Cancer stem cells: advances in knowledge and implications for cancer therapy

Abstract

Cancer stem cells (CSCs), a small subset of cells in tumors that are characterized by self-renewal and continuous proliferation, lead to tumorigenesis, metastasis, and maintain tumor heterogeneity. Cancer continues to be a significant global disease burden. In the past, surgery, radiotherapy, and chemotherapy were the main cancer treatments. The technology of cancer treatments continues to develop and advance, and the emergence of targeted therapy, and immunotherapy provides more options for patients to a certain extent. However, the limitations of efficacy and treatment resistance are still inevitable. Our review begins with a brief introduction of the historical discoveries, original hypotheses, and pathways that regulate CSCs, such as WNT/β-Catenin, hedgehog, Notch, NF-κB, JAK/STAT, TGF-β, PI3K/AKT, PPAR pathway, and their crosstalk. We focus on the role of CSCs in various therapeutic outcomes and resistance, including how the treatments affect the content of CSCs and the alteration of related molecules, CSCs-mediated therapeutic resistance, and the clinical value of targeting CSCs in patients with refractory, progressed or advanced tumors. In summary, CSCs affect therapeutic efficacy, and the treatment method of targeting CSCs is still difficult to determine. Clarifying regulatory mechanisms and targeting biomarkers of CSCs is currently the mainstream idea.

Introduction

Due to advances in cancer early detection and cancer treatments, cancer yearly mortality has been decreasing since 1995.¹ However, cancers still caused more deaths than COVID-19 and ranked as the second cause of death in the United States in 2020 and 2021.¹ The presence of cancer stem cells (CSCs) can be an essential factor that leads to failure of cancer treatments.

CSCs, first identified in 1990,² are a small group of cancer cells that possess properties of normal stem cells, such as self-renewal and pluripotency.³ The CSC model, also known as the hierarchical model, provides a paradigm for people to understand intratumoral heterogeneity, as they can differentiate into various phenotypes of cancer cells and maintain their population.⁴ CSCs are also characterized by enhanced ability to initiate tumor growth, proliferate, invade, migrate, and resist therapeutic effects.³ This implies a crucial role of CSCs in cancer development and makes CSCs an evaluable target for anti-cancer treatments. Therapeutic agents, such as monoclonal antibodies, tyrosine kinase inhibitors, chimeric antigen receptors (CAR) T cells, and tumor vaccines, targeting CSCs have been developed and tested in clinical trials.⁵

In recent years, studies have added knowledge in the origin, features, and especially therapeutic aspects of CSCs. Here, we summarize the research history, origins, properties, molecular regulations, mechanisms for therapeutic resistance, and treatment strategies of CSCs.

The development of the CSCs theory

The discovery and controversy of the CSCs

As early as 1855, the work of pathologist Rudolph Virchow illuminated that tumors stem from existing normal cells, sparking a scientific discourse about the origin of tumors (Fig. 1). Julius Cohnheim disagreed with that and contributed to his “embryonal cell rests” hypothesis in 1867.⁶ This posited that dormant embryonic cells within tissues could awaken into tumors.⁷ Spanning the 19th and 20th centuries, burgeoning research into the genetic underpinnings of cancer has fostered the prevailing notion that cancer arises from the accumulation of mutations in susceptible cells. However, given the terminal differentiation and quiescence of most body cells, their lifespan seldom permits the accrual of the requisite mutations to become cancerous.^8,9,10 Hence, cells endowed with the capacity for sustained proliferation are the likely precursors of tumors. This hypothesis gained traction and was bolstered with the discovery of Jacob Furth and Morton Kahn in 1937, which leukemia could be recapitulated in mice from single malignant cells.¹¹

Fig. 1

The Development of the CSC Theory. As early as 1855, in the discourse on the origins of tumors, Cohnheim posited that tumors stemmed from embryonic cells. Subsequent decades of genetic research concluded that tumor formation necessitates the accumulation of susceptibility genes, implying that the cells causing tumors must possess self-renewal capabilities. It wasn’t until 1937 that Furth demonstrated the potential of single malignant cells to induce tumors. This revelation spurred researchers to delve into the characteristics of such cells, encompassing self-renewal, aberrant differentiation, interaction with the microenvironment, and heightened plasticity. In 1997, John Dick identified leukemia stem cells. Since then, the theory of CSCs has basically taken shape. And people have begun to continuously isolate and prove CSCs from different tumor types

The understanding of this field deepened when James Till and Ernest McCulloch, in 1961, observed clone formation in the spleen during hematopoietic regeneration.^12,13 Moreover, these clones could form additional clones in other mice, laying the groundwork for understanding the self-renewing and differentiative capabilities of stem cells.^14,15 Schofield introduced the definition of “stem cell niche” in 1978, and further nuanced this, highlighting the role of microenvironment in nourishing and directing stem cells.^16,17 From the 1960s to the early 1990s, debates about the origin of tumors had oscillated between non-genetic “induction” or “niche destruction” versus proliferative single-cell mutations.^18,19,20

The landscape of stem cell research was revolutionized in 1997 when John and Bonnet first identified cells with extensive proliferative potential in acute myeloid leukemia (AML) and isolated CSCs characterized by the CD34⁺CD38⁻ phenotype.² This seminal breakthrough acknowledged the existence of leukemia stem cells and paved the way for the theory of “CSCs” in 2001.²¹ Tumors house a kind of rare cells with self-renewing potential named CSCs that drive tumorigenesis, akin to normal stem cells but with a role in cancer progression.²² The concept of CSCs has since expanded to various solid tumors. In 2003, Al-Hajj first isolated CD44⁺CD24^−/low CSCs from breast tumors, capable of significant tumorigenicity in mice. Their study had shown that 200 such cells could form transplanted tumors in recipient mice within 12 weeks. In the same culture time, 10,000 non-special breast cancer cells cannot form tumors.²³ In the same year, Sheila K Singh purificated a CD133⁺ CSC population from diverse brain tumors.^23,24 The discovery of these potent CSCs across both hematologic and solid malignancies has substantiated the CSC theory, with subsequent findings in prostate, colorectal, pancreatic, nasopharyngeal cancers, and so on.^{25,26,27,28,29,30}

Despite widespread support and experimental evidence for the theory of CSCs, xenotransplantation success rates and actual CSC percentages have fallen short of expectations.³¹ For instance, only a small fraction (<5%) of leukemia transplants in mice, as seen in Jacob Furth’s studies, successfully engrafted.¹¹ Hewitt’s research further corroborates the scarcity of successful transplantation, with colony formation in murine spleens documented at a mere 1% to 4%.³² Moreover, Park et al. in vitro clonal cultures from myeloma cells extracted from murine ascites exhibited clonal colony formation in only 0.01% to 1%.³³ Analogous low successful frequencies of tumorigenic cells are reported in vitro cultures of lung, ovarian, and neuroblastoma cancers.³⁴ Concurrently, more and more research reveals the diversity and instability of CSCs, with variations in cell origin, proportion, genetic makeup, and even phenotypic and functional traits.^35,36,37 Initially, CSCs exhibiting the CD34⁺CD38⁻ phenotype were linked to the etiology of AML. Subsequent findings, however, indicated that CD34⁺CD38⁺ cells also possess tumor-initiating potential in NOD/SCID mice lacking the Interleukin 2 Receptor Subunit Gamma (IL2RG) chain, suggesting that such activity might be independent of CD38 expression.^2,38 This phenomenon is mirrored in solid tumors, where certain CSC populations within the same tumor display distinct and non-overlapping marker profiles. Ginestier et al.‘s research in breast cancer revealed that cells exhibiting high Aldehyde Dehydrogenase (ALDH) activity not only demonstrated traits of tumorigenicity but also the capacity to self-renew and replicate the heterogeneity of the original tumor.³⁹ These cells also displayed minimal overlap with the previously characterized CD44⁺CD24^−/low phenotype breast CSCs, constituting less than 1% of the cancer cell population.³⁹ Research suggests that CD133 is a marker capable of identifying CSC populations across various solid tumors, including different forms of brain cancer.^40,41,42,43 However, subsequent studies have raised questions regarding the reliability of using CD133 to distinguish and isolate CSCs, indicating a degree of controversy in its application.^44,45,46 Firstly, CD133 may serve as a marker for glandular epithelium in certain tissues, complicating the distinction between CSCs and non-stem-like cancer cells. Secondly, research has demonstrated that CD133⁺ cell populations fail to replicate the morphology of the original upon xenotransplantation, suggesting the possibility of expression of CD133 on normal differentiated cells. Lastly, studies have shown that CD133^−/low populations have been shown to recapitulate the original tumor architecture, indicating that CD133 may not be the sole marker for identifying CSCs.^47,48,49 Moreover, plasma cells expressing the CD138 phenotype were found to only induce multiple myeloma (MM) in SCID-hu mice, failing to generate comparable tumors in NOD/SCID mice.^25,50,51,52 Similarly, analyses of samples from AML patients revealed distinct genetic and phenotypic characteristics of CSCs among individuals, highlighting the variability within the CSC population across different models and patient samples.^53,54 The proportion of CSCs within primary tumors is also highly variable, ranging from 0.2% to 82.5%.³⁵ For instance, CSCs with the CD34⁺ phenotype constitute less than 1% of AML cases, yet represent 82.5% in B-cell precursor acute lymphoblastic leukemia (ALL).^2,38,55 Conversely, the proportion of CSCs expressing the CD133⁺ phenotype in lung cancer ranges from a mere 0.4% to 1.5%, while in brain tumors and colorectal cancer, it can escalate to as high as 20%.^24,56,57 Furthermore, the frequency of CSCs may increase during tumor progression. Pece’s study found a higher proportion of CSCs in stage III breast cancer compared to stage I about 3 to 4 times on average.⁵⁸ The origins of CSCs remain elusive, with evidence from myeloid leukemia and brain tumors suggesting they may arise from normal stem cells, while findings from MM and ALL suggest alternative origins.^{2,24,42,59,60} The definition of CSCs becomes increasingly nebulous, raising doubts about the model itself. However, the methods used at that time could not account for cellular heterogeneity and proliferative potential within different tumor cell populations.^61,62 Moreover, there was heterogeneity in the analytical methods used.^63,64 Thus, debates over the CSCs model will persist until direct empirical evidence is presented. Nonetheless, the validity of the model should not be discounted due to the diversity and complexity that continue to emerge in experimental evidence.

