Robust nuclear lamina-based cell classification of aging and senescent cells

Cell senescence can be described by spatial changes in the nuclear lamina

During in vitro propagation, the hMSCs undergo cellular senescence within a few passages and with an associated reduction in cell doubling [11-13]. Cellular senescence of hMSCs is marked by an accumulation of p16^INK4a and a decline in hTERT accumulation (Figure 1A). Senescent hMSCs also exhibit changes in the shape of the nuclear lamina [4] and the nuclear lamina is deformed in cells with high p16^INK4a expression (Figure 1B). Similarly, misshaped nuclear lamina are exhibited in cells at passage 10 with undetectable hTERT expression (Figure 1B). This cell-based analysis indicates that these changes in lamina shape are associated with cellular senescence. An unbiased description of the lamina shape could help to identify these cells.

Recently we developed a method for quantification of the three-dimensional (3D) structure of the nuclear lamina [10]. This method was successful in discriminating between healthy and apoptotic cells that show massive changes in the nuclear lamina shape [10]. In the current study we have applied this method to fresh and senescent hMSCs, at passage 4 and 9, respectively. The nuclear lamina was visualized with lamin A-GFP, which was expressed with the lentivirus expression system. Z-stacks were taken from confocal images and 3D reconstruction revealed spatial changes in the nuclear lamina during cell senescence (Figure 1C). In fresh cells at passages 4-6, the nuclear lamina structure typically had an elliptical shape, whereas an irregular spatial lamina shape was found in senescent cells between passage 7 and 12 [4]. To examine spatial changes in lamin A the distributions of curvature and intensity in cross sections were compared between fresh and senescent cells. Typical nuclei from each passage are shown in Figure 1D. It is on these types of 3D images that we measure three features: the average normalized intensity (termed intensity), the skewness of the intensity distribution (termed skewness), and the normalized average absolute Gaussian curvature (termed curvature). Formal definitions of these features can be found in [10].

Vertical cross sections (y, z) revealed substantial changes in the distribution of intensity and curvature values between cells at passage 4 and passage 9 (Figure 1D). High values of normalized intensity were found mainly at the edges of the oblate spheroid in fresh cells (Figure 1D, PS4, (y, z)), but were distributed across the nuclear lamina in senescent cells (Figure 1D, PS 9 (y,z)). High intensity values of lamin A were also found in intranuclear structures (Figure 2, PS 9 (y,z)). Typically, intranuclear structures are formed in senescent hMSCs [4]. In most cases, regions with high intensity values of lamin A were surrounded by high curvature values (Figure 1D). Spatial changes of lamin A distribution were less pronounced in horizontal cross sections (Figure 1D; (x, y)), suggesting that the structural changes seen in horizontal slices show only part of the spatial changes of the nuclear lamina shape.

Next we investigated whether cell populations at passages 4 and 9 could be distinguished based on a quantitative description of the lamina shape. The values of the intensity, skewness, and curvature were plotted in 2D-scatter plots. Between passages 4 and 9 the distribution of intensity and mean intensity predominantly altered while the curvature changed only slightly (Figure 1E). The 2D-scatter plots revealed clustering of cell populations for any combination of two of the three features (Figure 1E). Classification errors between 7% ± 3.2% to 13% ± 4.1% were found using a linear classifier and cross-validation testing. The p-value of classification errors (P < 0.01) indicated that the classification is highly significant in all test combinations. The classification error drops to 6.1% ± 2.9% in a test combining all three descriptors. A few outlier cells were indentified (marked with arrows in Figure 1E). These could represent biological variations. As we aim to develop a method for cell classification that does not require manual manipulation of the dataset outliers were included in all analyses.

To validate the classification results, the lamina quantification method was applied to cell populations at passages 6 and 12 where endogenous lamin A was visualized with immunohistochemistry (Figure 1Fi). After quantification of the nuclear lamina, using the same procedure as described for living cells, a linear classification was found between cells at passage 6 and passage 12 (Figure 1Fii), and this, too, was significant, p < 0.01 (Figure 1Fiii). The classification error in the test for the two features, curvature and intensity, was similar between living and fixed cells, 13.0% ± 4.1% and 16.5% ± 6.2%, respectively. Moreover, in both living and fixed cells the values of mean normalized intensity are reduced in senescent cells. This indicates that changes in lamina structure during cell senescence are not caused by lamin A overexpression, but are induced by senescence.

