Cellular senescence in cancer: clinical detection and prognostic implications

The best known and most widely used marker of cellular senescence is enhanced activity of acidic lysosomal β-galactosidase in senescent cells [45, 46], as lysosomes increase in number and size when cells become senescent [47]. The senescence-associated beta-galactosidase (SA-β-Gal) activity is often considered the gold standard for identifying senescent cells, despite SA-β-Gal activity was reported as a non-universal marker for cellular senescence [48]. Although absent in most proliferating and quiescent cells [44], SA-β-Gal activity is expressed in certain cell types (i.e., macrophages [49], bone marrow cells [50], melanocytes and sebaceous and eccrine gland cells [48]) and in vitro cells under certain cell culture conditions (i.e., confluence and serum starvation [51‐53]) independent of a senescent cell state. Also, SA-β-Gal is not essential for senescence as cells can become senescent without expressing SA-β-Gal [54]. Of note, SA-β-Gal detection is only possible in fresh snap-frozen tissue samples [45], thus hampering its use in a clinical context.

To overcome the disadvantages of SA-β-Gal as marker for senescence, a biotin-linked Sudan Black B (SBB) analogue was designed to detect lipofuscin accumulation in senescent cells [45]. Lipofuscin is a non-degradable aggregate of oxidized lipids and proteins [55], that accumulate in lysosomes of senescent cells due to senescence-related lysosomal malfunction, and is considered a hallmark of cellular senescence [1, 56]. In contrast to the enzymatic SA-β-Gal activity, lipofuscin is preserved in fixed materials [51]. As such, detection of cellular senescence is feasible in formalin-fixed paraffin-embedded (FFPE) archival tissue samples using the SSB histochemical stain [45]. The interpretation of the assay requires some experience, as lipofuscin aggregates can be very small and background dirt can be wrongly interpreted as positive SBB-positive lipofuscin aggregates, comprising the overall sensitivity [45]. Interestingly, endogenous lipofuscin is linked to chronic liver disease and can be detected by autofluorescence in biopsied samples of human liver tissue [57]. As such, autofluorescence of lipofuscin in the context of cellular senescence could potentially be exploited to detect senescence in patient samples.

Other commonly used markers of cellular senescence are the cell cycle inhibitors p16^INK4a and p21^WAF1/Cip1, as most senescence-inducing triggers lead to the activation of the cell cycle inhibitor pathways p53/p21^WAF1/Cip1 and/or p16^INK4a [7] (Fig. 2). While p21^WAF1/Cip1 expression occurs early after senescence induction and is reversible upon tumor suppressor protein p53 inactivation, p16^INK4A expression is frequently induced late after senescence induction and is irreversible upon p53 inactivation [44, 58, 59]. p21^WAF1/Cip1 expression is therefore more likely to represent early cellular senescence, whereas p16^INK4a expression represents a more established and durable senescence response [44]. However, p21^WAF1/Cip1 can be expressed by non-senescent cells in case of DNA-damage [60] and the genes encoding for p21^WAF1/Cip1 and p16^INK4a (i.e., CDKN1A and CDKN2A, respectively) were not identified within the core transcriptome signature of senescent cells [61].

The senescence-associated cell cycle arrest, which occurs in G1 and possibly in G2 phase of the cell cycle [62], is marked by the absence of the proliferation marker Ki67 [1]. However, Ki67 is also absent in other cell states with a temporarily and durable cell-cycle withdrawal in G₀, such as quiescence and terminally differentiated cells, respectively. 5-ethynyl-2′-deoxyuridine (EdU) is another proliferation markers and a thymidine analog that, when administered to cells, can incorporate into DNA during replication [44]. Unfortunately, EdU is not applicable for ex vivo tissue samples as active proliferation is required after biopsy.

