The Escape of Cancer Stem Cells From Therapy
At the moment first-line therapeutic treatments in progressed tumors include in the most cases surgery, chemo- as well as radiotherapies (501) (compare Figure 2). Those have been shown to induce DNA damage and to trigger senescence in cancer cells, a process known as therapy-induced senescence (TIS) (10, 502, 503). TIS will cause a decreased tumor size and attracts immune cells such as neutrophils, monocytes as well as T-cells toward senescent tumor site (503). However, over a long-term period the anti-tumorigenic effects of TIS are lost and the cancer might gain stemness causing tumor relapses (Figure 2).
Figure 2  Kinetic of tumor development in pre-, early-, and late-therapeutic period upon application of chemo- and/or radiation therapy: current situation in the clinic. (A) In the pre-therapeutic situation, heterogeneous tumors are composed of several cell types, including CSC, tumor cells, TAMs, and CAFs; all characterized by biomarkers. (B) In the early post-therapeutic period, upon application of the first-line treatment (that currently uses mostly chemo- or radiotherapeutic regimens) several important changes occur in the tumor, in particular: tumor cells or CAFs die due to the therapy or become senescent, whereas CSCs mostly survive the treatment. Senescent cells (tumor cells and CAFs) attract immune cells toward the senescent site via SASP. The SASP therefore plays a positive role and attracts immune cells in this early post-therapeutic situation. Attracted immune cells promote the clearance of dead, of necrotic, and senescent tumor cells and CAFs. (C) In the late post-therapeutic situation uncleared senescent tumor cells and senescent CAFs and SASP thereof play a negative (pro-tumorigenic) role and support tumor development. SASP molecules provide stimulating factors for CSCs for further uncontrolled proliferation as well as their maintenance. Also, remaining senescent tumor cells acquire additional mutations that promote activation of a stemness phenotype and finally a tumor relapse.

Therapy-Induced Senescence: Its Hallmarks, Biomarkers, and Its Role in CSC Generation
Agents that induce DNA damage such as chemo- and radiation therapies have been identified to trigger senescence in differentiated cancer cells (10). TIS has been in the research focus, because it significantly contributes to the long-term outcome of patients (12). The DNA damage response ultimately activates one or several tumor suppressors pathways [p53, p16 (Ink4a), p21 (Waf1), and retinoblastoma (RB)], that trigger and maintain the senescence growth arrest (504). However, it is important to mention that the senescence phenotype can also be induced in cancer cells which lack functional p53 and RB protein (504). TIS and senescence in general, are recognized as a double-edged sword, that on the one hand leads to the attraction of immune cells, inflammation, and elimination of senescent tumor cells and correlates with a positive post-treatment prognosis and treatment outcome (505–507). On the other hand, senescence can play a strong pro-tumorigenic role that supports CSC generation, as described below.
Senescent cells are characterized by biochemical and morphological changes such as flattening and/or nuclear enlargement (508). There are several classical biomarkers of cellular senescence and they comprise: senescence-associated beta-galactosidase (SA-ß-gal) activity, expression of p53 protein, the amount of p53 in the nucleus, increase in expression of p14 (Arf), p16 (Ink4a) and p21 (Waf1), SASP, and often senescence-associated heterochromatic foci (SAHF) (12, 505, 507, 509–515). Furthermore, senescent cells display low Ki67 levels and show levels of heterochromatin protein 1 (HP1) gamma (516), as well as di- or tri-methylated lysine 9 of histone H3 (H3K9Me2/3) and histone H2A variant (macroH2A) (505, 517, 518). The usefulness of telomere length as a biomarker of senescence has been questioned (505).
Biomarkers that underline the effect of a therapeutic approach based on the induction of senescence have to be evaluated carefully and quite often simultaneously. The investigation of senescence markers after post-operative chemotherapy in muscle-invasive bladder cancer (MIBC) revealed that the simultaneous expression of several markers involved in the p53 pathway has to be checked to correctly assess the pathological outcome of MIBC (509). The analysis revealed that the expression of p14 (Arf) was associated with an impaired response to chemotherapy and poor prognosis, whereas p21 (Waf1) expression was related to reduced tumor cell proliferation (509).
