Could senescence phenotypes strike the balance to promote tumor dormancy?

It is well established that the immune system is critical for suppressing primary tumor formation and progression, and thus tumors must evade immune clearance at all stages of progression: establishment, metastasis, response to treatment, and relapse. It is self-evident that for a cell to survive in dormancy, it must avoid immune clearance, but moderate immune activity can also play a role in establishing tumor mass dormancy [ 35 , 183 - 185 ]. Senescent cells have been shown to interact with immune cells in a variety of contexts and by multiple mechanisms, including the SASP. These interactions can promote or inhibit the immune system. Thus, senescence and SASP modulation of the immune system could provide an attractive model for explaining a state of dormancy whereby a small tumor mass is kept in check, but also not eradicated. While the SASP constituents vary depending on context, stressors, and cell types, secretion of these factors allows senescent cells to interact with the immune system in multiple ways. Studies examining these interactions found that the SASP can regulate immune cell functions like proliferation [ 108 ], wound healing [ 106 ], and migration [ 186 ]. Secretion of certain factors may even promote clearance of senescent cells by various types of immune cells. For example, factors including IL-6, IL-8, IL1B, TGFβ, GM-CSF, MCP-1, MIP-1α, and GROα that are produced by senescent cells are recognized by receptors on NK cells, macrophages, and T-cells (reviewed in [ 187 ]). Some groups have demonstrated that upon inhibition of the SASP, phenotypes like immune surveillance become impaired [ 188 ]. Thus, how senescent cells are removed by different immune cell types upon SASP-mediated stimulation is an area of active research and is discussed below. Many studies have implicated NK cells in senescent cell immune surveillance, although most were performed using non-tumor models. Senescent cell modulation of NK cells could be mediated in several ways, including production of cytokines that potentiate NK cell function. Indeed, Krizhanovsky et al found that senescent cells that accumulated in fibrotic murine livers upregulated cytokines that promoted their NK cell cytotoxic clearance [ 189 ]. In examining how NK cells target senescent cells that produce these cytokines, Sagiv et al determined that granule exocytosis, but not death-receptor-mediated apoptosis, was required for senescent cell clearance by NK cells. This cytokine production by senescent cells was controlled by NF-kB activation [ 50 ], as well as BRD4-mediated chromatin remodeling [ 51 ]. Senescent cells would need to avoid NK cell mediated clearance if they were to persist in dormancy. In addition to NK-cell inducing cytokine production, senescent cells can also express ligands that promote surveillance or suppression by NK cells [ 190 ]. Many studies identifying these ligands implicate NKG2D and DNAM1 NK-cell activating receptors as the major receptors required for senescent cell clearance. Soriani et al found that myeloma cells made senescent upon treatment with doxorubicin or melphalan upregulate NKG2D ligands, which increase NK-cell degranulation [ 191 ]. Likewise, Sagiv et al determined that MICA and ULB2, which bind to NKG2D, are consistently upregulated by senescent cells and are necessary for NK cell-mediated cytotoxicity [ 192 ]. During pregnancy, senescent decidual cells are cleared by NK cells through NKG2D and DNAM1 receptors [ 193 ]. Most recently, one group found that activation of invariant NKT cells with a glycolipid antigen successfully eliminated senescent cells in two different senescence-associated disease models [ 194 ]. Though many studies have identified NK cells as the primary sentinel cell for senescent cell surveillance, this appears to be specific to certain tissues, such as liver. Expression of NKGD ligands by senescent cells could also potentially inhibit the NK cell response. Indeed, Stokes et al found that in a mouse model of lung cancer, NK cells limited the complete clearance of tumor cells made senescent by the experimental genetic activation of p53. Upon depletion of NK cells, infiltrating monocyte, neutrophil and macrophage populations expanded, leading to tumor clearance [ 195 ]. One effective tumor cell strategy to suppress NK cell function is the shedding of NKGD2 ligands after proteolytic cleavage by ADAM10 [ 196 ]. Both NKGD2 ligands and ADAM10 are upregulated after genotoxic stress caused by chemotherapy or radiation [ 197 ]. One recent study by Pereira et al found that HLA-E expression was increased on senescent cells, which interacts with the inhibitory receptor NKG2A expressed by NK cells. By blocking the interaction between HLA-E and NKG2A in vitro, senescent cells were able to be killed by NK cells [ 198 ]. Additionally, senescent cells can evade NK cell recognition by MMP-dependent shedding of NKG2D ligands, a process evident in residual tumors from large prostate and breast cancer patient cohorts treated with senescence-inducing chemotherapy [ 109 ]. The balance of negative and positive NKGD effects on the immune system, perhaps combined with the activity of SASP cytokines, could strike the necessary balance to maintain a senescent cancer cell in the dormant state for long periods of time. Other studies have demonstrated that macrophages are capable of interacting with and clearing senescent cells, but can also create an immune suppressive environment. This is primarily achieved through the secretion of macrophage-stimulatory factors like MCP-1, MIP-1α, and GM-CSF [ 45 ]. In vitro, one study showed that tissue-resident macrophages are attracted to and stimulated by conditioned