Advances in key technologies: from sorting to sequencing

The hypothesis of CSCs offers a pivotal theoretical framework for understanding tumor initiation and progression. Initially considered rare and dormant, forming a unidirectional hierarchy within tumors, CSCs were thought to generate all cell types within a tumor, occupying the apex of the tumor cell hierarchy.⁶⁵ However, further research has revealed the model of CSCs to be more complex and dynamic. CSCs exhibit phenotypic plasticity, transforming in response to the microenvironment, leading to genetically heterogeneous tumors.^66,67,68 Competitive interactions among various related but distinct subclones within tumors favor subpopulations with enhanced self-renewal capabilities and therapeutic resistance.³⁵ Initial research propelled by traditional cell sorting techniques, advancements, especially in sequencing technologies, have continually enriched and evolved our understanding of the CSC model. The evolution of cell sorting technologies has progressed from utilizing physical properties of cells, such as size, density, adhesiveness, and refractivity, to targeting cell surface antigen phenotypes and functional characteristics like dye efflux, calcium ion concentration, and pH.⁶⁹ Techniques include density gradient centrifugation, fluorescence-activated cell sorting (FACS), magnetic-activated cell sorting (MACS), and side population (SP) cell sorting (Table 1).

Table 1 Methods of isolating cancer stem cells

While density gradient centrifugation was initially designed for isolating mononuclear peripheral blood cells, its application quickly extended to stem cell separation.^70,71 Compared with a single-layer density gradient, the method of using a multi-layer discontinuous density gradient can separate CSCs more effectively. Percoll, a colloidal silica coated with polyvinylpyrrolidone (PVP), is preferred, though the toxicity of PVP in percoll limits its clinical safety.⁷²

In isolating CSCs, methods based on cell surface markers are prevalent, notably FACS and MACS.⁷³ MACS employs antibodies attached to magnetic beads to target cell membrane antigens, using magnetism to retain cells bound to beads within a column while unbound cells are washed away.⁷⁴ Despite its minimal impact on cell viability and suitability for large-scale sorting, MACS is limited by its reliance on single antigens, complex operation, and high costs.⁷⁵ FACS, on the other hand, utilizes fluorescently labeled antibodies to distinguish between CSCs and non-CSCs, offering higher specificity by screening multiple markers simultaneously. FACS can assess intracellular pathways and protein interactions and overcome the specificity challenges of CSC membrane antigens.⁷⁶ However, FACS requires stringent experimental conditions and precise cell pretreatment to maintain cell viability, posing challenges in terms of equipment cost and operational requirements.⁷⁵

The SP cell sorting method identifies CSCs using the Hoechst 33342 dye, capable of penetrating cell membranes.⁷⁷ Since the discovery of Goodell et al. in bone marrow studies in 1996, this technique has proven effective across various tumor cell lines.^78,79,80 SP cells, capable of asymmetric division and self-renewal, align with the characteristics of CSCs, suggesting SP sorting as a viable CSC enrichment strategy. Despite its simplicity, requiring only microscopy or flow cytometry to detect unstained cells, challenges include low separation efficiency and dye cytotoxicity. Nonetheless, its utility in sorting drug-resistant CSCs offers valuable insights for novel drug research. Researches indicated that when the activity of ABC transporters, such as ATP-Binding Cassette Transporter G2 (ABCG2), was inhibited, the SP phenotype cells decreased.^81,82 Conversely, an increase in expression lead to an augmentation of the SP phenotype cells.⁸³ Consequently, some researchers suggest that SP cells are not CSCs but rather a subset of cells capable of evading the cytotoxic effects of Hoechst dye.⁸⁴

Beyond the aforementioned methods, alternative approaches for isolating and identifying CSCs exist.⁸⁵ Drug selection separation gradually evolves cells towards drug resistance, isolating those capable of stable growth and passage, believed by some researchers to be CSCs.^86,87 Western blot analysis serves as a traditional identification method, prized for its simplicity, universality, and cost-effectiveness, though it risks false positives or negatives if improperly executed, typically serving as a technique for validation. In 2021, Han et al. developed a novel label-free, microfluidic technology for CSC sorting based on physical characteristics like size, elasticity, and adhesiveness, enabling stable, rapid, and efficient CSC selection and enrichment, offering a new platform for targeted drug screening and functional identification.⁸⁸

The integration of CSC sorting techniques with sequencing studies offers new insights into tumor complexity and heterogeneity. High-throughput technologies like RNA sequencing facilitate the monitoring of tumor microenvironment interactions and key gene expression dynamics. For instance, Chen discovered that under specific Trimethylation Of Lysine 4 On Histone H3 Protein Subunit (H3K4Me3) epigenetic modifications, the transcription factor MYC upregulates histidine decarboxylase, endowing glioblastoma stem cells (GSCs) with the ability to synthesize and secrete histamine. Histamine secreted by GSCs acts on the Histamine Type I Receptor (H1R) of vascular endothelial cells, activating the H1R/Ca²⁺/Nuclear Factor Kappa-B (NF-κB) signaling pathway to promote angiogenesis and advancing glioblastoma progression.⁸⁹ The combination of advanced imaging, short hairpin RNA (shRNA) technology, and subgroup analysis tools has also highlighted the critical role of tumor-associated antigens in GSC differentiation.^90,91 The development of multi-channel optical imaging systems has made it feasible to simultaneously monitor cell chemotaxis, proliferation, and NF-κB activity.⁹² In breast cancer, CSCs hyperactivate the Nuclear Respiratory Factors 2 (NRF2) pathway via the epigenetic reader Zinc Finger MYND-Type Containing 8 (ZMYND8), enhancing antioxidative capacity and evasion from oxidative damage and ferroptosis.⁹³ Erythropoietin-Producing Hepatocellular Carcinoma Receptor B2 (EPHB2) and Lysine-Specific Histone Demethylase 1 (LSD1) are noted for their roles in promoting CSC generation and drug resistance in hepatocellular carcinoma and thyroid cancer, respectively. The sequencing technology found that they interact with the T Cell Factor 1(TCF1)/EPHB2/β-Catenin signal pathway and Wingless-Type MMTV Integration Site Family (WNT)/β-Catenin, respectively.^94,95 Whole-genome sequencing of circular RNAs like circSLC4A7 has unveiled their interaction with Heat Shock Proteins 90 (HSP90), activating the Notch1 pathway and influencing gastric CSC progression.⁹⁶ In addition to solid tumors, leukemia stem cells have also been found to undergo specific ribosomal RNA methylation (2’-O-methylation) modifications. This methylation pattern can reshape ribosome function and protein translation, allowing leukemia stem cells to preferentially translate amino acid transporters, which facilitates the cells’ uptake of amino acids in the environment, thus improving the self-renewal and function of leukemia stem cells.⁹⁷ Sequencing studies were typically used to purify CSCs, focusing on molecular markers, which are involved in asymmetrical division, migration, and signaling pathways. Among them, MYC, Octamer-Binding Transcription Factor 4 (OCT4), Sex Determining Region Y 2 (SOX2), and ALDH are several key genes related to CSCs that are often focused on in research.^98,99,100

The complexity of tumors transcends single malignant cells, encompassing a diverse array of cell types such as immune and stromal cells, thereby exhibiting significant intra- and inter-tumoral heterogeneity.¹⁰¹ While traditional transcriptomic analyses have provided valuable insights into tumor growth and evolution, they may overlook signals from crucial cell groups or states.¹⁰² These pivotal cellular states, including CSCs and immune cells relevant to treatment responses, are essential for understanding and treating tumors. To surmount this limitation, scientists are adopting advanced technologies like single-cell RNA analysis (scRNA-seq) and spatial transcriptomics. These methods offer a refined understanding at the cellular and molecular levels, unveiling new dimensions of complex interactions and heterogeneity within tumors, thereby opening new avenues for cancer research and therapeutic strategy development.¹⁰³

Research into malignant gliomas has been at the forefront of single-cell analyses of brain tumor.¹⁰⁴ Utilizing scRNA-seq, researchers have uncovered a spectrum of stemness and differentiation potential in primary glioblastoma cells, revealing the importance of expression programs like POU domain, class 3, transcription factor 2 (POU3F2), Nuclear Factor I A (NFIA), and NFIB in regulating stem-like phenotypes.¹⁰⁵ Similar analyses of IDH mutant oligodendrogliomas and astrocytomas have disclosed comparable developmental hierarchies and gliogenic differentiation lineages, supporting the CSCs model. The model posits that the majority of cancer cells are well-differentiated, maintaining oligodendrocyte-like or astrocyte-like lineages, with a subset of undifferentiated cells exhibiting stem/progenitor traits.^106,107 Interestingly, higher tumor grades are associated with an enrichment of proliferative stem-like glioma cells, suggesting a significant role for a minority of cancer cells in the growth and progression of IDH mutant gliomas.^108,109 However, in primary H3K27M gliomas, lower differentiation correlates with a higher proportion of stem-like cells, indicating greater tumorigenic potential.¹¹⁰ Copy number variation (CNV) subclones and expression profiles inferred from scRNA-seq can also be used to study the relationship between genetic subclones and cellular state diversity within tumors.^111,112 In IDH1 or IDH2 mutant human oligodendrogliomas, different CNV subclones exhibit similar cellular hierarchies, suggesting that cellular status is primarily determined by developmental programs.¹⁰⁶ In contrast, IDH wild-type glioblastomas are characterized by four plastic and highly malignant cellular states, including neural progenitor cells (NPC-like), oligodendrocyte precursor-like cells (OPC-like), astrocytes like (AC-like) and mesenchymal-like (MES-like), and these states are not strictly determined by the CNV pattern.¹¹³

