2.1 Cellular Origin of CSCs

As the seed of a tumor, the source of CSCs has attracted long-term attention. Based on the previous research results, CSC can come from at least one of the following cell types: tissue stem cells, progenitor cells and mature cells [12]. In established tumor, CSCs can also be derived from differentiated tumor cells through plasticity.

2.2 Self-Renewal and Differentiation of CSCs

Self-renewal is a characteristic of cell populations that enables them to maintain their own stability, and different stem cells maintain self-renewal through different modes of cell division. Here, symmetric division of stem cells produces two stem cells or two non-stem cells, while asymmetric division produces one stem cell and one non-stem cell. A large number of lineage tracing results showed that LGR5⁺ ISCs mainly self-renewal through symmetric division, and the maintenance of the overall number of stem cells is determined by the specific microenvironment [33, 34]. CSCs and tissue stem cells have similar signaling pathways involved in self-renewal, such as the Wnt/β-catenin, Notch, and sonic hedgehog (Hh) pathways. However, CSCs may have different cell division mechanisms with tissue stem cells, the self-renewal of normal gastric tissue mainly depends on asymmetric division, which means that the two daughter cells are positioned differently relative to the basement membrane. In contrast, gastric stem cells change to symmetric division to generate more stem cells, during tumorigenesis. This process is influenced by intracellular factors such as SOX9, coregulation of tumor-inducing factors such as N-nitroso-methylurea, and microenvironmental factors such as gastrin generated by G cells [35, 36]. Nevertheless, the asymmetric division also exists in CSCs, which are considered an intrinsic property of CSCs due to the asymmetric distribution of their microenvironment [37]. We will elaborate on the intracellular and microenvironmental factors that regulate self-renewal of CSCs in subsequent sections.

Tumors harbor multiple cell types, and in addition to their self-renewal capacity, CSCs have the ability to differentiate into different cell types within tumors, and stemness of tumor is significantly positively correlated with heterogeneity [38, 39]. The multidirectional differentiation capacity of CSCs has been confirmed by multiple lineage tracing experiments in contrast to CD9^low cells that produce only CK19⁺ cells, CD9^high CSCs produce both CK19⁻ and CK19⁺ cells, indicating that CD9^high CSCs have multiple lineage differentiation capacity [40]. Transplanted CD133⁺ liver CSCs produce cells with different differentiated lineages, indicating that Prom1/CD133⁺ CSCs have differentiation or transdifferentiation potential [41]. LGR5⁺ CSCs in CRC are able to produce both LGR5⁺ cells and all adenoma cells, including Paneth cells, indicating that LGR5⁺ CSCs possess the ability of self-renewal and differentiation [42]. In addition to differentiating into tumor cells, CSCs are also able to transdifferentiate into other cell types to regulate tumorigenesis [43]. Transdifferentiation of CSCs into vascular endothelial cells (ECs) and promoting angiogenesis has been found in a variety of cancers, such as liver cancer [44, 45]. Unexpectedly, the CSCs of solid tumors also exhibit hematopoietic properties and are one of the sources of blood cells in tumors [46]. The multidirectional differentiation capacity of CSCs generates and maintains the diversification of cancer cells, playing a key role in shaping tumor composition, disease progression, and therapeutic resistance.

2.3 Heterogeneity of CSCs

Tumors are a mixture of cells with different molecular characteristics and medicine sensitivities, and the difference among different cells within a tumor is called intratumoral heterogeneity. Heterogeneity is the key cause of tumor drug resistance, recurrence, and difficulty in cure, which has a profound impact on tumor prevention, diagnosis, and treatment. Heterogeneity originates from the emergence and accumulation of events including genetic mutations and epigenetic modification's changes in different cells, as well as differences in the tumor microenvironment (TME). Increasing evidence shows that tumor heterogeneity is driven by CSCs [47]. In fact, the existence of CSCs is an important manifestation of tumor heterogeneity. Li and colleagues, using single-cell RNA sequencing and spatial transcriptomic analyses between colorectal cancer liver metastasis (metastases in the liver, LCRC) and ovarian metastasis (metastases in the ovary, OCRC), revealed the role of CSCs in heterogeneity of metastatic tumors [48].

Recently, an increasing number of studies have found that CSCs themselves are also heterogeneous. Single-cell omics, such as the emergence of single-cell transcriptome sequencing and single-cell ATAC sequencing, provide an ideal strategy for defining the heterogeneity of tumors and CSCs. Two studies utilized single-cell transcriptomic analysis of HCC formed by cell lines, freshly excised HCC samples, and HCC PDX to investigate CSC heterogeneity in HCC, and identified heterogeneity of HCC stem cells, with CSC subsets expressing different markers having different molecular characteristics and prognostic correlations [49, 50]. In the process of CRC development and treatment, CSCs and their microenvironment undergo dynamic changes, resulting in two types of CSCs in CRC, proliferating CSCs and resurrecting CSCs, which have significant differences in the regulatory levels of intracellular and microenvironmental factors [51, 52]. In addition, some new types of CSCs have been defined in digestive system tumors, which are described in detail below.

2.4 Plasticity of CSCs

The conventional view is that CSCs are a type of rare subset of cells with expression of specific markers. However, an increasing number of studies have found that CSCs are not constant, and under specific circumstances or stimuli, differentiated tumor cells are able to de-differentiate into stem-like cells, which indicates that CSCs present a state rather than a fixed condition [79]. The latest version of “Hallmarks of Cancer: New Dimensions” is proposed, and the “unlocking phenotypic plasticity” is the emerging hallmark of tumor, and the plasticity of CSCs is a significant embodiment of tumor plasticity. Plasticity is a key component of cancer pathogenesis and can manifest in various ways, including differentiation, dedifferentiation and transdifferentiation. It is also an important trigger for tumor metastasis, recurrence and drug resistance [80]. For example, the plastic transition of intestinal cancer cells to intestinal CSCs reactivates the Wnt pathway, promoting the clonogenic ability and metastasis of tumors [20, 81].

The plasticity of CSCs may originate from normal stem cells. CSCs and normal stem cells share the same surface marker, and both exhibit significant plasticity. Similar to normal stem cells, the loss of ISCs leads to various types of cells dedifferentiating into stem cells to replenish the stem cell pool, including absorptive progenitor cells, secretory progenitor cells, and Paneth cells [82, 83]. Correspondingly, when LGR5⁺ CSCs are cleared using DT, LGR5⁻ cells can dedifferentiate to form new CSCs, which subsequently establish a new tumor hierarchy and further drive tumor drug resistance and relapse. Moreover, similar mechanism is observed in the initiation process in which DSS and inflammation induced a depletion of LGR5⁺ cells, as well as in metastasis process in which the conservation from LGR5⁻ to LGR5⁺ cells plays an essential role in the expansion of metastatic loci [49, 50]. Plasticity plays a key role in the repair process following tissue homeostasis maintenance and homeostasis dysregulation, and accordingly, tumor cells increase their ability to resist multiple environmental perturbations and targeted therapies through the inheritance of normal tissue plasticity [38].

