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

1Aging is the leading risk factor for the chronic diseases that account for the bulk of morbidity, mortality, and health costs (Goldman et al., ). A fundamental aging mechanism that likely contributes to chronic diseases and age-related dysfunction is cellular senescence (Kirkland,,; Tchkonia et al.,; Kirkland & Tchkonia, ). Senescence refers to the essentially irreversible growth arrest that occurs when cells are subjected to potentially oncogenic insults (Tchkonia et al., ). Even though senescent cell abundance in aging or diseased tissues is low, achieving a maximum of 15 percent of nucleated cells in very old primates, senescent cells can secrete pro-inflammatory cytokines, chemokines, and extracellular matrix proteases, which together constitute the senescence-associated secretory phenotype or SASP (Herbig et al.,; Coppé et al.,; Kuilman et al., ). The SASP likely contributes to the correlation between senescent cell accumulation and local and systemic dysfunction and disease. Consistent with a role for cellular senescence in causing age-related dysfunction, clearing senescent cells by activating a drug-inducible ‘suicide’ gene enhances healthspan and delays multiple age-related phenotypes in genetically modified progeroid mice (Baker et al., ). Interestingly, despite only clearing 30 percent of the senescent cells, improvement in age-related phenotypes is profound. Thus, interventions that reduce the burden of senescent cells could ameliorate age-related disabilities and chronic diseases as a group (Tchkonia et al.,; Kirkland & Tchkonia, ).

2Despite their harsh internal and external microenvironments, senescent cells are viable. They survive even though they have active DNA damage responses, heightened metabolic flux, and increased local levels of SASP inflammatory cytokines and other factors that are able to induce apoptosis. Indeed, senescent cells are better able to withstand stresses such as serum deprivation than nonsenescent cells (Wang,; Fridman & Lowe, ). In vivo, senescent cells appear to be removed by the immune system (Xue et al., ), rather than apoptosis or necrosis. Therefore, we hypothesized that (i) anti-apoptotic, pro-survival mechanisms could be up-regulated in senescent cells and (ii) interfering with these protective mechanisms might achieve selective elimination of senescent cells. Based on these hypotheses, here we identified several clinically used drugs that induce apoptosis preferentially of senescent cells in vitro and in vivo, leading to improved cardiovascular function and exercise endurance, reduced osteoporosis and frailty, and extended healthspan in several murine systems.

3We first tested our hypotheses by comparing gene expression in senescent vs. nonsenescent cells using transcript array analysis (Fig.1A–C). We focused on fat cell progenitors, or preadipocytes, as they are arguably one of the most abundant types of senescent cells in humans (Tchkonia et al., ). Cultures were split and senescence induced in half of the cells using 10 Gy of ionizing radiation. Twenty-five days later, gene expression was measured using Affymetrix arrays and differential expression patterns identified by principal component analysis (see Supporting materials and methods Data S1 for details). Overall patterns of gene expression differed substantially between senescent and nonsenescent preadipocytes isolated from the same subjects (Fig.1A). Analyses of gene categories indeed revealed up-regulation of negative regulators of apoptosis (Fig.1B) and anti-apoptotic gene sets (Fig.1C) in senescent compared to nonsenescent cells (see also Supporting information Fig. S8).

4We next employed RNA interference to identify potential ‘senolytic’ targets. We used the following rationale for the selection of senescence-associated genes to target with siRNAs. (i) Senescent cells rely on anti-apoptotic, pro-survival defenses to a greater extent than nonsenescent cells. (ii) Senescent cells have much in common with cancer cells, such as active DNA damage responses (Ghosal & Chen, ), except senescent cells do not divide. Thus pro-survival pathways, which when inhibited drive cancer cell apoptosis, might be good targets as long as the pathway is not linked to cell proliferation. (iii) Similarly to cancer cells, senescent cells are metabolically active, with increased glycolysis (Dorr et al., ). Thus, senescent cells may be more dependent on pathways that protect against metabolically induced apoptosis than are nonsenescent cells. (iv) Dependence receptors promote apoptosis unless they are prevented from doing so by the presence of their ligands (Goldschneider & Mehlen, ). Thus, senescent cells may rely more on dependence receptor ligands than nonsenescent cells, as is the case in cancer cells (Goldschneider & Mehlen,; Xi et al., ). Therefore, we targeted anti-apoptotic pathway elements by RNA interference, including the dependence receptor ligands and metabolic pro-survival transcripts identified in oncology. Importantly, targets identified by this rationale have the potential to yield senolytics that also will have antitumor effects.

