MATLAB-based Methods Allow Precise,High-Throughput Quantification of Nuclear Morphology and Texture in Tachycardiomyopathy

Animal Model

Tachycardiomyopathy (T-CM) was induced in rabbits using previously established protocols.^16,20 Six rabbits were surgically implanted with right ventricular pacing devices (Advisa DR MRI SureScan, Medtronic, Minneapolis, MN). T-CM rabbits underwent incremental tachypacing, reaching up to 380 bpm over a 30-day period (Fig. 1). Pacemakers in the control group (SHAM) were not activated. Rabbits were monitored daily to assess overall health, including respiratory rate, body weight, and behavior. The animal study was approved by the institutional and governmental animal care committee (Ref. no. 55.2-2532-2-1121, Regierung von Unterfranken, Germany; University of Regensburg, Germany).

OPEN IN VIEWER

Figure 1.

Tachypacing induced systolic heart failure in rabbits. (A) Permanent pacemakers were implanted in rabbits. (B) Left ventricular specimens were stained with DAPI, WGA, and Isolectin B4. (C) Left ventricular fractional shortening declined in T-CM animals. (D) Exemplary depiction of ascites in T-CM animal. (E) Representative images of Masson’s trichrome staining in SHAM (left) and T-CM (right) animals (40× magnification). Abbreviations: T-CM, tachycardiomyopathy; PM, pacemaker; HF, heart frequency; LV, left ventricle; DAPI, 4′,6-diamidino-2-phenylindole.

Transthoracic Echocardiography

Standardized transthoracic echocardiography was performed as previously described.^16,20,21 Transthoracic echocardiography was used to assess cardiac morphology and function at baseline and on days 10, 20, and 30. Pacing was temporarily suspended during echocardiographic assessment. A HP Sonos 5500 equipped with a 12 MHz transducer (Philips Healthcare, Amsterdam, The Netherlands) was used. Echocardiography was performed with rabbits in a supine position from the left parasternal view. Left ventricular end-diastolic diameter (LVEDD) and left ventricular end-systolic diameter (LVESD) were measured by two-dimensionally guided M-Mode in the parasternal long axis. Systolic function was determined by calculating fractional shortening (FS) as FS = (LVEDD-LVESD)/LVEDD.

Tissue Fixation and Staining

The tissue samples for further analyses were removed from the anterior left ventricular wall, cutting a 1 × 1 cm large, transmural sample, left of the left anterior descending artery (LAD) and distal the first diagonal branch. The model used in this study is not a working heart model or a perfused heart model. We did not stop the heart in diastole and the heart was not perfusion fixed.

For fluorescence microscopy, left ventricular tissue was fixed in 4% formaldehyde solution (#F25, Waldeck GmbH & Co KG, Münster, Germany) for at least 24 hr and embedded in paraffin according to standard procedures. Thin 4 µm sections were cut with a microtome and stained with a Wheat Germ Agglutinin (WGA) Alexa Fluor 594 Conjugate (W11262, Thermo Fisher Scientific, Waltham, MA) for visualization of connective tissue. To display vessels/endothelial cells, biotinylated Isolectin B4 (B1205, Vector Laboratories, Newark, CA) and Streptavidin-Alexa Flour 488 (016-540-084, Jackson Immuno Research, West Grove, PA) were used. DAPI was used for nuclear staining (4′,6-diamidino-2-phenylindole, D1306, Thermo Fisher Scientific, Waltham, MA).

Image Acquisition and Processing

A standardized approach was employed for image acquisition, ensuring microscope settings were optimized according to the unique characteristics of each sample. A Zeiss Observer.Z1 microscope (Carl Zeiss AG, Oberkochen, Germany) was used to capture images at 40× magnification. The imaging process employed three specific filter sets: 49 DAPI, 44 FITC, and 71 HcRed. The software used for image acquisition was AxioVision (AxioVs40x64 4.9.1.0). To ensure consistency, regions containing transversely sectioned cardiomyocytes were specifically selected within the tissue.