medium from senescent but not non-senescent human fat cells [ 199 ]. One of the first in vivo studies of macrophage-mediated clearance, however, was performed in salamanders, in which macrophages were observed to clear senescent cells during limb regeneration [ 200 ]. In mice uterine stroma, macrophages clear senescent cells in the embryo implantation site [ 201 ]. In a model of tumor cell senescence, Kang et al found that pre-malignant, senescent hepatocytes secrete cytokines that trigger macrophage-mediated clearance of these cells [ 202 ]. Likewise, Lujambio et al showed in a model of liver fibrosis that senescent stellate cells release factors that polarize macrophages towards a tumor-inhibiting M1-state, which become capable of attacking senescent cells [ 203 ]. In another model of senescence, however, senescent thyroid cells secreted factors that skewed macrophage polarization to an M2 state [ 204 ]. These data suggest senescent cell interactions with macrophages could also create a balance of immune suppressive and promoting forces that create an environment favorable for dormancy. How senescent cells interact with T cells is an emerging area of great interest. Some early studies have shown that CD4+ T-cells can kill senescent cancer cells. Usually absent in normal cells, senescent cells can express MHC class II molecules, allowing for recognition by CD4+ T cells, and subsequent clearance by macrophages [ 202 ]. A recent study similarly identified an important role for CD4+ T cells in senescent cell surveillance, in which senescent liver cancer cells induced endothelial cell NF-kB signaling through the SASP, resulting in CD4+ T cells infiltration [ 205 ]. How host CD8+ T cells interact with senescent cells has been clarified by very recent studies. Amor et al explored the therapeutic concept that chimeric antigen receptor (CAR) T cells designed to target senescent cells could have senolytic activity. This study identified a cell-surface receptor broadly induced by senescent normal cells and showed that CAR T cells designed to target this receptor could efficiently kill senescent cells in vivo [ 206 ]. Future efforts are directed toward identifying cell-surface antigens specific to different subsets of senescent cells that can be targeted by antigen-specific CAR T cells. New studies have shown senescent liver tumor cells actually are primed to present antigen to T-cells, thus facilitating their removal [ 207 ]. In models of both normal and tumor cell senescence, Marin et al showed that senescent cells are highly immunogenic and engage the innate immune system by presenting antigen to T-cells [ 208 ]. Senescent cells also clearly persist in many cases, as would be required for long-term dormancy, and thus must evade immune detection. In fact, both chemotherapy-mediated senescent tumor cells and cells senescent in normal aging express “checkpoint” genes that inhibit T-cell effector functions [ 209 - 212 ]. In senescent normal cells, the checkpoint PD-L1 is predominantly expressed [ 211 , 212 ], and these cells can be eradicated with anti-PD-L1 therapy [ 211 ]. In senescent breast cancer cells, a host of redundant checkpoints are expressed, making their eradication more challenging [ 209 ]. Due to extremely complex and sensitive systems of regulation, the immune system is a prime candidate to maintain cells in long term dormancy. A balance between checkpoint expression and antigen presentation, for instance, could potentially direct the immune system to keep dormant, senescent cells in check but not eliminated. Experiments with CAR-T cells and immune checkpoint inhibitors suggest routes for therapeutic interventions. Many properties of senescent cells are consistent with a role in tumor dormancy. Chromatin changes occurring in senescence can mediate a stable, long-term arrest and expression of SASP-related cytokines ( Figure 1A ). Senescent cells also express genes that can modulate the immune system, either negatively or positively ( Figure 1B ). Different cytokines of the SASP can also stimulate or inhibit the immune system as well as angiogenic vascularization of the tumor ( Figure 1C ). A balanced interaction with the immune system and vasculature can keep tumor cells alive but dormant. Over time, SASP factors expressed by senescent cells could, either directly or thru neovascularization, promote non-cell autonomous escape from dormancy ( Figure 1D ). Alternatively, if the chromatin state that supports cell cycle arrest in a senescent cell is reversed, then proliferation could be restarted leading to a cell autonomous exit from dormancy ( Figure 1E ). While these properties have been rigorously described in many cell types and model systems, one must keep in mind that tumor dormancy is one of the most difficult phenomena to model in cancer. Tumor cells remain in dormancy in human patients for periods of time that exceed the lifespan of mice by many folds. The time lapse between any biopsy, treatment, and resection of the original tumor and the recurrence after a long dormancy creates obvious complications as well. Care must be taken when interpreting “dormancy” experiments. If the cell does not express the phenotype for more than a short period of time, is it modeling dormancy or arrest/quiescence/senescence followed by emergence? If tumor cells indeed enter senescence to survive in dormancy, then their eradication is of paramount clinical importance. Recent breakthroughs in the use of “senolytic” drugs have targeted multiple cancer cell types, including breast [ 213 - 221 ], but would these treatments be as effective at the site of dormancy? Additional investigations on the role of senescent cells in tumor dormancy and therapeutic targeting are warranted.