5Of the 39 transcripts selected for knockdown by siRNA transfection, at least 17 affected the viability of senescent cells more than the viability of nonsenescent cells (Supporting information Table S1). Of these, six triggered senescent cell death, but had little effect on proliferating, nonsenescent cells in two human cell types, preadipocytes (Fig.1D,F) and endothelial cells (Fig.1E,G). Interestingly, the senolytic transcripts included components of ephrin survival-regulating dependence receptor mechanisms (Goldschneider & Mehlen, ), ephrin ligand (EFN) B1, and EFNB3, as well as the cyclin-dependent kinase inhibitor 1A (p21), plasminogen-activated inhibitor-2 (PAI-2), the phosphatidylinositol-4,5-bisphosphate 3-kinase delta catalytic subunit (PI3KCD), a PI3K family member involved in regulating multiple cellular functions, including survival (Datta et al.,; Osaki et al., ), and BCL-xL, which regulates mitochondrial-dependent apoptosis and is the target of certain anticancer drugs (Minn et al.,; Leech et al., ).

6Interfering with expression of EFNB1 or 3, PI3KCD, p21, BCL-xL, or PAI-2 significantly reduced the viability (ATPLite intensity; Fig.1D) and survival (crystal violet; Fig.1F and Fig. S6) of senescent but not proliferating human abdominal subcutaneous preadipocytes. Reducing EFNB2 or 4 or PI3K isoforms other than PI3KCD had less or no effect (Fig. S2C; Table S1). siRNA transfection efficiencies and extent of mRNA knockdown were similar in senescent and proliferating preadipocytes (Fig. S1). Results were confirmed using second, distinct siRNAs or by Western immunoanalysis (Fig. S2A, B, & F). While proliferating human umbilical vein cells (HUVECs) tended to be generally susceptible to siRNAs under the conditions used, senescent HUVECs were more susceptible to EFNB1 and BCL-xL siRNAs than nonsenescent cells (Fig.1E,G). EFNB1 or 3 and PI3KCD siRNAs also interfered with the viability of preadipocytes made senescent by serial subculturing compared to nonsenescent cells (Fig. S2D) and did not interfere with the viability of quiescent, differentiated preadipocytes (Fig. S2E). Results were confirmed using crystal violet to measure cell survival (Fig.1G; Fig. S6).

7Based on potential associations among the genes targeted by senolytic siRNAs, we tested whether the gene products could be components of a common pro-survival signaling network (Fig.1H). Network analysis of these proteins using the STRING database suggested interaction of the anti-apoptotic proteins that we identified in siRNA assays.

8We next tested whether drugs that target gene products that protect senescent cells from apoptosis are senolytic in vitro. Of 46 agents tested, dasatinib (D) and quercetin (Q) showed particular promise in clearing senescent cells. D is a inhibitor of multiple tyrosine kinases, used for treating cancers (Montero et al., ), and is known to interfere with EFNB-dependent supprepression of apoptosis (Chang et al.,; Xi et al., ). D preferentially reduced viability and caused cell death of senescent human preadipocytes, but was much less effective on senescent HUVECs (Fig.2A). Note that by day 3, proliferating preadipocytes increased by 2-5-fold in number vs. day 0 in the presence of D. The viability of nondividing, senescent preadipocytes from the same subjects decreased by 30–40% in the presence of 50 nm or greater D, indicating selective reduction in the viability of senescent cells. Q, a natural flavonol, inhibits PI3K, other kinases, and serpines (Olave et al.,; Bruning, ). In contrast to D, at low concentrations, Q reduced the viability and caused cell death of senescent HUVECs to a greater extent than proliferating cells, but was less effective on preadipocytes (Fig.2B). Note that at 10 μm Q, nonsenescent HUVECs achieved a 2-3-fold increase in cell number between days 0 and 3, while parallel cultures of senescent cells were reduced by 50%, indicating selective killing of senescent cells. The combination of D+Q afforded selective killing of both senescent preadipocytes and endothelial cells (Fig.2C-F). By day 3, the viability of nondividing senescent preadipocytes exposed to D+Q was reduced by ∼70% compared to day 0, while nonsenescent, proliferating cells had increased by 2 - 4-fold. By day 3, the viability of senescent HUVECs treated with 10 μm Q and 100 nm D was reduced by ∼50% compared to day 0. Parallel cultures of nonsenescent, proliferating HUVECs increased in number by 1.5-fold over the same period of time. This suggests that the combination of D+Q selectively targets a broader range of senescent cell types than either agent alone. In both senescent and nonsenescent cultured preadipocytes, D and Q reduced expression of the anti-apoptotic regulator PAI-2 (Fig.2G,H).