The analysis of nuclear lamina requires successful segmentation of the 3D nuclear lamina from the rest of the image field-of-view [10]. Two differences were noted between segmentation of living and fixed cells. While 100% of living cells were successfully segmented, in fixed cells only 70% were correctly segmented. Unsuccessful segmentation means that the lamina contour was not completely found and these cells were excluded from further study. In addition, in tests that included the skewness, higher errors were found in fixed cells (Figure 1Fiii). This suggests that during the immunohistochemistry procedure structural artifacts could be generated. In a test including all three descriptors a significant classification was found. Together, following a successful segmentation, lamina-based cell classification between fresh or senescent cells is robust in fixed and in living cells.

Lamina-based cell classification is highly effective between cell populations

Changes in lamina structure are also induced by mutations in lamin A or lamin B genes. To evaluate the classification method on cells expressing mutations in a lamina gene, the hMSCs were transduced with a lentivirus expressing lamin A-R220Q-GFP mutant [4] or lamin B-L158D-GFP. Both mutant proteins leads to distorted nuclear shape [8]. Classification tests were performed between cells expressing these mutants at early passage number and a wild type lamin A-GFP expressing cells. At early passage number (passage 5 or 4), cells expressing lamin A-R220Q are significantly distinguished from those expressing wild-type lamin A (Figure 2Ai). Significant classification was not found, however, between cells expressing lamin A-R220Q at passage 5 and cells expressing WT-lamin A-GFP at passage 9 (Figure 2Aii). This is demonstrated by the high classification error >40%, p ≈ 0.5. Previously we showed that lamin A-R220Q-GFP expression in hMSCs leads to senescence-associated changes [4]. The lamina classification results suggest that in cells expressing lamin A-R220Q-GFP the nuclear lamina acquires a senescent shape. Cells expressing lamin B-L158D-GFP at passage 5 were significantly distinguished (p < 0.01) from young and senescent cells at passage 4 and 9, respectively (Figures 2Bi and 2Bii). Cells expressing lamin B-L158D-GFP show characteristics of apoptotic cells [8], therefore, a comparison to apoptosis-activated cells was preformed. The inducible caspase-8 (FKC8) was expressed in cells expressing lamin A-GFP and 16 hours after FKC8 activation cells were imaged, as previously described in [8]. No classification (p ≈ 0.5) was found between cells expressing lamin B-L158D-GFP and those undergoing apoptosis by the activated FKC8 (Figure 2Biii).

To further evaluate the robustness of the classification procedure, we performed a classification test on three cell populations: passage 4, passage 9 and FKC8-activated cells. The lamina was visualized with lamin A-GFP. A linear classifier was sufficient to distinguish between all three cell populations (Figure 3A). Evaluations of classification errors with a cross-validation test indicated that the classification is highly significant (Figure 3B). Based on the distribution of curvature and intensity values in the cell populations (Figure 3A), we suggest that when cells senesce changes in lamina structure are predominantly associated with a change in intensity, whereas in cells undergoing apoptosis curvature values are also increased.

A few cells at passage 9 (~6%) showed high curvature values similar to these measured in the FKC8-activated cells (Figure 3A, indicated with arrows). This suggests that apoptotic cells could be identified in a senescent cell population. The presence of apoptotic cells was confirmed by cleaved caspase-3 expression in protein extracts of cells at passage 12 (Figure 1A) where ~9% of the cells showed the expression (Figure 3C). Based on the lamina quantification method, 6% of the cells at passage 9 showed curvature values similar to those in FKC8-activated cells. This indicates that activation of apoptosis in senescent cells can be estimated with the lamina quantification method.

Cell classification between primary myoblasts from young and old donors

In muscle fibers, the shape of the nuclear lamina changes during aging [4]. To evaluate whether the lamina classification method is also suitable to distinguish cells from young and aging individuals, the nuclear lamina shape was examined in primary myoblasts from 37 year-old (myo-37y) and 65 year-old (myo-65y) donors. Fresh myo-65y cells showed key features of senescence including accumulation of p21^CDKN1A and p16^INK4a, heterochromatic foci, and reduced cell fusion, whereas these characteristics were not found in the myo-37y cells (Figures 4A and 4B).

Lamina quantification was performed on endogenous lamin A, which was visualized with immunofluoresence in fresh cultures. Based on the lamina classification method, the myo-37y and myo-65y cell populations were significantly distinguishable (Figure 4C). The intensity values in myo 65y were lower than in the myo-37y cell populations, while skewness and curvature values were higher in the 65y myoblasts, A similar change was found between fresh and senescent hMSCs (Figure 1D). This suggests that structural changes in the nuclear lamina during aging and cellular senescence are similar.