As SA-β-Gal activity [56], p16^INK4a and p21^WAF1/Cip1 expression [60] and absence of Ki67 expression [63] are neither specific nor universal for cellular senescence, the International Cell Senescence Association [1] and others [44] recommend combining different markers for the detection of cellular senescence with the highest accuracy. However, combining all the markers in the same tissue sample is not yet possible without incurring artifactual false positives or negatives [44]. Therefore, current senescence validation in patient tissue samples can be determined either in snap-frozen samples using a sequential staining for SA-β-Gal and Ki67 on independent sequential and adjacent sections, or in FFPE tissue samples using a double-staining for lipofuscin and/or Ki67, p21^WAF1/Cip1 or p16^INK4a combined with a sole staining of the remaining markers (Ki67, p21^WAF1/Cip1 or p16^INK4a) on sequential and adjacent Sects [44]. Additional immunohistochemical confirmation can be achieved by detection of components of senescence-associated heterochromatin foci (SAHF), including histone variant macroH2A [64], di- or trimethylated lysine 9 histone H3 (H3K9me2/3), heterochromatin protein (HP) 1 α, β and γ [64, 65] and high mobility group A (HMGA) proteins [64, 66, 67], and DNA damage foci such as phosphorylated H2AX (γH2AX) [68] (Fig. 1).

Using flow-cytometry, senescent cells can be identified and quantified on single-cell level by combining SA-β-Gal activity with staining of molecular markers for cellular senescence (e.g., γH2AX) and absence of Ki67 and/or high mobility group box 1 (HMGB1) protein [69]. Recently, a preliminary proof of concept method was developed to detect senescent cells with imaging flow cytometry based on measuring autofluorescence and morphological parameters, and on applying recent artificial-intelligence (AI) and machine learning (ML) tools [70], potentially facilitating cellular senescence detection without a multi-marker strategy.

Indirect markers of cellular senescence such as messenger RNA (mRNA) expression of p16^INK4a, p21^WAF1/Cip1 and lamin B1 (i.e., a nuclear lamina component and downregulated in case of senescence [71]), and selected senescence core genes can be determined by reverse transcription–polymerase chain reaction (RT-PCR) [44] (Fig. 1). However, both flow-cytometry and RT-PCR techniques require tissue dissociation and case-specific control samples, and do not provide any spatial information on senescent cells residing in the tissue, limiting their clinical utility for solid tumors. Although various senescence core genes have been determined [72], the senescent phenotype is dynamic and heterogeneous and depends on the tissue of origin and senescence-inducing trigger [72‐77]. It is currently unclear which specific core genes should be included in order to confirm cellular senescence in a disease- and tissue-specific context [44]. Recently, using machine learning, a gene expression classifier (SENCAN classifier) was developed for the detection of senescence in cancer samples. Using transcriptome data as input, the SENCAN classifier was able to classify whether cancer cells are senescent or not. Unfortunately, whereas the SENCAN classifier is able to accurately detect senescence in many cancer cells in vitro, its accuracy to detect senescence in in vivo cancer samples is still unclear [76].

Of note, uPAR was recently identified as a cell surface protein that is broadly and specifically upregulated in senescent cells of mice using RNA-sequencing datasets derived from three independent and robust mouse models of OIS as well as TIS [78]. uPAR is involved in many intracellular signaling pathways that promote cell motility, invasion, proliferation and survival [79] and is expressed by tumor and stromal cells in a wide variety of human cancers where its expression frequently indicates poor prognosis [79]. In this context, uPAR expression and secretion (i.e., soluable uPAR) could be an interesting biomarker of senescence in cancer patients detected by immunohistochemistry, flow-cytometry or enzyme-linked immunosorbent assay (ELISA) [78] (Fig. 1).

Detection of tumor-specific circulating material in patient fluids by means of liquid biopsy is an emerging field in oncology, with important clinical implications for personalized medicine [80]. In contrast to surgical or biopsy tissue samples, liquid biopsies are not subject to sampling bias, tumor heterogeneity and can be obtained repeatedly to monitor the evolution of the molecular profile of the tumor which may cause drug resistance [81]. However, the detection of cellular senescence via liquid biopsy in patient blood or urine requires specific senescence-associated circulating material including proteomes, extracellular vesicles (EVs) and circulating cell-free DNA (cfDNA).

