Mitochondrial Dysfunction Accounts for the Stochastic Heterogeneity in Telomere-Dependent Senescence

Central to current understanding of biological aging is the idea that limited evolutionary investment in mechanisms of cellular and molecular maintenance and repair causes a gradual accumulation of cellular damage, which in turn causes age-related frailty, disease, and death [ 1 ]. It follows that aging is not itself genetically programmed, although longevity is, through the genetic regulation of repair mechanisms. It also follows that multiple molecular mechanisms are expected to participate in cell aging and that the actions of these mechanisms are inherently stochastic, i.e., subject to the variations of chance. The essential multiplicity of mechanisms of aging is widely acknowledged, although in practice, it has not yet been much studied, whereas the stochastic nature of aging is evident in the pronounced variability that is a hallmark of the aging process [ 2 – 5 ]. In this paper, we describe a study of the multiplicity of mechanisms contributing to one of the best-studied cellular models of aging, i.e., the replicative senescence (loss of proliferative capacity) of human diploid fibroblasts (HDFs), which occurs after a number of population doublings in vitro [ 6 ]. We show that interactions between mechanisms whose connections have not been previously addressed in this system provide new insights into the underlying biology of replicative senescence. We also show that these interactions explain the marked cell-to-cell variation in replicative lifespan within a culture, which is seen even if cultures are grown clonally from single cell founders [ 7 ]. Replicative senescence in HDF is known to be caused ultimately by a DNA-damage response that is triggered by uncapped telomeres [ 8 , 9 ], which in turn result from replication-driven loss of telomere repeats in the absence of telomerase. This process is sometimes interpreted as “programmed aging” at the cell level, although the idea of an underlying program driving replicative senescence is hard to reconcile with its extensive, intrinsic heterogeneity. Furthermore, telomere-shortening rate and cell replicative lifespans can be greatly modified by DNA-damaging oxidative stress [ 10 ] via a telomere-specific repair deficiency, which causes stress-dependent accumulation of single-strand breaks [ 11 ] and accelerates telomere shortening during DNA replication [ 12 ]. This had led to the suggestion that telomere reduction is not strictly programmed, with telomere length acting as a mere cell-division counting device, but instead that telomeres act as sentinels for cumulative oxidative and/or environmental stress triggering division arrest when the damage burden (detected through telomere length) becomes too great [ 10 ]. Such an idea is fully consistent with the suggestion that replicative senescence serves a pleiotropic role in aging as an anti-cancer mechanism [ 13 ] for which there is growing evidence [ 14 – 17 ], since the damage that triggers senescence would otherwise pose an increased risk of malignancy. It is also consistent with recent evidence linking environmental stress to telomere length in humans [ 18 ], which could provide important insights into the nature of the connection between replicative senescence and organismal aging, as well as into the potential malleability of these processes. It has been recognized for some time that the heterogeneity in telomere-driven replicative senescence challenged the simple idea that in telomerase-negative cells, the telomeres shorten progressively through failure of complete chromosome-end replication. It has been suggested [ 19 ] that telomere uncapping, the process that leads to telomeres being recognized as inducers of a DNA-damage response triggering growth arrest, is itself stochastic, but such an effect does not appear sufficient to explain the observed heterogeneity in division capacity within cultures. Rubelj and Vondracek proposed a hypothetical model of “abrupt telomere shortening,” but this amounted only to speculation embodied in equations [ 20 ]. A data-driven modelling approach has shown how the effects of oxidative stress on telomere shortening can be used to study the dynamics of replicative senescence in silico and, in particular, how damage-induced mitochondrial defects might underlie stochastic cell-to-cell variation in oxidative stress, which by acting on telomeres can account for the observed heterogeneity [ 21 – 23 ]. Most of the evidence to date for mitochondrial involvement in aging has come from post-mitotic cells, although recently it was shown that extensive, stochastic variability in the presence and level of mitochondrial DNA mutations occurs in human colon stem cells, one of the most actively proliferating cell types in the body [ 24 ]. Motivated by this background of theoretical and experimental work, we have examined the possible role of mitochondria and oxidative stress in telomere-driven replicative senescence, something which until now has been unclear [ 25 ]. We first sho