All image processing was conducted in MATLAB (MathWorks, Natick, MA) R2021b. The images, stored in tagged image file format (.tiff), were separated into their red, green, and blue (RGB) channels. Before analysis, image matrices underwent contrast-limited adaptive histogram equalization to enhance contrast and optimize image quality.

Segmentation of Nuclei and Morphologic Evaluation

For the segmentation of DAPI-stained nuclei, the blue channel of the RGB image matrix was selected for further processing. The image was smoothed using a Gaussian filter to reduce noise. Due to light inhomogeneity, an adaptive thresholding technique was applied. This approach was chosen because, at lower thresholds, noise artificially increases the number of segmented regions. As the threshold increases, the influence of noise diminishes, leading to a reduction in the number of identified regions. However, further increases in the threshold result in the disconnection of nuclei, dividing them into regions of euchromatin and heterochromatin, which again increases the number of segmented regions (Fig. 2B). Nuclear shape and size were quantitatively evaluated using mean nuclear size, standard deviation of nuclear size, and eccentricity. The eccentricity was defined as the ratio of the minor and major axis length of an ellipse fitted onto the segmented nucleus. A value close to 0 indicates a circular shape and a value close to 1 indicates an eccentric shape.

OPEN IN VIEWER

Figure 2.

Image analysis workflow. (A) Image segmentation workflow. (B) Illustration of segmentation using adaptive thresholding with low point approximation. (C) Illustration of stages of analysis from left to right: DAPI nucleus, segmented nucleus, analysis of nuclear orientation and shape, selective analysis of nuclear texture. Abbreviations: DAPI, 4′,6-diamidino-2-phenylindole; RGB, Red, Green, Blue.

Evaluation of Nuclear Orientation and Its Heterogeneity

After successful nuclear segmentation, nuclear orientation was determined using the function regionprops(). From this, an orientation distribution histogram was generated, and histogram-derived parameters—including mean, standard deviation, kurtosis, and skewness—were collected. Furthermore, the Orientational Order Parameter (OOP, f2D) of the segmented nuclear image was calculated as described in Umeno et al. ¹² :

f 2 D = 2 ( ( c o s 2 ( θ − μ 0 ) ) − 1 2 )

where θ is the orientation of the considered cell and μ₀ is the mean orientation of all evaluated cells of the image. The term (cos ² ( θ − μ 0 ) ) describes the mean value of the entirety of evaluated cells. As the OOP obviates the pattern of orientation distribution in cellular clusters, we evaluated the co-occurrence of orientation of neighboring nuclei using a parameter we named orientation co-occurrence parameter (OCP) and defined it as follows:

O C P = 1 n ∑ i = 1 n cos ( θ n − θ n + i )

where θ n is the orientation of the analyzed nucleus and θ n + i is the orientation of the neighboring nucleus in a defined direction (90°) and n is the number of segmented nuclei. Furthermore, a second method was used to evaluate nuclear orientation. Here, a differential image was created from the segmented image to delineate sudden changes in pixel intensity (nuclear borders) from one direction. This process was repeated for 180 times from the angles 0° to 180° with an increment of 1°. For each direction, the number of pixels belonging to nuclear edges was counted and a histogram was derived containing the number of positive pixels for each degree (0°–180°). From this rotation histogram (Appendix Fig. 1), standard histogram parameters such as mean, standard deviation, kurtosis, and skewness were calculated.

Evaluation of Nuclear Texture

To selectively assess nuclear texture, a nuclear mask was created, retaining only pixel intensity values from the primary image corresponding to nuclear regions. From this image, a synthetic image was created only containing nuclear texture. A gray-level co-occurrence matrix (GLCM) was then constructed, and Haralick’s texture descriptors—contrast, correlation, entropy, and homogeneity—were computed using the function graycoprops().

