Spatial organization of fibroblast nuclear chromocenters: component tree analysis


  • Robert R. Snapp,

    1. Department of Computer Science, University of Vermont College of Medicine, Burlington, VT, USA
    Search for more papers by this author
  • Elyse Goveia,

    1. University of Vermont College of Medicine, Burlington, VT, USA
    Search for more papers by this author
  • Lindsay Peet,

    1. University of Vermont College of Medicine, Burlington, VT, USA
    Search for more papers by this author
  • Nicole A. Bouffard,

    1. Department of Neurological Sciences, University of Vermont College of Medicine, Burlington, VT, USA
    Search for more papers by this author
  • Gary J. Badger,

    1. Department of Medical Biostatistics, University of Vermont College of Medicine, Burlington, VT, USA
    Search for more papers by this author
  • Helene M. Langevin

    Corresponding author
    1. Department of Neurological Sciences, University of Vermont College of Medicine, Burlington, VT, USA
    2. Department of Orthopaedics & Rehabilitation, University of Vermont College of Medicine, Burlington, VT, USA
    3. Division of Preventive Medicine, Brigham and Women's Hospital, Harvard Medical School, Boston, MA, USA
    • Department of Computer Science, University of Vermont College of Medicine, Burlington, VT, USA
    Search for more papers by this author


Helene M. Langevin, Department of Neurological Sciences, University of Vermont College of Medicine, 89 Beaumont Ave., Burlington, VT 05405, USA. T: + 1 802 6561001; F: + 1 802 6568704; E:


The nuclei of mouse connective tissue fibroblasts contain chromocenters which are well-defined zones of heterochromatin that can be used as positional landmarks to examine nuclear remodeling in response to a mechanical perturbation. This study used component tree analysis, an image segmentation algorithm that detects high intensity voxels that are topologically connected, to quantify the spatial organization of chromocenters in fibroblasts within whole mouse connective tissue fixed and stained with 4',6-diamidino-2-phenylindole (DAPI). The component tree analysis method was applied to confocal microscopy images of whole mouse areolar connective tissue incubated for 30 min ex vivo with or without static stretch. In stretched tissue, the mean distance between chromocenters within fibroblast nuclei was significantly greater (vs. non-stretched, P < 0.001), corresponding to an average of a 500-nm increase in chromocenter separation (~10% strain). There was no significant difference in chromocenter number or average size between stretch and no stretch. Average chromocenter distance was positively correlated with nuclear cross-sectional area (r = 0.78, < 0.0001), and nuclear volume (r = 0.42, < 0.0001), and negatively correlated with nuclear aspect ratio (r = −0.65, P < 0.0001) and nuclear concavity index (r = −0.44, P < 0.0001). These results demonstrate that component trees can be successfully applied to the morphometric analysis of nuclear chromocenters in fibroblasts within whole connective tissue. Static stretching of mouse areolar connective tissue for 30 min resulted in substantially increased separation of nuclear chromocenters in connective tissue fibroblasts. This interior remodeling of the nucleus induced by tissue stretch may impact transcriptionally active euchromatin within the inter-chromocenter space.


A prominent feature of mouse connective tissue fibroblast nuclei is the presence of well-defined zones of heterochromatin, termed chromocenters. These are formed by centromeres from different chromosomes that coalesce during interphase to form dense clusters that co-localize with specific heterochromatin protein markers (Guenatri et al. 2004; Papait et al. 2008; Andrey et al. 2010). Genes within chromocenters are generally repressed and characterized by specific epigenetic signatures that establish a transcriptionally repressed chromatin state; in contrast, the space between chromocenters contains chromatin that is less dense and actively participates in gene transcription (Probst & Almouzni 2011).

We have previously reported that, in response to static stretching of mouse areolar connective tissue ex vivo, fibroblasts actively expand their nucleus, with a loss of nuclear invaginations (Langevin et al. 2010). In this study, we took advantage of the clear positional landmarks provided by 4′,-6-diamidine-2-phenylindole (DAPI; Invitrogen, Carlsbad, CA, USA) staining of chromocenters to measure the effect of static tissue stretch on the three-dimensional internal organization of mouse fibroblast nuclei. We used confocal images of whole connective tissue explants that had been stretched or non-stretched (~20% strain) for 30 min ex vivo, then fixed and stained with DAPI. We previously showed that this amount and duration of tissue stretch produced a ~20% increase in fibroblast nuclear cross-sectional area, ~10% decrease in nuclear thickness and ~30% decrease in nuclear concavity (Langevin et al. 2010). Such a large-scale flattening and smoothing of the nuclear surface could involve principally stretching of the nuclear periphery without extensive remodeling of the interior of the nucleus (Fig. 1A). On the other hand, it is also plausible that this nuclear remodeling could be more global, including the space between chromocenters (Fig. 1B).