The plasticity of tumors and CSCs is regulated by both intracellular and microenvironmental factors, reflecting the ability of tumor cells to transition between stem cell and differentiated states. Similar to tissue stem cells, the chromatin structure and epigenetic similarities among tumor cells are prerequisites for cell transdifferentiation and plasticity [84]. Under environmental stress (such as tumor occurrence, tumor metastasis, tumor treatment, nutrient deficiency, etc.), tumor cells and CSCs exhibit different cellular and molecular characteristics through chromatin and epigenetic remodeling, thereby offsetting these stresses and showing a high ability to adapt to the environment. Tumor develop with sustained inflammatory responses such as IFN-γ and regenerative pathway YAP1 activation, which generally induces a reduction or differentiation of LGR5⁺ cells and emerging of fetal-like cells [85]. Accordingly, Vasquez and colleagues demonstrated that the stemness of CRC is attributed to LGR5⁺ crypt base columnar cell (CBCs) and LGR5⁻ RSCs. Stemness of LGR5⁺ CBCs is enriched by APC mutation, whereas LGR5⁻ RSCs stemness is enhanced by IFN-γ, TGF-β or Kras or YAP activation induced by Kras/Braf mutation (Figure 1A) [25]. Combining scRNA-seq and in vitro organoid formation with various mutant ISCs or cocultured with various niche cells or niche factors, Qin and colleagues also reveal the continuous polarization of CSCs between proliferative colonic stem cells and revival stem cells, which are driven by niche fibroblasts or cell-intrinsic mutation such as APC or Kras (Figure 1B) [86]. IkBa deletion-induced NF-kB activation or Kras mutation trigger hyperactivation of Wnt/β-catenin in APC-mutant background, driving the dedifferentiation of non-CSCs to CSCs (Figure 1C) [20]. Some differentiated cells, such as Paneth cells, can dedifferentiate in an inflammatory environment (for instance DSS treatment) to form revival stem cells to compensate the loss of LGR5⁺ cells, which in turn drive intestinal carcinogenesis. About 25% of CRC patients suffer from this disease, which is much higher than IBD-type CRC (accounting for 1%–2%) (Figure 1D) [19]. During metastasis of CRC, metastatic cells leave the primary LGR5⁻ cells spread in the blood, and experience the fate transformation from LGR5^- cells to LGR5⁺ cells in the destination of tumor metastasis, indicating that cancer cells are able to transit between stemness and non-stemness states (Figure 1E) [26]. Chemotherapy is one of the most deeply studied predisposing factors of cellular plasticity. Sublethal doses (IC20-IC30) of 5-FU⁺ Iri treatment induce colorectal cells into persistent quiescent-like (PQL) state in P53- and YAP1-dependent manners, and this reprogramming enhances fetal-ISC-like characteristics and tumor initiating capacity [87]. Zapatero and colleagues analyzed drug responses in patient-derived organoids and revealed cancer-associated fibroblasts (CAFs)-mediated remodeling of CSCs during drug treatment from proliferative colon stem cells to chemotherapy-resistant resuscitating stem cells (Figure 1F) [51]. A similar phenomenon was found during breast cancer treatment, NF-kB-activated CAFs boost CSCs enrichment and chemoresistance via IL-6 and IL-8 [88]. The IFN-γ produced by immunotherapy can directly deliver non-CSC to CSCs through branched-chain amino acid transaminase1 (BCAT-1), resulting in the enrichment of CSCs after immunotherapy. On this basis, a combination treatment of BCAT-1 inhibitor and immunotherapy could inhibit the increase of CSCs after immunotherapy and ultimately promoted the tumor treatment effect (Figure 1G) [89]. In addition, type I interferons produced by tumor immunotherapy can promote the number and function of CSCs through KDM1B-mediated chromatin remodeling [90]. In the absence of ideal culture conditions, a population of MEX3A⁺ CSCs with slow proliferation and chemotherapy resistance emerges in tumor organoids. Moreover, the conversion from MEX3A^- LGR5⁺ cells to MEX3A⁺ LGR5⁺ cells in suboptimal conditions drives the acquirement of revival fetal-like state, also indicating that the regulation of CSCs plasticity mediated by environmental adverse factors promotes the adaptability of tumor cells to the harsh environment [69]. Upon isolation-induced stress among tumor initiation and metastatic seeding, therapeutic pressures induced by massive death of vulnerable cells, lacking of oxygenation and nutrients, pancreatic cancer cells were able to promote LPAR4 expression through downregulating miR-139-5p, further promoting fibronectin production through the LPAR4/AKT/CREB pathway, and finally enabling cells to cope with isolation stress by remodeling their own microenvironment and enabling LPAR4⁻ cells to acquire stress tolerance [91]. In conclusion, environmental and intracellular factors jointly drive the plasticity of CSCs, which adapt to the intracellular state and extracellular environment, thereby acquiring the ability of self-renewal, drug resistance, proliferation, and metastasis.

Recent cancer stemness theory holds the idea that CSCs are a mimicry adopted by cells to adapt to environmental changes when signaling pathways are destroyed by environmental factors such as therapeutic stress, including typical viral-like mimicry, vascular mimicry (VM), and immune cell mimicry [92]. In fact, CSCs themselves can be regarded as the ability of tumor cells to mimic stem cells, and thus have the intrinsic properties of stem cells, including increased autophagic activity, reduced iron death activity, induced dormancy, low expression of major histocompatibility complex (MHC), upregulation of immune evasion molecules, dysregulation of costimulatory signaling, and so on [68, 93-98].

2.5 CSCs and Tumorigenesis

CSCs are often referred to as the seeds of tumorigenesis, serving as the initial cells that give rise to tumors. Based on this, the “cancer stem cell” theory of tumorigenesis has been established. Together with the tumor mutation theory, it has become an important theory of tumorigenesis. Based on this concept, the “cancer stem cell” theory of tumorigenesis has been developed, standing alongside the tumor mutation theory as a major framework for understanding how tumors form. While the tumor mutation theory emphasizes the molecular timeline of tumorigenesis, the CSC theory focuses on the cellular events that drive this process. Recent lineage tracing studies have shown that only APC mutations in ISCs can lead to the formation of adenocarcinoma, whereas mutations in differentiated cells do not result in tumor formation [14]. The primary reason for this may be that these differentiated cells typically have a lifespan of only 3–5 days, which is insufficient for tumor development [99]. Although some recent studies have found that tumor cells can be derived from non-stem cells, such as papillary epithelial cells and Paneth cells, all these cells are transformed into long-term cells such as LGR5⁺ cells or proliferating stem cells, indicating the requirement of long-term mutant cells for tumorigenesis [19, 20]. Cancer stem cell-driven tumorigenesis is a complex process, involving the dynamic changes of various intracellular and microenvironments and the regulation of cell status, including the acquisition of competitive advantages of cancerous cells, the escape of immune surveillance, and the formation of immunosuppressive microenvironment. The detailed regulatory mechanism will be discussed in parts 3–5.

2.6 Therapeutic Resistance of CSCs

Drug resistance and recurrence after tumor treatment are almost inevitable, occurring universally across chemotherapy, radiotherapy, targeted therapy and immunotherapy. Under therapeutic stress, CSCs can evade treatment-induced cell death through various pathways [100, 101]. Additionally, CSCs possess strong tumor-initiating capabilities, allowing a small number of surviving CSCs to rebuild treatment-resistant tumors within just a few days to months. This makes them a key factor in tumor resistance and recurrence. Even after long-term treatment without any visible tumor, some called persister cells persist and acquire proliferative capacity, making the tumor increasingly resistant, and these persister cells prove to be a specific cancer stem cell type or CSCs state [102].

2.6.1 Chemotherapy Resistance

CSCs highly express the ABC transporter, such as ABCG2, with a stronger drug efflux capacity, by protecting genomic stability and inducing drug resistance (Figure 2A) [103, 104]. In fact, ABCG2 has been regarded as one of the classical markers of CSCs [105]. Due to the high expression of drug resistance molecules such as ABCG2 and the drug resistance of CSCs, researchers have also established a classic research system for CSCs, termed “side population” (SP) cells, a subpopulation of cells that resist chemotherapy in digest system tumors [106, 107]. On the other hand, the existing chemotherapeutic drugs mainly target rapidly proliferating cells, while CSCs are relatively resting and can escape the killing of chemotherapeutic drugs, and the resting nature of CSCs is widely considered the source of tumor drug resistance. Recently, resting LGR5⁺p27⁺ CSCs were found to resist chemotherapy and promote tumor recurrence in CRC (Figure 2B) [64]. Similarly, LGR5⁺p57⁺ slowly cycling CSCs was also in the resting state and promoted tumor resistance and recurrence, whereas DT treatment for deletion of p57-DTR cells leads to impaired therapy resistance [108]. Due to the drug-resistance properties of CSCs, drug treatment will be accompanied by the enrichment of CSCs [109].

2.7 The Role of CSCs in Tumor Metastasis

Tumor metastasis is a complex process that involves the invasion of primary tumor cells, their spread to distant sites, seeding at these locations, clonal expansion at the distant site, and the transitions between different tumor cell states, such as epithelial-mesenchymal transition (EMT) and mesenchymal-epithelial transition (MET) [121]. CSCs have multiple cellular differentiation capacity and high plasticity, can shuttle between different cell states to meet the needs of the metastatic process, for example, CD26⁺ CD44v6⁺ CSCs has stronger metastatic potential [81, 122]. The animal model data also demonstrated that the clearance of LGR5⁺ CSCs was able to inhibit intestinal cancer metastasis, confirming the essential role of LGR5⁺ CSCs in CRC metastasis [26, 27, 123].

Molecularly, the genes driving tumor stemness and tumor metastasis interact to a large extent, leading to the stronger tumor metastasis ability of CSCs. First, certain gene or stimulation regulates both tumor metastasis and tumor stemness [124]. As one of the most important transcription factors for tumor metastasis, ZEB1 regulates the self-renewal of CSCs in multiple tumor types [125]. In turn, multiple genes specifically expressed in CSCs also have important regulatory effects on tumor metastasis, such as MAFF, cis-HOX, mcPGK1, rtcisE2F, and so forth [126-129]. Second, the genes controlling stemness and transfer have the same upstream regulators [130]. Sublethal chemotherapy-induced YAP1-dependent fetal reprograming drives tumor initiating and metastasis [87]. These cross-coupling of stemness and metastasis-related pathways drive the highly metastatic signature of CSCs.