First, SASP proteins can be measured by mass spectrometry or immunoassays (ELISA, western blot) in patient plasma [44, 74]. A proteomic atlas of core SASP secreted proteins originating from multiple senescence inducers and cell types was recently determined [74], enabling senescence detection by the presence of core SASP proteins, such as growth/differentiation factor 15 (GDF15), matrix metalloproteinase-1 (MMP1), stanniocalcin-1 (STC1), tissue inhibitor of metalloproteinases 1 and 2 (TIMP1 and TIMP2) [44] (Fig. 1). However, several of these core soluble SASP proteins have also been identified as biomarkers of human disease [75] and are positively associated with age, frailty and adverse post-surgery outcomes [82]. For example, GDF15, MMP1 and STC1 have been identified as a biomarker for cardiovascular disease [83], several cancers [84] and for Alzheimer’s disease [85], respectively. SASP biomarkers currently lack sensitivity to detect and attribute senescence in patient plasma to specific pathologies, including cancer [75]. However, SASP protein profiles differ among cell type, senescence-inducing trigger and interval after senescence induction [62, 74, 75], as well as age category [82] as senescent cells accumulate with increasing age [6]. By determining disease-, tissue-, and inducer-specific SASP factors as well as robust core SASP factors secreted by senescent cells in multiple contexts [75], it will become possible to attribute the secretion of certain SASP factors to the presence of senescent cells in patients in the near future.

Next to SASP proteins, senescent cells of human origin (i.e., foreskin primary [86], normal lung [87] and diploid [88] fibroblasts, prostate [87, 89], hepatocellular [87] and triple negative breast [90] cancer cells, retinal pigment epithelial cells [88] and human chondrocytes [91]) are capable of releasing EVs in patient fluids such as blood and urine [92]. (Fig. 1). EVs are small, lipid-bilayer enclosed, cell-derived particles that bear surface molecules that allow them to target recipient cells and contain transmembrane and enclosing cytosolic proteins and RNA [75, 93]. Once internalized, EVs release their content into the cytosol modifying the physiological state of the recipient cell [93] and enabling cell communication with neighboring as well as distant cells. As for SASP proteins, EV production and content drastically differ in physiological [94] and pathological [92] conditions, making EVs an additional interesting source of disease biomarkers. EVs of senescent cells are capable of transmitting paracrine senescence to neighboring cells [86, 91], contain chemotherapy and key proteins involved in cell proliferation after chemotherapeutic challenge [90] and can even promote cancer cell proliferation [88]. Interestingly, protein content of EVs secreted by senescent cells differs from secreted SASP proteins [74, 86], suggesting that SASP and EVs do not act as surrogate biomarkers and have different clinical significance and value [75]. However, as for SASP proteins, it is still unclear how EVs secreted by senescent cells exactly behave in physiologic and disease-specific contexts, and to what extend they depend on age, tissue and senescence-inducing trigger [75].

Detection and analysis of cfDNA by means of next-generation sequencing, digital droplet polymerase chain reaction, exome or whole-genome sequencing [95] could be a third appealing strategy to detect and monitor the senescence burden in cancer patients. Senescent cells exhibit a DNA methylation pattern of promoter hypermethylation mainly involving metabolic regulators, whereas transformed cells exhibit a DNA methylation pattern of promoter hypermethylation involving primarily pro-survival and developmental genes [96] (Fig. 1). Also, using a machine learning based approach trained with different early passage and senescent cells, a DNA methylation fingerprint of cellular senescence (DNAmSen) was developed and validated in clinical patient samples, such as whole blood and skin tissue [97]. With this approach, clear and robust correlations were found between the patient’s age and DNAmSen present in the corresponding sample. Interestingly, also elevated DNAmSen were observed in lung samples from patients with COPD and lung cancer compared to those of healthy controls [97]. Not surprisingly, the release of cfDNA is affected by type of treatment and timing from treatment exposure, and also heavily depend on the cellular response to treatment. Of note, it is thought that blocks the release of cfDNA whereas apoptosis and necrosis are key contributor of its release [98].