Figure 1.

Cartoon illustration of the hypothesis tested in this study. (A) Nuclear expansion takes place at the nuclear periphery and chromocenters remain in a fixed position relative to one another. (B) Chromocenters move further apart from each other when the nucleus expands.

Component tree analysis is an image segmentation algorithm that detects high intensity voxels that are topologically connected (Guillataud 1992; Hanusse & Guillataud 1992; Mattes & Demongeot 2000; Najman & Couprie 2006). Adaptation of this method to whole tissue mounts stained histochemically (e.g. with DAPI) is particularly challenging due to variable penetration of stain into the connective tissue matrix, which introduces some inevitable variability in staining intensity between tissue samples. The goals of the current study were to (i) develop a component tree image analysis method with a built-in adaptive threshold feature that could quantify the spatial distribution of fibroblast chromocenters in histochemically stained samples of whole mouse connective tissue mounts and (ii) use this method to measure the effect of tissue stretch on the number, size and three-dimensional spatial distribution of chromocenters within fibroblast nuclei.

Materials and methods

The experimental protocols used in these experiments were approved by the University of Vermont IACUC Committee. This study examined a set of images obtained in a previously published study (Langevin et al. 2010) in which tissue samples from 28 C57BL/6 male mice (19–21 g) were excised and randomized to either stretch (n = 14) or no stretch (n = 14) ex vivo for 30 min followed by tissue fixation in 3% paraformaldehyde (PFA), staining of the nucleus with DAPI and imaging with confocal microscopy.

Tissue sample preparation and stretching ex vivo

Tissue samples were harvested immediately after death as previously described (Langevin et al. 2005, 2010). Whole skin flaps (8 × 3 cm) containing dermis, subcutaneous muscle and subcutaneous tissue were dissected away from the abdominal wall musculature, excised, and placed between stainless steel grips submerged in HEPES-physiological saline solution (HEPES-PSS) pH 7.4 at 37 °C. Tissue grips were connected to a 500-g capacity strain gauge transducer. Samples randomized to stretch were elongated at a rate of 1 mm s−1 by advancing a micrometer connected to the distal tissue grip until a load of 0.02 N (corresponding to 15–25% tissue strain relative to non-stretched length) was registered, and then maintained at that length for 30 min. Non-stretched samples were incubated in grips for 30 min at the length corresponding to the length of the tissue laying flat. At the end of incubation, samples were fixed in 3% PFA at the stretched (or non-stretched) length. After fixation, three samples of areolar connective tissue were dissected for each animal. Samples located between the superficial (subcutaneous) fascia and the deep (muscular) fascia were dissected (sample dimensions 20–30 mm in width, 60–80 mm in length and 20–50 μm in thickness), and mounted on glass slides.

Histochemical and immunohistochemical staining

Histochemical staining of the nucleus was performed with DAPI nucleic acid stain at a dilution of 1 : 1000 for 5 min at room temperature. Samples were counterstained with Texas Red conjugated phalloidin, a specific stain for polymerized actin, (Invitrogen) at a 1 : 25 dilution for 40 min at 4 °C for visualization of cell bodies. Samples were overlaid with a glass coverslip using 50% glycerol in phosphate-buffered saline (PBS) with 1% N-propylgallate as a mounting medium.

Confocal scanning laser microscopy

Tissue samples were imaged with a Zeiss LSM 510 META confocal scanning laser microscope. Four to six fields per sample were first imaged at 63× by an operator unaware of the study condition (stretched vs. non-stretched). One to two cells per field were then imaged at higher magnification using a PlanApochromat 100X (1.4 N.A.) oil immersion lens, focusing directly on the nuclei using a zoom factor of two. Thus, a total of 12–15 cells were imaged at both magnifications per animal. Z-stacks were acquired such that the entire thickness of each nucleus was captured. Each optical section had an optical thickness of 0.7 μm for all laser lines and a 0.33-μm inter-image interval. A total of 306 nuclei were imaged with a typical voxel resolution 0.045 μm × 0.045 μm × 0.33 μm.