Beyond gliomas, CSCs-like subpopulations have been identified in other solid tumor types. In advanced prostate cancer, the growth of CSCs correlates with diminished androgen response and enhanced expression of cell cycle-related genes, promoting androgen-independent plasticity.¹¹⁴ In breast cancer, mesenchymal/stem-like tumor cells are present in patients who respond to Epidermal Growth Factor Receptor (EGFR) inhibitors, and an EGFR-high-expressing subpopulation displays enhanced stem-like characteristics, reflecting an EGFR-dependent hierarchy.¹¹⁵ Chung et al. found characterized expression features promoting metastatic progression in rare subgroups of primary triple-negative breast cancer via scRNA-seq, uncovering pronounced epithelial-mesenchymal transition (EMT) and stemness traits driving tumor advancement and metastasis.¹¹⁶ Similarly, metastatic breast cancer cells display overarching EMT and stemness characteristics, though with distinct marker gene expressions.¹¹⁷ The scRNA-seq analysis of hepatocellular carcinoma also reveals heterogeneity in phenotype, function, and transcriptome of CSCs.¹¹⁸ Velten et al. combined scRNA-seq with lineage tracing using nuclear and mitochondrial somatic mutations to identify leukemia stem cell gene expression programs in AML, marked by transcriptional dysregulation and co-expression of stem and myeloid priming genes.¹¹⁹ Differentiated AML cells express various immunoregulatory genes, inhibiting T-cell activity in vitro.¹²⁰ In chronic myelogenous leukemia (CML), researchers identified unique molecular features of CSCs, revealing heterogeneity. A CSC subgroup in CML, characterized by distinct molecular traits, persists selectively during prolonged tyrosine kinase inhibitors (TKI) treatment, featuring quiescence-related gene expression and dysregulated genes and pathways.¹²¹ These insights deepen the understanding of cellular and molecular mechanisms underlying CML treatment resistance. Unlike other tumor cells, CSCs in head and neck tumors show extreme genomic instability, including chromosomal gains and losses.¹²² Ren et al. proposed a differentiation trajectory from CSC-like ductal cells to invasive ductal cells in pancreatic cancer, identifying five genes significantly associated with CSC prognosis.¹²³ Wu et al. found common mutations in signaling pathway genes in different colorectal cancer cell clones, providing evidence for monoclonal CRC origin and subsequent subclonal evolution.^124,125 Leung et al. demonstrated, through single-cell sequencing, exome sequencing, and targeted deep sequencing, that colorectal cancer metastasis follows a late dissemination model, with tumor cells evolving and acquiring mutations that enable clonal spread at the primary site.¹²⁶

Research into CSCs is an evolving and deepening field. Despite challenges such as the lack of a clear definition for CSCs and the need for integrating various experimental methodologies, continuous research and technological advancements hold promise for a deeper understanding of cancer’s essence. This progress is anticipated to unveil novel strategies for cancer treatment, navigating through the complexities of tumor biology to illuminate new pathways for intervention.

Overview of cancer stem cells

Origin hypothesis of cancer stem cells

Differentiated cells

Dedifferentiation is a reversed process by which differentiated cells return to a less differentiated stage within the same lineage.¹²⁷ Dedifferentiation represents a common biological phenomenon in several physiological processes, such as cardiac regeneration and wound healing.¹²⁸ By dedifferentiation, cells can gain stem-like properties, such as self-renewal and pluripotency, so this process also implies CSC formation and tumorigenesis (Fig. 2).¹²⁸ Taking advantage of scRNA-seq and lineage tracing techniques, a study reveals a trajectory of dedifferentiation that PROM-1⁺ hepatocellular CSC follows, which strongly supports the role of dedifferentiation in CSC formation.¹²⁹

Fig. 2

Origin, formation and/or maintenance of CSCs. CSCs originate from differentiated normal/cancer cells, stem/progenitor cells, or cell-cell fusion of cancer cells with stem cells or cancer cells with differentiated cells. The microenvironment of the CSC niche plays an essential role in the formation and maintenance of CSCs. MSCs, TAMs, MDSCs, and CAFs can secrete cytokines and chemokines that induce and/or maintain stem-like properties of cancer cells. Besides, CAFs can also modulate stemness by secreting EVs, and MSCs can regulate stemness through direct contact with CSCs. Finally, hypoxia and high nitric oxide (NO) concentration also support the CSC niche

Genetic or post-transcriptional alteration can lead to dedifferentiation of normal cells into CSCs. The combined loss of p16^INK4a and p19^ARF along with EGFR activation triggers the dedifferentiation of astrocytes and the genesis of glioblastoma.¹³⁰ Besides astrocytes, terminally differentiated neurons can also undergo dedifferentiation to neural stem cell following shNF1-shp53 virus injection and induce gliomas.¹³¹ In intestinal crypts, ablation of Leucine-Rich Repeat-Containing G-Protein Coupled Receptor 5 (LGR5⁺) stem cells leads to dedifferentiation of daughter crypt cells and replenishment of stem cells, which is dependent on the transcription factor Achaete-Scute Homolog 2 (ASCL2).¹³² Mature pigment-producing melanocytes can also dedifferentiate into tumor progenitors of cutaneous melanoma induced by mutant BRAF.¹³³ PGC7 induces promoter demethylation of transcription factors, such as GLI1 and MYCN, and facilitates dedifferentiation of hepatocellular cancer cells.¹³⁴ Downregulation or loss of Bcl3 leads to dedifferentiation of pancreatic cancer cells and expansion of the CSC population.¹³⁵ A study shows that RNA slicing also plays an important part in tumor cell dedifferentiation, as the splicing factor SRSF1 maintains stemness in colorectal cancer.¹³⁶ Also, downregulated miR-613 expression is associated with liver cell dedifferentiation and CSC formation, which is mediated by increased SOX9 expression.¹³⁷

Several signaling pathways are involved in the regulation of dedifferentiation in terms of CSC formation. For instance, activation of the WNT pathway and loss of Sterile Alpha Motif Domain (Smad4) drive differentiated intestinal epithelium to stem cell-like status and initiate colon cancer growth.¹³⁸ The WNT pathway is also associated with the dedifferentiation of breast cancer bone metastases into CSCs.¹³⁹ Activation of NF-κΒ leads to enhancement of the WNT signaling, which further supports dedifferentiation of intestinal villus cells and acquisition of stem cell markers and tumor-initiating capabilities of these cells.¹⁴⁰ The activation of the Transforming Growth Factor β (TGF-β) signaling pathway can also convert colorectal cancer cells into CSCs, which is dependent on the transcription factor Twist-Related Protein 1 (TWIST1).¹⁴¹ The activation of Hypoxia-Inducible Factor 1α (HIF-1α)/Notch pathway leads to dedifferentiation of pancreatic cancer cells and formation of stem-like cells.¹⁴² Extracellular Signal-Regulated Kinase (ERK) inhibition promotes cancer cell dedifferentiation and expands the CSC population in non-small cell lung cancer (NSCLC).¹⁴³ Fibroblast-released IL-6, activin-A, and Granulocyte Colony-Stimulating Factor (G-CSF) induce Signal Transducer And Activator Of Transcription (STAT3) and Smad activation, which consequently activate the WNT, Notch, and hedgehog pathways and induce dedifferentiation of lung carcinoma cells.¹⁴⁴

Environmental factors, including hypoxia, cytokines, and NO, also relate to the dedifferentiation in CSC formation. Under hypoxia, glioma, lung cancer, and hepatoma cells express high levels of stemness-associated transcription factors and CSC markers.¹⁴⁵ In lung adenocarcinoma, CSCs can be formed through dedifferentiation induced by Insulin-Like Growth Factor-II (IGF-II) secreted from cancer-associated fibroblasts, where the transcription factor Forkhead Box M1 (FOXM1) is involved.^146,147 In nasopharyngeal carcinoma cells, Epstein-Barr virus (EBV) latent protein Latent Membrane Protein 1 (LMP1) induces dedifferentiation to form stem-like cells through transcriptional inhibition of CCAAT Enhancer Binding Protein Alpha (CEBPA).¹⁴⁸ Exposure to progranulin leads to dedifferentiation of breast cancer cells and expansion of the CSC population.¹⁴⁹ Exosomes from GSCs can cause dedifferentiation of surrounding non-CSCs by activating the Notch1 pathway.¹⁵⁰ Finally, by stabilizing OCT4, a crucial transcription factor in CSCs, NO can induce the formation of lung or endometrial CSCs from differentiated cells.^151,152

Non-malignant stem/progenitor cells

CSCs are generally functionally and structurally like normal stem cells, such as the ability of self-renewal and multipotent differentiation and similar transcriptional profiles.¹²⁸ For instance, prostate CSCs share a conserved transcriptional program with normal prostate basal stem cells.¹⁵³ Also, based on results from immunohistochemistry and double-fluorescence immunostaining, hepatocellular cholangiocarcinoma shares a similar set of markers with hepatic progenitor cells.^154,155 These observations indirectly support that these CSCs can derive from tissue-resident stem/progenitor cells (Fig. 2).

Several studies have succeeded in the transformation from induced pluripotent stem cells (iPSCs) to CSCs. When cultured in a conditioned medium of mouse Lewis lung cancer,^156,157 pancreatic carcinomas,^158,159 hepatocellular carcinomas,¹⁶⁰ or prostate cancer cell lines,¹⁶¹ iPSCs obtained CSC features and higher tumorigenicity in vivo. Similar results can be obtained by culturing iPSCs with Lewis cell-derived extracellular vesicles¹⁶² or recombinant human Fibroblast Growth Factor 2 (FGF2).¹⁶³ Moreover, mouse embryonic stem cells can also get converted into CSCs in conditioned medium from mouse Lewis lung cancer or melanoma cells.¹⁶⁴ These studies provide evidence that CSCs can be induced from normal stem cells, although iPSCs are not equivalent to normal somatic stem cells. Ewing Sarcoma Breakpoint Region 1 (EWS)-Friend Leukemia Integration 1 (FLI-1) fusion gene and miR-145 in human pediatric mesenchymal stem cells drive their reprogramming into CSCs by increasing the expression of SOX2.¹⁶⁵ Although iPSCs do not fully represent adult stem cells, these studies show a possibility that CSCs can be induced from normal stem cells.