9In anticipation of testing D+Q in a preclinical model, the drugs were tested for the efficacy in reducing the viability of senescent murine cells. The combination of D+Q led to a significant reduction in the number of senescent, C12FDG-positive, primary mouse embryonic fibroblasts (MEFs) compared to either drug alone (Fig.3A). Likewise, Q alone or D+Q caused a significant reduction in the number of senescent bone marrow-derived murine mesenchymal stem cells (Fig.3B). These data and those in Fig.2 demonstrate that both D and Q are able to selectively kill senescent cells in two species, albeit with distinct cell-type specificity. We tested whether D+Q administered by oral gavage was senolytic in vivo. We initially tested D+Q in old mice (> 24 months-old), as senescent cell burden increases in fat tissue with aging and both preadipocytes and endothelial cells contribute to senescent cell burden with aging in fat (Tchkonia et al., ). A single dose of D+Q (D: 5 mg kg−1 body weight and Q: 50 mg kg−1 by oral gavage here and in the following studies), a drug ratio that was most effective in senescent MEFs (data not shown), reduced SA-βgal+ cells (Fig.3C) and p16 mRNA (Fig.3D) in fat from old mice within 5 days. D+Q also reduced p16-positive cells in liver from old mice (Fig.3E-F). As with AP20187, the drug that activates selective killing of cells expressing p16 in transgenic INK-ATTAC mice (Baker et al., ), not all senescent cells were removed by D+Q., yet functional improvement was seen (see below) Following irradiation of one leg of wild-type mice, a single treatment with D+Q reduced p16 expression in muscle (Fig.3G) and SA-βGal+ cells in fat (Fig.3H,I) at the site of localized ionizing radiation exposure in these mice.

10Cellular senescence is associated with cardiovascular dysfunction in humans (Tchkonia et al.,; Kirkland & Tchkonia, ), a major cause of morbidity and mortality in the elderly. While only mild cardiac dysfunction has been reported in old mice (Dai et al.,; Roos et al., ), substantial impairment in vascular reactivity is observed in aged mice (Roos et al., ). We tested whether treating 24-month-old mice with D+Q would improve cardiac ejection fraction (the fraction of heart volume pumped during each heart contraction) and vascular responses to acetylcholine, nitroprusside, or U46619 [endothelium-dependent relaxation, smooth muscle vascular reactivity to nitric oxide, and smooth muscle contractile function, respectively (Roos et al., )]. To allow time for senescent cells to apoptose and exclude potential ‘off-target’ effects of the drugs on nonsenescent cell types, which require continued presence of the drugs, for example, through direct vasoactive/antioxidant effects or through changing NAD+ (Chen & Pace-Asciak,; Ajay et al., ), we gave a single dose of the drugs and waited 5 days before assaying cardiac function. D and Q are cleared within 48 h of the last dose (Luo et al.,; Piantelli et al., ).