Image analysis

A custom designed program was used for measurement of the spatial organization of chromocenters. An important aspect of the component tree analysis method used in this study was its automated nature including an adaptive threshold selection that minimizes subjectivity. Each image was preprocessed by first eliminating essentially black voxels with DAPI staining below 15 (8 bit gray-level intensity). A histogram was then constructed for each image stack, and a preliminary threshold θ was obtained that corresponded to the 95th percentile of voxel intensity. The component tree was constructed by organizing the voxels with intensity greater than or equal to θ according to their intensity and topological adjacency. The tree was first initialized by selecting the highest intensity voxels as leaf nodes so that adjacent voxels of the same intensity were grouped together. The remaining voxels were then added to the tree in order of decreasing intensity, becoming new leaf nodes if they were topologically isolated from the current voxels in the tree, or as parents to any existing nodes for which they were neighbors (Fig. 2).

Figure 2.

Illustration of method used to for component tree analysis of confocal image stacks. (A) Horizontal (XY) projection of fibroblast nucleus viewed after each component tree was truncated using an adaptive threshold operation that detects connected components containing 50 voxels or more. Different colors distinguish topologically disjoint components. (B,C) Given the image in (A), with intensities in the range from 9 (bright) to 0 (dark), a component tree is obtained under the assumption that image pixels are topologically adjacent if they share a common edge. (Thus Ax and Ay are considered to be adjacent, but Ax and By are not.) The tree in (C) is constructed in a top-down fashion, beginning with leaf nodes for brightest pixels Ax and Cy, each having intensity 9. Next, the two pixels with intensity 8 are added to the tree. Since Az is not adjacent to the current components (Ax or Cy) it is added as a new leaf node. Since in the current tree, Pixel Cx is only adjacent to Cy, it becomes a parent of the latter. Next, Ay (with intensity 7) is added to the tree as the parent of its neighbors in the tree, Ax and Az, in effect merging these two components into one. This process continues until all 16 pixels have been included. Representing the image in this manner facilitates an efficient assessment of different image thresholds. For example, since a horizontal cut at level 5 intersects three branches of the tree, it is immediately evident that a threshold of five results in exactly three components: {Bw}, {Ax, Ay, Az} and {Cx, Cy, Dx, Dy}. Likewise, a threshold of one results in a single component containing 14 pixels.

The second step consisted in ‘pruning’ the component tree to eliminate ‘fused’ components that would spuriously merge more than one chromocenter (a large component could represent a large chromocenter or an aggregation of two or more nearby small chromocenters). An adaptive thresholding operation was applied in which a candidate threshold is temporarily increased which results in removing some voxels from the large component. If this operation resulted in creating additional components, such that the second largest resulting sub-component was larger than 25% of the original component, then the higher candidate threshold was adopted. If this was not the case (e.g. the large component remains single or the second largest sub-component was less than 25% of the original), the initial threshold was retained. Finally, components with volumes less than 0.0332 μm3 (typically fewer than 50 contiguous voxels) were eliminated. See Data S1 for detailed description of the computational method. Quantitative measurements were recorded (e.g. inter-chromocenter distances, number of chromocenters and their volumes), and a two-dimensional projection of each nucleus with its constellation of segmented components was generated (Fig. 2A). The efficacy of the method was validated by a blinded investigator who reviewed all results and identified images that contained either non-detected chromocenters or fused chromocenters. Overall, 97.7% of 4726 subjectively visible chromocenters were detected.

Another advantage of using component trees for image segmentation was that each node could be augmented to include a variable that described relevant attributes of the corresponding component. For this application we included the centroid as a three-dimensional vector, the component size (i.e. the total number of voxels) sj and second-order descriptors of the shape. During the construction phase, these variables could be quickly updated as new voxels were added to each component. When a new parent node was added to the tree, these variables could be initialized using the variables of its immediate children. For example, the initial centroid of a parent node was obtained as a weighted average of the centroids of its children, where the weights corresponded to the sizes of the children. By this means the centroid of each component (as a three-dimensional vector) and the component size (as the total number of voxels included), were computed for each node in the component hierarchy. Once the threshold had been selected for each component by the procedure described above, the number of components, n (i.e. chromocenters) was counted, and the Euclidean distances between each pair of centroids, di,j were computed.