Although the tumor stemness pathways promote tumor metastasis in various ways, it is not true that CSCs themselves establish tumor metastasis because CSCs can't cover every step in complicated cascade progress during metastasis. Metastasis-initiating cells (MICs) seeded as transfer are able to undergo a drastic phenotypic transition, adapt to the transcriptional cascade and the static metastasis [131]. Some studies have shown that metastatic stem cells are special types of CSCs with the same gene expression characteristics and markers, but they may be different cell types. Using emerging technologies such as lineage tracing and in situ imaging in vivo, Rheenen found that LGR5⁻ cells are more diffuse compared with LGR5⁺ CSCs, and the tumor cells in blood are also LGR5⁻ cells, indicating the inconsistency of MICs and LGR5⁺ CSCs [26]. In bowel cancer metastases, the surface markers of disseminated cells are mainly L1CAM and EMP1, instead of the marker LGR5 of the CSCs [132, 133]. In fact, in essence, distant tumor colonization and expansion are also a tumor initiation process, which also requires the tumor initiation ability of CSCs, so they need to reacquire the CSCs characteristics [26, 134].

2.8 Biomarkers for CSCs

The concept of CSCs emerged since 19 century, through an intensive investigation of hematological malignancy stem cells, when Dick and colleagues, among others, confirmed the tumor cell heterogeneity [135], and the first identification of CD34⁺CD38⁻ CSCs in acute myeloid leukemia (AML), marked a major breakthrough in CSC's research [136]. These findings in hematologic malignancies provide a paradigm for research of CSCs: whether a specific cell is a CSC can be evaluated by sorting based on specific markers and tumor initiation experiments. In fact, the markers of CSCs can be divided into cell surface markers and intracellular markers, and have received more attention because they facilitate cell type identification and sorting, and hold more potential in clinical diagnosis and treatment [137].

Using the fluorescence-activated cell sorting/magnetic-activated cell separation (FACS/MACS) and gradient dilution tumor formation experiments, many subsequent studies have defined the surface markers of CSCs in digestive system tumors. In liver cancer, CD133, CD13, CD90, and EpCAM are a common surface marker of liver CSCs, and CD133⁺CD13⁺ CSCs have a stronger tumor-initiating capacity [138-142]. For colorectal cancer, CD133 is an excellent marker for human cancer but not for mouse tumor [143, 144]. CD44 is another marker for colorectal CSCs which contributes to cancer metastasis [81]. Interestingly, it seems that CD133 and CD44 are universal markers for digestive system tumors, as they also serve as surface markers of gastric cancer, esophageal cancer, and pancreatic cancer [145-149]. The similarity of CSC markers between various tumor types indicates the shared signaling pathways and expression landscapes between these tumors.

Due to the role of CSCs in tumor therapy resistance, in addition to conventional surface molecules such as CD44 and CD133, attention has been paid to factors related to clinical features, such as drug pump molecules directly related to drug resistance, and immune checkpoint factors related to tumor immunity. ABCG2 is a key drug transporter molecule that has been identified as a surface marker for a variety of tumors, including pancreatic cancer, liver cancer, intestinal cancer and other digestive system tumors [150]. PD-L1 is an important receptor protein for tumor escape immune killing in intestinal cancer, liver cancer, and gastric cancer. The high expression of ABCG2 and PD-L1 is closely related to the tolerance of CSCs to chemotherapy and immunotherapy [151-153].

Identification of CSCs markers is not only fundamental for the identification of CSCs, but also provides the possibility for specific targeting of CSCs, such as antibodies targeting α2δ1 that can target HCC stem cells [154]. However, there are great similarities between existing CSCs markers and normal stem cells, such as CD133, which is highly expressed in normal ISCs [155]. Therefore, a comprehensive understanding of the differences between CSCs, non-stem cells, and normal stem cells is of great significance [156]. We will address this issue in more detail in Section 5.

2.9 Methods for CSC Investigation

As we know, there are multiple methods to evaluate certain specific biological characteristics of CSCs, such as “sphere formation” for self-renewal, “lateral population cells” for resistance, “trans-well” for metastasis. However, “gold standard” assay is FACS and xenotransplantation [80]. For example, the earliest identified CD34⁺ CD38⁺ CSC were able to engraft into immunodeficient NOD/SCID mice to form heterogeneous tumors [135, 136]. Although this approach advances our understanding of CSCs, the method has some technical and scientific problems that may obscure the accurate understanding of CSCs, including (1) variations such as xenograft sites, dilution reagent and recipient mice leading to conflicting results; (2) this assay actually detect the ability of tumor cells to survive in harsh processing steps and established tumors in recipient mice; (3) the cell–cell and cell–niche interactions are ignored and CSC niche are totally different from the naive niche.

Genetic markers, lineage tracing, and elimination become key technologies for the identification of CSCs, allowing the self-renewal and differentiation capacity of CSCs, both of which are important features of stemness. Through intraperitoneal injection or oral administration, Hans Clevers team achieved fine control of Cre activity in Ahcre; APC^flox/flox and found transformation of non-stem cell rarely drives neoplasia, which further proved that APC mutation in LGR5⁺ cells was the source of tumorigenesis [14]. CD133^-Cre-LacZ; Rosa26^-YFP mouse model proved that CD133⁺ stem cells could form all neoplastic cells in the tumor and retain a subpopulation of CD133⁺ cells, indicating that CD133 is also the marker of intestinal CSCs [144]. Of note, CD133 and LGR5 have similar tissue localization, both located at the base of crypts. In 2012, the Hans Clevers team used the “lineage retracing” technology to prove that the intestinal LGR5⁺ cells in mouse intestinal adenomas possess the ability of self-renewal (generating other LGR5⁺ cells) and differentiation (generating all other adenoma cell types), which firstly demonstrated the self-renewal and differentiation characteristics of CSCs in an in situ environment [42, 157]. On this basis, several studies utilize suicide genes such as DTR or thymidine kinase to deplete specific population cells and then determine the importance of this group of cells in tumorigenesis. For instance, iCaspas9-mediated LGR5⁺ cell ablation leads to tumor regression and subsequent reversion from KRT20⁺ cells [28]. Similarly, depletion of CSCs or CSC subsets reduces tumorigenesis and progress, such as Dclk1 in intestine tumor, LGR5 in gastric cancer, Prom1 in liver cancer [41, 123, 158]. To date, lineage tracing and lineage clearance have become the core technology for the study of CSCs maintaining the naive niche of tumor.

To detect CSCs behavior and characteristics in intact tumors, lineage tracing has been widely used for CSCs research. However, the technical requirements for lineage tracing are relatively high, particularly since it can only be performed on animals such as mice, which limits the study of human CSCs. In recent years, emerging organoid technology has addressed this limitation, becoming a key method for studying human CSCs and also serving as a major approach for mouse CSC research [28, 64]. Advantageously, tumors of the digestive tract and digestive system such as esophagus, stomach, and intestine can easily form organoids [159]. Significant discoveries have been made in the field of CSCs through the organoid research system, such as the resistance to chemotherapy of p57⁺ resting CSCs in intestinal cancer, effect of LGR5⁻ cells on intestinal cancer metastasis, effect of AQP5⁺ CSCs on gastric cancer formation, and role of MSI2⁺ pancreatic ductal adenocarcinoma on tumor initiation [15, 26, 108, 160, 161]. Of note, organoids are capable of performing functional studies of single factors in vitro, for example, investigating the regulatory effects of specific signaling pathways or genes on stem cell self-renewal and differentiation [99]. For example, Tape group utilized organoid or tumor treatment approaches to analyze the impact of various microenvironmental cells and factors on the cellular composition and states of tumor organoids during tumorigenesis, and found dynamic changes in proCSC and revCSC during tumorigenesis and tumor treatment [51, 86]. Along with multidisciplinary and standardized next-generation cancer organoids, organoids will play an increasingly important role in cancer and CSCs [162-164].

Traditional research is very focused on specific surface markers, and more and more people study CSCs as a state with CSC functionalities and molecular characteristics, so it is very useful to study the characteristics of CSCs through molecular profiling. Stem cell characteristics and stemness indices can be obtained by machine-learning methods or through the established markers from related stem cells [25, 165]. In recent years, single-cell sequencing technology has become a great tool for cancer stem cell omics research, which can realize the identification and functional characterization of subsets of CSCs, and largely avoid the inconsistent results caused by different single markers. For example, PTPRO/ASCL2 overexpressing CSC subset drives metastasis of CRC cells, and-specific subsets are also able to be associated with organ-specific metastasis of bowel cancer, LEPR⁺ CSCs interact with the microenvironment to drive squamous cell carcinomas, colorectal adenomas are generated from stem cells but serrated polyps are not [48, 166-168]. The development of single-cell sequencing technologies and the enrichment of analytical measures, such as RNA velocity, CytoTRACE, CellChat and CellPhoneDB, single-cell sequencing will surely deepen our understanding of CSCs [169-172].

