Monitoring Actin Cortex Thickness in Live Cells

Abstract

Animal cell shape is controlled primarily by the actomyosin cortex, a thin cytoskeletal network that lies directly beneath the plasma membrane. The cortex regulates cell morphology by controlling cellular mechanical properties, which are determined by network structure and geometry. In particular, cortex thickness is expected to influence cell mechanics. However, cortex thickness is near the resolution limit of the light microscope, making studies relating cortex thickness and cell shape challenging. To overcome this, we developed an assay to measure cortex thickness in live cells, combining confocal imaging and subresolution image analysis. We labeled the actin cortex and plasma membrane with chromatically different fluorophores and measured the distance between the resulting intensity peaks. Using a theoretical description of cortex geometry and microscopic imaging, we extracted an average cortex thickness of ∼190 nm in mitotic HeLa cells and tested the validity of our assay using cell images generated in silico. We found that thickness increased after experimental treatments preventing F-actin disassembly. Finally, we monitored physiological changes in cortex thickness in real-time during actin cortex regrowth in cellular blebs. Our investigation paves the way to understanding how molecular processes modulate cortex structure, which in turn drives cell morphogenesis.

Introduction

The shape of animal cells is primarily determined by the cell cortex, a cross-linked network of actin, myosin, and associated proteins that lies directly underneath the plasma membrane (1,2). The cortex enables the cell to resist externally applied forces and plays a central role in cell shape change. Local modulation of cortex mechanics has been shown to drive cell deformations during division, migration, and tissue morphogenesis, under the control of precisely regulated molecular pathways (3–6). Molecular regulators determine key mechanical properties of the cortex, such as tension and viscoelasticity, by changing the spatial organization of the cortical network (7–10). Thus, understanding the regulation of cell morphogenesis requires understanding cortex network architecture. However, almost nothing is known about the spatial arrangement of cortical actin, and even the most basic parameter, cortex thickness, has not been directly measured in live cells.

Transmission electron microscopy studies suggest a cortex thickness of ∼100 nm in Dictyostelium discoideum (11) and in retracting blebs in human melanoma cells (12). Although sample preparation for electron microscopy can perturb actin networks, these studies indicate that cortex thickness is below the resolution limit of conventional light microscopes, and close to that achieved by contemporary superresolution setups (13). As a result, the resolution of cortex structure using contemporary imaging techniques is challenging, and the contribution of changes in thickness to cortex-driven deformations is poorly understood.

To address this, we have developed a method to measure cortex thickness in live cells. Our method is inspired by single-molecule high-resolution colocalization (SHREC), which has been used to investigate the relative positions of single proteins and protein clusters (14–16). SHREC takes advantage of the fact that although the spatial dimensions of an object below the resolution limit cannot be resolved, the position of a point-like object can be determined with nanometer precision, provided a high signal/noise ratio (17). Here, we expand upon this technique and apply it to the study of a non-point-like (i.e., extended) object, enabling us to infer cortex thickness from the relative localization of cortical actin and the plasma membrane. Specifically, we label the cortex and plasma membrane with chromatically different fluorophores and develop a theoretical framework relating the relative positions of the resulting intensity peaks to cortex thickness. We then validate our method using computer-generated cell images. We show that perturbing actin depolymerization in live cells leads to an increase in cortex thickness. Finally, we monitor cortex thickness dynamics at the membrane of cellular blebs and find that cortex thickness increases during bleb retraction, demonstrating that our method can be used to investigate thickness changes during live cell deformations.

Materials and Methods

Cell culture and experimental treatments

HeLa cells were cultured and treated as described in detail in the Supporting Material. The GFP-Actin HeLa line was a gift from the lab of Frank Buchholz. Wild-type HeLa cells were a gift from the MPI-CBG Technology Development Studio (Dresden, Germany). Detailed information about plasmids and treatments can be found in the Supporting Material. EGFP-CAAX was a gift from J. Carroll. EGFP was replaced with mCherry by restriction digest by M. Bergert to create the mCherry-CAAX fusion. The Lifeact-EGFP plasmid was a gift from R. Wedlich-Söldner. The Lifeact-mCherry plasmid was a gift from N. Herold (lab of H.G. Kraeusslich). Methyl-β-cyclodextrin (Sigma-Aldrich, Hamburg, Germany) and Jasplakinolide (Invitrogen/Life Technologies, Darmstadt, Germany) were added to final concentrations of 5 mg/mL and 20 nM, respectively, ∼20 min to ∼1 h before imaging. CFL1 siRNA was transfected at 10 nM 48 h before imaging. Knock-down was performed in parallel for imaging experiments and Western blotting, as described in the Supporting Material.