Selective Nuclear Analysis of Cardiomyocytes

Selective extraction of nuclear features of cardiomyocytes as opposed to other nuclei such as endothelial nuclei was implemented using a distance transform matrix from the vessel image. Nuclei centered in between vessels were deemed cardiomyocyte nuclei. Nuclei in close proximity to vessel were excluded.

Evaluation of Extracellular Orientation

Evaluation of the extracellular orientation was performed on the red channel of the acquired RGB image. As the WGA staining of cardiomyocytes and extracellular matrix span a network occupying the whole image, an object-centric orientation evaluation was not performed. The staining was segmented using an adaptive thresholding algorithm. Subsequently, a rotation histogram (Appendix Fig. 1) was created and histogram parameters such as mean, standard deviation, kurtosis, and skewness were calculated from it.

Evaluation of Extracellular Texture

To evaluate morphologic properties of the extracellular structures contained in the image quantitatively, an image mask only containing regions of WGA staining was created using adaptive thresholding. From this mask, a gray-level co-occurrence matrix was created and Haralick’s texture descriptors contrast, correlation, entropy, and homogeneity were calculated using the function graycoprops(). Values for thickness were drawn through calculation of a distance transform matrix from the image mask.

Evaluation of Vessel Density

To assess vessel density, images corresponding to vessel staining were segmented. The total area occupied by vessels was measured as well as their number and spatial distribution.

Statistical Analysis

Data are presented as mean values including standard deviation or 95% confidence interval (95% CI). A two-sided Student’s t-test was performed for group comparisons of normally distributed quantities. For all cases in which the t-test’s assumptions were not met, a non-parametric unpaired Mann–Whitney test was performed. The strength of association between quantitative measures was assessed by Pearson’s (r, normally distributed data) correlation coefficient and Spearman’s correlation coefficient for the cases Pearson’s criteria were not met. In case of a normal distribution, we report the mean and standard deviation (Mean ± SD). For all non-normally distributed data, results are presented as the median along with the 25th and 75th percentiles [Median (25P.: 25th percentile, 75P.: 75th percentile)]. Statistical analysis was performed in GraphPad Prism (version 7 and 8, GraphPad Software Inc., San Diego, CA). For all statistical tests, a two-sided p-value <0.05 was considered statistically significant.

Validation

To validate the newly introduced OCP, we utilized synthetic phantom images with predefined nuclear orientations (Appendix Fig. 3). In addition, manual validation for all automatically calculated parameters was performed through manual delineation of nuclear shape, orientation, and membrane on randomly selected images.

Automated nuclear counts closely matched manual counts, demonstrating high concordance (Mean manual nuclear count per slide = 81.3 ± 3.8, mean automated nuclear count per slide = 81.9 ± 4.3, r = 0.981, p<0.001, Appendix Fig. 2A). Automated segmentation also showed strong agreement with manual delineation. Furthermore, T-CM and SHAM did not differ in number of nuclei per slide (Mean number of nuclei per slide T-CM = 84.9, Mean number of nuclei per slide SHAM = 81.9, p=0.287, Appendix Fig. 2B).

To assess the characteristics of OCP and OOP, we analyzed phantom images with predefined orientation properties (Appendix Fig. 3C–G). We could determine that the automatically assumed orientation was in concordance with the manually defined orientation (Appendix Fig. 3B).

Tachypacing Induces Systolic Heart Failure

Incremental tachypacing for 30 days induced severe left ventricular dysfunction (Fig. 1C). T-CM rabbits exhibited hallmark signs of advanced heart failure syndrome, including pleural and pericardial effusion as well as ascites (Fig. 1D).

Despite the presence of severe heart failure, Masson’s trichrome staining did not reveal detectable fibrosis (Fig. 1E).

