The concept of cancer stem cells (CSCs) emerged from investigations into the origins of malignant tumor formation. Landmark studies on acute myeloid leukemia provided evidence of CSCs and their fundamental role in tumor initiation and progression (Lapidot et al., 1994[70]). Subsequently, CSCs were identified in solid tumors such as breast cancer (Al-Hajj et al., 2003[3]), glioblastoma (Singh et al., 2004[121]), and later, colon cancer (Ricci-Vitiani et al., 2007[111]). The identification of CSCs relies on biological and molecular characteristics associated with a stem cell phenotype, including the expression of specific clusters of differentiation (CD) markers, such as CD133. Remarkably, studies demonstrated that as few as 100 cells expressing CD133 were sufficient to initiate brain tumor growth, while cells lacking CD133 expression, even in quantities ranging from 50,000 to 100,000, failed to induce tumor growth (Singh et al., 2004[121]). Consistent findings were observed in studies involving xenografts of human bladder, colorectal, and pancreatic tumors (Baccelli and Trumpp, 2012[10]). In leukemia, the number of CSCs ranges from approximately 0.1 % to 1 %, while in breast cancer it is around 2 %. Similarly, colorectal cancer exhibits CSC proportions of 0.5 % to 1 %. In particular, in triple negative breast cancer, characterized by its high aggressiveness, the proportion of CSCs can exceed 10 % of tumor cells (Jagguppilli and Elkord, 2012[55]). Figure 2(Fig. 2) (References in Figure 2: Capp, 2019[20]; Lachat et al., 2021[67]) shows a timeline that highlights key milestones in the history of CSC research and its involvement in metastasis.

The discoveries mentioned above laid the foundation for the CSC hypothesis, which forms the core of the hierarchical model of cancer (Rich, 2016[112]). CSCs have become a focal point of intensive biomedical research aimed at comprehensively understanding their characteristics. These cells share numerous traits with normal stem cells, including the expression of various surface markers (Atashzar et al., 2020[8]; Walcher et al., 2020[139]). Despite the increasing knowledge about CSCs, their origin remains elusive. Recently, attention has shifted to the potential involvement of the dedifferentiation process, wherein mutant progenitor cells or mutant differentiated somatic cells transform into a cancer stem-like state (Senga and Grose, 2021[115]; Walcher et al., 2020[139]).

Like normal stem cells, CSCs possess the ability for self-renewal and unlimited proliferation (Batlle and Clever, 2017[12]; Walter et al., 2021[140]). However, CSCs exhibit genetic instability, such as mutations, and undergo numerous epigenetic changes, including the down- or upregulation of non-coding RNAs (ncRNAs) (Bryl et al., 2022[17]; Khan et al., 2019[62]). These alterations contribute to the high plasticity of CSCs, both metabolically and phenotypically, enabling them to undergo epithelial-mesenchymal transition (EMT) (Gupta et al., 2019[46]; Lachat et al., 2021[67]; Thankamony et al., 2020[131]). The strong association between CSC stemness and the ability to undergo EMT is corroborated by the expression of genes such as Snai1, Twist, Zeb1/2, and the activation of signaling pathways like Notch, Hedgehog, MAPK, PI3K/AKT, Wnt, as well as the HIF-1 (hypoxia-inducible factor 1) pathway (Lachat et al., 2021[67]).

Many of the hallmarks of cancer (Hanahan and Weinberg, 2000[50], 2011[49]; Hanahan, 2022[48]) can be attributed to CSCs. These cells can stimulate VEGF-dependent neoangiogenesis and actively participate in alternative tumor vascularization, known as vasculogenic mimicry (VM) (Knopik-Skrocka et al., 2017[64]). CSCs interact with the microenvironment through various mechanisms and evade immune surveillance (Tsuchiya and Shiota, 2021[134]; Xie et al., 2022[154]; Yang and Teng, 2023[156]). Consequently, CSCs exhibit high resistance to chemotherapy, radiotherapy, and immunotherapy. Therefore, innovative anticancer therapies should target CSCs.