3.1 Signaling Pathways in CSCs

The unique stem characteristics of CSCs are determined by various specific signaling pathways within the cells. The number and function of CSCs are precisely regulated by multiple signaling pathways, among which the Wnt/β-catenin, Notch and Hh signaling pathways are the most central. Recently, the importance of Hippo/Yap, PI3K/AKT, NF-κB, and JAK/STAT pathways in CSCs has also received increasing attention [207].

Wnt signaling is a conserved and complex pathway, consisting of 19 Wnt ligands and more than 15 receptors [208]. An elegant study by Hans Clevers' team found that Wnt3a is mainly distributed on the cell membrane of intestinal crypt stem cells, binds to Fzd molecules on the stem cell surface, travels away from its source in a cell-bound manner through cell division rather than through diffusion, since cell division is an important factor in Wnt diffusion [209]. Wnt signaling pathway can be divided into the β-catenin-centered canonical Wnt signal transduction and the noncanonical Wnt signal transduction involving downstream effectors such as PCP/RTK/Ca2⁺ [210]. Aberrant Wnt signaling activation is present in almost all cancers of digestive system and CSCs [211, 212]. The Wnt signaling can be activated in multiple ways, including constitutive activation due to gene mutation and dynamic activation caused by molecular network, epigenetics and microenvironment. The Wnt component mutations include APC mutation in CRC, β-catenin mutation in stomach and liver cancer and AXIN mutations in gastrointestinal cancers [213-216]. In terms of genetic regulation within the signaling network, activation of Wnt/β-catenin can be driven by loss of βII-Spectrin, high expression of SCD and inhibition of GSK3β [179-181]. In recent years, noncoding RNAs and epigenetic factors have become important regulators of Wnt/β-catenin via modulating the expression of Wnt/β-catenin components APC, TCF, β-catenin [142, 203]. In terms of environmental factors, stemness factors such as Wnt and Rspo, and even immune factors and neurotransmitters in the microenvironment can promote self-renewal of CSCs by regulating the Wnt/β-catenin signaling pathway [217-220]. That supported the importance of cell-type plasticity in the occurrence of intestinal tumors [20]. Even in colorectal cancer with APC mutations, Wnt/β-catenin can still be regulated by microenvironment cells such as myofibroblast [221]. These different levels of regulatory mechanisms reveal the complexity of Wnt/β-catenin signaling pathway.

Like Wnt/β-catenin, the Notch signaling pathway is also a conserved signaling pathway in evolution and development, consisting of Notch receptors, Notch ligands, CSL, DNA-binding molecules and other regulatory molecules. There are four Notch receptors, namely Notch1–4, and five Notch ligands, namely DLL1, DLL3, DLL4, Jagged 1, and Jagged 2 in human beings [222]. The activation of Notch requires tight cellular connections, therefore, an artificial Notch reporter system can assess direct cell–cell contact [223]. The engagement of receptors and ligands causes the cleavage of the Notch receptor, leading to the expression of target genes such as HES and HEY by the Notch intracellular domain [224]. Due to the different expression patterns of Notch ligands and receptors in specific tumor types, Notch signaling pathway exerts protumorigenic or tumor suppressor roles via tumor-type dependent manner. Notch-related genes are highly expressed in gastric and colon cancers, promoting tumorigenesis, while their expression is low in liver cancer [225-227]. However, Notch-signaling pathway is generally recognized as a positive modulator in stemness regulation in CSCs, via promoting CSC survival, self-renewal and metastasis. DLL4 is highly expressed in gastric CSCs and enhances tumor angiogenesis and metastatic ability in CSCs [182]. The expression of Notch1 and Jagged1 in colorectal CSCs is essential for their self-renewal [228]. Even in HCC stem cells, Notch2, a special Notch molecule, is elevated and drives the expression of downstream HES1, HEY1, and NRARP, ultimately promoting the self-renewal of HCC stem cells [187]. Of note, Notch pathway activation also exists for HES1–ATOH1-mediated lateral inhibition, which plays a central role in the quantitative maintenance of intestinal absorptive cell-secretory cells, the similar mechanism may also be used by CSCs for quantitative control [229].

The Hh-signaling regulatory network includes the extracellular Hh ligand, the transmembrane Hh receptor PTCH, the transmembrane protein SMO and the downstream molecule GLI1 [230]. When there is no Hh ligand, PTCH induces the degradation of SMO, thereby inhibiting its function, while Hh ligand is present, the binding of Hh to PTCH leads to the degradation of PTCH, resulting in the activation of SMO, this ultimately activates the Hh signaling pathway through downstream signaling molecules such as GLI1. Many target genes of Hh pathway are markers of CSCs, such as Bmi-1, ALDH1, CD44, Nanog, Oct4, and Snail. Therefore, Hh pathway is a critical pathway for promoting the self-renewal and metastasis of CSCs [231-233]. In turn, the activation of Hh signal is also regulated by other genes, SPOP was reported to mediate the degradation of GLI2, thereby inhibiting the Hh signaling pathway and gastric cancer developments [183]. Aberrant activation of the Hh signaling pathway is a common feature of various tumors, including pancreatic cancer, liver cancer and gastric cancer [231, 234-236].

The above signals are not isolated; instead, the signaling pathways regulating CSCs involve complex interactions, with different signaling pathways sharing the same transcriptional regulatory elements [20]. Whether crosstalk between signaling pathways promotes or inhibits often depends on the environment, such as the interactions between Wnt/β-catenin and NF-κB, Wnt/β-catenin and YAP, and Wnt/β-catenin and Notch.

Wnt/β-catenin and NF-κB are both critical modulators in CSC survival and proliferation, and their crosstalk is intensively investigated. (1) TNFRSF19 is a target gene of β-catenin. Its variant, TNFRSF19.2, can regulate NF-κB activity and colorectal carcinogenesis by encoding the protein containing a TRAF-binding site, indicating the regulatory role of Wnt/β-catenin on NF-κB [237]. On the contrary, NF-κB can also regulate the activity of Wnt/β-catenin pathway, P65 and TCF have the same transcriptional activating element, CBP. Abnormal activation of P65 can amplify the Wnt/β-catenin signal activated by APC loss, resulting in dedifferentiation of non-stem cell regions and “top-down” colorectal carcinogenesis [20]. At the same time, NF-κB/P65 can regulate the expression of GSK3β and promote the activation of Wnt/β-catenin by stabilizing β-catenin and facilitating its nuclear translocation [238]. Additionally, there are negative regulatory relationships between Wnt/β-catenin and NF-κB, for example, in liver and colon cancer cells, β-catenin negatively regulates NF-κB activity, while activation of NF-κB can also inhibit β-catenin/TCF activity [239, 240].

There is also crosstalk between Shh and Wnt/β-catenin signaling pathways. The Wnt/β-catenin target gene CRD-BP associates with the mRNA of GLI1, the most important component for Hh activation. CRD-BP enhances the expression and transcriptional activity of GLI1 by stabilizing GLI1 mRNA, and ultimately promotes the self-renewal of colorectal CSCs [241]. In turn, Hh signaling can positively regulate the Wnt signaling pathway and promote the self-renewal and survival of colon CSCs [242]. In addition to crosstalk with Wnt/β -catenin signaling, there is also crosstalk between the Hh pathway and PI3K/AKT/mTOR, which synergistically regulate the survival, self-renewal and metastasis of pancreatic CSCs [243].

3.2 Metabolic Characteristics of CSCs and Their Roles in Maintaining Stem Cell Properties

CSCs are a type of cells within a tumor that possess special functions, characterized by unique microenvironmental and intracellular features, including metabolic features. Tumor metabolism and metabolites play a crucial role in the quantitative maintenance and functional regulation of CSCs. Accumulating evidence demonstrates the relationship between CSC self-renewal and metabolism. Studies have shown that normal stem cells mainly use glycolysis rather than oxidative phosphorylation (OXPHOS) to reduce ROS production, thereby maintaining their stem cell state, while CSCs rely on glucose consumption function [244]. Compared with non-CSCs, CSCs also exhibit enhanced lipid metabolism and iron metabolism, maintaining their stemness [245, 246]. Of note, the metabolite ceramide is increased in CSCs, inhibits tumor microimmune activity and plays a role in driving the immune evasion of CSCs and tumor progression [247].