TIS can play an anti-cancerous role (503). As demonstrated in our studies in premalignant and malignant liver disease, the induction of senescence leads to a so-called “senescence surveillance” mechanism, which relies on innate and adaptive immune cells. These cells clear senescent premalignant cells, thereby protecting premalignant liver from cancer development (535, 536). Interestingly, in further studies, we could show that the chemokine (monocyte chemoattractant protein 1, MCP-1) axis is of importance for the induction and maintenance of senescence and for the sufficient immune surveillance in the liver (525). Several biomarkers of senescence were found to correlate with a disease-free survival or with an improved OS in several solid cancers (516, 524). One such indicator, a lysosomal-beta-galactosidase (GLB1) that hydrolyzes beta-galactose from glycoconjugates and represents the origin of SA-ß-gal, was reported as a reliable senescence biomarker in prostate cancer (516). Inhibition of the cyclin-dependent kinase 4/6 (CDK)-RB pathway by a novel drug, SHR6390, resulted in reducing the levels of Ser780-phosphorylated RB protein and correlated with the G1 arrest as well as with cellular senescence in a wide range of human RB+ tumor cells in vitro (520). Xiang et al. identified seven senescence-associated genes (SAGs, Table 8) significantly decreased in senescent cells and increased in HCC tissues (524). Interestingly, those SAGs were strongly associated with OS, especially in Asian populations, and had a better predictive accuracy in comparison to serum AFP in predicting OS of HCC patients (524). Recently, Smolle et al. reviewed and underlined the role of members of the inhibition of growth (ING) family. These act as tumor suppressors, regulating cell cycle, apoptosis, and cellular senescence. The authors proposed them as clinically useful biomarkers in the detection and prognosis of lung cancer (523).
Table 8  Biomarkers of therapy-induced senescence (TIS).
Biomarker   References
Senescence-associated beta-galactosidase (SA-β-Gal)  (12, 14, 504, 505, 510, 516–520)
P53  (14, 504, 520, 521) (507)*
Retinoblastoma (RB) Protein  (CDKN2A; Ser780phosphorylated RB protein; cyclin-dependent kinase (CDK) 4/6-retinoblastoma)  (12, 14, 504, 519–521) (507)*
P14 (human)  P19 (mouse)  (12, 509, 510, 514, 515, 519)  (505, 507, 513)*
P16 (INK4A; CDKN2)  (12, 14, 509, 512, 515, 519, 522)  (505, 507, 513, 514, 516)
P21 (WAF1)  (14, 509, 522)  (505, 507, 513)*
Senescence-associated heterochromatic foci (SAHF)  (12, 509, 510, 515, 519)  (505, 507, 513)
Heterochromatin protein 1 (HP1) gamma  (509, 516, 518)
Telomere length  (505)*
Di- or tri-methylated lysine 9 of histone H3 (H3K9Me2/3)  (505, 517, 518)
Histone H2A variant (macroH2A)  (505, 517, 518)
Lysosomal-beta-galactosidase (GLB1)  (516)
Inhibition of growth (ING) family of proteins (ING−1,−2,−3,−4,−5)  (523)
Senescence-associated genes (SAGs) family: [18B (KIF18B), Citron kinase (CIT), Centrosomal protein 55 (CEP55), minichromosome maintenance complex component 5/7 (MCM), Cell division cycle 45 (CDC45), enhancer of zeste homolog 2 (EZH2)]  (524)
Senescence-associated secretory phenotype (SASP)  (12, 14, 510, 519)  (505, 507, 509, 522)
Soluble TNF-receptor-II  (11, 523)
Chemokine (C-C motif) receptor/ligand 2, (CCR2/CCL2); Monocyte chemoattractant protein 1 (MCP-1) axis  (525)
IL-1  (526)
IL-6  (527–531)
IL-8  (528, 531, 532)  (526, 527)
Regulated on activation, normal T cell expressed and secreted (RANTES)  (533, 534)
Examples of the most important biomarkers of TIS are listed. Stars indicate reviews (*). In line with the positive role of senescence, evidence exists regarding the role of TIS in turning “cold” tumors toward a “hot” phenotype that results in activating immune responses against tumor antigens (503). As reported in Her2+ breast cancer patients treated with Trastuzumab and chemotherapy, the treatment-induced epitope spreading was characterized by increased antibody responses not only to the tumor antigen Her2, but also to endogenous CEA, insulin-like growth factor-binding protein 2 (IGFBP2), and p53 (521).