Confocal imaging, ablation, and chromatic shift correction

For all live cortex thickness measurements save for bleb experiments, two-color image stacks around the cell equator of ∼30–70 z slices were collected for each cell with 0.1 μm step size. For chromatic shift correction, 200 nm diameter multicolor beads (Tetraspeck microspheres; Invitrogen/Life Technologies) were imaged using settings as for cell imaging. Chromatic shift was calculated and corrected for using Huygens Professional software (Scientific Volume Imaging, Hilversum, The Netherlands). After correction, a single equatorial plane for each cell image was selected using FIJI image analysis software (18). For bleb experiments, the average chromatic shift vector was determined before imaging to enhance acquisition speed; two z slices separated by 0.35 μm (the average z shift) were acquired and then aligned using FIJI. Microscope specifications, ablation settings, and red/green magnification correction are described in the Supporting Material. Contrast was adjusted for the example images shown in the figures for display purposes. Neither contrast nor brightness levels were adjusted before image analysis.

Cell segmentation and linescan parameter extraction

Image segmentation and extraction of linescans were performed in a semiautomated fashion using custom image analysis software written in PYTHON (www.python.org). For initial segmentation, pixels corresponding to the location of the membrane are first chosen based on intensity thresholding with high tolerance. Inappropriate points, such as those in the cytoplasm or in microvilli (MV), are then iteratively removed. After manual correction, points are fit to a Fourier decomposition with a variable number of modes to accurately account for high spatial frequencies when appropriate, while still providing a smooth and continuous contour. After segmentation, linescans perpendicular to the contour are generated by linear interpolation of surrounding pixels. A new linescan is drawn at each point where the contour encounters a new pixel in the original image, resulting in a total of ∼1000–2000 linescans per cell image (∼20 linescans per μm around the cell border). The resulting linescans have a pixel size equivalent to the original image and are 100 px in length for all experiments in this work. To generate average linescans, the mean intensity at corresponding positions from individual linescans is calculated. For binned linescans (see Fig. S3 in the Supporting Material), individual linescans are grouped into overlapping bins of different sizes and the mean intensity at corresponding positions is calculated. Bin size corresponds to the number of linescans that are averaged; bin size can also then be expressed as the length of the cell border along which linescans are averaged. For bin sizes <200 linescans, bins overlap by one-half of the bin size; for bin sizes ≥200 linescans, each bin is separated by 100 linescans.

Peak positions and intensities are determined by fitting a Gaussian function to the five points surrounding and including the pixel with highest intensity. Intracellular and extracellular background intensities are determined by taking a mean of 10 points starting at 1 μm to the left and right of the fit peak position, respectively (see Fig. S1).

Cortex thickness extraction

Extraction of cortex thickness h and cortex intensity i_c was performed by least-squares fitting. Values for h were constrained between 0 and 1000 nm and i_c was constrained between i_in and 500 AU. During fitting, guesses for h and i_c were used to determine x_c and i_p using Eqs. 1, 2, and S6 (see the Supporting Material) with σ (see Fig. S2) and the following input parameters obtained from linescans: i_in, i_out, and x_m. The differences between x_c and i_p from fits and x_c and i_p extracted from linescans were the residuals minimized during fitting. For bleb cortex extraction, constraints for h and i_c were broadened during fitting to allow extraction of lower values of h.

Synthetic image generation and analysis

High resolution two-dimensional synthetic images with an effective pixel size of 4 nm were generated as described in the Supporting Material. After convolution with a two-dimensional symmetric Gaussian with zero mean and standard deviation σ = 170 nm, these images were downsampled to an effective pixel size of 69 nm to match experimental imaging parameters. An additional frame containing an image of the cell contour used to draw the synthetic image was generated. This contour frame was used for cell segmentation using the custom image analysis software. We did not observe a significant difference in the extracted values if we used a semiautomated segmentation (as for experimental images) instead of using the contour employed for drawing the image (see the Supporting Material and Fig. S8). After linescan acquisition, extraction of linescan parameters and cortex thickness h was performed as described for experimental images above. For extraction of cortex thickness over a range of h values (see Fig. 4 C), constraints for h and i_c were removed during fitting to allow extraction of h under 100 nm.

Figure 4.

Evaluating cortex thickness measurements by model convolution. (A, left) Multiple magnifications of a synthetic cell image capturing main cortex features, drawn with a pixel size of 4 nm. (Middle) Intracellular background intensity is added and the image is convolved with a Gaussian with σ = 170 nm, comparable to the PSF of the microscope. (Right) The image is downsampled to a pixel size corresponding to experimental imaging parameters (69 nm). Gaussian noise is added to mimic experimental noise. (Insets are denoted by boxes.) Scale bar = 10 μm. (B) Cortex-PM peak separation, Δ, and cortex thickness, h, for synthetic images plotted as a function of background intensity ratio, i_out/i_in. (Dotted lines) Linear fits to the data. (C) Plot comparing h extracted from linescan analysis to the cortex thickness values specified during synthetic image generation (h imposed) for synthetic cell images with i_out/i_in = 0.25. (Points) Mean ± SD for n ≥ 25 synthetic cells for each imposed thickness. The identity function (gray line) is plotted for comparison.