Currently, there is no established method to detect cellular senescence in vivo in patients. Detection can be achieved by chromogenic [99, 100] or fluorogenic [101‐107] probes, preferentially hydrolyzed by SA-β-Gal, resulting in color- or fluorescence-enhanced senescent cells. Chromogenic [108] or fluorogenic [109] probes hydrolyzed by other lysosomal hydrolases overexpressed in senescent cells, such as α-L-fucosidase [110], can be used as well. Nanoparticles containing fluorescent dyes and probes have been developed that selectively release their content when the cap of the nanoparticle is hydrolyzed by SA-β-Gal after endocytosis [111‐113], or by interaction with CD9 receptors [114] or β2 microglobulin [115], both preferentially expressed by senescent cells. Due to low tissue penetrance and autofluorescence the clinical use of these fluorescent probes and nanoparticles may be limited in patients [116]. An alternative method could be the detection of endogenous lipofuscin as endogenous lipofuscin, next to ex vivo in patient samples, can also be monitored in vivo and non-invasively via imaging, as has been shown in mice with chronic liver disease [57].

It should be noted that all these approaches were validated only in vitro or in vivo in mice and its use in patients should be further investigated. Currently, there is one first-in-human trial in cancer patients evaluating the safety and imaging characteristics of a novel senescence-specific radiotracer [18F]FPyGal (i.e., a radioactive form of SA-β-Gal) that can be tracked non-invasively in the body through positron emission tomography (PET) imaging (SenPET; NCT04536454) [117] (Fig. 1). Whether this strategy is sufficient to detect all senescent cells [118] due to the aforementioned limitations of SA-β-Gal as a specific marker of senescence has yet to be determined.

The senescence-associated cell cycle arrest is considered fundamentally tumor-suppressive and the induction occurs through the involvement of different signaling and downstream cell cycle inhibitor pathways. Genotoxic stress induced by anticancer therapies results in a DNA-damage response (DDR) which leads to p53 and p21^WAF1/Cip1 activation whereas oncogenic signaling and tumor suppressor inactivation results in downstream activation of both p53/p21^WAF1/Cip1 and p16^INK4a via participation of the DDR and the Ras-Raf-MEK-ERK, PI3K/AKT/mTOR and p38/MAPK signaling pathways [6, 7, 62]. As such, TIS is primarily induced through p53/p21^WAF1/Cip1 pathway activation whereas OIS is induced through either p53/p21^WAF1/Cip1 and/or p16^INK4a pathway activation (Fig. 2A). Upregulation of functional tumor suppressor proteins p53, p21^WAF1/Cip1 and p16^INK4a inhibit downstream cyclin-dependent kinase (CDK)—cyclin complexes, such as CDK2—cyclin E and CDK CDK4/6—cyclin D, preventing phosphorylation of the retinoblastoma protein [7]. Hyperphosphorylation of this tumor suppressor protein blocks S-phase entry [120] and is responsible for the induction of senescence [6] (Fig. 2A). Of note, despite tumor suppressor proteins p53, p21^WAF1/Cip1 and/or p16^INK4a are primarily involved in TIS and OIS (Table 1), senescence can be induced [121] as well as bypassed [7, 16, 122‐124] independent of p53/p21^WAF/Cip1 and/or p16^INK4a pathway activation and inactivation/abrogation, respectively. Also, p21^WAF1/Cip1 can be activated by pathways that are independent of p53 [125].

There is abundant ex and in vivo evidence in several tumor types that OIS acts as a tumor-suppressive mechanism preventing the expansion of pre- or fully malignant cells (Table 1). OIS is found in precursor lesions and in low TNM stage tumors with more favorable clinicopathologic features [13, 202, 209], whereas in full-blown malignant lesions OIS-related markers are often dysregulated or completely lost [173] and correlate with higher TNM stage tumors and poor clinicopathological parameters [136, 164, 211, 212] (Table 1). Tumor suppressor protein p16^INK4a often comes forward as the main regulator for maintaining the OIS-associated cell cycle arrest which is considered to be more crucial for maintaining the senescence-associated cell cycle arrest whereas p53/p21^WAF1/Cip1 pathway activation is more involved in the initiation of senescence [59]. Mutations, deletions, secondary alterations and/or promoter silencing of cellular control genes (i.e., TP53, CDKN1A and CDKN2A) encoding for tumor suppressor proteins p53, p21^WAF1/Cip1 and p16^INK4a may result in inadequate senescence induction or escape from senescence due to absent or dysfunctional cell cycle inhibitor pathway activation and absent or dysfunctional tumor suppressor proteins to induce or maintain OIS. Hence, dysregulated (i.e., decreased or overexpressed) expression or complete loss of tumor suppressor proteins p16^INK4a, p21^WAF1/Cip1 and p53 are often correlated with increasing grade of malignancy and tumor progression, and associated with a negative prognostic outcome [139, 140, 158, 170, 187, 205‐207, 213, 214, 236] (Table 1) (Fig. 2B).