Nuclear Remodeling in T-CM

First, we hypothesized that T-CM entails specific nuclear remodeling. To investigate this, we quantitatively evaluated nuclear size, texture, and orientation. Nuclear sizes in left ventricular specimens differed between SHAM and T-CM animals [Mean SHAM (a.u.) = 2364 ± 295.2, Mean T-CM (a.u.) = 2612 ± 254.6, p=0.0042], where the nuclei in T-CM animals were notably larger. This includes nuclei from cardiomyocytes, epithelial nuclei, and nuclei of infiltrating immune cells. Their intrasample variability of size, however, did not differ [Mean SHAM (a.u.) = 1582 ± 276.4, Mean T-CM (a.u.) = 1689 ± 257.8, p=0.1637]. Nuclear eccentricity did not differ between the two experimental groups [Mean SHAM (a.u.) = 0.6552 ± 0.0319, Mean T-CM (a.u.) = 0.6567 ± 0.0284, p=0.6466]. When nuclei of cardiomyopathy were exclusively analyzed (Fig. 3), they were enlarged in T-CM (Fig. 3A). Nuclei from cardiomyocytes in the T-CM group also showed a slightly more homogeneous size distribution indicated by the relative standard deviation of nuclear size (Fig. 3B). The nuclear eccentricity of cardiomyocytes, however, was not different (Fig. 3C). The staining intensity of DAPI and WGA was similar in SHAM and T-CM animals [DAPI: Median SHAM (a.u.) = 102 (25P.: 89, 75P.: 126), Median T-CM (a.u.) = 113 (25P.: 110, 75P.: 116), p=0.1186; WGA: Median SHAM (a.u.) = 149 (25P.: 136, 75P.: 170), Median T-CM (a.u.) = 150 (25P.: 132, 75P.: 172), p=0.6615].

OPEN IN VIEWER

Figure 3.

Descriptors of morphology, texture, and vessel architecture in selective cardiomyocyte analysis. (A) Mean nuclear size in DAPI-stained samples from left ventricular tissue of SHAM and T-CM rabbits. (B) Relative standard deviation of nuclear size in DAPI-stained samples from left ventricular tissue of SHAM and T-CM rabbits. (C) Mean nuclear eccentricity in DAPI-stained samples from left ventricular tissue of SHAM and T-CM rabbits. (D) Nuclear contrast in DAPI-stained samples from left ventricular tissue of SHAM and T-CM rabbits. (E) Extracellular contrast in WGA-stained samples from left ventricular tissue of SHAM and T-CM rabbits. (F) Extracellular kurtosis in WGA-stained samples from left ventricular tissue of SHAM and T-CM rabbits. (G) Number of vessels per slide in SHAM and T-CM animals. (H) Median size of vessel [a.u.] in SHAM and T-CM animals. (I) Relative standard deviation of vessel size in SHAM and T-CM animals. Abbreviations: T-CM, tachycardiomyopathy; SD, standard deviation; DAPI, 4′,6-diamidino-2-phenylindole; a.u., arbitrary unit. Mean ± standard deviation.

In summary, our imaging platform provides a precise tool for quantifying nuclear size across all cells or specifically within cardiac myocytes, as well as measuring nuclear eccentricity. Notably, nuclear size was enlarged in T-CM compared with SHAM.

In addition to nuclear size, we wanted to assess whether more could be learnt on nuclear remodeling in T-CM by quantifying nuclear texture. For this purpose, gray-level co-occurrence matrix (GLCM)-derived parameters were calculated. Selective texture analysis of both nuclear and extracellular staining revealed differences in the uniformity of pixel intensity distribution between left ventricular specimens from SHAM and T-CM animals. These differences were assessed using parameters such as contrast, correlation, energy, and homogeneity.

DAPI-stained nuclei exhibited decreased nuclear contrast in T-CM animals upon selective nuclear analysis (Fig. 3D). This difference in nuclear texture was further supported by additional GLCM-derived parameters (Appendix Table 1). Interestingly, in the selective analysis of cardiomycyte nuclei, nuclei from T-CM animals showed increased contrast [Median SHAM (a.u.) = 0.0169 (25P.: 0.0103, 75P.: 0.0225), Median T-CM (a.u.) = 0.0247 (25P.: 0.0179, 75P.: 0.0316), p=0.0149]. The texture of WGA staining exhibited decreased contrast (Fig. 3E). These differences in texture could—in part—be found using histogram-derived descriptors, such as kurtosis of extracellular staining (Fig. 3F).