11Despite the fact that mice are relatively resistant to the development of age-related systolic dysfunction, treatment of 24-month-old mice with a single dose of D+Q significantly improved left ventricular ejection fraction (Fig.4A) and fractional shortening (Table S3), effects that were mediated by reductions in end-systolic cardiac dimensions (Fig.4C) but not cardiac preload (Fig.4B) or alterations in cardiac mass (Table S3). Consistent with previous reports from our group and others (Roos et al.,; Gioscia-Ryan et al., ), carotid arteries from aged mice displayed markedly impaired vascular relaxation in response to endothelium-dependent and -independent vasodilators compared to young mice. While D+Q elicited somewhat variable but statistically significant improvements in vascular endothelial function (Fig.4D), a complex amalgam of nitric oxide, endothelium-derived hyperpolarizing factors, and other vasoactive substances (Feletou & Vanhoutte, ), D+Q, yielded physiologically important and consistent improvements in vascular smooth muscle sensitivity to nitroprusside (Fig.4E). Interestingly, senescent cell clearance did not alter smooth muscle contractile function (Fig.4F). Collectively, these data suggest that senescent cells likely exert deleterious effects on cardiovascular function with chronological aging and that acute clearance of senescent cells may be a novel therapeutic approach to improve cardiovascular function and reduce morbidity and mortality from cardiovascular disease in the elderly.

12To test further the hypothesis that D+Q functions through elimination of senescent cells, we tested the effect of a single treatment in a mouse leg irradiation model. One leg of 4-month-old male mice was irradiated at 10 Gy with the rest of the body shielded. Controls were sham-irradiated. By 12 weeks, hair on the irradiated leg turned gray (Fig.5A) and the animals exhibited reduced treadmill exercise capacity (Fig.5B). Five days after a single dose of D+Q, exercise time, distance, and total work performed to exhaustion on the treadmill was greater in the mice treated with D+Q compared to vehicle (Fig.5C). Senescent markers were reduced in muscle and inguinal fat 5 days after treatment (Fig.3G-I). At 7 months after the single treatment, exercise capacity was significantly better in the mice that had been irradiated and received the single dose of D+Q than in vehicle-treated controls (Fig.5D). D+Q-treated animals had endurance essentially identical to that of sham-irradiated controls. The single dose of D+Q had no effect on endurance 7 months later in sham-irradiated controls vs. vehicle. Thus, a single dose of D+Q 7 months previously led to sustained improvement in function, consistent with the ability of D+Q to clear damaged senescent cells acutely, resulting in enhanced later-life physical endurance.

13To demonstrate that treatment with D+Q can extend healthspan, we used the Ercc1−/Δ mouse model of accelerated aging. These mice, which model human XFE progeria, have features resembling accelerated aging with a lifespan of 6 months (Dolle et al., ) and spontaneously develop progressive age-related chronic degenerative diseases (Gregg et al., ). MEFs deficient in ERCC1 or bone marrow-derived MSCs from Ercc1−/Δ mice have increased senescence and are more susceptible to elimination by D+Q (Fig.3A,B). Ercc1−/Δ mice were treated with 5 mg kg−1 D plus 50 mg kg−1 Q weekly by oral gavage or vehicle only (10% PEG400 in water) beginning at 4–6 weeks of age. Symptoms associated with aging were measured biweekly by an investigator blinded as to the treatment groups. Animals were euthanized after 10–12 weeks of treatment and molecular and histopathological endpoints measured. D+Q resulted in reduced expression of senescence markers in several tissues (Fig. S9and S12). This correlated with a significant reduction in a composite score of age-related symptoms (Fig.6A,B), including kyphosis, dystonia, tremors, loss of grip strength, coat condition, ataxia, urinary incontinence, impaired gait, hind limb paralysis, and poor body condition (Fig. S10). This reduction in symptoms indicates an extension of healthspan due to both the delay in onset of symptoms and attenuation of their severity (Fig.6B). In particular, the mice showed reduced dystonia and delayed onset of ataxia and gait disorders (Fig.6C and Fig. S10). In addition, quantitative computed tomography (pQCT) of lumbar spine demonstrated improved bone parameters in 16-week-old Ercc1−/Δ mice treated with D+Q compared to animals treated with vehicle only (Fig.6D). Similarly, the level of proteoglycans in the nucleus pulposus of the intervertebral disk, a marker of age-related disk degeneration, was significantly increased in mice treated with D+Q, suggesting that treatment with D+Q can slow age-related dysfunction even of a relatively avascular tissue (Fig.6E). Finally, sections of liver, kidney, and the femoral bone space were stained with H&E and scored for age-related pathology by two pathologists blinded to the treatment groups. Composite pathology scores for sibling groups revealed reduced pathology in most animals treated with D+Q compared to siblings treated with vehicle only (Fig.6F). Remarkably, the sibling groups identified as having the most dramatic differences in pathology were identical to those identified as having the greatest difference in aging score (Fig.6B and Fig. S11), demonstrating a close correlation between pre- and postmortem endpoints. Taken together, these data demonstrate that periodic treatment with senolytics is sufficient to reduce the burden of senescence markers, reduce frailty, and extend healthspan significantly.