Some studies give more direct evidence that CSCs may originate from non-malignant adult stem cells. For instance, following Adenomatous Polyposis Coli (Apc) depletion, LGR5⁺ intestinal stem cells transform into CSCs, fueling the unimpeded growth of adenomas.¹⁶⁶ Hepatocellular CSCs are found to derive from hepatic progenitor cells when the TGF-β or the WNT pathway is constantly activated in mice.^167,168 Mouse primary hepatic stem/progenitor cells, when transduced with oncogenic genes, acquire CSC markers, self-renewal ability, and pluripotency.^169,170 It is also notable that lineage-committed hepatoblasts and differentiated adult hepatocytes also gain stemness after the process.¹⁶⁹ Finally, following deletion of Brca1, mouse mammary epithelial luminal progenitors get the ability to generate basal-like breast tumors.¹⁷¹

Cell-cell fusion

Cell-cell fusion is commonly involved in several physiological processes, including fertilization, muscle maturation, development of bones and placenta, and immune responses.^172,173 For instance, a sperm and an egg fuse into a fertilized egg, a set of mononucleated myoblasts form a string of muscle fibers, trophoblasts fuse to form syncytiotrophoblasts, and several macrophages combine to make giant cells.^172,173 Bone marrow cells can adopt the phenotype of other cells, such as embryonic stem cells, through cell-cell fusion.¹⁷⁴ Similarly, cancer cells can also fuse with other cancer cells or non-malignant cells, forming tumor hybrid cells.^175,176 Particularly, fusion of cancer cells with non-malignant cells often gives rise to their malignancy and potentiates tumor heterogeneity.¹⁷² For instance, melanoma cells can gain phenotypes of fibroblasts and monocytes by cell-cell fusion,¹⁷⁷ and co-grafting of bone marrow-derived mesenchymal/stromal stem cells (BM-MSCs) and murine prostate cancer cells in vivo leads to enhanced tumor growth by cell-cell fusion.¹⁷⁸ Clinically, a study suggests that the number of tumor hybrids (fusion of cancer cells and leukocytes) in peripheral blood correlates with cancer stage and patients’ survival.¹⁷⁹

Given these properties of tumor hybrid cells, several studies support that cell-cell fusion can be one of the origins of CSCs (Fig. 2). MSCs have been recognized as an essential component in the tumor microenvironment and actively participate in tumor progression.¹⁸⁰ Fusion between BM-MSCs or embryonic stem cells and cell lines of breast cancer, NSCLC, liver cancer, ovarian cancer, or gastric cancer upregulated their stem cell markers and enhanced tumorigenicity abilities in vitro.^{181,182,183,184,185,186} Similarly, hybrids from human/mice liver cancer, breast cancer, or lung cancer cells and BM-MSCs exhibited mesenchymal features and demonstrated enhanced stemness and metastatic capabilities in vivo.^187,188,189 Besides solid tumors, cells of hematological malignancies, such as multiple myeloma, can also fuse with BM-MSC to gain stemness and stronger resistance to treatments.¹⁹⁰ However, the fusion of BM-MSCs with cancer cells does not always produce CSCs, such as esophageal CSCs,¹⁹¹ indicating that cell fusion is not the only mechanism of CSC formation at least in certain tumor types. Plus, human umbilical cord MSCs can also fuse with gastric cancer cells to enhance cancer proliferation, migration, and stemness.¹⁹²

In addition to MSCs, cancer cells can also fuse with other types of non-malignant cells and gain stem-like properties in the process. CD34⁺ liver CSCs can be formed by fusion of hepatobiliary stem/progenitor cells with CD34⁺ hematopoietic precursor-derived cells,¹⁹³ suggesting that tissue-resident stem cells are also able to fuse with other cells and form CSCs. The fusion of prostate cancer cells with muscle cells increased the number of CD133⁺ stem-like cells.¹⁹⁴ Hybrids from fusions of non-malignant human breast epithelial cells and breast cancer cells exhibit CSC properties,¹⁹⁵ which is dependent on the transcription factor Zinc Finger E-Box Binding Homeobox 1 (ZEB1).¹⁹⁶ Hybrids from tumor-associated macrophages and breast cancer cells also exhibit CSC phenotype and promote cancer metastases in a mouse model.¹⁹⁷ Furthermore, CSCs can also fuse with other cells and obtain higher malignancy. For instance, hybrids of CSCs and monocytes gain highly invasive capacities,¹⁹⁸ and those of BM-MSCs and SU3-RFP human glioma stem cells (GSCs) exhibited enhanced angiogenic effects compared to the parental cells.¹⁹⁹

Additionally, a study shows that the fusion of two human lung fibroblast cell lines, E6E7 and RST, results in hybrids with elevated ALDH activity, which is a CSC marker.²⁰⁰ This study suggests that hybrids from two non-malignant cells may also lead to CSC formation.

Environmental factors in cancer stem cell formation and/or maintenance

It is commonly believed that CSCs reside in niches of tumors, and their microenvironment, which is generally characterized by hypoxia, aberrant angiogenesis, and chronic inflammation, has great impacts on the formation and maintenance of CSCs (Fig. 2).²⁰⁰

Hypoxia/angiogenesis

Hypoxia and aberrant angiogenesis have been identified as two crucial features of the tumor microenvironment.^201,202 Tumor angiogenesis can be a consequence of hypoxia since hypoxia serves as a potent stimulus of Vascular Endothelial Growth Factor (VEGF) production, and disorganized vessels in tumors can aggravate hypoxia and vice versa.²⁰³ Under hypoxia, the HIF system is activated.²⁰⁴ And upregulated HIFs can promote the dedifferentiation of pancreatic cancer cells by activating the Notch pathway¹⁴² or that of melanoma cells by upregulating OCT4 expression.²⁰⁵ Also, hypoxia induces upregulation of SOX2, OCT4, KLF-4, Nanog, and Lin-28A, which are transcription factors contributing to dedifferentiation, and formation of stem-like cells in glioma, lung cancer, and hepatoma cells.¹⁴⁵ Likewise, VEGF, an essential pro-angiogenic molecule, can interact with the VEGFR family or the neuropilin (NRP) family and promote stemness of skin/breast cancer cells and extend the CSC pool in the niche.^206,207 Hypoxia can also indirectly regulate the stemness of cancer cells by altering functions of surrounding stromal cells, such as cancer-associated fibroblasts (CAFs)^208,209 and myeloid-derived suppressor cells (MDSCs).²¹⁰

Cancer-associated fibroblasts

CAFs are a group of interstitial cells of a mesenchymal lineage that are not epithelial, endothelial, or immune cells found in or adjacent to tumors.²¹⁰ Compared to normal fibroblasts, CAFs are hyperproliferative and have a unique secretion pattern that contributes to tumor angiogenesis and metastases.²¹¹ Some believe that CAFs can also promote dedifferentiation and CSC formation by activating the WNT or the Notch pathway.⁴ WNT5a from surrounding CAFs induce dedifferentiation of ovarian cancer and gastric cancer cells and maintain the undifferentiated state of ovarian CSCs by activating a noncanonical WNT pathway.^212,213

Moreover, several studies show that secretomes from CAFs promote the stemness of cancer cells. Head and neck squamous cell carcinoma and scirrhous gastric cancer cells express higher CSC markers when cultured in a CAF-derived conditioned medium compared to that from normal fibroblasts.^214,215 CAFs secret IL-6, activin-A, G-CSF, and IGF-II that mediate the dedifferentiation of lung cancer cells into CSCs.^144,146 Periostin from podoplanin-positive CAFs facilitates stem-like properties of gastric cancer cells by activating the Focal Adhesion Kinase (FAK)/Yes-Associated Protein (YAP) signaling.²¹⁶ Leukemia Inhibitory Factor (LIF) and Gremlin 1 from CAFs can promote Nanog and OCT4 expression along with stem cell markers CD24⁻/CD44⁺ in breast cancer cells.^217,218 CAF-derived HGF and IL-6 enhance the stemness of CD24⁺ liver cancer cells by activating the STAT3 pathway,²¹⁹ and IL-6 from CAFs also induce Chromobox 4 (CBX4) expression, which is a CSC phenotype regulator, in skin squamous cell carcinoma.²²⁰ Additionally, Matrix Metallopeptidases (MMPs) from activated CAFs induce EMT and enhance the stemness of prostate cancer cells.²²¹ However, when liver cancer or pancreatic ductal adenocarcinoma cell lines are cultured in a conditioned medium from CAFs, they have distinct expressions of CSC markers and aggressive phenotypes in a cell-line dependent manner,^222,223 suggesting that effects of CAFs on stemness of cancer cells may vary depending on types and subtypes of cancers.

Moreover, CAF-derived exosomes also participate in CSC formation and/or maintenance. MiR-146a-5p in CAF-derived exosomes can promote the stemness of bladder cancer cells by activating the Mammalian Target Of Rapamycin (mTOR) pathway.²²⁴ CircHIF1A in exosomes from hypoxic CAFs can sponge miR-580-5p in breast cancer cells and increase their expression of CD44, which is a CSC marker for breast cancers.²⁰⁸ And small extracellular vesicles with low level of miR-7641 are associated with activation of the HIF-1α pathway, which promotes stemness of breast cancer cells.²²⁵ Likewise, loss of miR-34c-5p in exosomes from CAFs maintains the stemness of laryngeal cancer cells.²²⁶ In summary, CAFs regulate the stemness of cancer cells through paracrine mechanisms that involve cytokines or extracellular vesicles.