Research on the origin of CSCs is strongly related to the elucidation of cancer initiation and progression. A stochastic model has prevailed for many years, assuming every cancer cell has tumorigenic potential. The discovery of CSCs resulted in a change in this model, and the so-called hierarchical model was proposed, stating that only CSCs can initiate tumor growth at both the primary and metastatic sites (Fanali et al., 2014[39]). To date, it has not been possible to determine the origin of CSCs. It is assumed that they may come from normal stem cells, progenitor cells, or differentiated somatic cells undergoing mutations, as well as from normal cancer cells (Figure 3(Fig. 3); References in Figure 3: Khan et al., 2019[62]; Walcher et al., 2020[139]). Regarding the high capacity of cancer cells to change their state, a dormancy process can be important as a protective mechanism against stress (Talukdar et al., 2019[127]).

With their long lifespan, stem cells are susceptible to accumulating mutations, potentially leading to cancerous transformation. CSCs are characterized by activating genes such as c-Myc, KLF-4, Sox-2, and OCT-4 (Villodre et al., 2019[137]), which may contribute to oncogenesis under favorable microenvironmental conditions. Notably, OCT-4 serves as a biomarker in seminomas, Sox-2 acts as a "key driver" in breast cancer and brain tumors, while KLF-4 is associated with colorectal cancer (Senga and Grose, 2021[115]).

The origin of CSCs from stem cells may be supported by their shared expression of surface markers typical of normal stem cells (Atashzar et al., 2020[8]). After undergoing prior mutation, progenitor cells can be reprogrammed into CSCs (see Figure 3(Fig. 3)). Notably, normal cancer cells can reversibly transition into a parent state, indicating that CSCs can arise from non-CSCs (Lambert and Weinberg, 2021[69]). Mutations and pro-inflammatory factors can facilitate this transition between parental and non-parental phenotypes, with the acquisition of the cell's ability to undergo EMT being a crucial determinant in this process (Khan et al., 2019[62]).

Moreover, differentiated somatic cells could serve as another source of CSCs (Khan et al., 2019[62]). The role of dedifferentiation in CSC formation is evidenced by studies conducted on glioblastoma, colorectal cancer, and pancreatic cancer. For instance, the dedifferentiation of normal intestinal epithelial cells may lead to the generation of tumor-initiating stem-like cells (Shih et al., 2001[119]). According to this hypothesis, CSCs in colorectal cancer may arise from a "top-down" process involving the dedifferentiation of a mature intestinal epithelial cell from the top of the crypt. Similarly, the ability of differentiated epithelial cells to dedifferentiate towards intraepithelial neoplasia has been observed in pancreatic follicular cells (Senga and Grose, 2021[115]).

As previously mentioned, CSCs represent a small fraction of the tumor cell population and exhibit some similarities to normal stem cells, posing challenges in their isolation and identification. Presently, CSCs are isolated by exploiting the affinity of antibodies against CSC surface markers or based on physical parameters, such as density or size (Babaei et al., 2021[9]). Methods like FACS (Fluorescence-Activated Cell Sorting) and MACS (Magnetic-Activated Cell Sorting), which utilize surface antigens, have proven highly effective. In fact, the initial isolation of CSCs was accomplished through FACS using antibodies and fluorescent markers (Lapidot et al., 1994[70]). Table 1(Tab. 1) (References in Table 1: Atashzar et al., 2020[8]; Kim et al., 2020[63]; Walcher et al., 2020[139]) illustrates the primary surface markers of CSCs.

One of the most known surface markers of CSCs is CD133, which is also present in normal hemopoietic and neural stem cells (Glumac and LeBeau, 2018[42]). CD133 is a poor prognostic factor (Park et al., 2019[97]); its overexpression correlates positively with the tumor stage. CD133 also mediates the drug resistance of CSCs. Studies in lung cancer showed that CSCs characterized by a CD133+ phenotype exhibited increased expression of some ATP-binding cassette transporters (ABC) and resistance to paclitaxel and platinum (Li et al., 2021[75]). Zhou et al. (2022[161]) showed that silencing CD133 expression in gastrointestinal adenocarcinoma cells increased the sensitivity of these cells to chemotherapeutics and inhibited CSC invasiveness.