14We previously demonstrated that the healthspan of transgenic mice can be enhanced by selectively killing senescent cells using a transgenic suicide gene (Baker et al., ). The identification of approaches to remove damaged, senescent cells would have a tremendous impact on quality of life and burden of age-related chronic diseases. To identify agents able to kill senescent cells, we hypothesized that senescent cells, like cancer cells, are dependent on anti-apoptotic pathways to ensure their survival following stress and damage. Based on this hypothesis, here we demonstrate that senescent cells indeed are susceptible to selective clearance by targeting pro-survival mechanisms using siRNAs and drugs, even at doses insufficient to kill normal proliferating or differentiated, quiescent cells. This observation opens up new approaches to develop clinically relevant small molecules or biologics that selectively eliminate senescent cells from nongenetically modified individuals, acting as senolytic agents. The prototype senolytic agents identified here, dasatinib and quercetin, have the ability to alleviate multiple aging phenotypes, as would be predicted if they truly act by eliminating senescent cells (Kirkland,; Kirkland & Tchkonia, ).

15Interfering with expression of the ephrin dependence receptor ligands, EFNB1 or EFNB3, induces selective loss of senescent cells. Ephrin receptors are the largest family of receptor tyrosine kinases (Boyd et al., ). Together with ephrin ligands, these receptors coordinate tissue and organ patterning, cell positioning, and cell survival during development and tissue turnover in a cell type-specific manner (Xi et al., ). Ephrin B ligands, which span cell membranes, can act as both ligands and receptors, making contact with ephrin receptors on adjacent cells. EFNB ligands can participate in dependence networks that constrain both the cells they are located on and adjacent cells from undergoing apoptosis (Furne et al., ). Ephrin signaling has been linked to cellular senescence: EFNB3 overexpression can induce p21, PAI-1, the SASP, and SA-βGal activity during wound healing (Hafner et al., ). Interfering with expression of EFNB3 in cancer cells can induce apoptosis and disrupt pro-survival networks (Stahl et al., ). Similar to cancer cells, here we demonstrated that silencing EFNB ligands induces apoptosis selectively in senescent as compared to nonsenescent cells.

16Silencing EFNB3 expression down-regulates AKT in cancer cells (Stahl et al., ). AKT is involved in regulating FOXO1 and mTOR, among other key pro-survival and metabolic homeostasis mechanisms (Chandarlapaty, ). PI3K is upstream of AKT, and the PI3KCD (catalytic subunit δ) is specifically implicated in the resistance of cancer cells to apoptosis. PI3KCD inhibition leads to selective apoptosis of cancer cells (Cui et al.,; Xing & Hogge, ). Consistent with these observations, we demonstrate that siRNA knockdown of the PI3KCD isoform, but not other PI3K isoforms, is senolytic in preadipocytes (Table S1).