Mitochondrial dysfunction induces a major reprogramming of nuclear gene expression, termed the retrograde response [ 27 ], which is a feature of replicative aging in budding yeast [ 28 ] and can be induced by severe mtDNA depletion in human cells [ 29 ]. Retrograde response includes dysregulation of Ca 2+ -dependent signalling, up-regulation of mitochondrial biogenesis [ 30 ], and major metabolic and anti-apoptotic adjustments [ 27 ]. However, induction of retrograde signalling in mammalian cellular senescence has not been demonstrated so far. An essential trigger for retrograde response is believed to be the increased cytoplasmic [Ca 2+ ] i levels due to low Ca 2+ storage capacity of dysfunctional mitochondria with low membrane potential [ 27 ]. Senescent fibroblasts displayed significantly higher cytoplasmic Ca 2+ levels under basal conditions as well as slower recovery after a Ca challenge as measured by live cell confocal imaging using the Ca 2+ -sensitive dye Fluo3 ( Figure 1 E and 1 F). Senescent fibroblasts also show increased mitochondrial mass and number ( Figure S1 ) consistent with activation of mitochondrial biogenesis [ 30 ]. To further characterize retrograde response in fibroblast senescence, we performed an RNA microarray expression analysis (see Protocol S1 ). KEGG pathway analysis comparing young with senescent cells indicated significant enrichment of differentially expressed genes in various major Ca 2+ -related signalling pathways as well as in energy (nitrogen and sulphur), nucleotide, and amino acid metabolism ( Table S2 ). Expression of 2,610 gene spots changed at least 2-fold between young and senescent cells, with the majority of these (1,522) down-regulated in senescence as compared to young fibroblasts ( Figure S2 ). A limited number of these changes (mostly up-regulations) were confirmed in cells that reached senescence under hyperoxia ( Figure S2 ). From all genes that were changed similarly under both normoxic and hyperoxic senescence, 92 genes (120 probe sets) could be assigned to one of four processes that are known to be activated in retrograde response: Ca 2+ -binding and Ca 2+ -mediated signalling ( Figure 1 G), glycolysis and Krebs cycle enzymes ( Figure 1 H), mitochondrial biogenesis and function ( Figure 1 I), and stress response ( Figure 1 J). The majority of these genes (69) are up-regulated in senescence. In particular, we noted activation of enzymes involved in calcium/calmodulin-dependent signalling, including PRKC A and H, PLCB4, EREG, CEBPE, CREB3L1 and -L4, TGFβ2, and IGFBP3 ( Figure 1 G). A number of enzymes involved in glucose metabolism, such as HK2, PDP2, and PDK4 ( Figure 1 H), glutamate metabolism, such as GCLM and GLUL ( Figure 1 H), and lipid metabolism, such as CPT1A, PTE1, and FABP4 ( Figure 1 H), were up-regulated, as well as enzymes with major functions in mitochondrial metabolism, such as AK3, PDK4 ( Figure 1 H), and MAOA ( Figure 1 I). Up-regulation of some of these genes in senescence and under hyperoxia was confirmed by reverse-transcriptase (RT)-PCR ( Figure S3 ). Increased resistance of cells to stress and apoptosis has also been described as part of the retrograde response. We observed increased mRNA expression of the antioxidant SOD2 and the anti-apoptotic BCL-2 family member BCL2L1 (BCL-X), and a decrease in the expression of the pro-apoptotic BCL2L11 (BIM) ( Figure 1 J). Taken together, the fact that mitochondrial dysfunction during senescence of human MRC5 fibroblasts is accompanied by increased mitochondrial biogenesis, compromised [Ca 2+ ] i handling, and transcriptional up-regulation of genes involved in Ca-mediated signalling, stress response, glycolysis, and mitochondrial function suggests that cellular senescence might be associated with the induction of retrograde response, which in turn is likely to be caused by mitochondrial dysfunction.