Results

Subresolution imaging of the actin cortex and plasma membrane

To measure cortex thickness, h, we determined the relative positions of the actomyosin cortex and the plasma membrane (PM; Fig. 1 A). To this aim, we coexpressed GFP-Actin and the plasma membrane-labeling probe mCherry-CAAX (19) in HeLa cells and blocked the cells in prometaphase using the Eg5 inhibitor S-Trityl-L-cysteine. At this stage in the cell cycle, HeLa cells display a rounded morphology and have a particularly well-defined cortex (Fig. 1 B). We acquired two-color confocal stacks around the cell equator and corrected image stacks for chromatic aberration (see Materials and Methods). We then segmented the cell boundary and acquired fluorescence intensity linescans perpendicular to the resulting contour using custom-made image analysis software.

Figure 1.

Relative localization of the actin cortex and plasma membrane. (A) Schematic representation of the experimental approach. The actin cortex and plasma membrane are labeled with chromatically different fluorophores (here, GFP-Actin and mCherry-CAAX, respectively; top). From the fluorescence intensity linescans across the cell border (bottom), the distance between the fluorescence intensity peaks of the two probes, Δ, can be measured and related to cortex thickness, h. (B) A representative HeLa cell stably expressing GFP-Actin and transfected with mCherry-CAAX blocked in prometaphase and imaged by confocal microscopy. (Inset) High-zoom image of area in box. (C) Linescans orthogonal to the cell border obtained based on segmentation of the cell shown in panel B. Scale bars for panels B and C = 10 μm. Higher zoom images for regions with many (a) or few (b) microvilli are shown in panel C′, with actin, PM, and merge images in the same order as in panel C. Scale bars for panel C′ = 2 μm. (D) Average linescans of actin and the PM, calculated by averaging individual linescans from panel C. Fluorescence peak position and intensity were determined by fitting several points around the pixel with maximum intensity (blue lines).

This provided a straightened image of the cell border (Fig. 1, C and C′; see also the Supporting Material). From this straightened image, we computed an average linescan across the cortex by taking the mean of the individual linescans along the cell contour. The resulting average linescans displayed Gaussian-like peaks of width comparable to the resolution of the microscope (Fig. 1 D, and see Fig. S2), indicating that the thicknesses of both the actin cortex and the PM are below the resolution limit. After accounting for differences in magnification between the red and green channels (see the Supporting Material), we localized the positions of the intensity peaks for actin and CAAX at subpixel resolution by fitting several points around the peak (Fig. 1 D). Finally, we determined the separation between the positions of the cortex and PM linescan peaks, Δ, to be 128 ± 28 nm (mean ± SD, n = 44 cells).

The separation between the actin cortex and plasma membrane peaks depends on the thickness of the cortical network, h (Fig. 1 A). However, as a result of diffraction, Δ also depends on relative background intensities. Due to the presence of cytoplasmic actin monomers, intracellular background, i_in, is higher than extracellular background, i_out, in the actin linescan (〈i_out/i_in〉 = 0.37 ± 0.13 [± SD]; Fig. 1, B and D). Thus, the actin intensity peak will be shifted toward the cytoplasm. To evaluate the extent to which relative background intensities affected Δ, we added Alexa488-Dextran to the cell medium, thus changing i_out/i_in without perturbing cortex thickness or the actin distribution. We first imaged a single mitotic HeLa cell, progressively adding increasing amounts of Alexa488-Dextran (Fig. 2 A). We observed that Δ was highly dependent on i_out/i_in; Δ decreased from ∼110 nm to ∼60 nm as i_out/i_in increased from ∼0.4 to ∼2.0 after addition of external dye (Fig. 2 B). We further probed the relationship between Δ and i_out/i_in by measuring Δ for many cells in the presence of extracellular fluorescent dye and observed a significant correlation between Δ and i_out/i_in (Fig. 2 C). Together, these observations show that the relative intra- and extracellular background intensities must be accounted for to extract cortex thickness.

Figure 2.