TIS is a very important protective mechanism that is induced immediately after chemo- or radiation therapy. TIS mediates the recognition and clearance of senescent tumor cells by immune cells (503, 535). Induction of TIS after the therapy is associated with a better prognosis and OS (524). However, if senescent tumor cells are not cleared in a timely fashion, their role at a later time points shifts from positive to negative, as discussed in the section below.

Negative Role of TIS: Cancer Progression
Several studies report a pro-tumorigenic effect of TIS leading to cancer recurrence and support of tumor development (503). Uncleared senescent cells acquire additional mutations, thereby escaping the cell cycle arrest and transform into malignant cells (536). Moreover, factors secreted by senescent cells are also reported to play a strong tumor-promoting role (526).
Was et al. suggested that senescent human colon cancer cells (HCT116) that appear during a doxorubicin-based therapy enter a “dormant” cellular state, survive the treatment, and cause tumor re-growth (537). Importantly, the recent findings by Scuric et al. suggest a long-term effect of chemotherapy and/or radiation exposure upon TIS (11). In this study, markers of cellular senescence, including higher DNA damage and lower telomerase activity, were observed many years later in breast cancer survivors (11). Elevated levels of a soluble tumor necrosis factor (TNF)-receptor-II, a pro-inflammatory biomarker and one of the main SASP molecules, were also detected (11). A negative effect of SASP was correlated to a p53 single-nucleotide polymorphism (SNP) at codon 72 which is correlated to increased risk of breast cancers (538). Using a humanized mouse model, Gunaratna et al. showed that SASP caused an increased invasion of pro-inflammatory macrophages (522). However, the inflammation proceeded into a chronic inflammation with pro-tumorigenic action and tumor-associated macrophages (TAMs) contributed to angiogenesis and increased tumor growth rates (522). Also, senescent cancer-associated fibroblasts (CAFs) and, in particular, expression of Caveolin-1 (CAV1) promote tumor invasion in pancreatic cancer (539). Moreover, in clinicopathological characteristics of patients, a high CAV1 expression directly correlates with higher levels of serum tumor antigens, with the rate of advanced tumor stage, and with significantly worse outcomes in both overall and disease-free survival (539).
It has been suggested that cancer therapies, especially chemo- and radiotherapies, possess long- and late-term pro-tumorigenic side effects and could therefore contribute to the relapse of the malignant disease they were intended to treat (540). Such long-term effects could be caused by the decreased removal of senescent cells, as described below.

Cancer Stemness: Senescence Escape
As mentioned above, cells undergoing senescence can still escape the senescence program and become malignant while acquiring additional mutations (519, 535, 536) (Figure 2). In our studies, we observed a spontaneous mutation [a deficiency in p19 (Arf)] in Ras-expressing hepatocytes, which resulted in a full-blown HCC development using a Ras-induced precancerous liver disease model (535, 536). The reversibility of TIS can be caused through the inactivation of tumor suppressors p53, p16 (Ink4A), p19 (Ink4d), and/or RB (504, 507, 519). Additionally, the over-expression of CDC2/CDK1 and survivin can promote cancer stem cell survival and can also promote the development of polyploidy (507). In general, mutations in CDKN2A, coding for p16 (Ink4a, CDKN2A), p21 (Waf1, CDKN1A), and p27 (Kip1, CDKN1B) as well as E2F3 and EZH2, and a high c-MYC expression might result in low percentages of senescent cells (504, 519). Moreover, particular mutations completely protect melanoma cells from cell cycle arrest upon chemotherapy: DMBC29 melanoma cells that carried a EZH2S412C mutation, expressed c-MYC at a low level and a wild type of CDKN2A did not undergo senescence, in contrast to many melanoma cells treated with vemurafenib and trametinib (519).
An escape of cells from senescence was also identified by Milanovic et al. in B-cell lymphoma studies (14). In those studies, the researchers showed that senescent cells substantially upregulated an adult tissue stem cell signature and activated Wnt signaling (14). This senescence-associated stemness was an unexpected cell-autonomous phenotype that caused the generation of cells with a higher tumorigenic potential in vitro (14).