The Warburg effect is the main feature of metabolic reprogramming in tumor cells, and the energy source of most cancer cells is mainly aerobic glycolysis rather than OXPHOS [248]. However, several studies have shown that CSCs prefer OXPHOS [249]. The stemness characteristics of Nanog⁺ liver CSCs depend on the activity of complex-I and the enhancement of mitochondrial function, which are further regulated by the SIRT1/MRPS5 pathway [250]. In another study, it was found that CD133⁺ CSCs in HCC exhibit impaired mitochondrial OXPHOS and enhanced mitochondrial fatty acid oxidation, which is further induced by TLR4-E2F1 signaling [251]. Liver CSCs are also associated with metabolic reprogramming involving ANGPTL4 and mitochondrial PDK4. Restoring mitochondrial pyruvate oxidation can enhance the effectiveness of chemotherapeutic drugs targeting liver CSCs [252]. Pancreatic CSCs preferentially use OXPHOS rather than glycolysis to consume galactose. Activation of OXPHOS promotes stemness features of CSCs, including elevated marker expression, enhanced tumorigenicity and immune evasion [253]. Cholangiocarcinoma stem cells enriched by sphere culture also exhibit enhanced OXPHOS levels [254]. The above studies have shown that OXPHOS is commonly activated in CSCs, but its function varies among different tumors and different CSC types.

Multiple studies have demonstrated that amino acids enhance and maintain stemness properties in CSCs [255]. In Kras^mut colorectal cancer cells, epigenetic remodeling induced by SLC25A22-dependent glutamine metabolism promotes WNT/β-catenin pathway and expression of colorectal CSC marker LGR5 [256]. CSCs in APC^Min/+ CRC also show increased expression of adenosylhomocysteinase (AHCY), an enzyme involved in the methionine cycle. Therefore, there was a significant co-expression of LGR5 and AHCY, and inhibition of AHCY y is able to reduce LGR5⁺ CSCs [257]. Xanthine oxidoreductase is lost in CD13⁺ CD133⁺ liver CSCs, and its loss blocks the accumulation of ROS in liver CSCs for CSC propagation [258]. ASCT2 is essential for the intracellular transport of alanine, serine, and cysteine. The interaction between ASCT2 and CD9 can promote the cellular uptake of glutamine and drive the malignant propagation of CD9⁺ PDAC CSCs [40].

The adipogenic pathway is activated in drug-resistant CSCs, and among these, the research on SCD is the most extensive. The expression of SCD is increased in EpCAM⁺ liver CSCs. SCD induces the transition from liver fibrosis to HCC via a Wnt positive-signaling loop by stabilizing lipoprotein-receptor-related proteins 5 and 6 [180, 259]. At the same time, SCD-mediated ER stress can also promote function of CSCs on sorafenib resistance [260]. High-fat diet plays a key role in maintaining the stemness of colorectal CSCs and is an important risk factor for CRC [261].

In CRC, LGR5⁺ CSCs can be divided into an active cycle cell population and a quiescent cell population according to their states, and the quiescent cell population has been found to specifically express p57, which plays an important role in maintaining the stability of the tissue stem cell population. When the p57 gene is deleted, the quiescent state of normal tissue stem cells will be disrupted, and their differentiation ability will also be impaired. In colorectal cancer, LGR5⁺ colorectal CSCs expressing p57 have little effect on tumor growth during the tumor stable period. However, chemotherapy will activate these cells, and they will greatly promote tumor regrowth [108].

3.3 The Role of Transcription Factors in the Self-Renewal and Stemness Maintenance of CSCs

As mentioned above, CSCs are a cellular state, and transcriptional regulation plays a central role in cell fate state regulation. OCT4, SOX2, Nanog, KLF4, c-MYC and other core transcription factors of embryonic stem cells and IPSCs have been extensively studied in the regulation of CSC stemness [262, 263]. The role of other types of transcription factors in the stemness regulation of CSCs has also been reported.

Highly expressed in liver cancer and liver CSCs, OCT4 is essential for the self-renewal and tumor-initiating capacity of liver CSCs. Interestingly, the activation of OCT4 is mediated by ZIC2, a transcription factor that binds to the promoter region of OCT4. ZIC2 recruits the NURF complex to initiate transcription of Oct4 and ultimately drive self-renewal of liver CSCs [184]. c-MYC is one of the most frequently activated oncogenes in human tumors. Highly expressed in various cancers, c-MYC is one of the most potent oncogenes and plays an important role in maintaining the population of CSCs [264]. c-MYC deletion induced the differentiation of liver CSCs and the formation of bile duct structures, accompanied by the loss of α-fetoprotein and increased expression of CK19 [265]. Similarly, the expression of MYCN was positively correlated with the activation level of the Wnt/β-catenin pathway and markers of liver CSCs (AFP, EpCAM, CD133), but negatively correlated with differentiation markers (including those of mature hepatocytes and bile duct epithelial cells). This indicates the relationship between MYCN and liver CSCs [266]. Of note, c-MYC is not only an important driver of tumorigenesis, but its activation in normal cells can also lead to cell proliferation arrest, senescence and apoptosis, indicating its role as the proliferation-promoting factor or apoptosis-promoting factor [267]. Nanog is highly expressed in HCC cells and is directly related to tumor metastasis and staging, meanwhile, Nanog is associated with the survival of gastric and pancreatic cancer, and patients with high expression of Nanog tend to have a poor prognosis [268-270]. The role of KLF4 in tumors and CSCs remains unclear, KLF4 can function as either an oncogene or a tumor suppressor depending on the tumor type, serving as a positive or negative modulator of transcription. Although KLF4 may act as an oncogene in certain types of tumors, it generally functions as a tumor suppressor in the digestive system, including cancers such as colorectal, gastric, liver, and esophageal cancer [271-274].

Besides, some less-studied transcription factors also play roles in maintaining the self-renewal of CSCs. Through transcriptome sequencing, Zhu and colleagues reveal that ZIC2 is among the most highly expressed transcription factors in liver CSCs, which drives liver CSC self-renewal via the OCT4 pathway [184]. PBX3, a target of tumor suppressor miRNAs including let-7c, miR-200b, miR-222, and miR-424 is critical for the stemness maintenance of α2δ1⁺ liver CSCsr through regulating several key stemness genes, including SOX2, Notch3, and EpCAM [185]. SOX9 is highly expressed in liver CSCs and promotes symmetrical cell division of CSCs through negative regulation of Numb, which is critical for liver CSC self-renewal and tumorigenesis. Meanwhile, low expression of Numb promotes Notch pathway activity, which further enhances the self-renewal of CSCs [186]. Notch pathway activation in liver CSCs also requires the inactivation of certain tumor suppressor genes, such as C8orf4. C8orf4 interacts with the Notch2 intracellular segment to block its nuclear localization and Notch pathway activation. Low expression of C8orf4 in HCC stem cells relieves its inhibitory effect on Notch pathway, thereby activating the Notch pathway in CSCs [187]. SALL4, a member of the zinc finger transcription factor family, is highly expressed in liver cancer and liver CSCs. SALL4 enhances the self-renewal and tumor-initiation potential of CSCs by driving the protein expression levels of several stemness-associated genes such as EpCAM, ABCG2, and CD44 [188]. In turn, SALL4 itself is regulated by OCT4 at transcriptional level. SALL4 promoter is demethylated in HBV⁺ HCC, and the binding of OCT4 and STAT3 to SALL4 promoter is enhanced, driving SALL4 transcription [275]. Besides liver cancer, in other types of digestive system tumors, the role of transcription factors in regulating CSCs is also crucial. For example, FOXP4, as a downstream target of the Hippo-YAP signaling pathway, can upregulate the expression of SOX12, thereby enhancing the stemness of CSCs in gastric cancer [276]. FOXO activates ST3GAL3/4/5 and ST6GALNAC6, thereby increasing the sialylation level in CCKBR⁺ CSCs to maintain their stemness and promote the occurrence of gastric cancer. (CCKBR⁺ cancer cells contribute to the intratumor heterogeneity of gastric cancer and confer sensitivity to FOXO inhibition) BATF2 enhances 5-Fu reactivity by inhibiting ABCG2 drug transporter protein and promoting PTEN stability, reduced expression of stem cell markers CD44, SOX2, and Nanog, and ultimately reducing the drug resistance and stem cell-like characteristics of gastric cancer cells [277]. HOXA 5 induces differentiation of colorectal CSCs by inhibiting the Wnt signaling pathway [190].