Outcome measures

The following outcome measures were recorded for each nucleus: (i) the mean inter-chromocenter distance,math formula; (ii) the maximum inter-chromocenter distance, maxi<j di,j ; (iii) the mean minimum inter-chromocenter distance, ∑ i minji di,j /n; (iv) the number of chromocenters, n; and (v) the mean chromocenter size, ∑i si /n.

In addition, we calculated correlations between these measurements and previously published measurements of nuclear cross-sectional area, volume, aspect ratio and concavity index that were based on this dataset (Langevin et al. 2010).

Statistical methods

Comparisons between stretch and non-stretch condition on derived quantitative chromocenter characteristics were performed based on nested mixed-model analyses of variance (sas, proc mixed). The model contained the experimental fixed factor, stretch condition and nested random factors: animal (within stretch condition), slide (within animal) and cell (within slide). F-tests corresponding to differences between stretch conditions, utilized animals within stretch condition as the error term. Correlation coefficients (i.e. Spearman's r) reflect the degree of association of quantitative attributes across cells, not animals. Statistical analyses were performed using sas statistical software (SAS Institute, Carey, NC, USA). Statistical significance was determined using α = 0.05.


Figure 3 shows examples of chromocenters stained with DAPI (Fig. 3A,B), as well as resolved by the component tree algorithm (Fig. 3C,D). Note that chromocenters represented by different colors indicate correspondence to topologically disjoint components of the tree.

Figure 3.

(A,B) Morphological appearance of fibroblast nuclei in non-stretched (A) and stretched (B) mouse subcutaneous areolar connective tissue (30 min ex vivo) fixed with 95% ethanol and stained with DAPI. Each image represents the XY projection of 15 confocal images of 0.7 μm thickness. (C,D) XY projection of image stacks after the chromocenters have been segmented by the component tree algorithm. Unique colors are randomly chosen to label the voxels that belong to the same components. Thus, two objects with different colors were resolved by the algorithm as being separate chromocenters.

Figure 3 also illustrates that chromocenters in fibroblast nuclei appeared further apart in stretched (Fig. 3B,D) than non-stretched (Fig. 3A,C) tissue. The mean inter-chromocenter distance within fibroblast nuclei was significantly greater in tissue stretched for 30 min ex vivo (P < 0.001) compared with tissue incubated for the same duration without stretch (Fig. 4A). Mean ± SE inter-chromocenter distance was 4.9 ± 0.1 μm vs. 5.6 ± 0.1 μm in non-stretched vs. stretched tissue, respectively. This corresponds to an average of a 500-nm increase in chromocenter separation in stretched tissue, which represents a ~10% strain in the inter-chromocenter space. The maximum inter-chromocenter distance also was significantly greater in stretched (12.1 ± 0.3 μm) than in non-stretched (10.9 ± 0.3 μm) tissue (P < 0.01) (Fig. 4B). There was no significant difference in chromocenter number, mean minimum inter-chromocenter distance, or average size between stretch and no stretch (Fig. 4C,D). Average inter-chromocenter distance was positively correlated with nuclear cross-sectional area (r = 0.78, P < 0.0001) and nuclear volume (r = 0.42, P < 0.0001), and negatively correlated with nuclear aspect ratio (r = −0.65, P < 0.0001) and nuclear concavity index (r = −0.44, P < 0.0001) (Fig. 5).

Figure 4.

Morphometric analysis of the effect of tissue stretch (30 min ex vivo) on subcutaneous areolar connective tissue fibroblast chromocenter distribution: (A) Mean inter-chromocenter distance, (B) maximum inter-chromocenter distance, (C) number of chromocenters and (D) mean chromocenter size. = 28 mice. Error bars represent SEM. Significantly different from non-stretched: **P < 0.001, *P < 0.01 (anova).

Figure 5.

Relationship between mean chromocenter distance and measurements of nuclear cross-sectional area (A), nuclear volume (B), nuclear aspect ratio (C) and nuclear concavity (D).