There is good evidence that reducing cellular ROS levels can extend the replicative lifespan of human fibroblasts [ 35 , 36 ], and delay telomere shortening [ 33 , 34 , 37 , 38 ], suggesting that mitochondrial dysfunction contributes to replicative senescence. However, other data indicate that mitochondrial ROS production might occur as a result of senescence ([ 39 – 41 ], unpublished data). To establish the existence and direction of a causal relationship, we slowed down mitochondrial superoxide production by mild uncoupling in young cells, because mitochondrial uncoupling has been proposed as a mechanism of cellular adaptation to protect against oxidative stress [ 42 ]. Treatment of MRC-5 cells with a 250 μM concentration of the uncoupler 2,4-dinitrophenol (DNP) resulted in mild mitochondrial uncoupling as measured by a decreased JC-1 fluorescence ratio (85.2% ± 0.4%, p = 0.05; Figure S4 A). This is important because severe uncoupling, for instance by overexpression of UCP-1 [ 43 ] or by using FCCP and DNP under conditions of high oxygen supply (unpublished data), can increase mitochondrial ROS production. Cellular ATP levels were decreased following DNP treatment and reached 75% ± 10% of those in untreated cells after 2 wk. Moreover, long-term DNP treatment prevented the further increase of mitochondrial mass that is characteristic for cells approaching senescence ( Figure S4 B). Importantly, uncoupling by DNP slowed the accumulation of superoxide in mitochondria of intact fibroblasts ( Figure 3 A). In comparison to controls, the replicative lifespan was extended by 4–5 PD, amounting to an increase of about 50% of the treated period ( Figure 3 B). Increased activity of Sen-β-Gal has been associated with cellular senescence [ 44 ], and this increase is also delayed under treatment with DNP ( Figure S5 ). Figure 3 Mild Uncoupling by DNP Prolongs Replicative Lifespan of Human Fibroblasts MRC5 fibroblasts were treated with 250 μM DNP from PD 35 till senescence. (A) MitoSOX fluorescence. Data are mean + Δm, n = 2. Fluorescence intensity at t = 0 was set to 100%. Solid line: linear regression, dotted lines: 95% confidence intervals. (B) Replicative lifespan. Data are from three (DNP) respective four (control) independent experiments. Hyperbolic best fits and 95% confidence intervals are shown. (C) γ-H2A.X staining. The bar denotes 20 μm. Insert: higher magnification showing focal staining. (D) Frequency of γ-H2A.X-positive fibroblasts. Data are mean ± s.d., n = 3. Differences are significant ( p < 0.05) between day 8 and 46. To establish the mechanism by which uncoupling extends replicative lifespan, we first measured the frequency of cells with nuclear foci containing the phosphorylated form of the histone variant H2A.X, γ-H2A.X ( Figure 3 C). Formation of foci containing γ-H2A.X can be initiated by DNA double-strand breaks or uncapped telomeres, and is a hallmark of the signalling pathway leading to senescence [ 8 , 45 – 47 ]. Mitochondrial uncoupling delayed foci formation in the majority of cells ( Figure 3 D). We next investigated the effect of mitochondrial ROS production on telomere maintenance. Telomere shortening is the major route to telomere uncapping in senescing fibroblasts [ 9 ], and the rate of telomere loss in telomerase-negative human somatic cells can be significantly modulated by either increasing oxidative stress [ 10 ] or by antioxidant protection [ 34 , 37 ]. However, no direct effect on telomere maintenance has been demonstrated before. Mitochondrial uncoupling by DNP resulted in improved telomere maintenance ( Figure 4 A). The telomere shortening rate of 80 ± 14 base pairs (bp)/PD in controls decreased to 9 ± 29 bp/PD in treated cells, so that DNP-treated cells senesced with longer telomeres than controls ( Figure 4 B). This effect of DNP was confirmed by measuring telomere length by quantitative fluorescence in situ hybridisation (Q-FISH) on metaphases ( Figure 4 C). The telomere shortening rate measured here in DNP-treated cells is lower than what would be expected if telomere shortening was governed by simple overhang resection [ 48 ]. However, overhang length and telomere shortening are not correlated in human fibroblasts [ 49 ]. Moreover, telomere shortening was similarly slow in other fibroblast systems under comparatively low oxidative stress [ 12 , 33 , 50 ]. To further corroborate the association between oxidative stress levels and telomere shortening, ROS levels in three different human fibroblast strains were modified by growth under high or low oxygen concentrations, treatment with the free radical scavenger PBN, or overexpression of SOD, and telomere shortening was measured by two independent methods (in-gel hybridisation and real-time PCR). A strong positive correlation was obtained ( Figure S6 ), suggesting that lowering the levels of ROS is relevant for the decreased telomere shortening rate under DNP treatment. Figure 4 Delayed Senescence in DNP-Treated Fibroblasts I