3.4 The Key Impacts of Epigenetic Regulation Alterations on the Number, Function, and Related Pathways of CSCs

Like genetic variants, alterations in epigenetic regulation are also important for the number and function of CSCs and are key regulators of the self-renewal and differentiation potential of CSCs. Multiple studies have shown that aberrant expression of epigenetic factors is common in CSCs. For example, BPTF, a core component of NURF complex, is highly expressed in liver CSCs and colon CSCs [184, 218, 236], EZH2, a component of the PRC complex, is highly expressed in liver CSCs and promotes the methylation of β-catenin [203]; SWI/SNF is also highly expressed in CSCs [142], more importantly, there is a shift of epigenetic complex types in CSCs, such as a transition from BRM-type SWI/SNF complex to BRG1-type SWI/SNF complex [202]. Besides abnormalities in chromatin remodeling complexes, dysregulation of epigenetic regulatory factors is also common in CSCs, including the posttranslational protein methyltransferase PRMT6, histone deacetylase SIRT1 and histone variant macroH2A1 [278-280]. Therefore, aberrant expression of epigenetic factors is a hallmark of CSCs, indicating their central roles in CSC function.

Epigenetic factors in CSCs not only have abnormal expression, but also exhibit abnormal binding sites in the genome, altered chromatin structure, and misregulation of target genes, which ultimately leads to the activation of oncogenic, self-renewal and antiapoptosis pathways, including MEK-ERK, PARP, NF-κB-/p65 and OCT4 signaling pathways. Due to the breakdown of DNMT1-mediated epigenetic silencing, BEX is highly expressed in liver CSCs, which promotes the activation of the β-catenin pathway, CSC stemness, and drug resistance [191]. Another independent study demonstrated that DNMT1, which can be inhibited by TGF-β, regulates CSC function and CD133 expression. Epigenetic reprogramming induced by a transient DNMT1 inhibition generates long-lasting cell context-dependent memory effects and influences the malignant properties of liver CSCs [192, 193]. LSD1 inhibits APC and PRICKLE1 expression by regulating H3K4me1 and H3K4me2 at the promoters, thereby promotes Wnt/β-catenin activation, and ultimately promotes the stemness and drug resistance of LGR5⁺ liver CSCs [194]. On the other hand, BMI1 is a crucial factor in maintaining drug-resistant SP cells in liver CSCs, and thus is essential for the stemness and drug resistance of liver cancer. Targeting BMI1 can inhibit the tumor-initiating ability of liver CSCs [281, 282]. Besides HCC, epigenetic regulation also plays a key role in CSCs of other digest system tumors, for example, MLL1 is an important factor for the stemness maintenance of colorectal CSCs and is critical for Wnt-mediated colorectal tumorigenesis [195]. BRD9 and SMAD2/3 jointly regulate the stemness of pancreatic CSCs, and the dysregulation of methylation can activate a fetal-like state and complement the stemness of CSCs [283].

M6A modification is a methyl group addition to the carbon-6 position of adenine in mRNA, and its role in mRNA stability, translation efficiency, and splicing has been widely reported [284, 285]. Recent studies have found that nascent RNAs in the nucleus can be modified by m6A, and these newly modified m6A can have a profound impact on chromatin structure and transcriptional activity. Therefore, m6A-mediated RNA epigenetics has become a new research field of epigenetics [286]. As a reader of m6A, YTHDF2 combines and stabilizes FZD10 mRNA in liver CSCs, and then activates Wnt/β-catenin and Hippo signaling pathway to initiate the stemness and Lenvatinib-resistance [114]. In addition, YTHDF2 promoted CSCs stemness through stabilizing OCT4 mRNA [287]. Unlike FZD10 and OCT4, binding of E2F6 and E2F3 mRNA to YTHDF2 results in mRNA degradation. IGF2BP2, another m6A reader, preferentially binds to E2F6 and E2F3 mRNA, and inhibits their association with YTHDF2, further activates Wnt/β-catenin through inhibiting E2F6/E2F3 mRNA decay in CSCs [129]. Similarly, m6A modification of TGF-β mRNA also inhibits its stability, therefore, RALYL-mediated reduction of m6A modification can promote TGF-β signaling and downstream PI3K/AKT and STAT3 signaling to promote the stemness of liver CSCs [196]. The above studies indicate that the same epigenetic factor has different regulatory effects on different target genes, even in the same CSC pool.

Besides N6-methyladenosine, N1-methyladenosine and other RNA modifications are also involved in the self-renewal of CSCs. m1A methyltransferase TRMT6/TRMT61A are highly expressed in liver cancer and CSCs, and involved in PPARδ translation and cholesterol synthesis via m1A modification of tRNA, promotes Hh signaling and liver CSC self-renewal of CSCs finally leads to poor survival [197]. Xinyuan Guan and colleagues demonstrate that ADAR1-mediated GLI1^R701G drives its cytoplasmic-to-nuclear translocation to drive self-renewal of liver CSCs [288]. Similarly, ADAR1-mediated A-to-I editing of SCD1 on 3'-UTR promotes mRNA stability to initiate gastric CSCs [100].

3.5 Diverse Regulatory Mechanisms of Noncoding RNAs in the Self-Renewal of Liver CSCs

With the development of RNA sequencing technology and the diversification of RNA-related research methods, a variety of functional noncoding RNAs have been identified in CSCs, among which microRNA, lncRNA and circRNA have been widely studied [289].

Hepatocellular carcinoma CSC is one of the most intensively studied types of noncoding RNA regulation. Multiple miRNAs regulate self-renewal of CSCs, including miR-130b, miR-148a, miR-192-5p, miR-429, and miR-1246 [198-201, 290]. These microRNAs function by regulating the stability of receptors on CSC surface or key genes in CSC signaling pathways, including miR155 for EpCAM, miR-1246 for AXIN2 and GSK3β, miR-181 for NLK, miR-148a for ACVR1 and miR-125b for SMAD, or by regulating the transcription of genes involved in differentiation or cell cycle. LncRNA are also important regulators in HCC stem cells, including lncTCF7, lncBRMlnca2, Lnc-β-Catm, lncDANCR, LncSOX4, PVT1 [142, 202-205, 236, 291]. CircRNAs involved in the self-renewal of HCC stem cells included circ-MALAT1, circIPO11, cia-MAFF, mcPGK1, and rtcisE2FF [126, 128, 129, 206, 292]. These lncRNAs and circRNAs exert their functions through various mechanisms, such as lncTCF7 recruiting SWI/SNF to the TCF7 promoter region, which promotes self-renewal of HCC stem cells through chromatin remodeling of TCF7 promoter, mcPGK1 encoded by mitochondria recruits PGK1 into the mitochondria and promotes the metabolic remodeling of liver CSCs, which further promotes the self-renewal of liver CSCs through the transcriptional regulation and protein stability regulation of β-catenin [142, 236]. Chromosome 8q24 region, including c-MYC and a non-coding RNA PVT1, frequently undergoes copy number amplification in tumorigenesis, PVT1 can maintain the stability of c-MYC, promote tumorigenesis and self-renewal of CSCs [293]. Of note, these noncoding RNAs may act together on the same signaling pathway, for example, lnncβ-CATM, lncTCF7, lncDANCR, mcPGCK1 and miR-1246 are all involved in regulating Wnt signaling pathway in HCC stem cells, illustrating the redundancy and complexity of noncoding RNA regulatory mechanisms.

4.1 Autocrine Regulation of CSCs

CSCs have a unique gene expression landscape and secretion characteristics. CD133⁺ liver CSCs express higher amounts of secretory proteins IL-8 and ANXA3 than CD133⁻ cells, and IL-8 efficiently promotes angiogenesis and self-renewal of CSCs. ANXA3 can activate Notch and JNK pathways, promoting self-renewal of liver CSCs [296, 297]. The above studies revealed that CSCs regulate their stemness in a positive feedback manner via secretory factors such as ANXA3. In fact, an ANXA3-neutralizing antibody has been developed, which shows an inhibitory effect on the propagation of HCC. Moreover, ANXA3 antibody increases the sensitivity of HCC to chemotherapy drugs such as cisplatin, sorafenib, and regorafenib via depletion of liver CSCs, indicating the clinical significance of blocking CSC autocrine regulation [298, 299]. Similarly, HCC progenitor cells, emerging several months before tumorigenesis, secrete IL-6 for malignant progression. IL-6 accumulation is induced by increased LIN28 expression, and activates JAK1/2–STAT3 signaling to further drive LIN28 transcription, forming a positive loop [18].