A theoretical description of cortex geometry accounts for the dependence of peak separation on background intensities. (A) Images of a single HeLa cell expressing GFP-Actin and mCherry-CAAX, blocked in prometaphase and in culture medium alone (left panel) or in the presence of increasing concentrations of extracellular Alexa488-Dextran (three right-most panels). Scale bar = 10 μm. Note: Even for the case where the background ratio, i_out/i_in, was as high as 2.0 (right panel), the peak could be fit; in this case, the ratio of peak intensity to extracellular background, i_p/i_out = 1.37. (B) Plot of the separation between the actin and PM intensity peaks, Δ, as a function of i_out/i_in. (Points) Experimental data. The leftmost point (i_out/i_in = 0.4) corresponds to no extracellular Alexa488-Dextran. (Dotted line) Δ values calculated from Eqs. 1 and 2 varying i_out and fixing h (200 nm), i_c (225.5 AU), and i_in (43.66 AU; see the Supporting Material for details). (C) Plot of Δ as a function of background ratio i_out/i_in for a population of mitotic HeLa cells in the presence of differing amounts of Alexa488-Dextran. Pearson correlation coefficient, r = −0.674 (p < 0.001). (D) A linescan reflecting a simplified model of cortex geometry (blue) and the resulting linescan after convolution to mimic the imaging process (green). The peak intensity of the convolved linescan, i_p, is lower than i_c, and the position of the peak, x_c, is shifted toward the side of higher background intensity (here, the cytoplasm) with respect to the center of the cortex, X_c. (E) Cortex thickness, h, extracted from the population of cells in panel C. The extracted h values were not significantly correlated with the background intensity ratio. r = −0.117 (p = 0.554).

Extraction of cortex thickness

We next sought to relate cortex thickness, h, to our measurements of Δ. To this aim, we developed a theoretical framework relating Δ to cortex geometry and accounting for the influence of background fluorescence levels on the position of the intensity peak of cortical actin. Electron microscopy observations suggest that the cortex is uniformly dense, decays abruptly at the cytoplasm interface, and is in direct contact with the plasma membrane (11). We thus approximated the actin distribution near the cell surface by a cortex of thickness h with homogeneous intensity, i_c, and a constant intracellular background, i_in (Fig. 2 D). Extracellular background, i_out, was also taken as constant. We mimicked the effect of diffraction during imaging by convolving this actin distribution using a Gaussian function with zero mean and standard deviation σ, reflecting the point spread function (PSF) of the microscope. The value of σ was determined experimentally to be ∼170 nm by imaging subresolution beads (see Fig. S2).

Convolution yields a linescan similar to experimentally measured linescans (Fig. 2 D). The position of the intensity peak of the convolved linescan, x_c, is shifted with respect to the center of the underlying actin cortex, X_c, by

$$\delta =X_c-x_c=\frac{{\sigma }^2}{h}\text{ln}\left(\frac{i_{\text{out}}-i_c}{i_{\text{in}}-i_c}\right)$$ (1)

(see the Supporting Material for details). Given that the internal and external background intensities are approximately equal in the PM linescan and that peak intensity, i_p, is much higher than background (〈i_out/i_in〉 = 1.31 ± 0.28 and 〈i_p/i_out〉 = 4.21 ± 1.41 [±SD] for mCherry-CAAX; Fig. 1, B and D), the peak position of the membrane, x_m, accurately reflects the effective center of the PM, X_m. Assuming that the thickness of the PM (∼5 nm) is negligible compared with that of the cortex leads to the following implicit definition of cortex thickness:

$$h=2\left(X_m-X_c\right)=2\left(\Delta -\delta \right),$$ (2)

where X_m > X_c (i.e., the membrane is located to the right of the cortex). We verified that this simplified model was sufficient to account for the observed dependence of Δ on relative background intensities (Fig. 2, A and B). Fitting the observed relationship between Δ and i_out/i_in with the equation Δ = h/2 + δ, derived from Eqs. 1 and 2, we could faithfully reproduce the data (Fig. 2 B, dotted line).

We then used this theoretical framework to extract h from single average linescans (Fig. 1 D). Because subresolution information is lost during imaging, directly fitting the actin intensity profile to our model (see Eq. S6 in the Supporting Material) results in nonunique sets of fit parameters. However, by imposing the constraint that the position of the PM coincides with the outer edge of the cortex (Eq. 2), we effectively make X_c and h dependent. This removes one independent variable for fitting, allowing us to robustly extract h. All of the parameters appearing in Eqs. 1 and 2, with the exception of h and i_c, can be directly obtained from the linescans. Fitting the equations to the data to extract h and i_c yields an average cortex thickness 〈h〉 = 186 ± 66 nm (±SD, n = 42 cells; see Materials and Methods and the Supporting Material for details).

