The Achilles' heel of senescent cells: from transcriptome to senolytic drugs

Abstract

The healthspan of mice is enhanced by killing senescent cells using a transgenic suicide gene. Achieving the same using small molecules would have a tremendous impact on quality of life and the burden of age-related chronic diseases. Here, we describe the rationale for identification and validation of a new class of drugs termed senolytics, which selectively kill senescent cells. By transcript analysis, we discovered increased expression of pro-survival networks in senescent cells, consistent with their established resistance to apoptosis. Using siRNA to silence expression of key nodes of this network, including ephrins (EFNB1 or 3), PI3Kδ, p21, BCL-xL, or plasminogen-activated inhibitor-2, killed senescent cells, but not proliferating or quiescent, differentiated cells. Drugs targeting these same factors selectively killed senescent cells. Dasatinib eliminated senescent human fat cell progenitors, while quercetin was more effective against senescent human endothelial cells and mouse BM-MSCs. The combination of dasatinib and quercetin was effective in eliminating senescent MEFs. In vivo, this combination reduced senescent cell burden in chronologically aged, radiation-exposed, and progeroid Ercc1(-/Δ) mice. In old mice, cardiac function and carotid vascular reactivity were improved 5 days after a single dose. Following irradiation of one limb in mice, a single dose led to improved exercise capacity for at least 7 months following drug treatment. Periodic drug administration extended healthspan in Ercc1(-/∆) mice, delaying age-related symptoms and pathology, osteoporosis, and loss of intervertebral disk proteoglycans. These results demonstrate the feasibility of selectively ablating senescent cells and the efficacy of senolytics for alleviating symptoms of frailty and extending healthspan.

Keywords: PI3K delta; dasatinib; dependence receptors; ephrins; p21; plasminogen-activated inhibitor; quercetin.

Fig 1

Senescent cells can be selectively targeted by suppressing pro-survival mechanisms. (A) Principal components analysis of detected features in senescent (green squares) vs. nonsenescent (red squares) human abdominal subcutaneous preadipocytes indicating major differences between senescent and nonsenescent preadipocytes in overall gene expression. Senescence had been induced by exposure to 10 Gy radiation (vs. sham radiation) 25 days before RNA isolation. Each square represents one subject (cell donor). (B, C) Anti-apoptotic, pro-survival pathways are up-regulated in senescent vs. nonsenescent cells. Heat maps of the leading edges of gene sets related to anti-apoptotic function, ‘negative regulation of apoptosis’ (B) and ‘anti-apoptosis’ (C), in senescent vs. nonsenescent preadipocytes are shown (red = higher; blue = lower). Each column represents one subject. Samples are ordered from left to right by proliferative state (N = 8). The rows represent expression of a single gene and are ordered from top to bottom by the absolute value of the Student t statistic computed between the senescent and proliferating cells (i.e., from greatest to least significance, see also Fig. S8). (D–E) Targeting survival pathways by siRNA reduces viability (ATPLite) of radiation-induced senescent human abdominal subcutaneous primary preadipocytes (D) and HUVECs (E) to a greater extent than nonsenescent sham-radiated proliferating cells. siRNA transduced on day 0 against ephrin ligand B1 (EFNB1), EFNB3, phosphatidylinositol-4,5-bisphosphate 3-kinase delta catalytic subunit (PI3KCD), cyclin-dependent kinase inhibitor 1A (p21), and plasminogen-activated inhibitor-2 (PAI-2) messages induced significant decreases in ATPLite-reactive senescent (solid bars) vs. proliferating (open bars) cells by day 4 (100, denoted by the red line, is control, scrambled siRNA). N = 6; *P < 0.05; t-tests. (F–G) Decreased survival (crystal violet stain intensity) in response to siRNAs in senescent vs. nonsenescent preadipocytes (F) and HUVECs (G). N = 5; *P < 0.05; t-tests. (H) Network analysis to test links among EFNB-1, EFNB-3, PI3KCD, p21 (CDKN1A), PAI-1 (SERPINE1), PAI-2 (SERPINB2), BCL-xL, and MCL-1.