CD44 is a receptor for many extracellular matrix components and a co-receptor for growth factors and cytokines (Li et al., 2021[75]). CD44 is profoundly involved in tumor initiation and progression. The ability of CD44 to receive signals from the microenvironment and transmit them to membrane-associated cytoskeletal proteins or the nucleus has been demonstrated, which affects gene expression and thus, cell behavior (Li et al., 2021[75]). CD44, like CD133, affects cell resistance. Two types of CD44+ and CD44- cells were identified in ovarian cancer studies. Only cells with CD44 expression resisted chemotherapy (Alvero et al., 2009[5]). Interestingly, CD44 can interact with hyaluronic acid, which leads to the activation of the Nanog and STAT-3 pathways and the expression of one of the ABC transporter, MDR1 (Price et al., 2018[105]). MDR1 is directly involved in cancer chemoresistance and is an essential marker of these cells due to high expression in CSCs (Muriithi et al., 2020[90]; Zhou et al., 2001[162]). Some ABC proteins can transport glutathione and thus, indirectly maintain redox homeostasis inside the cell. ABC transporters may thus protect CSCs from oxidative stress-related damage (Begicevic and Falasca, 2017[13]).

The results of the meta-analysis concluded that the presence of CD24 on the surface of CSCs is a factor associated with poor prognosis, metastasis, and shorter survival (Lee et al., 2009[73]; Yaghjyan et al., 2022[155]). Like CD44, CD24 with CD133 and CD166 are highly expressed in many types of cancer. In ovarian cancer, CD166 promotes invasion and metastasis and can inhibit apoptosis (Kim et al., 2020[63]). In lung cancer, high expression of CD44 and CD166 is correlated with high EpCAM expression, promoting metastasis (Walcher et al., 2020[139]).

Three models of tumor development were proposed - stochastic, hierarchical, and clonal evolution models (Cabrera et al., 2015[18]; Fanali et al., 2014[39]; Takebe and Ivy, 2010[126]). According to the stochastic model, all tumor cells have the same tumor-initiating activity, and the tumor mass is homogenous. The last two models assume the existence of CSCs and tumor heterogeneity. In the hierarchical model, only a small number of tumor cells (CSCs) show a capacity for tumor development and generate progeny cells, which lose tumorigenic potential (Lee et al., 2011[72]; O'Connor et al., 2014[95]; Rich, 2016[112]). The clonal evolution model indicates that CSCs can acquire selective oncogenic changes and then dominate other types of CSCs through active proliferation in a given tumor niche. Cells of the dominant subclones show similar tumorigenic potential (Nowell, 1976[94]). In each separate niche, such clonal expansions can occur independently (Takebe and Ivy, 2010[126]).

The lack of standardized, optimal study methods, unclear definitions, and nomenclature (Jordan, 2009[58]; Lathia, 2013[71]) seems to be the main problem in CSC studies. Different terms are used to describe cancer cells capable of self-renewal, such as stem-like tumor cells, tumor-initiating cells, or cancer stem cells (Lathia, 2013[71]). CSC best reflects the nature of these cells, which exhibit the ability to self-renewal and the capacity to recapitulate the parental tumor with cellular heterogeneity after transplantation to immunodeficient mice (Clarke et al., 2006[28]; Lathia, 2013[71]). However, the term cancer stem cell can be confusing (Jordan, 2009[58]). In light of data discussed in previous sections, CSCs can arise from stem cells and differentiated cancer cells forming a mass of tumors (non-CSCs). Studies on breast cancer (Chaffer et al., 2011[23]; Chaffer et al., 2013[24]) have revealed the acquisition of self-renewal capacity by non-CSCs. The switch to a CSC-like state depends on ZEB1, a transcription factor associated with EMT (Chaffer et al., 2013[24]). Interestingly, this conversion was mainly observed in basal, rather than in luminal type of breast cancer. It can explain the higher aggressiveness of the basal-like subtype of triple-negative breast cancer, compared to estrogen-dependent breast cancer (Dai et al., 2017[31]).