Autocrine mediates not only positive feedback regulation but also negative feedback regulation, which generally occurs in various cells. IFN-γ, secreted through immune cell activation, is the most important in vivo signal to initiate PD-L1, which is used to terminate or eliminate immune response [300]. Similarly, due to multiple Wnt inhibitory factors in Wnt target genes, activation of the Wnt pathway is accompanied by negative feedback regulation, for example, Wnt target genes RNF43 and AXIN2, which induce ubiquitination-induced degradation of FZDs (Wnt receptor) and β-catenin, respectively, therefore inhibit the Wnt pathway. As the most critical regulatory signal in CSCs, the Wnt pathway is also regulated by autocrine-mediated feedback inhibition loop. Notum, one of Wnt target genes, has recently been identified as a secretory feedback inhibitor of Wnt/β-catenin, when Wnt pathway is activated, high expression of Notum can deacylate Wnt ligand to diminish Wnt activation [301]. In tumorigenesis, gene or environmental disturbance may cause the failure of negative feedback regulation. For example, APC mutation, the most common type in colorectal cancer, causes deficiency of the β-catenin-degrading pathway, thus inducing constructive activation of Wnt/β-catenin signaling and subsequent Notum expression. Extracellular Notum deacylates Wnt ligands, which has little effect on APC^mut cells but inhibit Wnt/β-catenin activation in APC^WT cells. Accordingly, APC^mut cells obtain advantages in clonal expansion compared to APC^WT cells [173, 174]. A similar situation also exists in the mechanism of obtaining competitive advantage in Kras^G12D and PIK3CA^H1047R cells, in which BMPs secreted by LGR5⁺ cells drive the clonal advantage [175]. These studies suggest that autocrine is a key driver of CSC self-renewal and tumorigenesis.

4.2 Interaction Between CSCs and Tumor Non-Stem Cells

In addition to CSCs, tumor tissues contain a large number of non-stem cells. These non-stem cells are distributed around CSCs and play important roles in regulating CSCs. In fact, the ratio between CSCs and non-stem cells is always maintained at a certain level. When CSCs are depleted, tumor cells can dedifferentiate into CSCs, while tumor cells are cleared, CSCs can differentiate into tumor cells, indicating the inherent fate exchange mechanism between CSCs and non-stem cells [302].

Tumor non-stem cells reshape CSCs through direct contact or secrete signal factors. The initiation of Wnt signaling pathway in liver CSCs requires Wnt ligands in the environment and Wnt receptors on the surface of CSCs. FZD6 is highly expressed in CSCs, while Wnt5A is secreted from non-CSCs, the Wnt5A-FZD6 engagement drives Wnt/β-catenin activation and self-renewal of CSCs [217]. In homeostasis-state intestinal tissue, Paneth cells, a type of differentiated cell, provide factors such as Wnt ligands, EGF, and pyruvate for the self-renewal of ISCs [303, 304]. Besides, enteroendocrine and tuft cells serve as niche cells for stem cells when Paneth cells are absent [305]. Intestinal adenomas are reminiscent of normal crypt, and stem cells are adjacent to differentiated cells such as Paneth cells and KRT20⁺ epithelial cells [28, 42]. Therefore, it is expected that the maintenance of KIT⁺ stem cells in colorectal cancer requires stem cell factors produced by tumor non-stem cells [306]. The death of non-stem cells, generally induced by chemotherapy, radiotherapy and immunotherapy, significantly regulates the fate of stem cells via CSC-direct targeting or niche-dependent manner, for example, chemotherapy-induced cell death promotes the emergence of CD10⁺GPR77⁺ CAFs through activating NF-κB/P65 signaling, which further drives CSC self-renewal via IL-6 and IL-8 [88].

Cell competition occurs between CSCs and non-CSCs, or different CSCs with various intrinsic changes, such as mutations. CSCs often have a stronger competitive advantage than non-stem cells due to their greater environmental adaptability and plasticity. APC mutant CSCs maintain the advantage of self-renewal and clonal expansion by secreting Notum to inhibit the growth of surrounding normal cells [173, 174]. Similarly, Kras^G12D or PIK3CA^H1047R stem cells also outcompete neighboring cells via secreting BMP [175]. The surviving cells of gastric cancer after treatment usually have CSC capability, and the expression of SLC6A6 is upregulated, which escapes the killing of immune cells through taurine absorption, eventually leading to drug resistance and recurrence [307, 308].

4.3 Endothelial Cells

Lee M. Ellis and colleagues reveal that CD133⁺ liver CSCs and CD31⁺ ECs locate adjacently but without direct cell–cell contact, and ECs secreted soluble Jagged-1 to activate Notch signaling in CRC cells to promote CSC phenotype [309]. EC also secretes LRG1, which directly binds to and activates HER3, and activates protein synthesis through the PI3K–PDK1–RSK–eIF4B axis, thereby promoting CRC growth [310]. Vascular ECs promote CSCs to escape from radio-/chemo-therapy and promote, invasion and metastasis of CSCs [311, 312]. Some studies have shown that CSCs can differentiate into vascular ECs [313, 314]. This pattern of differentiation demonstrates the plasticity of CSCs and echoes the mimicry theory of CSCs as maintained above [92]. Interestingly, a recent lineage tracing study found that newly generated vascular ECs can be produced from within the tumor to the outside, indicating the tumor-intrinsic generation of ECs [223].

4.4 Cancer-Associated Fibroblasts

Activation and mass enrichment of fibroblasts are common features of various tumors, especially liver cancer, because most HCC develop in the context of cirrhosis [315]. α-SMA⁺ CAFs are directly related to the prognosis of HCC. CAFs secrete liver growth factor (HGF) to activate the MET–FRA1–HEY1 pathway, ultimately promoting the tumor initiation ability of hepatic CSCs and cirrhosis-dependent liver tumorigenesis [316]. Similarly, HGF produced by CAFs also promotes the expression of CK19, a marker of hepatic stem/progenitor cells and related to poor prognosis [317]. In addition to HGF, CAFs also secrete high levels of IL-6, which drives liver CSC self-renewal through IL-6–STAT3 pathway [318]. CAF-produced CLCF1 induces tumor cells to release TGF-β and CXCL6, further promotes tumor stemness via autosecretion and TAN-mediated paracrine [319]. CAFs also drive self-renewal of liver CSCs in Notch3-dependent manner through activating LSD1 [320]. Not only tumor-locating CAF cells, but the fibroblasts of adjacent tissues can secrete a series of cytokines, including HGF, IL-6, CCL2, CXCL1, CXCL8, and VEGF to recruit EpCAM⁺ liver CSCs for HCC metastasis [321]. Myofibroblasts deliver extracellular vesicles (EVs) to HCC cells, inducing the acquisition of tumor stem-like characteristics in HCC cells [322]. Hepatic astrocytes, which are fibroblasts located in the liver, can interact with monocyte, jointly maintain the immunosuppressed microenvironment and further promote tumor cell migration and globular formation [323].

In addition to liver cancer, the enrichment of fibrocytes is also reported in other tumors. Medema and colleagues found that Wnt/β-catenin activation was heterogeneous even in the context of APC^min, and tumor cells close to stromal myofibroblasts showed stronger Wnt signaling and stemness. Secretion of HGF by stromal myofibroblasts promotes Wnt/β-catenin activation to initiate CSC self-renewal, indicating a more complicated regulation network for Wnt/β-catenin activation [221]. The single-cell sequencing of mouse colorectal cancer revealed the presence of Ptges⁺ CAFs, which drives the expansion of Sca-1⁺ reserve-like stem cells and colorectal tumorigenesis via PGE2–Ptger4 engagement and subsequent YAP activation [324]. CAFs in CRC secrete TGF-β2, which together with HIF-1α, promotes the expression of GLI2 in CSCs, and further promotes the self-renewal and drug resistance of colorectal CSCs through the Hh pathway [325]. Contrary to the opinion that colorectal CSCs are located inside the tumor near blood vessels, Louis Vermeulen group reveals that functional CSCs are located at the edge of the tumor via constructing a marker-free lineage tracing assay. These marginal stem cells cannot be distinguished by traditional labeling systems such as LGR5 and are important drivers for tumor expansion [309]. Moreover, these CSCs located close to CAFs, which secrete osteopontin to drive clonogenicity and tumor expansion [326]. Treatment of pancreatic cancer cells with conditioned media from CAFs but not normal human fibroblasts increases tumor-sphere formation and CD44⁺ CSCs through SPP1 [327]. The regulative effect of CAFs on CSCs has also been validated in breast and pancreatic cancers [88, 328].

Multiple studies have revealed the key role of CAFs in tumors and tumor drug resistance. The drug resistance of CSC is partly driven by CAFs in the microenvironment. Christopher J. Tape's team used single-cell sequencing to analyze cell types in patient derived orgnoids (PDOs) coculture with CAFs or in response to multiple therapies, and identify the changes in the type of CSCs, with cell fate transition from MKI67⁺ proliferative CSCs to MKI67^- revival CSCs [51]. The damage response of chemotherapy and alteration of TME such as CAFs are vigorous inducer for stemness. For example, chemotherapy for breast cancer induces the activation of NF-κB pathway in CAFs, leading to the emergence of CD10⁺GPR77⁺ cells, which further drives CSC expansion [88].