17p21 and PAI-1, both regulated by p53, are implicated in protection of cancer and other cell types from apoptosis (Gartel & Radhakrishnan,; Kortlever et al.,; Schneider et al.,; Vousden & Prives, ). We found that p21 siRNA is senolytic (Fig.1D+F), and PAI-1 siRNA and the PAI-1 inhibitor, tiplaxtinin, also may have some senolytic activity (Fig. S3). We found that siRNA against another serine protease inhibitor (serpine), PAI-2, is senolytic (Fig.1D+F). Like PAI-1, PAI-2 can protect against apoptosis (Tonnetti et al.,; Delhase et al., ). EFNB1 and 3 expression appears linked to that of BCL-xL, PI3KCD, p21, PAI-1, and, indirectly, PAI-2, in published reports and by bioinformatics analysis (Fig.1H) (Hafner et al., ), suggesting these transcripts may be networked, a hypothesis that merits further study. Consistent with this hypothesis are our findings that D alone reduced the level of p21 protein, and furthermore, D plus Q reduced p21, PAI-2, and BCL-xL (Fig.2G,H).

18D and Q are both approved for use in humans and appear to be relatively safe. Interestingly, imatinib, which is very closely related to D, was not senolytic, at least in preadipocytes (Fig. S4). D and Q are promiscuous, like many drugs that affect signaling pathway kinases either directly or indirectly. Despite this, they appear to have more senolytic activity against some types of senescent cells than others, and overall they appear to work better in combination than individually. Thus, one strategy to follow in developing future senolytic agents will be to use promiscuous agents or combinations of drugs to target anti-apoptotic networks. Alternatively, additional senolytic agents could be developed by de-convoluting the mechanisms through which D, Q, or other senolytics exert their effects. Importantly, as we found that senescent cells originating from different types of cells vary in susceptibility to RNA interference and pharmacological interventions, it may be feasible to design drug strategies focused on specific indications by clearing senescent cells arising from particular cell types or in specific organs.

19Whether candidate senolytics actually alleviate phenotypes though removing senescent cells or through possible off-target effects on nonsenescent cells is an important and difficult issue to resolve. We first considered comparing effects of senolytics to those of removing senescent cells from INK-ATTAC mice. However, we felt that while this may indicate an association between phenotypic effects of removing senescent cells by candidate senolytic drugs and those of removal by activating a ‘suicide gene’ in senescent cells, this approach would not establish cause and effect. Even if candidate senolytic agents had effects resembling those due to genetic clearance of senescent cells, and even if effects of the drugs were not additive to effects of genetic clearance, off-target effects could still not be ruled out. For example, clearing senescent cells genetically could influence a critical effector protein also directly targeted by the drug, especially if studies involve continuous administration of drugs.

20We also considered ruling out off-target effects by expressing constitutively active targets of the candidate senolytic drugs in senescent cells of genetically modified mice and determining whether effects of the drugs are blocked in these animals. However, the targets of the senolytic agents we found have important functions in cell regulation, and constitutively expressing them would be anticipated to have many effects that could confound the experiment.

21Instead, to start to rule out off-target effects, we examined whether removing senescent cells has sustained effects for many weeks after the drug is no longer present. Apart from agents that permanently alter cellular or tissue composition, such as antimicrobials, anticancer agents, extracellular matrix modifiers, or teratogens, there are few drugs known to exert a sustained effect long after the drugs are no longer present. Indeed, our results demonstrated that a single treatment of D+Q had phenotypic effects persisting far after the drug is no longer present. For example, the treadmill endurance in mice in which one leg had been irradiated 3 months before a single dose of senolytics remained improved to the level of that in sham-irradiated controls for 7 months after treatment with vehicle or D+Q. In addition, the senolytic treatment did not affect endurance of the sham-irradiated controls. This long-lasting effect is more consistent with a change in cellular or tissue composition; in this case, a decrease in senescent cell burden, than an off-target effect on a metabolite, pathway, or physiological parameter that requires continued dosing with a drug.

22An important observation is that senolytics appear to alleviate multiple types of dysfunction. The senolytic agents used here enhanced cardiac and vascular function in aging mice, reduced dysfunction caused by localized irradiation, and alleviated skeletal and neurological phenotypes in progeroid mice. Remarkably, in some cases, these drugs did so with only a single course of treatment. In previous work, we and our collaborators found that genetic clearance of senescent cells slowed development of lordokyphosis, cataracts, and lipodystrophy in progeroid mice (Baker et al., ). Thus, the accumulation of senescent cells in association with a number of diseases, disabilities, and chronological aging likely contribute to the causation and pathophysiology of these problems or their symptoms. Together with chronic, ‘sterile’ inflammation, macromolecular dysfunction, and stem and progenitor cell dysfunction, cellular senescence may contribute to both aging phenotypes and increased susceptibility to a range of chronic diseases.