In contrast to nuclear texture, nuclear orientation did not differ between the two subgroups, quantified by the OOP [Median SHAM (a.u.) = 0.0564 (25P.: 0.0159, 75P.: 0.0954), Median T-CM (a.u.) = 0.0766 (25P.: 0.0394, 75P.: 0.1027), p=0.3948] or OCP [Median SHAM (a.u.) = 0.8193 (25P.: 0.7501, 75P.: 0.8523), Median T-CM (a.u.) = 0.8017 (25P.: 0.7386, 75P.: 0.8468), p=0.4930]. However, orientation analysis using the rotation vector revealed differences between the two subgroups [Median SHAM (a.u.) = 25743 (25P.: 24226, 75P.: 30306), Median T-CM (a.u.) = 23196 (25P.: 22539, 75P.: 24002), p<0.0001]. Together, our MATLAB-based imaging platform allows quantification of several aspects of nuclear remodeling, such as nuclear size, texture, and orientation. The most prominent differences observed were nuclear enlargement and increased nuclear contrast.

Vascular Architecture in T-CM

Given that T-CM is associated with left ventricular eccentric hypertrophy, we investigated whether vascular structure is altered in T-CM.

The total number of vessels did not differ between SHAM and T-CM animals (Fig. 3G). However, the median vessel size was increased in T-CM (Fig. 3H), while the overall size distribution remained comparable between groups (Fig. 3I). In total, vessels enlarged in T-CM.

Interrelations of Quantitative Morphologic Descriptors

To evaluate the parameters drawn from our script, it is important to cast light on selected interrelated parameters in our subset as well as methodologically related descriptors. Mean nuclear size exhibited a negative correlation with nuclear contrast (r = −0.4154, 95% CI: −0.6257 to −0.1489, p=0.0033, Fig. 4A). However, it is important to note that nuclear contrast did not correlate with DAPI staining intensity (r = 0.2160, 95% CI: −0.07261 to 0.4712, p=0.1404). The same cannot be noted for WGA staining intensity and extracellular contrast, which show a high correlation (r = 0.6133, 95% CI: 0.3989–0.7644, p>0.0001, Fig. 4B). Nevertheless, the extracellular contrast does not correlate with the number of stained pixels present (r = 0.1031, 95% CI: −0.1865 to 0.3762, p=0.4855). In addition, WGA staining intensity did not show a meaningful correlation with orientation parameters such as rotation (r = 0.2783, 95% CI: −0.006332 to 0.5212, p=0.0554). The textural descriptors have shown the greatest discriminatory effect between the two experimental groups. Due to their derivation from a GLCM, a certain interrelation is warranted. The least correlation could be observed between contrast and correlation for nuclear (r = −0.6691, 95% CI: −0.8010 to −0.4753, p<0.0001) and extracellular (r = −0.2075, 95% CI: −0.4643 to 0.0814, p=0.1570, Fig. 4D) staining. Therefore, it is noteworthy that extracellular contrast (see above) and extracellular correlation [Median SHAM = 0.9741 (25P.: 0.9716, 75P.: 0.9772), Median T-CM = 0.9781 (25P.: 0.9750, 75P.: 0.9817), p=0.0052] are strong discriminators of SHAM and T-CM animals even though they are not highly interrelated.

OPEN IN VIEWER

Figure 4.

Interrelation of selected parameters. (A) Nuclear GLCM-derived contrast versus mean nuclear size expressed in number of pixels per nucleus. (B) GLCM-derived contrast from a WGA mask versus the mean WGA staining intensity. (C) GLCM-derived nuclear energy versus contrast. (D) GLCM-derived correlation of WGA mask versus contrast. Abbreviations: GLCM, gray-level co-occurrence matrix; a.u.: arbitrary unit.