To determine whether the value of h was independent of relative background levels, we determined cortex thickness values for the population of HeLa cells treated with differing amounts of Alexa488-Dextran (Fig. 2 C). Unlike Δ, the extracted values of h were not significantly correlated with i_out/i_in (Fig. 2 E). We also found that the average cortex thickness value from dye-addition experiments (〈h〉 = 159 ± 48 nm [±SD, n = 28 cells]) was not significantly different from the 〈h〉 values from experiments with no external dye (186 ± 66 nm [±SD, n = 42 cells]).

To determine if our procedure for h extraction could also account for changes in intracellular background, we analyzed cells expressing the F-actin binding probe EGFP-Lifeact, which displays lower relative intracellular background than GFP-Actin (for EGFP-Lifeact, 〈i_out/i_in〉 = 0.50 ± 0.16, 〈i_p/i_out〉 = 2.57 ± 0.43; for GFP-Actin, 〈i_out/i_in〉 = 0.37 ± 0.13, 〈i_p/i_out〉 = 1.86 ± 0.35 [±SD]; compare Fig. 3, A and B, and Fig. 1, B and D). Consistent with the differences in intensity ratios, in cells labeled with EGFP-Lifeact, we observed a significantly lower $\Delta$ compared to cells expressing GFP-Actin (Fig. 3 C). However, the 〈h〉 values were not significantly different (Fig. 3 D). Taken together, our data strongly suggest that we can extract cortex thickness values that do not depend on relative intra- or extracellular background intensities and can thus be used to compare thickness values across different experimental conditions.

Figure 3.

Thickness extraction is robust to exchange of fluorescent probes. (A) Prometaphase HeLa cells expressing the F-actin-binding probe Lifeact and the PM probe CAAX. Scale bar = 10 μm. (B) Average linescans of GFP-Lifeact and mCherry-CAAX signal for cell in A (top panels). (C) Box plot comparing actin-PM intensity peak separation, Δ, between cells expressing GFP-Actin/mCherry-CAAX and GFP-Lifeact/mCherry-CAAX. ^∗∗∗p < 0.001 by Welch t-test. (D) Box plot comparing extracted cortex thickness, h, between cells expressing GFP-Actin/mCherry-CAAX (Actin/PM), EGFP-Lifeact/mCherry-CAAX (Lifeact/PM) and EGFP-CAAX/mCherry-Lifeact (PM/Lifeact [color swap]). p > 0.05 for all categories by Welch t-test.

Evaluation of noise and errors in cortex thickness measurements

To test our correction for chromatic aberration, we exchanged the colors of fluorophores used to measure cortex thickness. We thus expressed EGFP-CAAX and mCherry-Lifeact in HeLa cells (Fig. 3 A). After extraction of h, we found no significant difference in cortex thickness between cells expressing EGFP-Lifeact/mCherry-CAAX or EGFP-CAAX/mCherry-Lifeact (Fig. 3 D). This result indicates that we successfully account for chromatic aberration.

We next asked whether our thickness measurements were dependent on the length of the cell border over which we averaged individual linescans. To this aim, we split the linescans around the cell periphery into bins of variable size, averaged the linescans within each bin, and measured cortex thickness (see Fig. S3, A and B). We found that the mean thickness did not depend on bin size (see Fig. S3 C), and the standard deviation decreased with increasing bin size (see Fig. S3, C and D). For the smallest bin size tested, 50 linescans (∼2.5 μm), the SD in our h measurements reached ∼70–80 nm (see Fig. S3 D).

Our simplified model of cortex thickness assumes that actin intensity in the cortex is uniform. To test how a different cortical actin distribution could affect our measurements, we modified our theoretical description of the underlying cortex geometry (Fig. 2 D, blue curve), incorporating a linearly decreasing intensity gradient from the plasma membrane to the cytoplasm (see Fig. S4 A; see the Supporting Material for details). After convolution of cortex profiles with varying gradient steepness, the resulting convolved linescans were similar in shape (see Fig. S4 B). Even if cortex intensity decreased from the membrane toward the cytoplasm by 50%, our procedure yielded thickness values within ∼10% of the imposed thickness (see Fig. S4 C; see the Supporting Material).

Evaluating cortex thickness measurements by model-convolution

To verify our determination of h and our image analysis tools, we simulated microscopic images of the cell cortex using a model-convolution approach (reviewed in Gardner et al. (20)). We first generated high-resolution two-dimensional images of simplified cells that captured major features of the actin cortex (Fig. 4 A). The parameters used to generate synthetic images were chosen to accurately reflect experimental images and measurements and where possible, were based on electron microscopy studies (see the Supporting Material and Table S1 in the Supporting Material). Simulated cells were drawn at a resolution of 4 nm per pixel and contained an actin cortex of overlapping filaments and an overlying, undulated plasma membrane (Fig. 4 A, left). We then added background intensity inside the cell and convolved these high-resolution images with a Gaussian kernel with zero mean and σ = 170 nm, similar to the PSF of the microscope (Fig. 4 A, middle). To mimic imaging parameters, we downsampled the synthetic images to a 512 × 512 array of pixels of 69 nm and added noise (Fig. 4 A, right; see the Supporting Material for details).