Angiogenesis is the process by which new capillaries are generated from pre-existing vasculature [ 84 ]. Oxygen and essential nutrients in the blood necessary for sustaining growth and cell cycle progression are delivered to tumor cells through this process [ 36 ]. Failure to recruit or remodel vasculature creates a hypoxic environment that can cause cellular stress, leading to withdrawal from the cell cycle or cell death [ 85 , 86 ]. On the other hands, ample vasculature can create conditions where tumors can flourish and grow [ 85 ]. One way cells adapt to hypoxic conditions is through the activation of Hypoxia-Inducible-Factor-1 (HIF-1) transcription factor [ 87 ]. HIF-1 triggers the activation of downstream genes that regulate cell migration, cell proliferation, and angiogenesis [ 87 ]. If the cells in a small tumor mass are devoid of a vascular derived energy supply, survival long term is uncertain [ 85 ]. If a tumor mass is replete with energy and nutrients available from blood vessels, these conditions would favor cell proliferation, survival, and tumor growth, leading to rapid recurrence [ 85 ]. When the number of blood vessels is somewhere in between, an equilibrium can be reached between proliferating and dying cells resulting in a state of angiogenic dormancy [ 36 ]. Tumor cells can escape angiogenic dormancy through a process termed the “angiogenic switch” [ 88 - 91 ]. It is well known that blood vessel formation, including the angiogenic switch, is mediated by a plethora of factors including cytokines, growth factors, and immune cells [ 92 ]. When a tumor is adequately vascularized and oxygen is present, HIF-1 is inactive because its accumulation is prevented through polyubiquitination by the von Hippel-Lindau (VHL) protein and subsequent proteasomal degradation. If a tumor is poorly vascularized and oxygen levels decrease, there is a shift to the induction of HIF-1 and related downstream factors [ 93 , 94 ]. HIF-1 can induce growth arrest through the activation of p53, increase of p21, and reduction in Bcl-2 [ 95 ]. In this arrested pro-survival state, cells can produce SASP related growth factors and cytokines [ 95 - 98 ]. Angiogenesis is key to tumor cell survival in dormancy but must be regulated such that conditions do not overtly favor rapid relapse or cell death. Through the SASP, senescent cells produce factors that are both pro- and anti-angiogenic, and thus are compelling candidates to mediate angiogenic dormancy. The angiogenic switch abides by a strict regulatory network of pro- and anti-angiogenic factors [ 36 , 99 , 89 ]. Major pro-angiogenic factors include vascular endothelial growth factor (VEGF) and basic fibroblast growth factor (b-FGF), while anti-angiogenic factors include endostatin and thrombospondin-1 (TSP-1) [ 100 ]. If this delicate balance between pro- and anti-angiogenic factors is disrupted due to physiological stimuli, hypoxia, or inflammation, the system preferentially increases pro-angiogenic factors, which can then enable escape from angiogenic dormancy [ 92 , 90 ]. Indeed, SASP factors are known to mediate angiogenesis [ 43 , 101 , 102 ] through the production of VEGF, FGF, interleukins, chemokines, matrix metalloproteinases (MMPs), and proteins associated with chromatin remodeling [ 92 , 103 ]. Coppé et al, demonstrated that through production of VEGF, senescent HDFs were capable of inducing angiogenesis [ 104 ]. However, HIF1 did not increase in parallel with the senescence-associated increase in VEGF production. This indicates that production of VEGF by senescent cells cannot be solely attributed to the classic mechanism of hypoxia [ 105 , 104 ]. Moreover, when studied in vivo, there was a difference in vascularization between tumors developed in the presence of pre-senescent fibroblasts and those developed in the presence of senescent fibroblasts. Notably, tumors that developed in the presence of senescent fibroblasts were significantly more vascularized and larger in size [ 104 ]. Examining the function of senescent cells in wound healing, DeMaria et al showed that the production of platelet-derived growth factor AA (PDGF-AA) by senescent cells promoted angiogenesis and wound healing [ 106 ]. The ability of vasculature to form continuous networks is dependent on the reconstruction of pre-existing extracellular matrix components. To accomplish this, MMPs are present within the tumor microenvironment to aid in the breakdown of the extracellular matrix (ECM) and allow for the generation of new blood vessels [ 107 ]. Production of MMPs is a near universal feature of the SASP [ 12 , 108 ], including tumor cells made senescent by chemotherapy [ 109 ] that could potentially be dormant at a distal site. Studies in metastatic breast cancer have demonstrated that inhibition of MMPs, including MMP-2 and MMP-9, prevent metastasis, reduce ECM degradation, and inhibit angiogenesis by exposing or altering ECM receptors [ 107 , 110 , 111 ]. Oxidative stress and DNA damage caused by chemotherapeutics activate MMPs and can serve as targets for preventing metastasis and angiogenesis [ 107 , 112 ]. IL-6 and IL-8 are very well characterized, highly expressed products of the SASP across diverse cell types [ 108 , 113 - 115 ] and also play an important role in angiogenesis. IL-6 has been shown to promote angiogenesis in multiple cancers, including, glioblastoma, neuroblastoma, basal cell carcinoma, mesothelioma, gastric and cervical cancers [ 116 - 