Fig 2

Dasatinib and quercetin target senescent cells. (A) D is more effective in selectively reducing viability (ATPLite) of senescent preadipocytes than HUVECs. Preadipocytes and HUVECs were exposed to different concentrations of D for 3 days. The red line denotes plating densities on day 0 of nondividing senescent (set to 100%) as well as proliferating nonsenescent cells (also set to 100%). Preadipocyte data are means ± SEM of four experiments in each of four different subjects. HUVEC data are means ± SEM of five replicates at each concentration. (B) Q is more effective in selectively reducing viability (ATPLite) of senescent HUVECs than preadipocytes. Proliferating and senescent preadipocytes and HUVECs were exposed to different concentrations of Q for 3 days. Preadipocyte data are means ± SEM of four experiments in each of four different subjects. HUVEC data are means ± SEM of five replicates at each concentration. (C) Combining D and Q selectively reduced viability of both senescent preadipocytes and senescent HUVECs. Proliferating and senescent preadipocytes and HUVECs were exposed to a fixed concentration of Q and different concentrations of D for 3 days. Optimal Q concentrations for inducing death of senescent preadipocyte and HUVEC cells were 20 and 10 μm, respectively. (D) D and Q do not affect the viability of quiescent fat cells. Nonsenescent preadipocytes (proliferating) as well as nonproliferating, nonsenescent differentiated fat cells prepared from preadipocytes (differentiated), as well as nonproliferating preadipocytes that had been exposed to 10 Gy radiation 25 days before to induce senescence (senescent) were treated with D+Q for 48 h. N = 6 preadipocyte cultures isolated from different subjects. *P < 0.05; anova. 100% indicates ATPLite intensity at day 0 for each cell type and the bars represent the ATPLite intensity after 72 h. The drugs resulted in lower ATPLite in proliferating cells than in vehicle-treated cells after 72 h, but ATPLite intensity did not fall below that at day 0. This is consistent with inhibition of proliferation, and not necessarily cell death. Fat cell ATPLite was not substantially affected by the drugs, consistent with lack of an effect of even high doses of D+Q on nonproliferating, differentiated cells. ATPLite was lower in senescent cells exposed to the drugs for 72 h than at plating on day 0. As senescent cells do not proliferate, this indicates that the drugs decrease senescent cell viability. (E, F) D and Q cause more apoptosis of senescent than nonsenescent primary human preadipocytes (terminal deoxynucleotidyl transferase dUTP nick end labeling [TUNEL] assay). (E) D (200 nM) plus Q (20 μm) resulted in 65% apoptotic cells (TUNEL assay) after 12 h in senescent but not proliferating, nonsenescent preadipocyte cultures. Cells were from three subjects; four replicates; **P < 0.0001; anova. (F) Primary human preadipocytes were stained with DAPI to show nuclei or analyzed by TUNEL to show apoptotic cells. Senescence was induced by 10 Gy radiation 25 days previously. Proliferating, nonsenescent cells were exposed to D+Q for 24 h, and senescent cells from the same subjects were exposed to vehicle or D+Q. D+Q induced apoptosis in senescent, but not nonsenescent, cells (compare the green in the upper to lower right panels). The bars indicate 50 μm. (G) Effect of vehicle, D, Q, or D+Q on nonsenescent preadipocyte and HUVEC p21, BCL-xL, and PAI-2 by Western immunoanalysis. (H) Effect of vehicle, D, Q, or D+Q on preadipocyte on PAI-2 mRNA by PCR. N = 3; *P < 0.05; anova.

Fig 3

Dasatinib and quercetin reduce senescent cell abundance in mice. (A) Effect of D (250 nm), Q (50 μm), or D+Q on levels of senescent Ercc1-deficient murine embryonic fibroblasts (MEFs). Cells were exposed to drugs for 48 h prior to analysis of SA-βGal⁺ cells using C₁₂FDG. The data shown are means ± SEM of three replicates, ***P < 0.005; t-test. (B) Effect of D (500 nM), Q (100 μm), and D+Q on senescent bone marrow-derived mesenchymal stem cells (BM-MSCs) from progeroid Ercc1^−/Δ mice. The senescent MSCs were exposed to the drugs for 48 h prior to analysis of SA-βGal activity. The data shown are means ± SEM of three replicates. **P < 0.001; anova. (C–D) The senescence markers, SA-βGal and p16, are reduced in inguinal fat of 24-month-old mice treated with a single dose of senolytics (D+Q) compared to vehicle only (V). Cellular SA-βGal activity assays and p16 expression by RT–PCR were carried out 5 days after treatment. N = 14; means ± SEM. **P < 0.002 for SA-βGal, *P < 0.01 for p16 (t-tests). (E–F) D+Q-treated mice have fewer liver p16⁺ cells than vehicle-treated mice. (E) Representative images of p16 mRNA FISH. Cholangiocytes are located between the white dotted lines that indicate the luminal and outer borders of bile canaliculi. (F) Semi-quantitative analysis of fluorescence intensity demonstrates decreased cholangiocyte p16 in drug-treated animals compared to vehicle. N = 8 animals per group. *P < 0.05; Mann–Whitney U-test. (G–I) Senolytic agents decrease p16 expression in quadricep muscles (G) and cellular SA-βGal in inguinal fat (H–I) of radiation-exposed mice. Mice with one leg exposed to 10 Gy radiation 3 months previously developed gray hair (Fig.5A) and senescent cell accumulation in the radiated leg. Mice were treated once with D+Q (solid bars) or vehicle (open bars). After 5 days, cellular SA-βGal activity and p16 mRNA were assayed in the radiated leg. N = 8; means ± SEM, p16: **P < 0.005; SA β-Gal: *P < 0.02; t-tests.

