Automated image analysis of nuclear shape: What can we learn from a prematurely aged cell?

The nuclear shape of HGPS and normal fibroblast cells can be automatically extracted, visualized, and analyzed

In order to test our automatic analysis of nuclear shape, we first cultured fibroblasts from two HGPS fibroblast cell lines (HGADFN155-p15 and HGADFN167-p12, HGPS1 and HGPS2 respectively) and from one normal control (HGFDFN168-p14, normal). The cells were fed with fresh MEM medium containing 15% FBS and grown at 37°C. To visualize the nuclei, we performed immunofluorescence staining of the nuclear membrane with a mouse monoclonal antibody raised against lamin A/C (MAB3211). This antibody has been well characterized in HGPS cells and has also been used in studies on other laminopathies. Fluorescence images of about 100 randomly selected nuclei per cell line were taken with a Zeiss fluorescence microscope at 400X magnification (examples are shown in Figure 1a). A custom-written MATLAB program (details as described in Material and Methods) was used to extract nuclear shapes and properties of the shape, such as boundary curvature. In Figures 1a and b, the nuclear boundaries, which are colored by curvature, are shown overlaid on the microscopy images. Convex curvatures were kept positive, while concave curvatures were made negative. As shown by the color bar in Figure 1c, blue represents regions of large positive curvature, and red represents regions of large negative curvature. A blebbed cell, such as the cell shown in the top of Figure 1b, will have boundary regions that are dark red and dark blue, whereas a non-blebbed cell, such as the cell shown in the bottom of Figure 1b, will have boundary regions that are mostly blue and green with almost no red.

We next assembled plots that, for each cell line, display the boundary curvatures of all of the measured nuclei, as shown in Figure 1c. In these heat maps, each vertical line represents the boundary curvature of one nucleus. To construct such a plot, imagine cutting each colored boundary at the location farthest from the nucleus center (that location is labeled 0 on the heat map), pulling the boundary straight, and then lining it up next to the colored boundaries of the other nuclei. The heat maps of blebbed populations, such as the HGPS cell lines, have many red speckles, whereas the heat maps of unblebbed populations, such as the control cell line, have few red speckles.

Within each plot, the nuclei are ordered from left to right by increasing mean negative curvature (MNC), a measure of nuclear blebbing. We defined the MNC of each nucleus by averaging all negative curvatures, excluding the positive curvatures entirely, and taking the absolute value. As shown in Figures 1d and 1e, the HGPS1 and HGPS2 cell lines have larger MNCs, and hence are more blebbed, than the control cell line. HGPS1 also has a larger MNC than HGPS2, perhaps because HGPS1 is at a later cellular passage, and thus more senesced. We found that both HGPS MNC distributions are statistically different from the MNC distribution of the control (see Figure 1e for p-values).

To validate the automated nuclear shape analysis, we also assessed nuclear morphology using the standard technique: manual blind counting. Nuclei with protrusions or invaginations were counted as blebed, while other nuclei were counted as normal. We found that 73% of HGPS1 nuclei, 63% of HGPS2 nuclei, and 24% of normal nuclei were abnormal. These counts are in quantitative agreement with the MNC distributions of the respective cell lines (Figure 1e). In order to better evaluate how the results of manual counting correlate with quantitative shape metrics, we had experienced human counters label individual nuclei as either normal or blebbed, and analyzed the MNC of the two populations. The analysis shows a positive correlation between MNC and the percentage of nuclei that were labeled as blebbed (Figure S1).

Since the automated analysis extracts the boundary of each nucleus, we can assess nuclear morphology using various shape metrics besides boundary curvature. For each nucleus, we also calculated area, perimeter, number of invaginations, eccentricity and other metrics. In analogy to how microarray data is analyzed to find relationships between genes, we used correlation as a measure of interrelationship between the 15 different measures of nuclear shape determined in this study. We hierarchically clustered the 15 measures of nuclear shape and laminA/C fluorescence intensity (Figure S2). We found several families of nuclear measures that roughly correspond to size, extent of blebbing, eccentricity, laminA/C fluorescence intensity, and the standard deviation of fluorescence intensity. The clustering reassures us that the intensity, which is affected by immunostaining and imaging details, does not notably affect the measured MNC. The clustering also indicates that the standard deviation in MNC and the tortuosity are measures related to MNC. Also related to mean MNC is the solidity, which is the ratio of the measured area and the area of ‘convex hull,’ or the minimal convex shape that bounds the measured shape of the nucleus.