Mesenchymal stem cells

During tumor initiation and development, MSCs are believed to be constantly recruited to the tumor, making them an unneglectable group of cells in the tumor microenvironment (TME).²²⁷ Coculturing BM-MSCs with hypopharyngeal or prostate cancer cells induces expression of stemness markers in these cells,^228,229 indicating MSCs can be not only the origin of CSCs but also an ally in CSC formation and maintenance. MSCs can induce fatty acid oxidation by upregulating mitofusin 2, a mitochondrial fusion-inducible factor, Carnitine Palmitoyl Transferase 1 (CPT1), and lncRNA MACC1-AS1 in gastric cancer, which finally leads to enhancement of stemness.^230,231,232 Plus, the direct contact between MSCs and breast cancer cells upregulates the miR-199a in cancer cells, which subsequently represses the transcriptional regulator forkhead box P2 (FOXP2) and finally leads to higher stemness.²³³

Culturing colon cancer or melanoma cells in an MSC-derived conditioned medium increased the expressions of stemness markers of the cancer cells,^234,235 indicating the secretomes of MSCs can induce stemness of cancer cells. Platelet-Derived Growth Factor (PDGF) from MSCs gives rise to ALDH⁺ CSCs in a model of ovarian malignant ascites.²³⁶ MSC-derived IL-8 can induce stem-like properties of gastric cells and blocking Programmed Cell Death-Ligand 1 (PD-L1) undermines this effect by reducing the expression of the transcription factor CTCF.²³⁷ Conditioned medium or just IL-6 from MSCs increases expression of stemness markers, such as OCT4, Nanog, and SOX2, via NF-κB activation in osteosarcoma cells.^238,239 Prostaglandin E2 from MSCs also increases the level of ALDH-high CSCs in human colorectal carcinoma cells by activating the β-Catenin pathway.²⁴⁰ Exosomes from p53 deficient mouse BM-MSCs can internalize UBR2 into gastric cancer cells and increase their expression of CSC markers via the WNT/β-Catenin pathway.²⁴¹ Adipose- and placenta-derived MSCs increase the proportion of CD133⁺/CD44⁺ colon CSCs via the IL-8/Mitogen-Activated Protein Kinase (MAPK) pathway.²⁴²

However, MSCs do not always facilitate the stemness of cancer cells. For instance, endometrium‑derived MSCs suppress the stemness of endometrial cancer by inhibiting the WNT/β‑Catenin signaling pathway.²⁴³ MSC-derived exosomes reduce the proliferation, migration, invasion, angiogenesis-stimulating, and self-renewal abilities of hepatocellular CSCs by inducing ERK phosphorylation²⁴⁴ or pancreatic CSCs by inhibiting the β-Catenin signaling.²⁴⁵ Altogether, MSCs are an important component in the TME, some of which can promote the stemness of their surrounding cancer cells through their secretomes or exosomes, while some MSCs may reduce the stemness of the surrounding cells.

Macrophages

Tumor-associated macrophages (TAMs) represent one of the most abundant groups of immune cells in tumors.²⁴⁶ Based on their immune functions, TAMs can be simply classified into the M1 proinflammatory phenotype and the M2 anti-inflammatory phenotype, although this classification neglects the great diversity of TAMs.²⁴⁶ ScRNA-seq shows that the maintenance of stemness of hepatocellular cells is mainly based on M2 polarization rather than the recruitment of TAMs.²⁴⁷ In the spleen of a murine chronic myeloid leukemia model, red pulp macrophages provide a niche for leukemia stem cells and support their stemness.²⁴⁸

Coculturing of TAMs and pancreatic cancer cells or culturing oral squamous cell carcinoma in an M2 macrophage-derived conditioned medium promotes their expression of stemness-related genes.^249,250 Conditioned medium from TAMs promotes stemness of lung cancer cells by upregulating Ubiquitin-Specific Peptidase 17 (USP17), which subsequently disrupts the TNFR-Associated Factor (TRAF) 2/TRAF3 complex.²⁵¹ These results suggest that the secretomes of TAM are essential in the process. TAM-derived TGF-β1 promotes stem-like properties of esophageal squamous cancer cells,²⁵² glioblastoma cells,^253,254 pancreatic cancer cells,²⁵⁵ prostate cancer cells,²⁵⁶ hepatocellular cancer cells,²⁵⁷ and breast cancer cells.²⁵⁸ Additionally, M2-TAMs secretory pleiotrophin enlarges the CSC group in human diffuse large B lymphoma by upregulating the β-Catenin expression.²⁵⁹ TAM-derived interleukin-1β, TNF-α, and IL-6 promote stemness of Doublecortin Like Kinase 1 (Dclk⁺) colon tuft cells and initiate tumor growth.²⁶⁰ TAM-derived IL-6 also enriches breast CSCs by activating the STAT3 signaling,²⁶¹ and it also activates WNT and promotes stemness of ovarian cancer cells in 3D engineered microenvironments.²⁶² M2-TAM-derived IL-8 induces stemness of ovarian cancer cells in vitro by activating the STAT3 pathway.²⁶³ Macrophages can also secrete IL-10 to promote stemness of NSCLC cells by activating the JAK1/STAT1/NF-κB/Notch1 signaling.²⁶⁴ IL-33 can also recruit macrophages into the TME and stimulate the secretion of prostaglandin E2, which subsequently supports stemness of colon cancer cells.²⁶⁵ Inhibitor Of Differentiation 1 (ID1) from TAMs can inhibit transcription of two stemness inhibitory factors, SerpinB2 and CCL4, and lead to stemness enhancement.²⁶⁶ Besides, LSEC in TAMs can enhance breast cancer stemness by binding to Butyrophilin Subfamily 3 Member A3 (BTN3A3) on breast cancer cells,²⁶⁷ suggesting that direct cellular contact of TAMs and cancer cells can also enhance stemness. TAM secretory S100 calcium-binding protein can induce stemness of hepatocellular cancer cells by activating the NF-κB pathway in a calcium-dependent manner.²⁶⁸ M2-TAM secretory VEGF or EGF promotes stemness of breast cancer cells by activating the VEGF/NRP-1/GTPase Activating Protein And VPS9 Domains 1 (GAPVD1) axis or the EGFR/STAT3/SOX2 signaling, respectively.^207,269 M2-TAM-derived IGF-1 and IGF-2 promotes thyroid cancer stemness by activating the PI3K/AKT/mTOR pathway.²⁷⁰ Macrophage-derived glycoprotein nonmetastatic B induces the production of IL-33, an IL-1-like cytokine, via CD44 in a mouse lung cancer model, which in turn induces the CSC properties of these cells.²⁷¹ Likewise, in head and neck squamous cell carcinoma, macrophage-derived hyaluronic acid (HA) induces activation of the PI3K/Eukaryotic Translation Initiation Factor 4E-Binding Protein 1 (EIF4EBP1)/SOX2 signaling via CD44 and increases the density of CSCs in vitro.²⁷² M2-TAM-derived exosomal miR-27a-3p and the miR-17-92 cluster promote stemness of hepatocellular cancer cells by upregulating Thioredoxin-Interacting Protein (TXNIP) or disturbing the balance of the TGF-β1/Bone Morphogenetic Protein 7 (BMP-7) pathways.²⁷³

Several chemokines and chemokine ligands are involved in TAM-induced CSC formation or maintenance as well. TAM-derived Chemokine C-C Motif Ligand 2 (CCL2) activates AKT and increases the expression of β-Catenin in triple-negative breast cancer cell lines, which eventually induces their CSC properties.²⁷⁴ CCL8 promotes stemness of glioblastoma cells by activating ERK1/2.²⁷⁵ CXCL12 and TGF-β from M2 TAMs elevate DNA Topoisomerase II Alpha (TOP2A) expression and enhance stemness of hepatocellular cancer cells via the TOP2A/β-Catenin/YAP1 axis.²⁷⁶ Also, CXCL12 from M2 macrophages activate the WNT/β-Catenin pathway to facilitate stemness of colorectal cancer cells.²⁷⁷ Macrophage-derived CXCL7 fosters glioma stemness.²⁷⁸ Macrophage secretory IL-1β and CCL18 facilitate stemness of head and neck squamous carcinoma.^279,280

Notably, M1 macrophages can also induce a subgroup of CD44^high/CD24^−/low or ALDH1⁺ breast CSCs in vitro, although prolonged coculture finally endows the macrophages with M2 properties.²⁸¹ A study shows that breast CSCs respond more robustly to monocytes/macrophages than do differentiated non-stem cells through a juxtracrine mechanism, indicating monocytes/macrophages play an essential role in maintaining CSC niches.²⁸²

Myeloid-derived suppressor cells

MDSCs are a heterogeneous group of immune cells from the myeloid lineage that exert immunosuppressive effects.²⁸³ Coculture of ovarian cancer cells and MDSCs increases the expression of colony-stimulating factor 2 that activates the STAT3 and leads to upregulation of stemness markers.²⁸⁴ MDSCs can upregulate the expression of miR101 in ovarian cancer cells and subsequently repress the core-pressor gene C-terminal Binding Protein-2 (CtBP2), which restrains cancer stemness.²⁸⁵ Granulocytic MDSCs trigger piRNA-823 expression that promotes DNA methylation and maintains the stemness of multiple myeloma CSCs.²⁸⁶ Hypoxia can induce increased secretion of exosomes containing S100 Calcium-Binding Protein A9 (S100A9) from granulocytic MDSCs, which leads to enhanced stemness of colorectal cancer cells.²¹⁰ MDSCs cultured in CAF-derived conditioned medium express a higher level of 5-Lipoxygenase (5-LO) that induces synthesis of leukotriene B4, which finally results in enhanced stemness of intrahepatic cholangiocarcinoma.²⁸⁷ MDSCs also endow stemness to breast cancer cells by secretory IL-6 and NO that activate the STAT3 and Notch pathways, respectively.²⁸⁸ The Notch pathway can also be activated by granulocytic MDSCs and contributes to stem maintenance in esophageal squamous cell carcinoma,²⁸⁹ and the STAT3 pathway is also activated in pancreatic cancer cells in the presence of MDSCs.²⁹⁰ MDSC-derived PGE2 also increases the stem cell-like properties in epithelial ovarian cancer.²⁹¹

Besides, NO is frequently upregulated in cancers.²⁹² NO disturbs the ubiquitin-mediated prosomal degradation of OCT4 and induces dedifferentiation of human lung cancer cells.¹⁵¹ NO also promotes stem-like properties of mouse glioma cells by activating the Notch pathway.²⁹³ Ionizing radiation is one of the inducers of CSCs’ formation across several cancer types,²⁹⁴ which will be discussed in more detail in the CSCs and sensitivity/resistance to radiotherapy section.