However, escape from senescence is not the only pathway that promotes an increase in the cancer stemness phenotype. Stemness within the tumor tissue is also regulated indirectly by signaling molecules which support the maintenance of stemness in CSCs as well as non-CSCs, as described in the following sections.

Cancer Stemness: SASP and CSC Maintenance
The stemness phenotype within a tumor can also be mediated via SASP (526). Several studies address the strong pro-tumorigenic phenotype (526) whose cytokines can mediate the maintenance of CSCs. The most prominent interleukins (IL) of SASP are IL-1,−6, and−8 (526). These cytokines can influence the CSC phenotype and functionality and therefore influence the plasticity phenotype of CSCs.
Using breast cancer cell lines, Di et al. showed that an induction of senescence in mesenchymal stem cells by hydrogen peroxide treatment causes an increased secretion of the inflammatory cytokine IL-6, which led to a higher migratory capacity of breast cancer cells in vitro as well as in xenotransplants (541). An increase in the aggressive metastatic chemoresistant phenotype upon inflammatory cytokine stimulation such as IL-1ß, IL-6, and RANTES (regulated on activation, normal T cell expressed, and secreted) was also observed by others (533, 534). Our own work indicated that IL-8 blocks differentiation of hepatocellular premalignant cells, a pathway mediated via mammalian target of rapamycin complex 1 (mTORC1) kinase, that causes an increase in chemotherapy resistance (532). An increase in tumorigenicity and EMT of breast cancer cells has been shown to correlate to an increased expression of CD44 or CSC-like properties and be caused by the senescence-associated IL-8 and IL-6 (527–529). Pathways that might be involved in such cellular reprogramming processes are the JAK2/STAT3 signaling pathway (542), the IL-6/STAT3 and NOTCH cross-talk signaling (187, 530) as well NFκB-IL-6 signaling axis, responsible for the generation of CSCs (531). Interestingly, interference with those pathways by aspirin increased chemosensitivity and decreased self-renewal in breast cancer cells (531). In colorectal cancer cells the inflammatory cytokine IL-6 mediates deacetylation, which subsequently activates NANOG transcription and accumulation of stemness phenotypes, correlating with malignant progression and poor prognosis (543).
To summarize, TIS on the one hand has positive effects that eliminates differentiated tumor cells and also causes invasion of immune cells with anti-tumorigenic functions. On the other hand, senescence causes negative effects that are reflected by pro-tumorigenic functions causing CSC development and a gain of cancer stemness (Figure 2).
An additional level of complexity is added by the plasticity of CSCs as well as non-CSCs, which also causes increased cancer stemness, resistance, and relapse. Examples are given in the next paragraph.

Cancer Stemness: Plasticity of CSCs and Non-CSCs
Cancer stemness is not only triggered by senescence escape and acquisition of stemness phenotypes or supported by maintenance of stemness (544) but also by the plasticity of CSCs and non-CSCs, altogether causing tumor relapses after treatment, as described below.
Plasticity is regulated by the TME that is very heterogeneous and consists of CAFs, TAMs, and neutrophils as well as of cancer-associated adipocytes, tumor-infiltrating lymphocytes, and cancer cells with or without stem cell characteristics (545). Therefore, a clear separation between SASP effects and plasticity cannot be made as several direct and also indirect regulatory networks are involved (Figure 2).
Mechanistically, plasticity of cells is a characteristic that ensures robust tissue regeneration and homeostasis (546, 547) and describes the phenotypic and molecular changes of tumor cells increasing stemness and reflecting the tumor's ability to self-renew (18, 548). This phenotype is ultimately closely linked to EMT (15, 548). As described, the transition from the epithelial to mesenchymal state is associated with defined regulatory networks, chromatin remodeling and gene expression programs that are specific to the epithelial, mesenchymal or hybrid cellular state (15–18). Plasticity increases the complexity by suggesting that CSCs can switch between different cellular states, characterized by the expression of surface markers as well as transcription factors (18, 56). Examples for this come from the analysis of different tumor cells: Chaffer et al. demonstrated that CD44low cells (non-CSCs) can switch to a CD44high phenotype (CSCs) resulting in mammosphere formation, a phenotype that could be induced by upregulation of the zinc finger E-box binding homeobox 1 (ZEB1) protein expression induced by TGF-ß (548), which is a major cytokine of the TME (545). In NSCLC cell lines, two distinct CSC subpopulations have been described by expression of CD133 and the aldehyde dehydrogenase (ALDH) (549). ALDHs compose an enzyme superfamily with metabolic functions. The analysis of its activity is often used to identify CSCs (550, 551). Analyzing CD133 and ALDH activity, Akunuru et al. separated cancer stem/progenitor cells (CD133+, ALDHhigh) from non-CSCs (CD133− or ALDHlow) and showed that non-CSCs can interconvert into CSCs. The latter process is activated by TGF-ß signaling or signaling by the zinc finger protein SNAI (Snail) transcription factor family. The described process underlines the dynamic plasticity of CSC/non-CSCs cells (549). After TGF-ß treatment, the authors observed an increase in IL-1ß and IL-6 as well as an increase in CD133+ and ALDHhigh subpopulations (549).