As a control experiment, we tested whether the cell density would influence the MNC. We seeded cells from the same HGPS cell line at densities of 3000, 9000, and 27000 cells per well in 4-well chamber slides (Figure S3). The three densities did not appear to have different MNC distributions, nor were the measured MNC distributions statistically distinct.

RAD001 and rapamycin similarly reduce blebbing in HGPS fibroblast cells

Recent work has revealed that rapamycin, an mTOR (mammalian target of rapamycin) inhibitor, significantly reduces the phenotypic hallmarks of progeria in HGPS cells by down-regulating progerin [15]. Everolimus (RAD001), which is the 40-O-(2-hydroxyethyl) derivative of rapamycin, works similarly to sirolimus as an mTOR inhibitor but is better tolerated by patients. In order to compare the efficacy of RAD001 to rapamycin, we treated HGPS fibroblasts with rapamycin, RAD001, or mock, and then analyzed the nuclear morphology of each treatment group.

Cultured fibroblasts from an HGPS patient (HGADFN155-p15) and a normal individual (HGFDFN168-p15) were used in this experiment. The cells were fed every other day with fresh MEM medium containing 0.68 μM rapamycin, 0.1 μM RAD001, 0.5 μM RAD001, or the same volume of vehicle (DMSO, 0.025% v/v) for a duration of seven weeks. To examine the effects on nuclear morphology, we labeled cells with an antibody for lamin A/C and an antibody specific for progerin (Figure 2a). To judge the impact of rapamycin and RAD001, we first scored the percentage of nuclei with abnormal morphology in the usual way by manual blind counting. At least 200 randomly selected cells were scored by fluorescence microscopy for each cell line under each condition. In comparison with the passage-matched, mock-treated HGPS cells, the rapamycin or RAD001 treated HGPS cells exhibited a clear reduction in nuclear blebbing (Figure 2b). Since increased genome instability was reported in HGPS cells [15, 22, 23], we also examined whether RAD001 treatment can improve this phenotype. Using immunofluorescence staining, we observed a reduction in 53BP1 foci in rapamycin or RAD001 treated cells, indicating that inhibition of mTOR prevents DNA damage induced in prematurely senescent cells by progerin (Figure S4). Quantification of progerin protein by western blotting analysis also revealed an over 50% reduction in progerin levels in rapamycin and RAD001 treated HGPS cells (Figure 2c). We also detected a weaker progerin-staining signal in almost all of the rapamycin or RAD001 treated HGPS cells, and their nuclear morphology appeared substantially improved compared to untreated cells (Figure 2a).

As reported in Cao et al. [15], we again noticed a reduction in overall cell proliferation at the beginning of the treatments (less than two weeks) with either rapamycin or RAD001 compared to the mock treated samples. We used a tetrazolium salt-based cell proliferation assay (WST1 assay) to analyze this apparent cell growth inhibition for the concentrations of RAD001 used in the experiment: 0, 20, 60, 100, and 500 nM. Figure S5 shows that all treatments for both control and HGPS cell lines had a similar reduction in cell proliferation compared to the mock treatments (less than 40%), suggesting that any effective dose of RAD001 may have similar anti-hypertrophic effects [24].

In parallel to the blind counting, we took immunofluorescence images of about 100 randomly selected nuclei per treatment group and automatically analyzed their nuclear morphology. Heat maps, which display the boundary curvature of the treated HGPS cells, are shown in Figure 3a. From the heat maps we see that the mock treated cells are much more blebbed than the rapamycin or RAD001 treated cells, which is consistent with our blinded counting. Indeed, we found that the MNC distributions of the rapamycin and RAD001 treated cells were statistically different from that of the control group (Figure 3b). Similarly, our analysis showed a reduction in the number of invaginations in treated HGPS cells (Figure 3d).

Interestingly, we also found that the RAD001 and rapamycin treated nuclei had a smaller area than the mock treated nuclei (Figure 3c). Moreover, we noticed that the eccentricity, which is a measure of how elongated the nuclei are, did not change as a result of the rapamycin or RAD001 treatments (Figure 3e). Our analysis suggested that rapamycin or RAD001 treatments appear to locally improve abnormal morphology, without affecting the overall shape of the nuclei, though still altering nuclear size (see discussion). In summary, our data suggest that, similar to rapamycin, RAD001 can reverse the nuclear phenotypes in HGPS cells through promoting progerin clearance.