However, in certain tumor types, the (abundant) presence of OIS or expression of senescence-associated markers is also linked to worse prognosis [132, 197, 221, 237, 238, 248]. Perhaps even more surprisingly, both absence and extensive presence of senescence in CRC was associated with negative prognosis whereas moderate presence was associated with the best prognosis [174], demonstrating that an extensive senescence burden can paradoxically impair clinical outcome in contrast to a moderate senescence burden.

Concerning TIS, evidence demonstrates that TIS is an in vivo relevant outcome of various anticancer therapies in several tumor types (Table 1). For example, in breast, colorectal and prostate cancer TIS was observed after neoadjuvant genotoxic chemotherapy [150, 177, 178] and antihormone therapy [204]. Therapy-induced senescent cells were identified in residual drug-resistant tumors [150] and in samples with partial or incomplete pathological response to neoadjuvant therapy [149], suggesting TIS might persist after neoadjuvant therapy [203] and is responsible for incomplete tumor regression [204]. The presence of TIS is however linked to contradictory clinical outcomes and is associated with worse [135, 137] as well as improved [163, 176, 178] prognosis depending on tumor type. For example, while in non-small cell lung cancer (NSCLC) TIS is associated with worse OS [135], in CRC a higher proportion of therapy-induced senescent cells after chemotherapy treatment was associated with a longer progression-free survival (PFS) compared to when the proportion of senescent tumor cells did not change before and after chemotherapy [178]. Thus, regardless of the type of cancer, the senescence burden of OIS and TIS seems to be an important determinant affecting the outcome in cancer patients.

Whereas the senescence-associated cell cycle arrest acts tumor-suppressive, SASP factors secreted by senescent cells can be both tumor-suppressive and tumor-promoting [255]. The main signaling pathways involved in SASP regulation include NF-κB, p38, mTOR, C/EBPβ and JAK2/STAT3 [16, 256‐260]. Interleukin (IL)-1α is secreted by oncogene-induced and therapy-induced senescent cells and initiates the production of key SASP proteins such as IL-6 and IL-8 through activation of NF-κB and C/EBPβ [62, 261]. The senescent phenotype is subsequently enforced autocrinally by IL-6 [16] and IL-8 [122] and transmitted paracrinally to neighboring cells by IL-1α [262], further enhancing the production of these SASP factors. Abundant SASP factors IL-6 and IL-8 have both anti-tumorigenic and pro-tumorigenic effects [263]. For example, both interleukins mediate the recruitment of macrophages, T cells and natural killer (NK) cells supporting immune surveillance and elimination of senescent cancer cells [32] but also create a chronic inflammatory TME driving cancer development [73] and attract myeloid derived suppressor cells that suppress T [264] and NK cells [196] and blocks IL-1α signaling, preventing paracrine senescence in neighboring cancer cells [265]. Next to pro-inflammatory cytokines, the SASP may consist of a variety of chemokines (e.g., CCL2 and CXCL1), angiogenic factors (e.g., VEGF), growth factors (e.g., HGF, PDGF, EGF and TGFα), matrix-remodeling enzymes (e.g., MMP1 and MMP3) and bioactive lipids [62, 263]. However, its composition is highly dynamic [263], complex and variable and depend on the cell type, senescence-inducing trigger and type of senescence [62, 74, 75], resulting in cancer-specific and context-dependent effective SASP levels. Besides its variable composition, the SASP is suggested to have a time-dependent impact [31]. Whereas the short term presence of SASP is suggested to be primarily tumor-suppressive, the long term presence of pro-inflammatory SASP factors can drive cancer [31, 32]. Thus, depending on the secretion, composition and the duration of its presence, the net effect of the SASP may be tumor-suppressive or tumor-promoting, thereby either enhancing or opposing the tumor-suppressive property of the senescence-associated cell cycle arrest.