Interferon-ß (IFN-ß) as well as Oncostatin M (OSM), also cytokines within the TME, have been shown to regulate CSC phenotypes (552). Activation of IFN-ß signaling pathways in non-CSCs blocks the expression of CD44 and Snail, which causes a decrease tumor sphere formation and additionally inhibits invasion (552). In contrast, OSM induces a stemness phenotype in non-CSCs (552). One of the major regulators of colorectal tumor plasticity (either CSCs or cancer cells) are the Wnt-ß-catenin and the KRAS/BRAF/ERK pathways, which have been implicated to regulate tumorsphere formation, self-renewal as well as resistance, as reviewed by Pereira et al. (553) and Zhan et al. (554). Activation of Wnt-signaling increased sphere and clone formation as well as drug resistance (555, 556). Acquisition of stemness was also described by Perekatt et al. using transgenic mice to analyze the function of Wnt-signaling in tumorigenesis and de-differentiation in the gut (28). The authors show that the inactivation of Smad 4, a factor that regulates the differentiation program, promoted the development of adenomas with characteristics of activated Wnt signaling over long-term periods (28). Such Wnt activation can correlate with increased treatment resistance as reviewed by Mohammed et al. (557). Also in gastric cancer, activation of the Wnt pathway causes an increase in CD44 as well as Oct-3/4 expression and correlates with an increased proliferation (558).
As described above, a gain of stemness due to SASP and CSC maintenance or by plasticity of CSCs and non-CSCs, can cause increased resistance (Figure 2). CSCs (pre-existing or post-therapeutically generated de novo) can escape the treatment by the expression of drug exporters and detoxification proteins, entrance into dormancy as well as resistance to DNA damage induced cell death (4, 15, 185, 559, 560). Their survival causes tumor relapses (Figure 2). To interfere with the relapse, several strategies have been under investigation to block CSC resistance and growth (9, 13), as described below (Figures 3, 4).
Figure 3  Targeted personalized second-line therapy as a future perspective. (A) Analysis of post-therapeutic biopsy samples: follow-up studies need to be included into regular clinical post-therapeutic relapse analysis. After therapy, local biopsies of remaining tumor tissue and/or satellite tissue should be taken periodically (even after several years post-therapy) and a multivariant analysis for biomarkers has to be performed, including the analysis of CSC biomarkers, pro-inflammatory cytokines, senescent markers as well as markers for CAFs. (B) Targeted second-line therapy needs to be performed based on the analysis described in (A) and will include a specific targeted eradication of remaining cells that could promote tumor relapse and metastasis. Targeted therapies comprise CAR-based approaches targeting CSCs as well as senescent cells or CAFs and TAMs. They also include senolytic drugs to deplete senescent cells independent of CAR approaches.
Figure 4  Targeted personalized first-line therapy as a future perspective. (A) Pre-therapeutic period: local biopsies before the therapy would allow to determine the heterogenic composition of the tumor, consisting of several biomarkers to be analyzed (CSC, CAFs, and TAMs biomarkers, tumor cell antigens, as well as e.g., T-cell compositions). (B) First-line targeted personalized therapeutic approach—therapeutic regimens could combine several approaches: the chemotherapy and small molecules (both selected based on tumor genotype), combined with immunotherapies (antibodies and checkpoint inhibitors based on tumor and analysis of T-cell phenotype), as well as CAR cell-based therapies targeting CSCs, CAFs, and TAMs. Combination therapy will allow a precise and efficient targeting of the heterogenic tumor composition from the beginning on.