121 ]. In a nude mouse model, IL-8 generated by tumor cells mediated angiogenesis in a Ras-dependent manner, but required additional activating factors, such as those produced in the SASP, to tip the scales toward neovascularization [ 122 ]. Further, IL-8 binding to CXCR1/2 on endothelial cells enhanced proliferation, survival, and MMP expression [ 123 - 126 ] and promoted angiogenesis and tumor progression [ 127 , 128 ]. Huang et al demonstrated that elevated tumor IL-8 mediates a secondary angiogenic pathway following VEGF inhibition by sunitinib in treatment-resistant renal cancer patients. This resistance typically occurred over the course of a year [ 129 ], suggesting a potential link to the SASP. Cytokine and cell-specific SASP factor IFN-γ induces chemokines such as CXCL1, CXCL12, CXCL9, and CXCL10, that possess either pro-angiogenic (CXCL1/12) [ 130 - 133 ] or anti-angiogenic (CXCL9/10) properties [ 134 - 137 ]. CXCL9/10/12 are SASP factors expressed in senescent fibroblasts, colorectal cancer [ 135 ] and multiple myeloma [ 138 ] following treatment with chemotherapy. In senescent colorectal cancer cells, immune evasion may be aided by reconstructing pre-existing vasculature. This occurs through the downregulation of anti-angiogenic chemokines (CXCL9/10) and upregulation of pro-angiogenic CXCL12, thus altering normal blood flow [ 139 , 140 ] and the reduction of immune-mediating chemokines [ 141 , 135 ]. Various other small chemokines (CCLs) are induced in senescence as part of the SASP, including CCL-8, CCL-2, and CCL-7 [ 103 , 142 , 143 ]. CCL-2 was shown to mediate angiogenesis in the brain by binding to its receptor, CCR2, on endothelial cells and increasing pro-angiogenic factors [ 144 , 145 ]. Additionally, CCL-2 promoted angiogenesis in liver fibrosis [ 146 ], prostate cancer [ 147 ], and breast cancer [ 148 ] by upregulating VEGF. CCL-7 bound CCR1 promotes angiogenesis in colorectal [ 149 ] and liver cancer [ 150 ]. Another pro-angiogenic factor, CCL-8, helped facilitate angiogenesis in breast cancer [ 151 ], endometriosis [ 152 ], and in carotid plaques [ 153 ]. If the SASP produces factors that consistently and potently promote angiogenesis, this would highly favor a rapid tumor recurrence rather than long-term dormancy. Striking a balance with pro-angiogenic signaling, recent studies have shown the SASP can also negatively regulate angiogenesis. The induction of the senescence phenotype in colon cancer cell lines exposed to radiation led to decreased expression of pro-angiogenic factors and angiogenesis [ 154 ]. While not all angiogenic factors were downregulated as a result of radiation, pro-angiogenic VEGF-A was reduced compared with proliferating cells [ 154 ]. In SW480 cells, senescence markers such as p21 and SAβGal were upregulated after over-expression of caveolin-1 (CAV1). The CAV1 over-expression repressed angiogenesis by reducing PDGF-BB, but did not alter VEGF expression [ 155 ]. Other studies on adipose-derived mesenchymal stem cells demonstrated the neutralizing effect that anti-angiogenic factors, such as IL-4, have on pro-angiogenic factors like VEGF and FGF, thus negatively regulating angiogenesis [ 156 ]. Tumor growth factor-β (TGFβ) is another known SASP factor [ 157 ] that has been shown to regulate angiogenesis, but can either stimulate or inhibit the development of new blood vessels [ 158 - 160 ]. TGFβ’s inhibitory effect stems from its ability to suppress the expression of VEGFA [ 161 ]. SASP factor IL1B was also shown to inhibit angiogenesis [ 162 ]. Andriani et al showed paracrine IL1B was involved in senescence resulting from whole chromosome instability [ 163 ]. IL1B from tumor cells was shown to promote angiogenesis [ 164 - 166 ], but some studies also showed that IL1B could inhibit angiogenesis [ 162 , 167 ]. Volpert et al demonstrated the anti-angiogenic role of IL-4 [ 168 ], largely through the inhibition of VEGF and bFGF [ 169 , 170 ], while Baba et al suggest IL-4 in bone marrow cells serves a pro-angiogenic role [ 171 ]. Other cytokines involved in senescence as well as angiogenesis, include IL-10 and IL-12 [ 126 , 172 , 173 ]. IL-12 has been characterized as an anti-angiogenic factor in retinopathy, lung carcinoma, and sarcoma [ 174 - 176 ]. The role of IL-10 in angiogenesis is context dependent, where IL-10 is pro-angiogenic in eye-related diseases [ 177 , 178 ] and non-small cell lung cancer [ 179 ], but anti-angiogenic in pancreatic cancer [ 180 ], ischemic tissues [ 181 ], and ovarian cancer [ 182 ]. Taken together, these studies show that SASP-related proteins that are integral to cell viability and proliferation are also potent and very well characterized positive and negative regulators of angiogenesis. Under various stressors, the transition from a proliferating cell to a senescent cell may act as a means of evading cell death in hypoxic conditions or preventing immune detection. In the context of tumor mass dormancy, it is plausible that tumor cells in a low oxygen environment may enter a senescent state, or, have a senescent state reinforced, that allows them to produce pro- or anti-angiogenic SASP factors resulting in angiogenic dormancy. It is possible that following some long period of dormancy, SASP factors could trigger the angiogenic switch. With neovasculature reaching the hypoxic, arrested tumor, the influx of essential nutrients and growth factors would promote re-entry into the cell cycle and tumor recurrence.