We first used synthetic images to evaluate our image analysis tools and to test our approach to measure h. Using our image analysis software and the segmentation contour used to draw the image, we extracted average cortex and PM linescans and determined h for each synthetic image. When intra- and extracellular background intensities were equal (i_out/i_in = 1), for an imposed thickness of 200 nm, we measured an average separation between actin and PM peaks, Δ, of 105 ± 1 nm (± SD). We extracted an average h of 209 ± 2 nm (± SD), confirming that we were able to accurately extract cortex thickness using our image analysis tools and model. The small difference between the imposed and extracted cortex thickness is largely due to actin-filled undulations in the PM (Fig. 1 C′, and Fig. 4 A). We then generated a series of synthetic images, keeping all parameters constant with the exception of the background intensity ratio, i_out/i_in. Consistent with our experimental results, Δ was strongly correlated with i_out/i_in, and this correlation matched the predictions of our model. In contrast, the extracted cortex thickness h only weakly depended on i_out/i_in (Fig. 4 B). Even with i_out/i_in as low as ∼0.1, we extracted h values within 10 nm of the imposed 200 nm thickness. This analysis indicates that our extraction of h accurately accounts for the effect of differing background intensities.

We next used synthetic images to test the accuracy and precision of our thickness measurements. Accuracy reflects how close the mean of a measurement matches the true value, whereas precision is related to the variance of the measurement (21). We generated synthetic images with cortex thicknesses ranging from 20 nm to 500 nm, with i_out/i_in = 0.25 to best match typical experimental cell images. The mean h values extracted for all imposed thickness values tested, even very thin cortices (<50 nm), were all within 10 nm of the imposed thickness. However, for imposed cortex thickness values lower than 100 nm, the variance was higher than for thicker cortices (Fig. 4 C). Together, this indicates that our method remains accurate for thinner cortices, though the measurements are less precise than for higher cortex thickness values (Fig. 4 C). In summary, using a model-convolution approach, we established that we are able to extract h over a wide range of imposed cortex thickness and background ratios, with a precision and accuracy of ∼10–20 nm for cortices with uniform intensity.

Estimating the influence of microvilli

Another feature that could potentially affect the position of intensity peaks, and thus our measurements of cortex thickness, are microvilli (MV), actin-filled protrusions present at the cell surface (Fig. 1, B, C, and C′). To test the influence of MV on cortex thickness measurements, we first simulated images of cells with varying MV number and maximum length (see Fig. S5 A). After image analysis and thickness determination, we found that a higher number and maximum length of MV caused an increase in Δ (see Fig. S5, B and C). For MV parameters most qualitatively similar to experiments, the error in the extracted h value was <10 nm (see Fig. S5 D). Our synthetic MV analysis also suggested that for higher numbers of MV, our approach may underestimate h if MV are relatively short, but overestimate h if MV are relatively long (see Fig. S5 D). We thus tested experimentally whether the presence of MV affected our measurements of cortex thickness. To this end, we treated mitotic HeLa cells with Methyl-β-cyclodextrin (MBCD), a cholesterol-sequestering compound that has previously been shown to reduce the number of MV (22). Cells treated with MBCD had markedly fewer MV compared to controls, though average linescans were qualitatively similar (see Fig. S6, A and B; compare to Fig. 1, B–D). We did not observe a significant difference in h after MBCD treatment (see Fig. S6 C), strongly suggesting that the presence of microvilli does not influence cortex thickness measurements.

Stabilizing actin increases actin cortex thickness

The thickness of the actin cortex has been proposed to depend on the dynamics of cortical actin filaments (23). We therefore asked whether stabilizing cortical actin could cause an increase in thickness. We treated prometaphase HeLa cells expressing GFP-Actin and mCherry-CAAX with Jasplakinolide, a pharmacological agent that stabilizes actin filaments (24) (Fig. 5 A). Jasplakinolide-treated cells had an ∼50% thicker cortex compared with untreated controls (Fig. 5 C), suggesting that reducing actin disassembly indeed causes an increase in cortex thickness.

Figure 5.