4.5 Neuron

Due to the promotion of anxiety and other neurological factors contributing to a variety of tumors, the neuroregulation of tumors has received much attention from researchers.

In recent years, an increasing body of evidence showed that nerve cells are one of the most important microenvironment cells in CSCs [329]. Gastric and colorectal CSCs have the potential to differentiate into neuronal cells that form the nervous system in tumor tissues, and it has been shown that knocking down the neurogenic capacity of CSCs reduces tumor growth [330]. Dclk1⁺ tuft cells and nerve cells in gastric cancer secrete acetylcholine (ACh), which drives epithelial proliferation and tumorigenesis via the ACh receptor M3R. At the same time, Ach treatment can also stimulate gastric cancer cells to produce nerve growth factors, therefore accelerating nerve expansion and forming ACh-NGF feedforward [331]. In pancreatic cancer, enhanced cholinergic signals inhibit self-renewal of CSCs and tumorigenesis of pancreatic carcinoma through CHRM1 and subsequent inhibition of MAPK/EGFR and PI3K/AKT [332]. The controversial functions of cholinergic signals in gastric and pancreatic cancer may be due to their different receptors and downstream transmission of activation/inhibition signals. The undecapeptide substance P promotes AKT activation and inhibits apoptosis upon binding to the neurokinin-1 (NK-1) receptor, which is highly expressed in gastric and intestinal cancers. And NK-1 receptor antagonist exerts antitumor effects on colon CSCs [333]. In colorectal cancer, the content of 5-HT significantly increased due to the blocking of 5-HT degradation pathway. At the same time, the 5-HT receptors HTR1B, HTR1D and HTR1F are highly expressed in colorectal CSCs which can bind to enriched 5-HT in bowel cancer tissues to promote the activation of Wnt/β-catenin signals and self-renewal of colorectal CSCs [334].

4.6 Adipocytes

Metabolic abnormalities are typical features of tumors, meanwhile, obesity and fat cells have promoting effects on a variety of tumors including liver cancer, bowel cancer and pancreatic cancer. Fatty liver is one of the most primary risk factors in liver cancer, which is mainly characterized by the accumulation of fat cells. Lipids and immune factors secreted by adipocytes, including IL-6, IL-8, and MCP1, can promote the expansion, metastasis and resistance to sorafenib of liver CSCs [335]. High-fat diets also facilitate TWIST1 expression via TLR4-Nanog and LEPR-Stat3 signaling in CSCs [336]. Palmitic acid from high-fat diet also activates the PPAR-δ and Wnt/β-catenin signals of ISCs to accelerate the numbers and functions of ISCs and progenitor cells, further promoting the carcinogenesis of CSCs [261]. Visceral adipose stromal cells (V-ASCs) locating near tumor cells secrete IL-6 and HGF to promote the expansion and metastasis of CD44V6⁺ colorectal CSCs while CSCs in turn secrete NGF and NT-3 to recruit adipose stem cells which forms an ASC-CSC feedback loop [337].

4.7 Immune Cells Modulate CSCs

Immune cells in tumors are an important part of TME. CSCs have stronger tumor initiation abilities and their ability to escape immune killing is a key step in tumorigenesis, so CSCs have a unique advantage in escaping immune system monitoring. More and more studies have proved that immune cells have become an important microenvironment cell type in CSCs.

4.8 Microbial Niche Modulate CSCs

The TME contains not only a variety of cells but also a variety of types of microorganisms, some of which play crucial regulatory roles in the process of tumor development. In this part, we summarized the microbial signals that modulate stemness in the TME, including hepatitis viruses for HCC, Helicobacter pylori for gastric cancer, Clostridium nucleatum for colorectal cancer, and intramural microorganisms.

4.9 Impact of Niche Factors on CSCs Regulation

Besides wide diversity of cells, the TME also includes diverse noncellular characteristics, which involves extracellular matrix, hypoxia, nutrient deficiency, acid, inflammatory environment, and so forth. These microenvironmental factors modulate tumor stemness in a variety of ways (Figure 3B).

4.10 CSCs Remodel Niche

Tumor is a complex cell population, and there are many complex relationships between tumor cells and microenvironment cells, including parasitism, symbiosis, predation and mimicry. Tumor cells not only passively receive environmental signals, but also deeply reshape the microenvironment to create an environment suitable for their own survival, and eventually evade immune surveillance and gain growth advantages (Figure 4). The microenvironment can be regulated by CSCs through direct contact, paracrine or exosomes [410].

4.11 Systemic Regulation

For the past several years, the mutual interactions of cancer (stem) cells and systemic factors are intensively investigated. Here, we address the systemic regulations of tumors, including the interaction of distal tissues, endocrine factors, lifestyle and other diseases and tumors.

4.11.1 Impact of Distal Tissues on Tumor Regulation

Tumor cells and CSCs not only remodel the microenvironment, but also have a profound remodeling effect on the distal tissues of the individual, where remodeling of the pre-metastatic microenvironment by cancer (stem) cells is a prerequisite for tumor metastasis [121, 422]. Tumor cells undergo deep remodeling of distant organs through the secretome before reaching the metastatic site. The exosomes produced by CSCs contain special microRNAs and mRNAs before they reach the metastasis site, which have a stronger advantage in vascular growth and tumor metastasis [423]. In recent years, there has been increasing attention to the effects of tumors on the systemic immune system. Exosomes produced by cancer cells express high levels of PD-L1 on the surface, inhibiting the function of CD8⁺ T cells and promoting tumor growth [424]. For more details, you can refer to the relevant overview [425].

The intestine is an extension of the external environment to the body, colonizing a large number of microbes for normal physiological functions such as digestion and absorption. Recent studies have found that intestinal microbiota can affect tumors and immunity throughout the body, ultimately regulating tumorigenesis and CSCs, suggesting that gut microbiota acts as tissue-like components in tumors regulation. Formic acid, a metabolite produced by F. aggregatum, can promote the stemness and invasion of colorectal cancer cells via activating the AhR signaling pathway. Formic acid and Fn treatment induce more ALDH⁺ stem cells, with enhanced SOX2 expression. At the same time, formic acid treatment can also promote the expansion of Th17 cells (Figure 5A) [426]. Intestinal bacteria Klebsiella spp. produces DNA alkylating metabolite tilimycin, which induces DNA damage of circulating stem cells, inducing apoptosis of intestinal epithelium and colorectal colitis, which contributes to colorectal tumorigenesis [427]. In addition to inflammatory response and DNA damage, gut microbiota can also produce metabolites such as butyrate, which promotes hyperproliferation of MSH2 mutant colorectal cancer, while DNA-mismatch pathway can reregulate β-catenin activity and TA cell differentiation, indicating that intestinal flora also plays a key role in the regulation of Wnt pathway [428].

CSCs are highly expressed hormonal receptors to response systemic factor derived from distant tissues. Insulin, an endocrine factor produced by pancreatic β cells, plays a key role in blood glucose regulation, and is able to promote the self-renewal of CSCs through multiple mechanisms. CSCs express high levels of insulin-like growth factor (IGF1R) and insulin receptor (IR), which have a high corresponding effect on insulin-like growth factor and insulin [429]. Interestingly, insulin acts on the IR of CSCs, inhibits the activity of GSK3β through IRS2–PI3K pathway, prevents phosphorylation and ubiquitination of c-MYC, which maintains the stability and activity of c-MYC, and finally promotes the self-renewal of CSCs (Figure 5B) [430]. A variety of tumors exhibit sex differences, suggesting that CSCs are also regulated by sex hormones, including androgens and estrogens [431]. Adipocytes secrete adipokine adipsin, a complement factor for activation of alternative pathway, to promote the secretion of HGF, mediate the interaction between “adipose and cancer stem cells,” and ultimately promote the self-renewal of CSCs (Figure 5B) [432].

5.1 Targeted Therapy for CSCs

After years of exploration and research, researchers have developed multiple methods to target CSCs, including targeting surface markers and intracellular factors including stem cell signaling pathways, epigenetics, stem cell regulators, kinases and phosphatases (Table 2), the drugs undergoing clinical trials for therapy targeted to the CSC showing in Table 3.

5.2 Targeted Therapy for Niche of CSCs

The microenvironment plays a key role in the self-renewal and tumorigenesis of CSCs and is becoming a new target for tumor intervention. Tumor immunotherapy, which is essentially the remodeling of the immune microenvironment of CSCs, has achieved great success in clinical practice, suggesting that targeting the TME is an effective strategy. Here, we focus on recent advances in targeting the microenvironment of CSCs (Table 4), the drugs undergoing clinical trials for therapy targeted to the niche of CSC shown in Table 3.