Features of cancer stem cells

Self-renewal and pluripotency

Since the first identification of CSCs in 1997,² self-renewal and pluripotency have been considered two essential features of CSCs. This discovery comes from the observation that ALL cells are organized hierarchically, with a subset of cells that can replicate themselves and give rise to other malignant lineages, mimicking normal hematopoietic stem cells.² In solid tumors, CSCs were first identified in breast cancer, in which the CD44⁺ CD24^−/low lineage-cells underwent self-renewal and differentiation processes.²³ These two properties of CSCs are also the basis to explain the formation of intratumoral heterogeneity in the CSC hypothesis or the hierarchical model of tumorigenesis⁴.

Cancer stem cells in cancer development

Based on the hierarchical model of carcinogenesis or the classical CSC hypothesis, CSCs, originating from normal stem cells or progenitors, are the cellular origins of cancers that can self-renew and give rise to the cellular hierarchies that explain the intratumoral heterogeneities.²⁹⁵ Nevertheless, the observation that some cancer cells can interchange between differentiated states and stem-like states does not favor this hypothesis.²⁹⁶ Plus, some cancers do not follow the CSC model.²⁹⁷ Therefore, although the terms tumor-initiating cell (TIC) and CSC have been used interchangeably, CSCs are not necessarily the cell origin of cancers based on the cellular plasticity model.⁴ That is, some malignant differentiated cells with oncogenic mutations can undergo dedifferentiation and form stem-like cells that cause intratumoral heterogeneities.⁴ However, this does not eliminate the role of CSCs in cancer initiation supported by many studies. AML cells originate from a subgroup of stem-like cells, as mentioned above.² LGR5⁺ intestinal crypt stem cells, upon oncogenic mutations, serve as cells of origin of intestinal cancer.¹⁶⁶ Injection of CD133⁺ human brain CSCs into non-obese diabetic, severe combined immunodeficient mice causes tumor formation, while that of CD133⁻ does not.²⁴ Conversely, deletion of SOX2, which is essential in maintaining the stemness of CSCs, decreases the formation of skin squamous-cell carcinoma.²⁹⁸ Also, decreasing MYC activity that sustains stemness of hepatocellular CSCs attenuates hepatocellular carcinoma initiation.²⁹⁹

CSCs are generally characterized by vigorous proliferation.³ Cancer proliferation is heavily dependent on the activation of the AKT, mTOR, and MAPK/ERK, which result in upregulated expression of proteins responsible for the cell cycle.³⁰⁰ The signaling pathways that involve these molecules are also major signaling pathways,⁵ which we will introduce in detail in the following section. Indeed, the acquisition of stemness is usually accompanied by enhancement of proliferation.^{301,302,303,304} Conversely, interventions that inhibit stemness also impair the proliferative potential of the cells.^305,306

Cancer metastasis involves several biological processes that can be summarized into 5 essential steps, including cell escape, intravasation, survival maintenance, extravasation, and outgrowth.³⁰⁷ In epithelial malignancies, the EMT is a crucial event in metastasis.³⁰⁸ And the EMT can generate stem-like cells in human mammary epithelial cells,⁶⁶ indicating a strong correlation between the EMT and CSCs. Molecularly, the WNT/β-Catenin, Notch, PI3K/AKT, hedgehog, and NF-κB signaling pathways are involved in the acquisition of mesenchymal properties of cancer cells. The pathways are also crucial in inducing and maintaining the stemness of CSCs.³⁰⁹

CSCs are often indicated as a reason for multi-drug resistance. This is partially attributed to their capability to maintain quiescence to avoid the therapeutic effects of anti-cancer treatments.³¹⁰ For instance, CD13⁺ hepatocellular CSCs predominate in the G0 phase of the cell cycle and exhibit resistance to 5-fluorouracil treatment, as the mechanism of 5-fluorouracil primarily involves inhibition of DNA replication.³¹¹ In addition, CSCs can reduce intracellular accumulation of therapeutic agents by overexpressing ALDH and ATP-Binding Cassette (ABC) transporters.³¹² They also have better DNA repair capabilities and ROS clearance to avoid apoptosis induced by chemotherapy or radiation therapy stress.³¹³ Finally, CSC supports an immunosuppressive niche that can exclude therapeutic agents and impair the efficacy of immunotherapy.⁴ Detailed mechanisms for CSC-induced chemoresistance, radioresistance, and resistance to targeted therapy and immunotherapy will be introduced in the following sections.

Biomarkers of cancer stem cells

One of the most efficient ways to identify CSCs in tumors is to use biomarkers for CSCs. Based on their cellular distribution, CSC markers can be classified into intracellular markers and cell-surface markers. Intracellular markers include transcription factors that function in the nucleus and markers found in the cytoplasm. Tables 2 and 3 summarize frequently used CSC markers in solid tumors and hematopoietic malignancies, respectively. Among them, generally accepted markers are introduced below.

Table 2 Frequently used cancer stem cell markers for solid tumors

Table 3 Frequently used cancer stem cell markers for hematopoietic malignancies

OCT4, SOX2, and Nanog are the core transcription factors that regulate the embryonic stem cell state.³¹⁴ OCT4, SOX2, and Nanog are encoded by the Sex-Determining Region Y (SRY) gene,³¹⁵ the POU Domain, Class 5, Transcription Factor 1 (POU5F1),³¹⁶ and the Nanog gene,³¹⁷ respectively. They collaborate to positively regulate their promoters, activate the expression of genes necessary to maintain the embryonic stem cell state, and repress the expression of lineage-specific transcription factors.^205,314 Similar stemness-maintaining functions of OCT4, SOX2, and Nanog have also been determined in adult stem cells.^{318,319,320,321} Expression of these transcription factors in cancer also endows stem-like properties to the cancer cells, unsurprisingly making them classical markers for CSCs.^322,323,324 In addition, SALL4, encoded by a member of the Spalt-Like (SALL) gene family, SALL4,³²⁵ is also a transcription factor that regulates embryonic stem cell state by cooperating with Nanog.³²⁶ SALL4 expression is identified in several solid and hematopoietic malignancy types and correlates with CSC properties.³²⁵

Several cytoplasmic proteins are also identified as CSC markers. ALDHs refer to a group of enzymes that catalyze the oxidation of aldehydes to carboxylic acids, which can be further classified into 3 classes in mammals.³²⁷ Physiologically, ALDHs are present in most tissues of humans and have the highest concentration in livers, orchestrating drug metabolism.³²⁸ This also indicates an important role of ALDH in cancer drug resistance.³²⁹ ALDH activity has been considered a marker for not only normal stem cells but also CSCs of solid and hematopoietic malignancies.³³⁰

RNA-binding protein Musashi Homolog 1 and 2 (Musashi-1/2) are encoded by the MSI1 gene and the MSI2 gene, respectively.³³¹ Both are RNA-binding proteins involved in post-transcriptional regulations of gene expressions and expressed in stem cells and progenitors to maintain their self-renewal.³³² Musashi-2 also support hematopoiesis, which makes them a CSC marker in hematopoietic malignancies.^333,334,335

Leucine Zipper-EF-Hand Containing Transmembrane Protein 1 (Letm1) is encoded by the Letm1 gene, which is a transmembrane protein located in the inner membrane of mitochondria and functions as a Ca²⁺/H⁺ antiporter.³³⁶ In gastric, colorectal, and lung cancer, studies reveal a positive correlation between Letm1 and stemness-related signatures.^337,338,339 Furthermore, suppressing or elevating the Letm1 expression leads to inhibited or enhanced stemness of colorectal cancer or osteosarcoma cells, respectively.^340,341

Alpha-Fetoprotein (AFP, α-fetoprotein) is encoded by the AFP gene in humans, which is produced by the fetal liver and the yolk sac. The serum level of AFP peaks during embryogenesis and rapidly decreases after birth but re-increases in the presence of hepatocellular cancer or germ cell tumors, making it an evaluable biomarker for these two types of malignancies.³⁴² Cells with high AFP levels exhibit stem-like properties in pancreatic cancer, cholangiocarcinoma, and hepatocellular cancer, making it a potential CSC marker for these types of cancer.^343,344,345

Polycomb complex protein BMI-1, also known as polycomb group RING Finger Protein 4 (PCGF4) or RING Finger Protein 51 (RNF51), is encoded by the BMI-1 gene. BMI-1 takes part in the repair of DNA double-strand breaks by homologous recombination³⁴⁶ and is essential for self-renewal in stem cells,^347,348 which makes it also a marker for several types of solid tumor and hematopoietic CSCs.

Doublecortin-Like Kinase 1 (Dcamkl-1) is encoded by the DCLK1 gene,³⁴⁹ which is a microtubule-associated protein that was recently revealed to have a role in regulating inflammation.³⁵⁰ Also, a study reports that Dcamkl-1 marks intestinal CSCs but not normal CSCs, making it an ideal marker for colorectal CSCs.³⁵¹

A large variety of cell-surface proteins can be applied as CSC markers for solid tumors. C-X-C Chemokine Receptor Type 4 (CXCR4), also known as CD184, is a CXC chemokine receptor encoded by the CXCR4 gene.³⁵² The ligand for this receptor is CXCL12.³⁵³ CXCR4 is famous for its role as one of the receptors inducing the human immunodeficiency viruses (HIV) infection of T cells.³⁵⁴ CXCR4 is also involved in cancer progression for its role in activating the PI3K/AKT, PLC, hedgehog, ERK1/2, and JAK/STAT pathways.³⁵⁵

LGR5, also known as G-Protein Coupled Receptor 49 (GPR49) or G-Protein Coupled Receptor 67 (GPR67), is encoded by the LGR5 gene.³⁵⁶ LGR5 has been identified as a part of the WNT signaling complex to potentiate the WNT/β-Catenin signaling.³⁵⁷ Given the crucial role of the WNT signaling in cancer stemness, LGR5 has also been identified as a cell-surface marker for several solid tumor types (Table 2).