Fig 4

Effects of senolytic agents on cardiac (A–C) and vasomotor (D–F) function. D+Q significantly improved left ventricular ejection fraction of 24-month-old mice (A). Improved systolic function did not occur due to increases in cardiac preload (B), but was instead a result of a reduction in end-systolic dimensions (C; Table S3). D+Q resulted in modest improvement in endothelium-dependent relaxation elicited by acetylcholine (D), but profoundly improved vascular smooth muscle cell relaxation in response to nitroprusside (E). Contractile responses to U46619 (F) were not significantly altered by D+Q. In panels D–E, relaxation is expressed as the percentage of the preconstricted baseline value. Thus, for panels D–F, lower values indicate improved vasomotor function. N = 8 male mice per group. *P < 0.05; A–C: t-tests; D–F: anova.

Fig 5

Senolytic administration alleviates radiation-induced impairment in treadmill exercise endurance. (A–B) One leg of 4-month-old mice was radiated at 10 Gy. Three months later, hair on the irradiated leg had turned gray (A) and treadmill exercise capacity (B) was lower in irradiated (N = 13) than sham-irradiated mice (N = 14). **P < 0.002; t-test. (C) Five days after a single dose of D+Q, treadmill endurance was better than in vehicle-treated controls. D+Q had no effect in sham-irradiated controls. (N = 6-9 animals per group). Bars represent means ± SEM; *P < 0.05; **P < 0.001; anova; Tukey–Kramer test. (D) 7 months after a single dose of D+Q, treadmill endurance was again assayed. All groups ran on the treadmill on four occasions, each 1 week apart. Bars represent means ±SEM of the average performance of each group on each of the four occasions they ran. Endurance is shown as a function of the overall performance of all four groups on each occasion when mice ran (expressed as %: mean Joules per group/total Joules per all groups that day). * Different from the other groups; P < 0.05; anova; Duncan’s test.

Fig 6

Periodic treatment with D+Q extends the healthspan of progeroid Ercc1^−/Δ mice. Animals were treated with D+Q or vehicle weekly. Symptoms associated with aging were measured biweekly. Animals were euthanized after 10–12 weeks. N = 7–8 mice per group. (A) Histogram of the aging score, which reflects the average percent of the maximal symptom score (a composite of the appearance and severity of all symptoms measured at each time point) for each treatment group and is a reflection of healthspan (Tilstra et al., 2012). *P < 0.05 and **P < 0.01 Student’s t-test. (B) Representative graph of the age at onset of all symptoms measured in a sex-matched sibling pair of Ercc1^−/Δ mice. Each color represents a different symptom. The height of the bar indicates the severity of the symptom at a particular age. The composite height of the bar is an indication of the animals’ overall health (lower bar better health). Mice treated with D+Q had delay in onset of symptoms (e.g., ataxia, orange) and attenuated expression of symptoms (e.g., dystonia, light blue). Additional pairwise analyses are found in Fig. S11. (C) Representative images of Ercc1^−/Δ mice from the D+Q treatment group or vehicle only. Splayed feet are an indication of dystonia and ataxia. Animals treated with D+Q had improved motor coordination. Additional images illustrating the animals’ gait and body condition are in Fig. S10. (D) Quantitative computed tomography (QCT)-derived bone parameters at the lumbar spine of 16-week-old Ercc1^−/Δ mice treated with either vehicle (N = 7) or drug (N = 8). BMC = bone mineral content; vBMD = volumetric bone mineral density. *P < 0.05; **P < 0.01; ***P < 0.001. (E) Glycosaminoglycan (GAG) content of the nucleus pulposus (NP) of the intervertebral disk. GAG content of the NP declines with mammalian aging, leading to lower back pain and reduced height. D+Q significantly improves GAG levels in Ercc1^−/Δ mice compared to animals receiving vehicle only. *P < 0.05, Student’s t-test. (F) Histopathology in Ercc1^−/Δ mice treated with D+Q. Liver, kidney, and femoral bone marrow hematoxylin and eosin-stained sections were scored for severity of age-related pathology typical of the Ercc1^−/Δ mice. Age-related pathology was scored from 0 to 4. Sample images of the pathology are provided in Fig. S13. Plotted is the percent of total pathology scored (maximal score of 12: 3 tissues x range of severity 0–4) for individual animals from all sibling groups. Each cluster of bars is a sibling group. White bars represent animals treated with vehicle. Black bars represent siblings that were treated with D+Q. The √ denotes the sibling groups in which the greatest differences in premortem aging phenotypes were noted, demonstrating a strong correlation between the pre- and postmortem analysis of frailty.