Evidence from patients with various tumor types show that oncogene-induced senescent cells are capable of secreting a tumor-promoting and immune-suppressive SASP that is linked to impaired clinical outcome. NSCLC patients with an elevated senescence-related gene signature score overexpressed an immune-suppressive SASP and demonstrated decreased infiltration levels of cytotoxic T cells and NK cells and increased levels of immune-suppressive cells (i.e., neutrophils, cancer-associated fibroblasts, regulatory T cells, and resting NK cells), disease progression and worse OS [133]. The importance of the interaction between the SASP and immune surveillance of senescent tumor cells is further emphasized by the finding that, in CRC, both a lower average distance between senescent cells and T cells as well as a higher percentage of T cells near senescent cells were linked to improved survival, suggesting that the tumor-suppressive potential of cellular senescence is determined by the TME and immune cell-mediated elimination of senescent tumor cells [174]. The SASP of senescent cells can also direct neighboring cells and drive cell transformation and tumor initiation [231, 232] and mediate collective invasion and metastasis [219], as evidenced in adamantinomatous craniopharyngioma and papillary thyroid cancer. Also in precursor lesions of pancreatic ductal adenocarcinoma (PDAC) (i.e., pancreatitis and pancreatic intraepithelial neoplasias), a senescence-associated inflammatory SASP was linked to PDAC progression [189].

Increasing evidence demonstrates that also therapy-induced senescent cells can produce a tumor-promoting and immune-suppressive SASP that might impair clinical outcome. For example, in response to genotoxic chemotherapy, TIS and a protumorigenic SASP were observed in prostate cancer resection samples [73], and overexpression of SASP factor were associated with impaired outcome in head and neck squamous cell carcinoma patients [254] as early TIS and SASP production upon radiotherapy was demonstrated in a preclinical model. In an elucidative study, therapy-induced senescent cells of breast and prostate cancer patients were found to evade immune clearance by shedding of natural killer group 2D (NKG2D) ligands and paracrine suppression of NKG2D-receptor-mediated immunosurveillance [150]. Of importance, since TIS depends on p53/p21^WAF1/Cip1 pathway activation (Fig. 2), the tumoral p53 status indirectly determines SASP production and outcome after treatment with senescence-inducing anticancer therapies. This was illustrated in an in vivo p53 wild-type breast cancer model where TIS was induced instead of cell death after chemotherapy treatment and resulted in minimal regression of the tumor and early relapse through the secretion of protumorigenic SASP [266]. Accordingly, breast cancer patients harboring a TP53 mutation showed an improved response to anthracycline-based chemotherapy [267, 268].

In contrast, abundant evidence links both OIS [165, 202] and TIS [163, 176, 178] to improved outcome (Table 1). In this scenario, it is conceivable that oncogene-induced and therapy-induced senescent cells secrete moderate to low effective SASP levels and/or secrete a SASP with a net tumor-suppressive and immune-promoting effect. For example, in pilocytic astrocytoma (PA), a low grade glioma and most common brain tumor in children, SASP factors were upregulated, and high levels of IL-1β and SASP expression were associated with favorable PFS [224]. The SASP was therefore suggested to regulate OIS in PA [225] and held responsible for the slow growth pattern, the lack of progression to higher-grade astrocytomas and the high OS of affected patients [224]. In addition, no oncogene-induced or therapy-induced senescent cells were identified in chemotherapy-naïve and neoadjuvant chemotherapy treated breast cancer samples, suggesting tumoral senescent cells either were already cleared by the immune system or bypassed senescence [151].