The reprogramming of non-CSCs to stem-like cells was also observed in glioblastoma ( Suvà et al., 2014[125]). Four transcription factors (POU3F2, SOX2, SALL2, OLIG2) introduced into differentiated glioblastoma cells induced stemness phenotype. It suggests that this state can be transient and reversible in CSCs and non-CSCs (O'Connor et al., 2014[95]). Hence, an alternative plasticity model of reversible cellular plasticity of cancer cells was proposed. This model focuses on a dynamic, not a static, CSCs model. Cancer stemness is related to the state of the cell rather than its type, and bidirectional interconversions between CSCs and non-CSCs are possible (Donnenberg et al., 2013[36]; Gupta et al., 2009[44]; O'Connor et al., 2014[95]; Rich, 2016[112]). Tumor microenvironment factors, like hypoxia, can be the driving force that reverts non-CSCs to a functional CSC state (Lathia, 2013[71]; Qin et al., 2017[106]).

A small number of CSCs within the tumor, raised in the hierarchical model, was questionable. The number of CSCs can vary significantly in different tumor types (Baccelli and Trumpp, 2012[10]). In some cancers, like melanoma, the cells responsible for tumor initiation were quite numerous (Quintana et al., 2008[108]). The frequency of CSCs can be increased during tumor progression and differ between grade 1 and grade 3 breast tumors (Pece et al. 2010[99]). The functional test is considered a standard in cancer cell stemness studies, however, its relevance is still debated (Lathia, 2013[71]). With this method, it is possible to measure the ability of transplanted cells to develop tumors in mice, but not the frequency of CSCs in tumors in situ (O'Connor et al., 2014[95]). This means that more precise tests are necessary to establish the number of cancer cells with tumorigenic potential. Recently, MutaSeq and mitoClone were used in acute myeloid leukemia to distinguish CSCs from non-CSCs and normal stem cells (Velten et al., 2021[135]). These tests are based on the genomic and mitochondrial mutations, not on CSC surface markers. Surface markers do not appear to be specific in distinguishing tumorigenic from non-tumorigenic cells (Lathia, 2013[71]; Quintana et al., 2010[107]; Shackleton et al., 2009[116]). CD133 is not exclusive to CSCs and can also be expressed in non-CSC populations (Li, 2013[76]). The studies of Shmelkov et al. (2008[120]) showed that CD133 expression in the colon is not restricted to CSCs, and CD133 is expressed on differentiated colonic epithelium. The use of surface markers does not allow for the tracking of dynamic changes, such as EMT, which is associated with acquiring stemness phenotype (Shibue and Weinberg, 2017[118]). Studies using reporter systems Oct4GFP and CatulinGFP help to observe the changing states of invasive cancer cells (Gielata et al., 2022[41]). RNAseq analysis of these cells revealed high expression of genes critical to cellular movement, invasion, and VM.

The CSC hypothesis is still being rebuilt, and it may be that the clonal evolution and hierarchical models together may explain what happens during tumor development. Genetically and epigenetically distinct cancer stem cell subclones, can derive from the cell with the first oncogenic mutation. Next, the subclones can give rise to more differentiated non-CSCs. Under tumor microenvironment influence, dynamic changes between stem (tumorigenic) and non-stem cell (non-tumorigenic) states may follow due to high cell plasticity and adaptation capabilities.