In cases of clinical dormancy, the timing of arrest exit is critical. Dormancy can last for years, suggesting a very stable, long-term arrest. The nature of senescent phenotypes suggests this state could be a multifunctional contributor to cancer dormancy [ 37 ]. Senescent cells are arrested and have properties such as the SASP [ 12 ], phagocytosis proficiency [ 38 , 39 ], autophagy [ 40 ], and others that could enable tumor cell survival for years, but also provide compelling possibilities for tumor cells’ eventual emergence from dormancy [ 2 , 14 , 41 - 43 ] via both cell autonomous and non-cell autonomous mechanisms. For example, cancer cells have been shown to emerge from senescence (or, semantically, a senescent-like state) to proliferate again, at least in tissue culture [ 30 , 31 , 29 , 32 , 27 , 28 ]. Senescent cells could also potentially cause recurrence by promoting proliferation of neighboring, non-senescent cells. Indeed, this has been demonstrated [ 44 , 45 ], including a convincing recent in vivo study [ 46 , 47 ]. Many of the phenotypes of senescence are initiated and maintained by gene expression changes mediated by transcription factors such as p53, NF-kB, the Stat family, and the Rb family [ 11 , 14 , 48 - 52 ]. Transcriptional activity of the tumor suppressor p53 plays a critical role in the senescent phenotype, halting the cell cycle via induction of the cyclin dependent kinase (Cdk) inhibitor p21, among other targets [ 53 , 14 ]. In the case of cell autonomous emergence from dormancy, attenuation of the p53 signaling pathway can result in restart of proliferation [ 25 , 32 , 54 ]. If the arrest is indeed permanent, a non-cell autonomous role in driving recurrence is still possible. Recently it was shown in a mouse model of liver cancer that ectopic p53 activity induced by deletion of its negative regulator, Mdm2 (a genetic event that is an extremely potent activator of p53 [ 55 ]), caused widespread senescence with SASP and inflammation. These senescent cells accelerated tumorigenesis driven by oncogenic Kras in cells that did not have Mdm2 deletion [ 46 , 47 ]. This study clearly demonstrates, in vivo, in a genetic model, that senescent cells can promote tumorigenesis in neighboring, non-senescent cells, suggesting another potential mechanism for how senescent cells might promote emergence from dormancy. Transcription factors mediate, and are influenced by, post-translational histone modifications that control chromatin structure, accessibility of regulatory regions of DNA by other transcription factors, and creation of binding sites for transcriptional regulators [ 56 , 57 ]. Epigenetic alterations could contribute to cellular reprogramming and dedifferentiation [ 58 ] which eventually lead to reversion of senescence, thus making cells escape senescence and reenter cell cycle. The post-translational modification of histones has been well accepted as a dynamic process that modulates the transcriptional programs in cells [ 57 ], but some histone modifications are considered more durable. The presence or absence of these types of modifications make an attractive model for stable control of gene expression over the long term, such as would be required for a state of clinical dormancy [ 59 , 57 ]. Post-translational histone modifications play an important role in regulating transcriptional programs in senescent cells [ 60 , 61 ], including in chemotherapy treated tumor cells [ 48 ], and could potentially maintain the state of cellular dormancy. Further modification or reversal of these histone marks could explain eventual emergence from senescence. Histone acetylation and phosphorylation dynamically reduce basic histone charges and promote transcriptional activation [ 62 , 63 ]. Histone methylation, however, is believed to be a more static modification in regulating transcriptional programming. For example, H3K9me3, the gene silencing mark in heterochromatin, can be maintained and passed down after multiple generations of cell division [ 59 ]. H3K4me3, an active mark, is maintained in sperm at developmental genes and contributes to the gene expression patterns in pre-implanted embryos after fertilization [ 64 ]. Another active mark, H3K4me1, is present at enhancers in a primordial germ cell model and maintains germline competence [ 65 ]. The above studies support the idea of cellular plasticity mediated by dynamic and static histone modifications. These modifications could cooperate to regulate transcriptional programs that maintain dormancy in tumor cells for years, but still retain the capacity to revert or modify further, thus adjusting transcriptional programs in response to the ever-changing microenvironment. The transcriptional control of gene expression in senescent cells is mediated by both durable and more transient histone marks, thus presenting an attractive model for dormancy and cell autonomous recurrence after long periods of time. Genes whose expression promotes features of dormancy, such as Cdk inhibitors, would likely be enriched for histone modifications that mark actively transcribed genes, while genes that oppose the state of dormancy, such as cyclins, would be enriched for histone modifications that mark repressed genes. The durable and transient histone modifications that regulate gene expression programs in senescent cancer and non-cancer cells and how these modifications might regulate dormancy are discussed herein. Senescence-associated heterochromatin foci (SAHF) are present in senescent cells and regulate expression of genes involved in stable arrest. The SAHF are formed from concentric epigenetic layers with H3K9me3 heterochromatin cores