23An advantage of alleviation of symptoms by a single or few doses of senolytics is that they might be given during periods of generally good health, for example, before elective surgery or other circumstances where senescent cell generation could be beneficial. This may help reduce side effects yet still allow senescent cells to be generated when needed, for example, during wound healing (Demaria et al., ). This possibility merits further study in animal models. Additionally, as senescent cells do not divide, drug resistance would be expected to be less likely than is the case with antibiotics or cancer treatment, in which cells proliferate and so can acquire resistance (Tchkonia et al.,; Kirkland & Tchkonia, ).

24We view this work as a first step toward developing senolytic treatments that can be administered safely in the clinic. Several issues remain to be addressed, including some that must be examined well before the agents described here or any other senolytic agents are considered for use in humans. For example, we found differences in responses to RNA interference and senolytic agents among cell types. Effects of age, type of disability or disease, whether senescent cells are continually generated (e.g., in diabetes or high-fat diet vs. effects of a single dose of radiation), extent of DNA damage responses that accompany senescence, sex, drug metabolism, immune function, and other interindividual differences on responses to senolytic agents need to be studied.

25Detailed testing is needed of many other potential targets and senolytic agents and their combinations. Other dependence receptor networks, which promote apoptosis unless they are constrained from doing so by the presence of ligands, might be particularly informative to study, especially to develop cell type-, tissue-, and disease-specific senolytic agents. These receptors include the insulin, IGF-1, androgen, and nerve growth factor receptors, among others (Delloye-Bourgeois et al.,; Goldschneider & Mehlen, ). It is possible that more existing drugs that act against the targets identified by our RNA interference experiments may be senolytic. In addition to ephrins, other dependence receptor ligands, PI3K, AKT, and serpines, we anticipate that drugs that target p21, probably p53 and MDM2 (because they regulate p21 and serpines), BCL-xL, and related genes will also have senolytic effects. This is especially so as existing drugs that act through these targets cause apoptosis in cancer cells and are in use or in trials for treating cancers, including dasatinib, quercetin, and tiplaxtinin (Gomes-Giacoia et al.,; Truffaux et al.,; Lee et al., ).

26Effects of senolytic drugs on healthspan remain to be tested in chronologically aged mice, as do effects on lifespan. Senolytic regimens need to be tested in nonhuman primates. Effects of senolytics should be examined in animal models of other conditions or diseases to which cellular senescence may contribute to pathogenesis, including diabetes, neurodegenerative disorders, osteoarthritis, chronic pulmonary disease, renal diseases, and others (Tchkonia et al.,; Kirkland & Tchkonia, ).

27Like all drugs, D and Q have side effects, including hematologic dysfunction, fluid retention, skin rash, and QT prolongation (Breccia et al., ). An advantage of using a single dose or periodic short treatments is that many of these side effects would likely be less common than during continuous administration for long periods, but this needs to be empirically determined. Side effects of D differ from Q, implying that (i) their side effects are not solely due to senolytic activity and (ii) side effects of any new senolytics may also differ and be better than D or Q. There are a number of theoretical side effects of eliminating senescent cells, including impaired wound healing or fibrosis during liver regeneration (Krizhanovsky et al.,; Demaria et al., ). Another potential issue is cell lysis syndrome if there is sudden killing of large numbers of senescent cells. Under most conditions, this would seem to be unlikely, as only a small percentage of cells are senescent (Herbig et al., ). Nevertheless, this possibility needs to be tested.