Local accumulation of lamin A-GFP is associated with reduced protein mobility

Our studies here revealed that folding of the nuclear lamina is associated with changes in the mean normalized intensity and skewness change during senescence and aging. This suggests that redistribution of lamin A-GFP during cell senescence is associated with local protein accumulation in the nuclear lamina. To investigate this hypothesis, lamin A-GFP mobility was compared between regions with high and low fluorescence intensity (HFI and LFI respectively). Images of representative regions are shown in Figure 5A using the fluorescence recovery after photobleaching procedure (FRAP). As a control, FRAP was carried out on lamin A-GFP in the nucleoplasm (Figure 5A). The mean integrated fluorescence intensity before bleaching in regions with HFI are significantly higher than those of LFI or nucleoplasm regions (Figure 5B). Lamin A in HFI regions is significantly (P < 0.0001) less mobile compared with lamin A in LFI regions (Figure 5B). Confirming previous studies [4], lamin A at the nuclear envelope is far less mobile when compared to lamin A at the nucleoplasm.

To examine differences in lamin A-GFP mobility within a single nucleus, bleaching was carried out on HFI and LFI regions at opposite sides of the nucleus (Figure 5Ci). Statistical analysis of normalized intensity recovery plots revealed that T/2 and the immobile fraction significantly differed between HFI and LFI regions (Figure 5Cii). In HFI regions lamin A-GFP mobility was slower compared to the mobility in LFI regions. Together FRAP results show that local accumulation of lamin A at the nuclear lamina is associated with reduced mobility.

Cell culture and lentivirus transduction

The nuclear lamina was visualized with lamin A-GFP, which was expressed with a lentivirus vector previously described in [4, 8]. Activation of FKC8 was carried out with 100 nM AP20187 (ARIAD pharmaceuticals, Cambridge, UK), as described in [8]. Cells were seeded on glass plates prior to imaging. Cells with high lamin A-GFP overexpression, which does not correspond to endogenous lamin A, were excluded.

Human skeletal primary myoblasts from a 37-year-old and a 65-year-old donor (Tebu-bio, Le Perray en Yvelines, France) (named here as 37y and 65y, respectively) were cultured in a medium containing DMEM + 20% Fetal Calf Serum supplemented with an equal volume Skeletal Muscle Cell Media (PromoCell, Heidelberg, Germany) at 37°C under 5% CO₂. Cells were propagated for only one or two passages and subsequently were seeded on collagen-coated glass plates for imaging.

Protein detection and analysis

Immunofluorescence and western blotting were carried out as previously described [8]. For protein detection in western blots, aliquots of 50 μg were separated in 12% SDS-PAGE gel. Antibodies in this study are mouse anti-lamin A (Santa Cruz Biotechnology, Santa Cruz, CA, USA) or rabbit anti lamin A (Abcam, Cambridge, UK), rabbit anti-cleaved-caspase-3 (1:200, R&D Systems, Minneapolis, MN, USA), mouse anti-hTERT cloneY182 (Epitomics, Burlingame, CA, USA), mouse anti-human p16 and mouse-anti p21. (Both antibodies were kindly provided by A.G. Jochemsen, Leiden University Medical Center.)

Image processing and quantification

Cells were recorded using a Leica TCS/SP2 confocal microscope system equipped with a 100x/1.4 NA plan Apo objective. Living cells were imaged at 36.8°C and 5% CO₂. The spatial sampling distance for the Leica confocal system was Δz = 122 nm/voxel in the axial direction and Δx = Δy = 40 nm/voxel to 60 nm/voxel in the lateral directions. For three-dimensional reconstruction of images, Z-stacks were resampled as described in [25, 26]. The program for identification of the nuclear lamina and quantification of the structure is completely described in [10]. Statistical analysis and the classification tests are performed with PRTools (http://www.prtools.org). The leave-one-out cross-validation of classification error, tests the percentage of N cells that are classified wrongly when one cell is in the test set and N – 1 cells are in the training set. In this way all N cells are tested, one at a time, but at any given stage the cell under tested is not in the training set.

Fluorescence recovery after photobleaching (FRAP)

FRAP experiments were performed on a Leica TCS/SP5 confocal microscope as described in [17]. Prior to photobleaching, regions were selected based on lamin A-GFP intensity at the nuclear envelope. Fluorescence intensity was determined in single confocal cross sections, using Leica software, version 1.8.0 (Leica Microsystems GmBH, Wetzlar, Germany). Image recording at 0.555-second intervals and 4% laser transmission included 10 and 300 images before or after photobleaching, respectively. The recovery curves were calculated with Excel. Intensity values were corrected for background, fluorescence fading, and photo-bleaching during the image-recording period. P-values were calculated with the T-student test in Excel.