In the field of image analysis, component trees are frequently used as a connection filter, an image-processing algorithm that identifies connected groups of voxels or pixels that satisfy a prescribed criteria that may depend on intensity as well as the size, shape, orientation, and location of the component itself (Vincent 1993; Breen & Jones 1996). In this context, component trees (or max-trees) have been used to visualize cell structures imaged by confocal microscopy (Ouzounis & Wilkinson 2007). Other related algorithms include classic morphology filters based on structuring elements, inclusion trees and related hierarchical models (Salembier & Wilkinson 2009). Segmentation algorithms, such as the popular watershed transformation (Vincent & Soille 1991) and level-set methods (Sethian 1999), also can be applied to detect contiguous structures in images. Indeed, the latter two algorithms have been widely used to analyze distributions of cells and cell nuclei in histological sections (Ancin et al. 1996; Mukherjee et al. 2004; Kopec et al. 2011) In the current study, we have shown how component trees can be applied to automate measurements of the distribution of chromocenters in fibroblasts within whole connective tissue, with minimal user-interaction.

We chose to use DAPI staining for this study rather than a more specific immunohistochemical marker for centromeric heterochromatin (such as heterochromatin protein 1) (Guenatri et al. 2004) because of its more uniform penetration and staining in our whole tissue samples. A limitation of DAPI staining, however, is that a rim of peripheral heterochromatin also is present in mouse fibroblasts adjacent to the nuclear membrane that could potentially be falsely detected as centromeric heterochromatin. However, this peripheral DAPI staining typically did not reach the detection threshold used in our algorithm. Although our adaptive threshold procedure could be refined further, the method was able to resolve different chromocenters that were in close proximity, and allowed the detection of stretch-induced induced changes across our experimental groups. The high correlation between nuclear size and inter-chromocenter distance further supports the notion that the widened chromocenter spacing induced by tissue stretch is related to the underlying physiological response of the cell.

We found no difference in mean chromocenter number or size in stretched vs. non-stretched tissue, which is consistent with the known stability of condensed chromatin. On the other hand, the 500-nm increase in chromocenter separation found in this study suggests that a substantial remodeling of the inter-chromocenter space occurred during the 30 min of tissue stretch. The importance of such a rapid and large-scale remodeling of the nuclear interior is its potential effect on chromatin architecture at a molecular level. There is long-standing evidence that the shape and internal spatial structure of the nucleus is directly relevant to its overall function (Ingber et al. 1987; Misteli 2007). Importantly, however, although evidence is accumulating that local mechanical forces exerted by molecular motors participate in the complex process determining access to DNA at any point in time, it remains to be determined whether these processes are influenced by mechanical forces originating outside the nucleus (Brower-Toland et al. 2002; Pope et al. 2005; Gieni & Hendzel 2007; Sutherland & Bickmore 2009; Tan & Davey 2011). A recent statistical analysis of the distribution of chromocenters in interphase nuclei demonstrated that chromocenters are not randomly distributed within nuclei (Andrey et al. 2010) and computer simulation models suggest that the presence of euchromatin loops extending out from chromocenters may play a role in chromocenter spacing (Fransz et al. 2002; de Nooijer et al. 2009). Further investigation will be important to determine whether a change in fibroblast chromatin organization induced by stretching of the tissue could have a direct influence on gene transcription or, conversely, whether mechanisms are in place to shield euchromatin from such a large-scale remodeling of the inter-chromocenter space.

In conclusion, a novel image analysis method including component tree analysis and adaptive thresholding was successful in quantifying the spatial distribution of nuclear chromocenters in fibroblasts within whole mouse connective tissue. Furthermore, these results demonstrate that mechanically induced nuclear remodeling produces substantial strain within the inter-chromocenter space.


The authors thank Kirsten N. Storch for performing the original experiments that led to this study, and Cathryn Koptiuch for assistance with manuscript preparation, as well as Drs Douglas J. Taatjes, Nicholas H. Heintz, Alan K. Howe and Gary S. Stein for helpful discussions. This work was funded by the National Institutes of Health Center for Complementary and Alternative Medicine research Grant RO1-AT01121. The project described in this manuscript was also supported by Award Number 1S10RR019246 from the National Center for Research Resources for purchase of the Zeiss 510 META confocal scanning laser microscope. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the National Institutes of Health. The authors have no financial conflicts of interest.

Author contribution

Robert R. Snapp designed the component tree method and wrote the manuscript; Elyse Goveia and Lindsay Peet implemented successive iterations of the adaptive threshold method; Nicole A. Bouffard performed the nuclear measurements; Gary Badger performed the statistical analyses; Helene M. Langevin conceived the study, obtained funding and coordinated the data analysis and writing of the manuscript. All authors contributed to writing and editing the manuscript.