RAD001 induces a gradual change in nuclear morphology in a dosage dependant manner

Based on the above analysis, we suggested RAD001 could be used at 100 nM concentration to achieve similar beneficial effects in HGPS cell cultures as rapamycin at 0.68 μM as described in Cao et al. [15]. Next, we explored the sensitivity of the curvature analysis program, since quantitative image analysis is most useful if it can reveal small changes that are difficult to observe. Thus, we lowered the dosage of RAD001 to 20 or 60 nM, and shortened the duration of treatment to two weeks. An HGPS fibroblast cell line (HGADFN167-p15) and a control fibroblast cell line (HGFDFN168-p15) were fed with fresh MEM medium containing 20 nM RAD001, 60nM RAD001 or the same volume of vehicle (0.3% DMSO) every other day. Nuclear curvature outline and heat map analyses of MNC were carried out at the end of the two-week treatment (Figure 4a). Box plot analysis indicated a significant reduction of MNC in the HGPS cell line, even in the cells receiving 20 nM RAD001 (Figure 4b), while those minor morphological improvements were not visible with the traditional blinded counting method, suggesting that the automated analysis is more sensitive. Importantly, we noticed a dosage dependent reduction in area of the nuclei of both treated HGPS and treated control cells (Figure 4c): the area of mock treated nuclei was greater than both doses of RAD001 treated nuclei, but the nuclei that received the smaller dose of RAD001 (20 nM) had greater area than the nuclei that received the larger dose (60 nM). This result suggests the improvement in nuclear shape is a gradual process, the area reduction is primarily due to non-specific effects of the drug treatment, and incrementtal improvement during treatment can be captured and quantified by this curvature outline imaging analysis.

Cells, Cell Culture, and Treatments

Primary human dermal fibroblasts used in this study were obtained from the Progeria Research Foundation (PRF): two HGPS fibroblasts, HGADFN155 and HGADFN167, and a control normal fibroblast line, HGFDFN168. Fibroblasts were cultured in MEM medium (Invitrogen) supplemented with 15% FBS and 2 mM L-glutamine under 5% CO2 at 37°C. Normal and HGPS fibroblasts were replenished with fresh MEM medium containing 0.68 μM rapamycin/DMSO, or indicated concentration of RAD001/DMSO, every other day for up to 50 days. Control cells were treated with vehicle (DMSO) in MEM medium. Rapamycin (Sirolimus) was purchased from Sigma, and RAD001 (Everolimus) was obtained from Selleck.

Immunofluorescence Staining

For immunofluorescence, cells were seeded in 4-well chamber slides. After fixation in 4% paraformaldehyde/PBS at room temperature for 15 min, cells were permeabilized with 0.5% Triton X-100/PBS at room temperature for 5 min, followed by an overnight incubation in the blocking solution (4% BSA/TBS) at 4°C. Cells were then stained with primary antibodies for 3 hours at room temperature on the following day. The primary antibodies used in this study were: a rabbit polyclonal antibody against progerin (custom peptide antibody, Yenzym); a goat anti-lamin A/C antibody (N-18, Santa Cruz); and a mouse anti-lamin A/C antibody (MAB3211, Chemicon). After primary antibody incubation, primary antibodies were detected with Alexa Fluor-labeled secondary antibodies (Invitrogen). Slides were mounted with Vectashield mounting medium containing DAPI and observed with a Zeiss fluorescence microscope. Images were taken using a 40x objective (about 100 nuclei per condition). Exposure times and acquisition settings were established at the beginning of each set of experiments and kept constant for all treatments.

Extraction of Nuclei Boundaries

A custom-written MATLAB program was used to extract nuclei boundaries. In order to reduce image histogram variability both between and within images, we first used contrast-limited adaptive histogram equalization. An 8 × 8 grid of tiles, a clip limit of 0.02, and a Rayleigh distribution were employed. Next, we binarized the images using MATLAB's built-in thresholding function, which uses Otsu's method. Holes within bright regions were then filled, and regions that either overlapped the image boundary or were smaller than 800 square pixels were removed. In order to smooth and enlarge the regions, which correspond to nuclei, the images were morphologically eroded with a disk of radius 3 pixels and then morphologically dilated with a disk of radius 6 pixels. The convex hulls of these smoothed and enlarged regions were next calculated. These outer convex hulls were later used as an outer limit on the nuclei boundary positions. Next, the images were morphologically eroded with a disk of radius 2 pixels. The convex hulls of these smoothed and slightly enlarged regions were used to initialize an active contour based boundary extraction algorithm.