Epithelial Cell Adhesion Molecule (EpCAM), also known as CD326, is known for its role in cell-cell adhesion in the epithelia,³⁵⁸ but its roles exceed this in cancer. Upon cleavage, the intracellular domain of EpCAM forms a complex with FHL2 and β-Catenin, which, with interaction with Lef1, leads to transcription of oncogenes, such as c-Myc.³⁵⁹ Besides, EpCAM also facilitates EMT by inhibiting E-cadherin.³⁵⁹

CD24, also known as Heat Stable Antigen (HSA), is encoded by the CD24 gene in humans, which also functions as a cell-cell adhesion molecule.³⁶⁰ CD24 also mediates several signaling pathways that could lead to stemness enhancement of tumor cells.³⁶¹ Likewise, CD44, also known as Homing Cell Adhesion Molecule (HCAM) and Phagocytic Glycoprotein-1 (Pgp-1)also induces cell-cell adhesion and interactions.³⁶² It also takes part in activations of PI3K/AKT and Src/MAPK pathways and serves as a c-Met co-receptor.³⁶² Both molecules can individually or combinedly mark CSCs in several solid tumor types. Moreover, the combination of CD44⁺/CD24⁻ also marks CSCs in breast cancer, prostate cancer, head and neck squamous cell carcinoma, and ovarian cancer (Table 2).

CD133, also known as Prominin-1 (PROM1) and encoded by the PROM1 gene, belongs to the pentaspan transmembrane glycoproteins family.³⁶³ CD133 can activate the PI3K/AKT, Src, and β-Catenin signaling intracellularly to participate in cancer progression.³⁶³ CD133 is expressed in a wide range of human tissues and can serve as a CSC marker for various types of solid tumors and hematopoietic malignancies (Tables 2 and 3). The combined use of CD44 and CD133 as CSC markers has been reported in gallbladder cancer.³⁶⁴ Plus, CD44⁺/CD133⁻ and CD44⁻/CD133⁺ cells both can represent CSCs in colorectal cancer.^365,366

The differences between CSC markers for solid tumors and those for hematopoietic malignancies mainly lie in the variation of cell surface markers (Table 3) (Fig. 3). Interleukin-1 Receptor Accessory Protein (IL1RAP), encoded by the IL1RAP gene, is a receptor for interleukin-1.^367,368 It has been identified as a CSC marker for myeloid leukemia.³⁶⁹ Similarly, CD25, a receptor for interleukin-2, and CD123, a receptor for interleukin-3, are also identified as CSC markers for AML or CML.^{370,371,372,373} CD70, expressed on the surface of various cells, and CD27, expressed on the T cell surface, are a pair of costimulatory molecules. The CD70/CD27 signaling is found activated in acute or chronic myeloid leukemia stem cells and contributes to the stemness formation of these cells by activating the WNT pathway.^374,375,376 CD34⁺/CD38⁻ is also identified as a marker for myeloid leukemia stem cells and has been widely used.² Compared to myeloid leukemia, CSC markers for lymphoblastic leukemia are hardly reported. A study suggests that CD90 and CD110 correlate with stemness of ALL cells and might be a CSC marker.³⁷⁷

Fig. 3

Biomarkers for CSCs in solid tumors and hematopoietic malignancies. Biomarkers for CSCs in solid tumors (left), hematopoietic malignancies (right), or both (center). The biomarkers can be classified into cell-surface markers and intracellular markers. Intracellular markers can be further classified into transcription factors that function in the nucleus and molecules that are found in the cytoplasm. Cell-surface markers make up the main differences between markers of solid tumors and those of hematopoietic malignancies

Notably, a single CSC marker or a pair of CSC markers might not be sufficient to identify CSCs. For instance, while CD133⁺, CD166⁺CD44⁺, and CD24⁺CD44⁺ phenotypes of human colorectal cells do not correlate with stem cell properties, these 3 sets of markers are reported as CSC-specific in colorectal cancer.³⁷⁸ Plus, ALDH1 alone does not correlate with stem cell-like features in hepatocellular cancer cells,³⁷⁹ but CD133⁺ALDH⁺ cells are significantly more tumorigenic than their CD133⁻ALDH⁺ or CD133⁻ALDH⁻ counterparts,³⁸⁰ suggesting that combined use of CD133 and ALDH can better distinguish hepatocellular CSCs. Conversely, the absence of a CSC marker does not always indicate the absence of stemness. For instance, CD44⁻ head and neck squamous carcinoma cells also have stem-like features, although CD44 is a CSC marker for this type of cancer.³⁸¹ This phenomenon indicates that these CSCs may have distinct origins. Indeed, in glioblastoma, CD133⁺ and CD133⁻ CSC respectively resemble fetal neural stem cells and adult neural stem cells, both of which exhibit stem-like properties.³⁸² It is also noteworthy that certain CSC markers do not apply to every type of malignancy, even though it expressed in a wide range of tissues. For example, although ALDHs are present in most human tissues and represent a CSC marker for several cancer types, their activities play no functional role in stem cell-like properties in anaplastic thyroid cancer cells.³⁸³ Also, the CSC markers, CD133 and CD44, are generally overexpressed in gastrointestinal stromal tumors (GISTs) and cannot be used to distinguish CSCs from non-CSCs.³⁸⁴

Molecular regulations in CSCs

WNT/β-Catenin pathway

The WNT/β-Catenin signaling pathway, known for its involvement in various physiological processes and diseases, is evolutionarily conserved.³⁸⁵ Recent evidence highlights its crucial role in maintaining the stemness of CSCs. Chen et al. demonstrated its significance in converting mouse-iPSCs into CSCs.³⁸⁶ This pathway regulates stemness in CSCs across diverse cancer types, including lung, liver, thyroid, colorectal, cervical, and glioblastoma. For instance, in cervical cancer, cells with elevated Leucine-Rich Repeat-Containing G-Protein-Coupled Receptor 6 (LGR6) exhibit enhanced stemness, as LGR6 activates the WNT/β-Catenin pathway, forming a positive feedback loop with Transcription Factor 7-Like 2 (TCF7L2).³⁸⁷ Similarly, LSD1 maintains stemness in thyroid cancer by targeting Adenomatous Polyposis Coli 2 (APC2) or indirectly regulating Dickkopf WNT Signaling Pathway Inhibitor 1 (DKK1) via the HIF-1α/miR-146a axis to antagonize the WNT pathway.⁹⁴ In liver cancer, EPHB2 sustains tumor stemness by activating the SRC/β-Catenin cascade. The WNT/β-Catenin pathway, in turn, upregulates EPHB2 expression in a TCF1-dependent manner, forming a positive feedback loop linked to liver CSCs.⁹⁵ Furthermore, non-coding RNAs play a pivotal role in stemness maintenance. For example, Protein Kinase Membrane-Associated Tyrosine/Threonine 1 (PKMYT1) associated lncRNA sponges miR-485-5p to upregulate PKMYT1, inhibiting β-transducin repeat containing protein 1 (β-TrCP1)-mediated β-Catenin degradation and activating WNT signaling in NSCLC stem cells.³⁸⁸ Similarly, in liver CSCs, lncRNA Small Nucleolar RNA Host Gene 5 (lncSNHG5) activates the WNT/β-Catenin pathway by inhibiting Upstream Frameshift 1 (UPF1), sustaining stemness.³⁰² Additionally, overexpression of LINC00839 in GSCs via Methyltransferase-Like 3 (METTL3)-mediated m6A modification enhances c-Src-driven phosphorylation of β-Catenin, activating WNT signaling and promoting stemness.³⁸⁹ Likewise, in colorectal cancer, Sec62, induced by METTL3-mediated m6A modification, enhances β-Catenin nuclear translocation, reducing its ubiquitination degradation and promoting cancer stemness.³⁹⁰

The involvement of the WNT/β-Catenin signaling pathway in CSCs contributes to malignant behaviors such as tumorigenesis and differentiation. Kim et al. showed that colorectal cancer cells expressing CD44 and CD133, markers of CSCs, exhibit strong tumor-initiating effects, accompanied by significant activation of the WNT/β-Catenin pathway.³⁹¹ Furthermore, a CD44⁺Cellular Prion Protein (PrPc⁺) LGR4⁺ CSC subpopulation in colorectal cancer demonstrates high metastatic potential, with LGR4 and PrPC activating the WNT/β-Catenin pathway.³⁹² Far Upstream Element-Binding Protein 1 (FUBP1) upregulation in colorectal cancer activates the WNT/β-Catenin cascade, enhancing stemness and potentially driving tumorigenesis.³⁹³ In breast cancer, Calreticulin (CALR) promotes a stem cell phenotype, with upregulation by HIF-1 activating the WNT/β-Catenin pathway to facilitate tumor initiation.³⁹⁴ Piwi-Like RNA-Mediated Gene Silencing 2 (Piwil2)-overexpressing cervical cancer cells exhibit strong stemness, partly attributed to the WNT/β-Catenin pathway, inhibition of which induces cell differentiation and suppresses tumorigenicity.³⁹⁵

The WNT/β-Catenin signaling pathway in CSCs is involved in the metastasis process. Husain et al. demonstrated that Farnesyl Dimethyl Chromanol (FDMC), an inhibitor of the WNT/β-Catenin pathway, suppresses the stemness and metastatic potential of colorectal CSCs, inducing their apoptosis.³⁹⁶ Colorectal cancer exhibits overexpression of Disheveled3 (DVL3), activating the WNT/β-Catenin/c-Myc/SOX2 signaling cascade, thereby enhancing stemness and metastatic potential.³⁹⁷ In gastric cancer, ST2⁺ serves as a functional marker of CSCs and activates the WNT signaling pathway, promoting metastasis through interaction with BCL-XL.³⁹⁸ Similarly, in pancreatic cancer, upregulated Frizzled-7 (FZD7) promotes CSC phenotype and liver metastasis via the canonical WNT/β-Catenin pathway.³⁹⁹ Additionally, polychlorinated biphenyls 2,3,5-trichloro-6-phenyl-[1,4]-benzoquinone (PCB29-pQ) activates the WNT/β-Catenin pathway, enhancing breast cancer stemness and metastasis.⁴⁰⁰

Most studies have consistently shown a positive correlation between the activation of the WNT/β-Catenin pathway and the malignant behavior of CSCs. However, in radioresistant glioblastoma, the expression of N-cadherin correlates positively with the inhibition of the WNT/β-Catenin signaling pathway. N-cadherin binds to β-Catenin in the cytoplasm, inhibiting neuronal differentiation mediated by the WNT signaling pathway and maintaining a stem-like phenotype.⁴⁰¹ Conversely, in ameloblastoma, β-Catenin expression is negatively correlated with the CSCs’ marker SOX2. Exogenous activation of the WNT/β-Catenin signaling pathway leads to the inhibition of tumor stemness and invasiveness.⁴⁰² These findings suggest that the role of the WNT/β-Catenin signaling pathway in CSCs is complex and may vary across different cancer types and states.