Since the pioneering work of Harman [ 56 , 57 ], a wealth of data has been accumulated that supports a causal role for mitochondrial dysfunction and ROS production in aging of postmitotic cells [ 26 ]. Mitochondrial dysfunction induces retrograde response, a pathway that signals electron transport chain disruption to the nucleus, thus causing wide-ranging adaptations. This is an important part of replicative aging in yeast [ 58 , 59 ]. Interestingly, induction of the retrograde response plays partially contradictory roles in yeast aging: as a part of the normal aging process, it can extend lifespan if activated early, but it also contributes to genomic instability and thus curtails longevity [ 58 ]. Retrograde signalling has been described in human cells deprived of mitochondrial DNA or following mitochondrial uncoupling [ 30 , 60 ]. Disruption of the MMP results in elevated cytoplasmic free [Ca 2+ ] i and activation of Ca 2+ -dependent signalling including calcineurin and the NFkB pathway, PKC, CREB, and the JNK/MAPK pathway. Retrograde signalling includes up-regulation of mitochondrial biosynthesis via PGC-1 and PPAR-γ, major metabolic and anti-apoptotic adjustments, and possible interaction with TOR signalling [ 27 ]. However, there appears to be a large amount of plasticity in this response, depending on, among other things, the specific treatment and cell line examined. Moreover, the role, if any, of mitochondrial dysfunction and retrograde signalling in mammalian cellular senescence is poorly understood [ 25 ]. Our data showing decreasing mitochondrial membrane potential, increased mitochondrial biogenesis, decreased capacity for cytoplasmic [Ca 2+ ] i regulation, up-regulation of various genes involved in Ca 2+ -dependent signalling pathways, metabolic readjustment, and activated anti-apoptotic response are consistent with induction of retrograde signalling in near-senescent human fibroblasts. It should be stressed that the specific subset of genes identified in Figure 1 G– 1 K cannot be taken as an unbiased representation of retrograde response, because they haven been isolated a posteriori from a sample of differentially expressed genes, and they have not been shown to be individually related to this process. However, they might represent a first approach towards candidates for “signature” genes of retrograde response in senescent fibroblasts. Our data are in accord with recent data showing metabolic impairment in senescent cells, including a significant decline in ATP levels [ 61 ]. Thus, we propose that mitochondrial dysfunction and resultant retrograde signalling might be a conserved mechanism in cellular replicative aging from yeast to man. Interestingly, our results suggest that the disruption of the MMP in senescent fibroblasts is not simply the result of direct ROS-mediated damage. Rather, it is at least in part an adaptation mediated by increased expression of UCP-2 and, possibly, UCP-3. Low MMP together with high ROS levels was found before in postmitotic cells [ 62 ]. Activity of uncoupling proteins, including UCP-2, is induced in response to lipid peroxidation products [ 63 ] or to superoxide anion in vitro [ 32 ] and in vivo [ 64 ]. UCP-2 increases the proton leak and decreases MMP and mitochondrial ROS production [ 65 ]. This led to the hypothesis of mitochondrial uncoupling as an adaptive mechanism to oxidative stress with potential relevance for aging of postmitotic cells [ 42 ]. However, this hypothesis has recently been challenged by showing that physiologically controlled overexpression of SOD-2 in mice, although resulting in decreased mitochondrial superoxide production, did not alter mitochondrial coupling, MMP, or activity of uncoupling proteins in brown fat or skeletal mitochondria [ 66 ]. There are conflicting data regarding the interaction between mitochondrial