28Senescent cells have been identified at sites of pathology in multiple diseases and disabilities or may have systemic effects that predispose to others (Tchkonia et al.,; Kirkland & Tchkonia, ). Our findings here give support for the speculation that these agents may one day be used for treating cardiovascular disease, frailty, loss of resilience, including delayed recovery or dysfunction after chemotherapy or radiation, neurodegenerative disorders, osteoporosis, osteoarthritis, other bone and joint disorders, and adverse phenotypes related to chronologic aging. Theoretically, other conditions such as diabetes and metabolic disorders, visual impairment, chronic lung disease, liver disease, renal and genitourinary dysfunction, skin disorders, and cancers could be alleviated with senolytics. (Kirkland,; Kirkland & Tchkonia,; Tabibian et al., ). If senolytic agents can indeed be brought into clinical application, they would be transformative. With intermittent short treatments, it may become feasible to delay, prevent, alleviate, or even reverse multiple chronic diseases and disabilities as a group, instead of one at a time.

29Detailed descriptions of our preadipocyte, HUVEC, MEF, and MSC culture methods are in Data S1 and publications (Tchkonia et al.,; Wang et al., ). The protocol was approved by the Mayo Clinic Foundation Institutional Review Board for Human Research.

30Preadipocytes or HUVECs were irradiated with 10 Gy of ionizing radiation to induce senescence or were sham-irradiated. Preadipocytes were senescent by 20 days after radiation and HUVECs after 14 days, exhibiting increased SA-βGal activity and SASP expression by ELISA (IL-6, MCP-1). Where indicated, senescence was induced by serially subculturing cells.

31Microarray analyses were performed using the r environment for statistical computing (http://www.R-project.org). Array data are deposited in the GEO database, accession number GSE66236. Gene Set Enrichment Analysis (version 2.0.13) (Subramanian et al., ) was used to identify biological terms, pathways, and processes that were coordinately up- or down-regulated with senescence. The Entrez Gene identifiers of genes interrogated by the array were ranked according to the t statistic. The ranked list was then used to perform a pre-ranked GSEA analysis using the Entrez Gene versions of gene sets obtained from the Molecular Signatures Database (Subramanian et al., ). Leading edges of pro- and anti-apoptotic genes from the GSEA were performed using a list of genes ranked by the Student t statistic.

32Cellular SA-βGal activity was quantitated using 8–10 images taken of random fields from each sample by fluorescence microscopy.

33Primers are described in Table S2. Cells were transduced with siRNA using RNAiMAX and harvested 48 h after transduction. RT–PCR methods are in our publications (Cartwright et al., ). TATA-binding protein (TBP) mRNA was used as internal control.

34Data on protein–protein interactions (PPIs) were downloaded from version 9.1 of the STRING database (PubMed ID 23203871) and limited to those with a declared ‘mode’ of interaction, which consisted of 80% physical interactions, such as activation (18%), reaction (13%), catalysis (10%), or binding (39%), and 20% functional interactions, such as post-translational modification (4%) and co-expression (16%). The data were then imported into Cytoscape (PMID 21149340) for visualization. Proteins with only one interaction were excluded to lessen visual clutter.

35Mice were male C57Bl/6 from Jackson Labs unless indicated otherwise. Aging mice were from the National Institute on Aging. Ercc1−/Δ mice were bred at Scripps (Ahmad et al., ). All studies were approved by the Institutional Animal Care and Use Committees at Mayo Clinic or Scripps.

36Four-month-old male C57Bl/6 mice were anesthetized and one leg irradiated with 10 Gy. The rest of the body was shielded. Sham-irradiated mice were anesthetized and placed in the chamber, but the cesium source was not introduced. By 12 weeks, p16 expression is substantially increased under these conditions (Le et al., ).

37Rings from carotid arteries were used for vasomotor function studies (Roos et al., ). Excess adventitial tissue and perivascular fat were removed, and sections of 3 mm in length were mounted on stainless steel hooks. The vessels were maintained in an organ bath chamber. Responses to acetylcholine (endothelium-dependent relaxation), nitroprusside (endothelium-independent relaxation), and U46619 (constriction) were measured.

38High-resolution ultrasound imaging was used to evaluate cardiac function. Short- and long-axis views of the left ventricle were obtained to evaluate ventricular dimensions, systolic function, and mass (Roos et al., ).

39As a measure of physical function, exercise capacity was determined on a motorized treadmill (LeBrasseur et al., ). Running time was recorded, and running distance (a function of time and speed on the treadmill) and work were calculated.