CSCs are characterized by high metabolic and phenotypic plasticity (Gupta et al., 2019[46]; Thankamony et al., 2020[131]). The concept of CSC plasticity has become a paradigm, guiding efforts to enhance our understanding of initiation processes, tumor progression, and the transition of cells into a dormant state under challenging conditions. This transition often results in resistance to treatment (Gupta et al., 2019[46]; Pastushenko et al., 2018[98]; Talukdar et al., 2019[127]; Thankamony et al., 2020[131]). The induction factor and driving force of CSC plasticity is hypoxia (Lachat et al., 2021[67]). Oxygen deficiency in the tumor leads to activation of HIF-1α and HIF-2α (hypoxia-inducible factors) (Hajizadeh et al., 2019[47]). In the normoxic state, the HIF-1α subunit is inactivated, while under hypoxic conditions, HIF-1α forms a complex with HIF-1β, acting as a transcription factor (Zhang et al., 2021[159]). HIF-1α preferentially increases the activity of genes associated with glycolysis, whereas HIF-2α activates genes mainly related to the cell cycle and stemness (Hajizadeh et al., 2019[47]). Table 2(Tab. 2) (References in Table 2: Chatterjee and Sil, 2019[25]; Clark and Palle, 2016[28]; Dong et al., 2016[35]; Gupta et al., 2019[46]; Hajizadeh et al., 2019[47]; Jin et al., 2020[56]; Sharma et al., 2022[117]; Sun et al., 2020[124]; Wang et al., 2021[143]; Wicks and Semenza, 2022[150]; Zhang et al., 2021[159]) shows some of the effects of hypoxia induced gene expression.

Recent years have shown that drugs such as salinomycin may contribute to the elimination of CSCs (Huczynski et al., 2016[54]). Salinomycin is an ionophore antibiotic isolated from the bacterial strain Streptomyces albus. The effects of salinomycin include interfering with the Na⁺/K⁺ ion balance, inhibiting the Wnt pathway, increasing caspase activity, activating the MAPKp38 pathway, and inhibiting NF-κB (Huczynski et al., 2016[54]). The main effect is the induction of cell apoptosis. Studies using the CD44^high/CD24^low breast cancer cell line revealed that the antibiotic is almost 100 times more effective in eliminating CSCs compared to drugs such as paclitaxel (Gupta et al., 2009[45]). Due to its molecule size, salinomycin is not subject to pumping out of the cell via the MDR1 protein, and actually inhibits this protein. Hence, salinomycin can induce apoptosis in those cells that have acquired resistance through high expression of MDR1 and anti-apoptotic proteins such as bcl-2 (Dewangan et al., 2017[33]). Regarding the crucial role of CSC interactions in the TME, CSCs-derived Exos seem to be a good target for innovative therapies (Yang and Teng, 2023[156]).

Some surface markers of CSCs may be used as a target for anti-CSC therapy (Eid et al., 2023[37]). Current attention is on anti-EpCAM antibodies (Catumaxomab) (Kubo et al., 2018[66]), or regulation of CD133 levels by the beta-2 isoform of PLC (PLC-β2) (Brugnoli et al., 2019[16]).

Surface antigens of cancer cells have become a tool of innovative CAR-T cell therapy. The groundwork for CAR-T (chimeric antigen receptor) therapy began in 1989 when the “two-chain” CAR receptor antibody TCRaV H + TCRbV L was first obtained in vitro, followed in 1993 by the first cancer-specific CARs demonstrating in vitro cytotoxicity (Eshhar, 2014[38]; Styczynski, 2020[123]). These findings are the basis of the development and subsequent use of the first CAR-T cell drugs targeting CD19, present on the surface of acute lymphoblastic leukemia and lymphoma cells. Currently, markers present on the surface of solid tumor cells, like EGFR, Her-2, Muc16, CEA, or TAA, are tested in preclinical and clinical trials (Marofi et al., 2021[87]; Masoumi et al., 2021[89]; Veschi et al., 2023[136]).