uncoupling and ROS generation and its role for aging in vivo: Mitochondrial uncoupling levels correlated positively with lifespan in outbred mice [ 67 ], and overexpression of the human UCP-2 in adult neurons decreased cellular oxidative damage and extended the lifespan of flies [ 68 ]. However, mitochondrial coupling decreased with age in mice skeletal muscle concomitant with altered cellular metabolism and energetics, but without observable changes in the expression of UCP-3, the major uncoupling protein in muscle [ 69 ]. Moreover, long-term caloric restriction increased rat skeletal muscle UCP-3 content and decreased mitochondrial H 2 O 2 production, but did not increase leak-dependent (state 4) respiration [ 70 ]. Thus, a thorough study of the effects of ROS level variation on the expression of uncoupling proteins and MMP in a well-defined cellular system was warranted. We found here significant inverse correlations between both mitochondrial superoxide and cellular peroxide levels and MMP, which were independent of the specific type of intervention used to modify ROS levels. This confirms a primary signalling rol

To measure mitochondrial superoxide, cells were stained in 5 μM MitoSOX Red (Molecular Probes, http://www.invitrogen.com ) for 10 min at 37 °C, and FL3 median fluorescence intensity was measured by flow cytometry (Partec PAS, http://www.partec.com ). Specificity of MitoSOX for superoxide has been shown by the manufacturer, and its mitochondrial localisation was tested by co-staining with Mitotracker Green (unpublished data). Cellular peroxide levels were assessed by staining with 30 μM DHR (Molecular Probes) for 30 min at 37 °C and analysis of FL1 fluorescence. As a positive control, cells were treated with H 2 O 2 (0 to 400 μM for 30 min) before DHR staining (unpublished data). MMP was measured as the FL3/FL1 ratio after staining cells with 1-μg/ml JC-1 (5,5',6,6'-tetrachloro-1,1',3,3'- tetraethylbenzimidazolylcarbocyanine iodide; Molecular Probes) in phenol-free RPMI 1640 (Sigma) for 30 min at 37 °C. Uncoupling with carbonyl cyanide p-trifluoromethoxyphenylhydrazone (FCCP) (20 μM for 30 min) was used as positive control, resulting in a 60% decrease of the JC1 ratio. Mitochondrial mass was measured as FL1 fluorescence after staining of cells with 10 μM nonyl acridine orange (NAO; Molecular Probes) for 10 min at 37 °C in the dark. The flow cytometer was calibrated using fluorescent microspheres. All data are mean ± s.e.m. from n independent experiments with measurements in duplicate and 10 4 cells per measurement. Cells growing on coverslips were stained as above and observed using a Zeiss LSM 510 Meta confocal microscope (Carl Zeiss, http://www.zeiss.com ). Conventional transmission electron microscopy was used to confirm NAO staining. Twelve sections of cells with central nuclei per group were systematically random sampled to measure the volume density of mitochondria per cell.

Telomere restriction fragment length was measured by in-gel hybridisation following pulsed-field gel electrophoresis (CHEF3; BioRad, http://www.bio-rad.com ) as described [ 51 ]. The amount of loaded DNA and its integrity was controlled by EtBr staining of total DNA in the gel and by re-hybridisation with a (CAC) 8 probe as described [ 34 ]. The average telomere length was calculated as weighted mean of the peak using AIDA v3.11 image analysis software (Raytek, http://www.raytest.de ). In some experiments, telomere length was additionally measured using real-time PCR as described [ 51 ]. The telomere shortening rate per PD was calculated as the slope of the linear regression between telomere length and PD.

Metaphases were prepared by treatment of subconfluent cells with 10-μg/ml Colcemid for 2 h at 37 °C, followed by 60 mM KCl for 15 min at room temperature and fixation in ethanol: acetic acid (3:1). Slides were air dried and baked at 60 °C for 1 h, re-hydrated in 1 ml of 2× SSC at 37 °C for 2 min, and dehydrated. A total of 20 μl of Cy-3–labelled telomere-specific (C 3 TA 2 ) 3 peptide nucleic acid (PNA) probe (4 ng/μl) (DakoCytomation, http://www.dako.com ) was applied to the cells followed by co-denaturation at 85 °C and hybridisation for 2 h at room temperature in the dark. Cells were washed with 2× SSC/0.05% Tween for 10 min at room temperature. Telomere signals from at least 14 metaphases per group were quantified using TelomereQuant v 1.0 (Dako).

