the cell nucleus.

Fig 1. Cell migration through microfluidic constrictions.

(A) Cells expressing NLS-GFP and H2B-tdTomato migrating through a microfluidic device. Scale bar: 50 μm. (B) Time series of a nucleus squeezing through a constriction. Scale: bar 20 μm. (C) Time series of a NE rupture event. NLS-GFP leaks into the cytoplasm upon NE rupture and is reimported into the nucleus as the NE is repaired. Scale bar: 20 μm.

Fig 2. Flowchart of automated analysis steps.

Steps the program takes when analyzing an image sequence are detailed, including image processing (top) and post-processing (bottom).

Fig 3. Examples of nuclear identification and tracking.

(A) Merged image of the transmitted light and tdTomato channels. Nuclei (red) can be seen in the migration device. Scale bar: 50 μm. (B) Binarized version of red channel of image A. Each nucleus is identified as a separate object (white). (C) Example of tracking results. Nuclei (red) have been identified, and their centroid positions during migration are shown as yellow tracks. For clarity, tracks displayed here are limited to data for only the last six hours.

Fig 4. Detection of NE rupture.

(A) During NE rupture (arrow), NLS-GFP (green) spreads throughout the cytoplasm, causing the nuclear NLS-GFP signal to lose intensity. In contrast, the H2B-tdTomato signal (red) remains approximately constant. (B) Normalizing these two signals against one another (H2B/ NLS) significantly reduces the effects of noise and allows for more accurate NE rupture detection. (C) Steep increases in the H2B/NLS ratio, which correspond to high values of Δ(H2B/NLS), plotted in (D), indicate the start of a NE rupture event. The data shown here are for a representative cell.

Fig 5. Verification of automated image analysis by comparison to manual analysis.

(A) Automated analysis results plotted against manual analysis results (mean ± s.e.m. from four observers) for individual cells in two separate image sequences, each of which corresponds to a single section of a microfluidic device. For perfect agreement, the regression line plotted through these points would have a slope of one. Only one automated-analysis result substantially deviated from the manual reference, indicated by an asterisk. The manual analysis determined the nucleus to make two attempts to pass through the constriction, failing the first but succeeding the second time. The program identified this as a single, longer attempt. (B) Constriction transit times (mean ± s.e.m.) determined by four manual analysts and the automated analysis for the cells in the same two image sequences analyzed for panel A. Cells are BT-549 breast cancer and are either overexpressing lamin A (Image sequence 1) or an empty vector (Image sequence 2). Overexpression of lamin A results in less deformable nuclei and longer transit times through narrow constrictions (*, p < 0.05; **, p < 0.001 as calculated by ANOVA followed by Tukey’s multiple comparison test; n = 23–26 and 20–24 (depending on the analyst), respectively).

Fig 6. Application of automated image analysis program.

(A) A549 cells depleted for lamin A/C (n = 40 cells across 4 microfluidic device sections) pass faster through small constrictions than non-target controls (n = 26 cells across 4 device sections). Transit times through larger openings were not statistically different (*, p < 0.05 as calculated by t-test; n = 21, 14, respectively). (B) HT-1080 cells depleted for CHMP7 (n = 65 cells across 3 device sections) took longer to repair their NE and restore nucleo-cytoplasmic compartmentalization than non-target controls (**, p < 0.01 as calculated by Mann-Whitney t-test; n = 48 cells across 3 device sections). (C) Dynamics of NE rupture and repair visualized by the ratio of nuclear H2B-tdTomato/NLS-GFP fluorescence for the CHMP7-depleted and non-target control cells. The H2B/NLS signal is expressed relative to its value at t = 0, i.e., immediately prior to rupture, and normalized to reach a peak value of 1 for each nucleus. CHMP7-depletion results in slower return to baseline, indicating delay in NE repair. Error bars represent mean ± s.e.m.