The problem in developing an optimal CAR-T construct that acts selectively on CSCs is that some CSC markers are also present on the surface of normal stem cells. However, the results of phase I clinical trial NCT02541370 (Wang et al., 2018[145]) demonstrated the positive effect of CD-133⁺-CAR-T in patients with hepatocellular carcinoma, and metastasis did not develop. In thyroid cancer studies, the phosphorylation in CD133⁺CSCs promotes their self-renewal and immune escape via induction of PD-L1 expression (Wang et al., 2020[146]). High expression of PD-L1 interacting with PD-1 on Tcyt can induce anergy of both CAR-T cells and non-genetic modified T cells (Johnson et al., 2022[57]). Targeting only immune checkpoints, including PD-L1, is insufficient to achieve the full efficacy of anti-tumor therapy. Combination therapy using immune checkpoint inhibitors with chemotherapy (Grodzka et al., 2023[43]) or with CAR-T cells (Dianat-Moghadam et al., 2022[34]) is recommended.

The immunosuppressive microenvironment and tumor cells' high plasticity can be significant obstacles to CAR-T cell effectiveness (Table 4(Tab. 4); References in Table 4: Cheever et al., 2022[26]; Masoumi et al., 2021[89]) (Masoumi et al., 2021[89]). Increasing TAA expression and using bispecific CAR-T can be helpful (Catamero et al., 2022[22]). Many clinical trials are being conducted with this aim, among them is a phase I/II clinical trial (NCT04077866) with an intratumorally administered B7-H3-CAR-T construct in patients with glioblastoma multiforme. A phase I clinical trial, NCT04227275, with CAR-T-PSMA-TGFβRDN cells, is dedicated to prostate cancer patients (Kankeu Fonkoua et al., 2022[61]). These CAR-T cells have information about the prostate-specific membrane protein antigen (PSMA) and TGFβRDN. As a result, the block of TGFβ immunosuppressive effects is observed. This action may contribute to “bypassing” the immunosuppressive effects of the microenvironment (Cheever et al., 2022[26]).

Combining CAR-T cells with epigenetic reprogramming, epi-immunotherapy is now considered a promising research direction (Veschi et al., 2023[136]). Such compounds as DNMT inhibitors, BET, HDAC, EZH, and SIRT-1 agonists, among others, have received attention. DNMTi modulates macrophage activity and promotes Tcyt activity, and BET reduces PD-L1 expression. Acute myeloid leukemia studies suggest that DNMTi may also increase TAA expression on the surface of tumor cells, which may provide greater CAR-T cell efficacy in target recognition (Veschi et al., 2023[136]). A similar effect was described by Kailayangiri and colleagues (2019[60]) after using CAR-T cells specific for GD2, a marker of Ewing's sarcoma tumor cells. Regarding CAR-T cell effectiveness, epi-immunotherapy opens possibilities for epigenetic reprogramming with modifiers acting via miRNAs at the DNA level (Alvanou et al., 2023[4]; Li et al., 2016[74]). In addition to miRNA-based drugs targeted to cancer cells (Bryl et al., 2022[17]), ncRNAs, like miR-28, can be used to protect T cells from anergy by silencing PD-1 and regulating cytokine secretion. miRNA138 can silence PD-1, as well as CTLA-4 (Akbari et al., 2021[2]). Another example is miRNA155, whose expression in CAR-T cells increases the cytolytic activity of CD19-CAR-T cells (Zhang et al., 2022[158]).

Much research over the last few decades has led to the breakthrough discovery of CSCs, their biology, and their role in tumor initiation, development, metastasis, as well as cancer resistance to treatment. However, the origin of cancer stem-like cells is still unclear, and the CSC model is still in debate. The lack of consensus on reliable CSC markers and definition complicates translating the study results into clinical applications. For this reason, further research on CSC and cancer cell heterogeneity should be a priority. The most promising therapeutic approach is the development of combination therapies in which a cancer stem-like cell-targeted strategy could be used with chemotherapy, radiation therapy, and/or immunotherapy. There are high hopes for strategies based on ncRNAs or genetically modified T cells. However, researchers and clinicians may face many limitations and challenges regarding the complex biology and dynamic changes of cancer cells and within the tumor microenvironment.