There is mounting evidence that senescence also occurs in the TME and has prognostic implications. In gastric cancer, CRC and esophageal cancer patients, an enriched senescence gene signature in noncancerous cells, but not in cancerous cells, of the TME (e.g., endothelial cells, enteroendocrine cells, macrophages and fibroblasts) resulted in a longer disease-free survival and OS [168]. In contrast, identification of senescent cancer-associated fibroblasts (CAFs) adjacent to epithelial ovarian cancer cells in ovarian cancer specimens were suggested to promote ovarian cancer tumorigenesis [161]. The presence of a senescence-associated gene signature in peritumoral tissue of hepatocellular carcinoma (HCC) patients was also associated with early recurrence and poor survival as peritumoral OIS induced an accumulation of C–C chemokine receptor 2⁺ myeloid cells through secretion of C–C motif chemokine ligand 2, resulting in NK cell inhibition and enhanced HCC growth [196]. Interestingly, in a murine rectal cancer and patient-derived tumor organoids model, IL-1α was found to predispose inflammatory CAFs to p53-mediated TIS upon irradiation, which in turn resulted in chemoradiotherapy resistance and disease progression through the secretion of cytokines and extracellular matrix constituents supporting the invasion and metastasis of cancer cells and counteracting the irradiation-induced tumor cell death [184]. Consistently, the presence of inflammatory CAFs in pre-therapeutic patient biopsies resulted in poor chemoradiotherapy response and low IL-1α receptor antagonist serum levels, which enhances IL-1 signaling and predisposes inflammatory CAFs to TIS, correlated with poor prognosis in rectal cancer patients [184].

Hence, cellular senescence is not solely limited to cancerous cells but also occurs in cells of the TME as well as of the immune system [269].

Evidence of several cancer types (presented in Table 1) suggests that the prognostic implications of OIS and TIS are highly context-dependent and primarily depend on the (i) the senescence burden; (ii) the secretion; and (iii) the composition of the SASP and/or duration of SASP presence. Therefore, in a simplified schematic model, we present different scenarios that could provide a rationale for the differential outcomes of cellular senescence observed in cancer patients, based on the interplay between these three factors, i.e., (i) the senescence burden (i.e., low, moderate or high); (ii) the secretion (i.e., low or high); and (iii) the composition of the SASP (i.e., net tumor-promoting and immune-suppressive or net tumor-suppressive and immune-promoting) and/or duration of SASP presence (i.e., short term or long term) (Fig. 3).

The senescence-associated cell cycle arrest is considered tumor-suppressive whereas the composition of the SASP and/or duration of SASP presence determines whether the senescence-associated cell cycle arrest is reinforced or opposed. The degree of SASP secretion and levels, which depend on the senescence burden, determines to which extent the senescence-associated cell cycle arrest is reinforced or opposed. As such, OIS and TIS can have tumor-suppressive and tumor-promoting properties (Fig. 3A-F).

The proposed model provides a rationale for the differential outcomes of OIS and TIS observed within the same cancer type, such as in [174], as well as between different types of cancer (Table 1). Accordingly, cancer cells originating from urinary systems, glands and soft tissues (e.g., prostate cancer, adenoid cystic carcinoma, RCC and melanoma) exhibited relatively higher cellular senescence gene scores than tumors originating from reproductive organs (e.g., breast cancer, cervical SCC and OC) [136], and were correlated with SASP factors and immune related genes, suggesting SASP-induced immune infiltration. The infiltration of immune cells varied however in a cancer-specific pattern [136], contributing to the context-dependency of the proposed model and the interplay between the senescence burden, the secretion and composition of the SASP.

Of note, as senescent cells can reinforce their senescent phenotype in an autocrine fashion [16, 122] and paracrinally transmit the senescent phenotype to adjacent malignant and non-malignant cells [262, 270], the tumoral senescence burden can increase over time resulting in altered tumoral repercussions (Fig. 3, vertical arrow hourglasses) in addition to the differential time-dependent impact of SASP (Fig. 3, horizontal arrow hourglasses). Thus, cellular senescence in cancer should be considered as a dynamic, rather than an irreversible, static condition [271], with antitumorigenic and protumorigenic features that can change over time.