Cortex thickness increases after treatments reducing actin disassembly. (A) Prometaphase HeLa cells expressing GFP-Actin and mCherry-CAAX treated with DMSO or Jasplakinolide (Jas.). Scale bar = 10 μm. (B) GFP-Actin and mCherry-CAAX-expressing HeLa cells after transfection of scrambled (Scr.) siRNA or siRNA targeted against cofilin 1 (cofilin 1 knock-down, CFL1 KD). Scale bar = 10 μm. (C) Box plot comparing cortex thickness, h, between untreated control cells and cells treated with DMSO or Jasplakinolide or transfected with scrambled or anti-CFL1 siRNA. ^∗∗p < 0.01, ^∗∗∗p < 0.001 by Welch t-test. (D) Western blot confirming knock-down of CFL1 after transfection of anti-CFL1 siRNA. GAPDH was used as a loading control.

To further probe the relationship between actin depolymerization and cortex thickness, we reduced expression of cofilin, an actin severing and depolymerizing protein (25,26). We knocked down cofilin by transfecting cells with siRNA targeted against CFL1, the human gene that encodes cofilin, as well as scrambled control siRNA (Fig. 5, B and D). We then blocked the cells in prometaphase and measured cortex thickness. Cells transfected with CFL1 siRNA had an ∼60% thicker cortex than control cells (Fig. 5 C), further suggesting that reducing actin disassembly results in an increase in cortex thickness.

Cortex thickness increases during bleb retraction

Finally, we asked whether our method could be used to follow changes in cortex thickness during dynamic cell shape change. To this aim, we monitored the growth and retraction of cellular blebs, transient membrane protrusions in which de novo cortex assembly can be observed (27). We hypothesized that cortex thickness at the bleb membrane would begin near zero and gradually increase during bleb retraction. Bleb formation was induced in a controlled manner by local laser ablation of the cortex (23). We followed the progression of bleb growth, arrest, and retraction (Fig. 6 A and see Movie S1 in the Supporting Material) and acquired linescans around the bleb periphery from the first sign of cortex regrowth (Fig. 6 B). From these linescans, we extracted cortex thickness, h, at the bleb membrane over time (Fig. 6, C and D). Bleb cortex thickness values were relatively noisy, likely due to the small area over which linescans were averaged. The amplitude of the noise (∼75 nm) was consistent with our measurements of cortex thickness in bins of comparable size (2–5 μm) around the cell periphery (see Fig. S3). Smoothing these values, we observed that h was initially lower than preablation cortex thickness, and in approximately half of the cells, h increased gradually to the preablation value during retraction (Fig. 6 C). In the other half of the cells, bleb cortex thickness increased to a value beyond preablation thickness and relaxed back to the preablation value (Fig. 6 D). Taken together, these data suggest that 1), The cortex is polymerized from the membrane in blebs, as bleb cortex thickness is initially close to zero and increases over time, and; 2), A cellular mechanism exists to tightly control cortex thickness, as thickness ultimately returns to its preablation value. Furthermore, this analysis demonstrates that our approach can be applied to monitor thickness dynamics in live cells undergoing shape change, and the time resolution of such studies is limited only by the acquisition time.

Figure 6.

Monitoring cortex thickness in blebs. (A) Montage from a timelapse image series of bleb growth and retraction after laser ablation of the cortex in a prometaphase HeLa cell expressing GFP-Actin and mCherry-CAAX. Scale bar = 5 μm. See also Movie S1. (B) Average linescans across the bleb border at different time points from panel A. (C) Plot of cortex thickness during bleb retraction shown in panel A. (D) A time course from another bleb experiment in which bleb cortex thickness exceeded preablation thickness and subsequently relaxed back to the initial value. For panels A–D, time 0 corresponds to ablation time. For panels C and D, bleb cortex thickness data (points) were smoothed with a moving average of window size 11 points (lines).

Discussion

In this report, we combine subpixel image analysis with theoretical modeling to measure the thickness of the actin cortex. We extract cortex thickness through a precise analysis of linescans across the cell border, labeling the cortex and PM with chromatically different fluorophores. We measure an average cortex thickness of ∼190 nm in mitotic HeLa cells and show that reducing actin disassembly leads to an increase in thickness. By following actin cortex regrowth in cellular blebs, we also demonstrate that this assay can be used to monitor thickness in real time.