We next processed the original images for use by the active contour based boundary extraction algorithm. First, the brightness and contrast of the image was adjusted so that 1% of the pixels was saturated at the lowest intensity and 1% was saturated at the highest intensity. For each nucleus, any pixel outside of its outer convex hull, which was found as described above, was set to zero, and the brightness and contrast were again adjusted as before. We next calculated the binarization threshold using MATLAB's built-in thresholding function, but did not binarize the image. Instead, we nearly binarized the image by setting any pixel whose value was less than 70% of the threshold value to the lowest intensity and any pixel whose value was greater than 130% of the threshold value to the highest intensity. The remaining, non-saturated pixel intensities were then stretched to fill the entire intensity range. The holes in this gray-scale, nucleus image were next filled.

An active contour, or snake algorithm, was used to extract nuclei boundaries with sub-pixel resolution, as described in Xu and Prince [28]. The gradient vector flow (GVF) field of the processed, nucleus image was calculated. The inner convex hull of the nucleus, found as described above, was next interpolated and used as the initial position of the active contour. The contour, which is a polygon, was then repeatedly deformed 75 times and interpolated until the change in area from one set of deformations to the next was no more than 10 square pixels. Contours were not allowed to deform more than 50,025 times. The GVF, snake interpolation and snake deformation functions are from Xu and Prince [28]. (The following parameters were supplied to these three functions: number of GVF iterations, 80; d_min, 0.5; d_max, 2; tension or alpha, 0.02; rigidity or beta, 0.05; step size or gamma, 1; external force or kappa, 0.6.) The contour was interpolated a final time, resulting in an outputted polygon with sides of constant length (d_min).

Some contours do not correspond to single nuclei, but rather are multiple, overlapped nuclei or are auto-fluorescent regions of cells. The user is next given an opportunity to remove such unwanted contours. Each analyzed image is sequentially displayed and polygons clicked on by the user are removed from further analysis.

Analysis of Boundary Shape

We calculated the boundary curvature at each boundary point by fitting a circle to that boundary point and the two points 25 boundary points away from it. The curvature was then calculated as the reciprocal of the radius of that circle. Convex curvatures were kept positive, while concave curvatures were made negative. For each nucleus, the boundary point farthest from the centroid was labeled boundary point 0. When visualized with color, curvature values were cut-off such that magnitudes above a cut-off value were set to that cut-off value. For each nucleus, the number of invaginations was calculated by simply counting the number of boundary regions, of any length, where negative curvature was uninterrupted by positive curvature, and eccentricity was defined as the eccentricity of an ellipse with the same second-moments as the nuclear shape. The eccentricity of an ellipse describes how elongated the ellipse is; a circle would have an eccentricity of 0, and a line segment has an eccentricity of 1. We have previously similarly analyzed the shape of migrating amoebae [29]

WST-1 Cell Proliferation Assay

A WST-1 cell proliferation assay (Roche) was used to analyze the effects of RAD001 on cellular growth. HGPS cell line HGADFN167 p12 and control cell line HGADFN168 p14 were seeded in separate standard 24-well plates at 10,000 cells in 500 μl fibroblast medium per well. Wells were treated with 0, 20, 60, 100 and 500 nM RAD001/DMSO in triplicate and the solvent controlled at 0.1% for all wells. The cells were then incubated with treatment for 72 hours. The medium was removed from each well and 500 μl of 10% WST-1 reagent in fibroblast medium was applied to each well following the incubation. Three blanks, consisting of 500 μl of 10% WST-1 reagent in fibroblast medium, were also created. The absorbance (450nm) of each well was read after 3 more hours of incubation using a SpectraMax M5^e plate reader (Molecular Devices), and the average absorbance of the blanks was subtracted from each measurement.

Cell numbers were calculated from the absorbance values using a standard curve established by repeating the experiment without treatment and seeding at 1000, 2000, 4000, 8000 and 16000 cells per well in duplicate. The percent survival was calculated for each sample by the equation: percent survival = 100 * (cell #)/(average cell # of 0 nM treatment), then averaged by treatment. The error was calculated using the standard deviation of the percent survivals of the 3 samples for each treatment.

Statistics

Statistical analysis of MNC and other values was performed using the ttest2 function in MATLAB v7.12.0 that performs a Student's t-test. The parameters were set to a 1-tailed t-test, assuming unequal variances, and a p-value of 0.05 or less was considered significant.