surrounded by H3K27me3 rings. The euchromatin with active histone marks, such as H3K36me3, is excluded from the SAHF [ 66 ]. H3K9me3 and H3K27me3 are well known repressive histone marks in gene regulation and, thus, the expression of genes located within SAHF are downregulated during senescence. For example, SAHF form at E2F responsive promoters (i.e., cell cycle genes) in senescent HDFs, which contributes to the silencing of E2F target genes and maintenance of prolonged cell cycle arrest [ 67 ]. The importance of this modification in cancer is exemplified by the finding that oncogenic stress fails to cause senescence when Suv39h1, the H3K9 methyltransferase responsible for the Rb mediated changes, is absent in mouse lymphoid cells. Failure to create these histone modifications in response to oncogene activation accelerated lymphoma formation in vivo [ 58 ]. Overexpression of histone H3K9 demethylases can also prevent senescence in melanoma cells [ 68 ]. Expression of H3K9 demethylases Lysine-Specific Demethylase-1 (LSD1) and Jumonji C Domain-Containing Moieties (JMJD2C), inhibited OIS in mouse embryo fibroblasts (MEFs) and enabled melanoma development by cooperating with BRAF in mouse and zebrafish models. OIS was restored through specific inhibition of LSD1 or JMJD2C [ 69 ]. In another example of H3K9 modification regulating senescence, Macha et al showed that head and neck squamous cell carcinoma (HNSCC) cell lines can be made senescent by knockdown of MUC4, a membrane-bound mucin overexpressed in several kinds of cancer. When transplanted into immunocompromised mice, these cells form smaller tumors compared to controls. The team further showed that H3K9ac, a histone mark present in regions of active transcription, was reduced at the cyclin E promoter [ 70 ]. In another study, H3K9ac was significantly increased at promoters of NF-κB target genes, leading to senescence in SIRT6-deficient HeLa cells. The hyperacetylation on H3K9 was related to enhanced NF-κB mediated gene modulation and senescence induction, which contributed to aging phenotypes and short life span in mice [ 71 ]. H3K9me2, a repressive mark, also regulates senescence gene expression. H3K9me2 was depleted at the promoter of SASP genes IL6 and IL8 during oncogene-induced senescence (OIS) in HDFs. Proteasomal degradation of the histone methyltransferases G9a and GLP was at least partially responsible for the loss of H3K9me2 [ 72 ]. In cell autonomous emergence from dormancy, these histone marks would likely have to be reversed or overcome. Alternatively, durable presence of these histone marks would promote continued expression of genes such as IL6 that drive non-cell autonomous emergence over time. H3K27me3 is a histone mark present in regions of actively repressed transcription and is redistributed in the genome of senescent cells to produce their dramatically altered gene expression patterns [ 73 ]. In human gastric cancer cells, which lack the histone methyltransferase EZH2, H3K27me3 was mostly absent from the promoters of Cdk inhibitor genes CDKN2A and CDKN2B , allowing for expression of INK4A/ARF, p15, and p16 and thus promoting cell cycle arrest during senescence [ 74 , 75 ]. H3K27me3 was also reduced at the promoter region of CDKN2A and SASP gene TIMP4 through downregulated expression of EZH2 in aging-induced senescent mouse atrial fibroblasts (AFs). The importance of H3K27me3 in senescence was further confirmed by the accelerated senescence in early passage AFs after knockdown of EZH2 , while overexpression of EZH2 slowed senescence in late passage AF cells [ 76 ]. In senescent HDFs, Shah et al showed extensive epigenetic reorganization. Expansive stretches enriched in H3K4me3 and H3K27me3 histone marks were formed, as well as stretches of H3K27me3 loss. These regions were in and around the location of Lamin B associated domains and are likely to regulate expression of SASP genes, among others [ 73 ]. H3K4me3 is a histone mark that is present at transcriptionally active promoters. The transcription-associated histone methyltransferase MLL1 was shown to control expression of proliferation genes and SASP genes by epigenetic regulation in senescent HDFs. H3K4me3, with DNA damage marker γH2AX, were shown to be present at the promoter and transcriptional start site of SASP genes, such as MMP1 , MMP10 , SERPINB2 , CXCL3 , and CCL3 , and this was mediated largely through the activity of MLL1 [ 77 ]. Activator protein 1 (AP-1) was also reported to induce active de novo histone marks (H3K4me1 and H3K27ac) at specific enhancers in RAS-induced senescent HDFs. This alters the landscape of activated enhancers and organizes the regulation of the transcriptional program of genes related to senescence [ 78 ]. Changes in activity of MLL1 or AP-1 in long term dormancy could potentially promote escape from senescence and thus a cell autonomous mechanism of relapse from dormancy. The active histone mark, H3K79me2/3 was shown to be enriched at the locus of the SASP gene IL1A in RAS-induced senescent HDFs. This occurred through the activity of methyltransferase Disruptor of Telomeric Silencing 1-like (DOT1L). This DOT1L-mediated IL1A expression promoted downstream activation of SASP genes such as IL6 , IL1B and CXCL8 , through the activity of transcription factor C/EBPβ. These data, suggest an important role of H3K79me2/3 in regulating SASP during senescence [ 79 ]. Deacetylation of the active histone mark, H4K16ac, through the histone deacetylase SIRT2, promotes DNA compaction and SAHF formation during replicative senescence or OIS in HDFs [ 80 ]. Beyond regulating the cell cycle and SASP, post-translational histone modifications