Figure S1 Senescent MRC5 Fibroblasts Contain More Mitochondria Than Young Ones (A) Electron micrographs of perinuclear areas in proliferating (PD 35, left) and near-senescent (PD 46, right) cells. Note the perinuclear accumulation and degenerative changes of mitochondria, such as formation of cup-shaped (c) and circular mitochondria containing inclusions of dense osmiophilic material, in the senescent cell. n, nucleus; m, mitochondrion. (B) Confocal images of NAO-stained fibroblasts taken under identical imaging conditions. Note the intense perinuclear staining in the senescent cell. (C) Mitochondrial volume density V V mitochondria (in percentage of cell volume) as measured by electron microscopic morphometry. Data are mean ± s.e.m. from 12 cells per group. (D) Quantification of NAO fluorescence by FACS, indicating mitochondrial mass per cell. Data are mean ± s.e.m., n = 4. (E) Relative mtDNA copy number as measured by quantitative real-time PCR. Data are mean ± s.e.m., n = 3. (426 KB JPG) Click here for additional data file. Figure S2 Hierarchic Cluster Analysis of Gene Expression in Senescent MRC5 Cells Young fibroblasts arrested by confluency in G0/G1 (four independent experiments, YOUNG) were compared to fibroblasts grown to senescence under normoxia (three experiments, SEN). A total of 2,610 probe sets (accession numbers) were changed at least 2-fold between YOUNG and SEN, with the majority (1,552) of these being down-regulated in senescent cells. Only a limited number of these changes were confirmed in fibroblasts grown to senescence under hyperoxia (two experiments, HYPER). Expression levels are colour-coded as shown on the right. (673 KB JPG) Click here for additional data file. Figure S3 Up-Regulation of Signalling and Metabolic Enzymes in Senescence and under Hyperoxia RT-PCR was performed for the genes indicated on the right using conditions indicated in Table S1 . GAPDH was measured as loading control. H, cells kept under hyperoxia for 47 d; S, senescent cells (PD 48); Y, young cells (PD 25). (139 KB JPG) Click here for additional data file. Figure S4 Effect of Long-Term DNP Treatment on MMP and Mitochondrial Mass (A) DNP treatment decreases MMP in pre-senescent, but not in senescent cells. MMP was measured as FL3/FL1 ratio of JC-1–stained cells treated with 250 μM DNP for the indicated times. Data are mean ± s.e.m., n = 3. Stars denote significant differences to controls with p = 0.05 (Paired t -test, two-tailed). (B) Long-term DNP treatment partially compensates for the increase in mitochondrial mass during senescence. Mitochondrial mass measured as FL1 fluorescence of NAO-stained cells. Data are mean ± s.e.m., n = 3. Star denotes significant differences to controls with p = 0.003 (Paired t -test, two-tailed). (147 KB JPG) Click here for additional data file. Figure S5 DNP Slows Down the Accumulation of Sen-β-Gal–Positive Cells Control cells at PD 46 (A) and DNP-treated cells at PD 50 (B) were both stained for Sen-β-Gal at day 56 of the experiment. (237 KB JPG) Click here for additional data file. Figure S6 Correlation between Cellular ROS Levels and Telomere Shortening Rates in Human Fibroblasts ROS levels were varied in three human fibroblast strains (P43, upward triangles; P100, downward triangles; and MRC5, circles) by growing cells until senescence under either normoxia (filled symbols), hyperoxia (open symbols), hypoxia (red symbols), or hypoxia plus addition of 400 μM α-phenyl-butyl-nitrone (PBN), a free radical scavenger (pink symbols). MRC5 cells were not grown under hypoxia, but were treated with 250 μM DNP (green symbol) or were transfected with an SOD3 transgene [ 34 ] (yellow symbol) instead. Cytoplasmic peroxide levels were measured by DHR flow cytometry. DNA samples were taken every 2–3 PD (a total of between three and 14 samples per strain and condition), and telomere length was measured by in-gel hybridisation and (for P100 cells) by real-time PCR. Measurements were performed in quadruplicate gels or PCRs, and telomere shortening rates were calculated by linear regression. Data are mean ± s.e.m. The correlation between DHR fluorescence and telomere shortening rate was modelled by an exponential regression with an r 2 = 0.956. (171 KB JPG) Click here for additional data file. Protocol S1 Supplemental Protocols (35 KB DOC) Click here for additional data file. Table S1 PCR Conditions (28 KB DOC) Click here for additional data file. Table S2 KEGG Pathway Functional Enrichment Analysis: 20 Most Statistically Significant Annotation Terms (42 KB DOC) Click here for additional data file.