To achieve subresolution thickness measurements, we employ an image analysis approach inspired by the SHREC technique. SHREC has been used to observe the hand-over-hand movement of a processive motor protein (14) and to characterize the subresolution architecture of kinetochores in budding yeast (15) and cytokinetic nodes in fission yeast (16). In these studies, the structures that were investigated could be considered to be bright, point-like fluorescent objects in a homogenous background. Localization was thus accomplished by fitting the resulting fluorescent signal with a Gaussian function. Here, we investigate the spatial extension, rather than the position, of a subresolution object. We extract cortex thickness from measurements of the distance between the peak fluorescence intensities of cortical actin and the PM, which requires assumptions about the underlying geometry of the cortex (Fig. 2 D). Our analysis, which assumes a uniform cortex intensity, allows us to extract cortex thickness values that do not depend on relative background intensities, the choice of fluorophores, or the probe used to label actin. Furthermore, we show that for linear gradients in cortical actin where intensity decreases up to 50% from the membrane to the cytoplasm, we expect to underestimate thickness by less than ∼10% (see Fig. S4). Together, this strongly suggests that our assumption of a uniform cortex and our simple description of the imaging process (Fig. 2 D) are sufficient and provide a definition of cortex thickness that can be measured robustly and compared between experimental conditions.

We then adopted a model-convolution approach (20) to evaluate our thickness measurements using synthetic cell images, where cortex geometry and background intensities are well controlled. The synthetic images also allowed us to test how changing MV density and length affect our thickness measurements. Our model-convolution analysis further supported the efficacy of our thickness extraction method and indicated that in our experimental conditions, the presence of actin-filled MV at the cell surface did not significantly perturb our measurements of cortex thickness. In addition, we found that reducing MV experimentally did not affect our thickness measurements (see Fig. S6). Finally, using synthetic images, we estimated the accuracy and precision of our thickness measurements to be ∼10–20 nm for a cortex with uniform intensity.

Other experimental factors, including optical aberrations due to imaging away from the coverslip (28) or imperfections in cell segmentation, could potentially affect the accuracy and precision of our approach. Such factors would effectively add additional blurring, or convolution, to the linescans, increasing the effective width, σ, of the PSF (see also the Supporting Material). However, increasing σ to values above what we measure experimentally does not significantly change our cortex thickness measurements or the independence of h on the background intensity ratio, i_out/i_in (see Fig. S7). We therefore do not expect that such experimental factors significantly affect the accuracy or precision of our assay.

In addition to allowing the extraction of cortex thickness, our approach provides, to our knowledge, a novel way to differentiate between changes in cortex thickness and density, which are likely to influence cortex mechanics in different ways. During processes in which changes in cortical intensity have been observed, such as accumulation of actomyosin in the cleavage furrow during cytokinesis (29), it is unclear whether such changes reflect differences in cortex thickness or density. By simultaneously extracting cortex thickness, h, and intensity, i_c, which is related to cortex density, we now have the potential to discriminate between changes in these different structural properties of the cortex.

Our approach to measure cortex thickness relies on a combination of standard confocal imaging and subresolution image analysis. Despite the recent emergence of several superresolution microscopy techniques, the measurement of cortex thicknesses of ∼200 nm in live cells is not possible with contemporary superresolution setups. The cortex is most prominent in rounded cells, precluding the use of total internal reflection fluorescence-based techniques for thickness measurements. The resolution achieved for live imaging of intracellular structures in three dimensions is, as of this writing, still limited to ∼100 nm (13) and is thus insufficient to resolve the changes in cortex thickness reported here (Figs. 5 and 6). Contemporary superresolution setups allowing for a resolution under 100 nm have image acquisition times of several minutes and would require cell fixation. In fixed cells, stochastic optical reconstruction microscopy has been used to uncover the presence of two separate layers of actin, each ∼30-40 nm thick and separated by ∼100 nm, throughout the lamellum and possibly extending into the lamellipodium (30). It will be interesting to explore whether this type of imaging, combined with our method using dual-color fluorescence peak localization, could provide further insight into the architecture of the cortical actin network.

The global mechanical properties of the cortex, such as tension and viscoelasticity, arise from microscopic network organization and dynamics. However, because the architecture of the cortex is poorly understood, contemporary physical models of cortex mechanics remain speculative with respect to how molecular-scale events control larger-scale mechanical properties (5). For example, depletion of the actin severing protein cofilin has been shown to increase cortical tension (23). Coarse-grained theoretical models indicate that cortex tension is directly proportional to thickness (31); based on this prediction, it has been proposed that cofilin may control tension by regulating cortex thickness (23). However, this has never been tested experimentally. We now show that depletion of cofilin indeed causes an increase in cortex thickness sufficient to account for the observed change in tension (Fig. 5). Our results represent an important step in bridging scales between molecular processes and emerging mechanical properties of the actomyosin cortex.

Conclusion

Many morphogenetic processes, including cell division, migration, and epithelial shape change are driven largely by the cortex. Our study therefore opens new avenues for the quantitative investigation of how regulation of cortex thickness in space and time drives cellular morphogenesis. More broadly, our method to measure cortex thickness can be extended to investigate how cortical components are distributed throughout the cortical network, and how, together, they give rise to the emergent physical properties of the cell surface.