contribute to apoptosis evasion observed in senescent cells. For example, in senescent HDFs, the repressive mark H4K20me3 was increased on pro-apoptotic gene Bax , whereas the active H4K16ac mark was decreased. The opposite result was found on the anti-apoptotic gene Bcl2 , suggesting that histone modification plays a multifunctional role in senescent cell evasion of apoptosis [ 81 ], and thus provide a mechanism for long-term survival in a dormant state. The histone variant, macroH2A1 was shown to be depleted from SASP genes, such as IL6 , IL8 , CXCL1 , CXCL6 , during OIS in HDFs, leading to activation of SASP factors. Activation of these SASP factors led to further removal of macroH2A1 in a paracrine and autocrine manner, generating a positive feedback loop that reinforced OIS [ 82 ]. The histone marks with known roles in senescent cell phenotypes are summarized in Table 1 . Most of the senescence-related histone modifications were identified in HDFs, while modifications in cancer cells are relatively understudied. The senescence-related histone modifications are, theoretically, reversible through interactions between proteins, non-coding RNAs, and stimuli from different microenvironments [ 22 , 83 ]. Histone modifications that promote chromatin structures that favor the expression of senescent transcriptional programs would likely require reversal for escape from senescence in dormancy. Whether stability of chromatin structure prevents relapse could be tested in models of dormancy by using inducible expression of histone modifiers such as Suv39h1, Sirt2, DOTL1, and MLL1 that are known to be involved in senescence (discussed above). Alternatively, conditional, floxed alleles could be combined with inducible Cre systems to delete these same histone modifiers in senescent cells in vivo in models of dormancy. Such studies would enable temporal expression/deletion in cancer cells that are senescent and dormant, and then allow a simple determination of whether the time to relapse is shorter or longer. One would surmise that, for instance, deletion of a histone modifier involved in repressing cell cycle genes might promote cell autonomous emergence from dormancy, while over-expression of a histone modifier that promotes SASP expression might promote non-cell autonomous emergence from dormancy. In sum, epigenetic changes that mediate cellular plasticity and the senescent state suggest the possibility of a long-term arrest and survival state consistent with clinical dormancy. This rigorous, stable, epigenetic programming is balanced by the potential for a rare reversal of the chromatin state, leading to loss of senescent phenotypes (including cell cycle arrest) and tumor recurrence.

Senescence is one potential way cells can respond to various cellular stresses [ 1 , 2 ]. This stress can be caused by intrinsic sources, such as reactive oxygen species generated during metabolism or hyper-replication driven by oncogenes, or can be extrinsic, coming from DNA damaging agents or mitotic spindle poisons such as those used in cancer treatment [ 3 ]. In experimental models, entry into senescence preserves viability in an arrested state for long periods of time, up to years in many cases [ 4 , 5 ]. Under stress conditions, cancer cells can adopt many, if not all, the features of senescence including cell cycle inhibitor expression, lysosome expansion and senescence-associated beta galactosidase (SAβGal) positivity, senescence-associated secretory phenotype (SASP) expression, Lamin B loss, etc. [ 6 , 7 ], and could potentially exist in this state and contribute to late relapse. Because of these characteristics, entry into senescence presents an intriguing potential mechanism for observations of clinical dormancy in cancer treatment. This review will focus on the context of cancer cell responses to chemotherapy and how entry into senescence could enable a dormant state for extended periods followed by eventual relapse. We examine the properties of senescent cells that could contribute to dormancy. These properties have been characterized in numerous normal or tumor cell types and model systems that have been exposed to a variety of stressors, not necessarily tumor cells treated with chemotherapy. That said, many phenotypes of senescence, including those related to gene expression, morphology, survival, etc., are common to all or most cells made senescent by any stress. In response to DNA damaging chemotherapy or mitotic stress, the two primary modes of chemotherapy action, cells can avoid intrinsic cell death by entering a state that has many features of senescence [ 8 , 7 ]. Senescence was originally defined as a “permanent cell cycle arrest” in normal human diploid fibroblasts (HDFs) passaged in culture [ 4 , 9 ]. In these specific conditions, emergence from senescence is exceedingly rare [ 9 ]. The HDF model has been used to define virtually all parts of the senescent phenotype, including the stable arrest, the changes in cellular morphology, the genetic requirements, including TP53 and RB1 , and more recently, the SASP and interactions with the immune system [ 10 ]. Subsequent research in many model systems has revealed that a program similar to that characterized in HDFs is activated in a wide range of cell types in response to many different stresses, both in vitro and in vivo. It is not surprising that senescence programs differ in diverse cell types, but the fundamental properties are the same: a stress stimulus activates the program, then cells enter a stable cell cycle arrest with extensive gene expression changes that include a secretory phenotype [ 11 ]. The specific mediators of arrest (p21, p16, p15) and SASP factors produced (peptides including IL6, CSF1, IGFPB7, and dozens more [ 12 ], as well as lipids including prostaglandins [ 13 ]) will naturally vary by cell type and stress. Cancer cells were discovered to enter a senescent-like arrest in vitro after exposure to chemotherapy drugs such as doxorubicin or taxol [ 6 , 5 ]. Subsequent research has characterized this arrest as being similar to any other senescent arrest [ 14 ]. Recent evidence suggests senescence as a survival mechanism is a clinical problem, particularly in breast cancer, and could contribute to dormancy. When breast tumors are treated with DNA damaging chemotherapy, wild type p53 protein can direct a cell cycle arrest program that allows the tumor cell to avoid entering mitosis with lethal DNA double strand breaks, thus evading death from mitotic catastrophe [ 15 ]. The cell cycle arrest program directed by p53 also enables expression of other senescent phenotypes, such as SAβGal positivity and the SASP [ 11 , 15 ]. The clinical relevance of the p53-mediated program is evident in clinical studies: in breast cancer patients treated with chemotherapy, those with p53 wild-type tumors capable of entering senescence in response to chemotherapy, pathological complete response (pCR) is rarely observed [ 16 - 19 ], and median overall survival is roughly 4 years [ 20 ]. For patients with p53 mutant tumors, pCR is much more common, and patients survive on average more than 20 years [ 20 , 21 ]. If senescent cells were harmless and indeed permanently arrested, then they would not contribute to relapse. But cells can overcome senescence via two mechanisms, bypass or escape [ 22 ]. Cells bypass senescence by continuing to proliferate without adopting senescent phenotypes even though the stress stimulus still exists. Senescence bypass usually happens in less differentiated cell types, such as stem cells and some cancers, that have high expression of epithelial-mesenchymal transition transcription factors (EMT-TFs). In these cells, EMT-TFs not only contribute to high cellular plasticity but also inhibit p16/Rb pathway, thus attenuating the activation of senescence and cell cycle arrest [ 23 ]. Whether cells escape senescence is somewhat dependent on semantics. Using criteria derived from early studies in HDFs to define senescent cells as “permanently arrested”, no cell that restarts proliferation would be considered to have ever been senescent, no matter what other phenotypes were expressed. Over time, this narrow view has been challenged. Some more differentiated cell types including epithelial cells enter senescence or senescent-like arrest with the expression of senescent phenotypes but can restart proliferation [ 24 ]. Indeed, even classically senescent HDFs exit their senescent state with minimal genetic perturbation [ 25 ]. In a population of treated cancer cells that are mostly senescent, proliferative clones can appear over time [ 26 - 28 ]. Further investigation has shown cancer cells can enter a senescent-like state, adopt morphological changes, and then reenter the cell cycle and proliferate [ 29 - 32 ]. This phenomenon has been relatively understudied, and more research is necessary to define the state of cells that exit an arrest that is, at least superficially, quite senescent-like. Indeed, the program of senescence shares features with another mechanism of arrest utilized by tumor cells to avoid cell death: a diapause-like state [ 33 , 34 ]. Similarities include a slow or non-cycling population of cells that avoids mitotic related cell death from chemotherapy and has altered levels of BCL2 family proteins that promote survival. Nonetheless, entering a senescent-like arrest will prolong the survival of the damaged or stressed cell, which can ultimately result in one of two fates: the cell is destined to never divide again, perhaps removed by the immune system; or, the senescent cell avoids immune detection, remains viable, and eventually emerges from its senescent state to proliferate again (cell autonomous effect), or influences neighboring, non-senescent cells to proliferate, survive, or metastasize (non-cell autonomous). Tumor dormancy is characterized by the presence of cells at either a primary or distal site that, at one point in time, are not detectable or symptomatic, but eventually reveal themselves after a recurrence that can occur years or decades following treatment and surgical resection [ 35 , 36 ]. Research has suggested multiple models to explain tumor dormancy [ 35 ]. Tumor mass dormancy posits that proliferation of cancer cells at some specific site is offset by cell death, thus creating a net zero gain in actual tumor mass. Tumor mass dormancy can be mediated by at least two different means, immunomediated dormancy or angiogenic dormancy [ 35 ]. Cellular dormancy is defined as when a cell or mass of cells is in proliferative arrest or a bare minimum of proliferation, but also experiencing little concurrent cell death [ 35 ]. As with tumor mass dormancy, the tumor is neither growing nor shrinking, but is only evident when this arrest is exited, cells proliferate, and the tumor recurs. It is self-evident that many senescent phenotypes, including growth arrest, SASP, and immune modulation, might be favorable to dormant cells, or could even be the primary driver of dormancy. It should be noted, interpreting data that address a potential role for senescence in mediating dormancy requires care. Modeling actual clinical dormancy is intractable due to the small number of cells (indeed, undetectable clinically) that persist and the length of time (years or decades) the cells remain dormant.