Hedgehog pathway

The classic hedgehog signaling pathway encompasses several cascades. Initially, Patched (PTCH) binds to the hedgehog ligand, relieving the inhibition of Smoothened (SMO). This event further facilitates the dissociation of the Suppressor of Fused (SuFu) from GLI, allowing GLI activators to regulate target genes.⁴⁰³ Yan et al. demonstrated that the interaction between glioma cells and endothelial cells activates the hedgehog pathway, promoting the transformation of glioma cells into a GSC phenotype.⁴⁰⁴ The regulation of the hedgehog pathway in CSCs is intricately linked to their emergence and various malignant biological behaviors.⁴⁰⁵

The hedgehog signaling pathway plays a pivotal role in maintaining the stemness of CSCs. Kelch Domain-Containing 8 A (KLHDC8A) has been identified in GSCs as an upstream factor involved in maintaining stemness by activating the hedgehog signaling pathway through ciliogenesis.⁴⁰⁶ Similarly, Liu et al. revealed the existence of the ISL1/sonic hedgehog (SHH)/GLI1 axis, which promotes GSCs’ stemness.⁴⁰⁷ The elimination of the liver CSCs’ stemness maintainer Ubiquitin-Like With PHD And Ring Finger Domains 1 (UHRF1) results in extensive DNA hypomethylation, ultimately upregulating CEBPA to inhibit the hedgehog pathway.⁴⁰⁸ Additionally, miR-324-5p weakens the function of multiple myeloma stem cells by inhibiting the hedgehog signaling pathway.⁴⁰⁹ Guen et al. demonstrated the connection between the EMT program and stemness, showing that tumor-initiating cells activate the hedgehog pathway through the EMT program to enhance stemness.⁴¹⁰

Furthermore, the hedgehog signaling pathway is implicated in the tumor-initiating function of CSCs. In liver CSCs, the circIPO11/Topoisomerase 1 (TOP1)/GLI1 axis associated with liver cancer initiation has been identified. TOP1 is recruited to the GLI1 promoter by circIPO11 to activate the hedgehog pathway, promoting stemness and tumor initiation.⁴¹¹ Mok et al. revealed that cholesterol-related pathways are significantly upregulated in liver CSCs compared to normal stem cells. The hedgehog signaling pathway is activated in hepatic CSCs as a downstream factor for cholesterol synthesis mediated by the caspase-3/Sterol-Regulatory Element-Binding Protein 2 (SREBP2) axis, ultimately maintaining stemness and tumorigenicity.⁴¹² Similarly, TRNA Methyltransferase 6 (TRMT6)/TRMT61A-mediated N1-methyladenosine methylation in liver CSCs promotes cholesterol metabolism and activates the hedgehog pathway to maintain stemness and enhance tumorigenicity.⁴¹³ In breast CSCs, the activated hedgehog signaling pathway is positively associated with stemness maintenance and tumorigenicity. Overexpression of Tetraspanin-8 (TSPAN8) relieves the inhibition of SMO by PTCH1 and phosphorylates SMO by promoting the binding of PTCH1 to SHH1, recruiting Ataxin-3 (ATXN3) to reduce the ubiquitination degradation of the SHH/PTCH1 complex, ultimately promoting GLI1 transcription.⁴¹⁴ Additionally, Polypeptide N-Acetylgalactosaminyltransferase 1 (GALNT1)-mediated glycosylation of SHH in bladder cancer activates the hedgehog pathway, increasing the stemness and tumorigenicity of CSCs.⁴¹⁵ Immunity may also play a significant role in influencing the effects of the hedgehog pathway in CSCs. IL-25, an intrinsic hedgehog pathway agonist, promotes CSCs’ function, increasing colitis-related tumorigenesis through the accumulation of GLI1.⁴¹⁶

It is widely recognized that CSCs participate in the process of metastasis by activating the hedgehog signaling pathway.⁴¹⁷ Upregulated Ubiquitin-Specific Peptidase 37 (USP37) in breast CSCs binds and stabilizes GLI1 to activate the hedgehog pathway, which further regulates the stemness and metastatic potential of CSCs.⁴¹⁸ GLI1 was identified as a key regulatory gene for colorectal cancer stemness, and activation of the Hh/GLI1 signaling cascade was positively correlated with the invasiveness of colorectal CSCs.⁴¹⁹ Disc Large Homolog 5 (DLG5), an activator of the hedgehog signaling pathway in glioblastoma. DLG5 prevents ubiquitination and degradation of GLI1 to promote the migration and stemness maintenance of GSCs.⁴²⁰

Notch pathway

The Notch pathway comprises several main components: the Notch receptor, Notch ligand, CBF-1, suppressor of hairless, Lag (CSL), DNA binding protein, and downstream target genes. Initially discovered by Drosophila,⁴²¹ the Notch pathway has been shown to play a crucial role in promoting the formation of medulloblastoma stem cells.⁴²² It is implicated in maintaining the stemness of CSCs, as evidenced by its upregulation in supratentorial ependymoma and mucoepidermoid carcinoma, where it correlates positively with the expression of CSCs’ markers.^423,424 Additionally, syndecan-1 in inflammatory breast CSCs acts as a molecular marker maintaining their stem phenotype by activating the Notch pathway.⁴²⁵ While most studies support the positive relationship between Notch pathway activation and stemness maintenance, Högström et al. reported that upregulation of the Notch pathway attenuated the stemness of Prospero Homeobox 1 (PROX1⁺) colorectal cancer cells.⁴²⁶

Moreover, activation of the Notch pathway in CSCs has been associated with metastasis in various tumors such as breast cancer, glioma, renal cancer, and ovarian cancer. In breast cancer, Bone Morphogenetic Protein 4 (BMP-4) promotes stemness and EMT programs by activating the Notch pathway in a Smad4-dependent manner.⁴²⁷ Similarly, Signal Peptide CUB Domain And EGF-Like Domain Containing 2 (SCUBE2) overexpression in breast cancer cells enhances tumorigenicity and metastatic potential by activating the Notch pathway.⁴²⁸ Family With Sequence Similarity 129 Member A (FAM129A) prevents ubiquitination and degradation of Notch1, upregulating the Notch pathway to maintain the stemness and metastatic potential of GSCs.⁴²⁹ Notably, the upregulated Notch pathway in renal CSCs contributes to multiple malignant biological behaviors, including metastasis, stemness maintenance, and tumorigenesis.⁴³⁰ Additionally, glycosyltransferase GnT-III-mediated bisecting glycosylation of Notch1 effectively activates the Notch pathway, supporting stemness maintenance and metastasis.⁴³¹

Activation of the Notch pathway in CSCs is associated with tumorigenesis, differentiation, and immune regulation. Liposarcoma cells with continuous activation of the Notch pathway exhibit overexpression of CSCs’ marker genes, leading to enhanced tumorigenesis compared to cells with normal Notch activity.⁴³² Speckle-Type POZ Protein-Like (SPOPL), a stemness maintainer highly expressed in GSCs, activates the Notch pathway, thereby increasing tumorigenicity.⁴³³ Inhibition of the Notch pathway in GSCs induces significant neuronal differentiation and reduces stemness.⁴³⁴ Similarly, lncRNA FOXD2 Adjacent Opposite Strand RNA 1 (FOXD2-AS1) recruits TATA-Box Binding Protein Associated Factor 1 (TAF-1) to the promoter of Notch1, initiating the Notch signaling pathway in GSCs. Inhibition of FOXD2-AS1 induces the apoptosis and differentiation of GSCs while attenuating their stemness.⁴³⁵ Additionally, the Notch pathway plays a crucial role in immune system regulation. Expression of histone methyltransferase G9a in GSCs positively correlates with stemness characteristics. G9a binds to the Notch suppressor F-Box And WD Repeat Domain Containing 7 (FBXW7), upregulating the Notch pathway and enhancing the expression of PD-L1 in GSCs. This, in turn, weakens the function of T lymphocytes, creating an immunosuppressive microenvironment.⁴³⁶

NF-κB pathway

The NF-κB pathway, consisting of several cascades, is activated when cells encounter various stimuli, leading to the degradation of I-kappa B (IκB) protein by IκB kinase activation. This degradation releases NF-κB dimers, which are further activated through various post-translational modifications and translocated to the nucleus. There, they bind to target genes, promoting the transcription of these genes.⁴³⁷ NF-κB activation plays a critical role in the formation of breast CSCs.⁴³⁸ Pathway analysis of CSCs isolated from prostate cancer and NSCLC patient revealed the specific activation of the NF-κB pathway, suggesting its potential as an effective therapeutic target.^439,440 Evaluation of the NF-κB signature in patient-derived GSCs can accurately predict the prognosis of low-grade glioma.⁴⁴¹

Moreover, the NF-κB pathway is implicated in maintaining stemness. Calcium Calmodulin-Dependent Protein Kinase II γ (CaMKIIγ), identified as a marker of AML stem cells, maintains stemness by activating the 5-LO/NF-κB pathway.⁴⁴² In ovarian CSCs, NF-κB pathway-related proteins are highly expressed, and inhibiting the NF-κB pathway reduces the CSC population.⁴⁴³ The lncRNA ASB16 Antisense RNA 1 (ASB16-AS1) cooperates with ATM kinase to phosphorylate Tripartite Motif Containing 37 (TRIM37), activating the NF-κB pathway and promoting gastric cancer cell stemness.⁴⁴⁴ Additionally, the Let-7a/Ras/NF-κB axis acts as a stemness antagonistic pathway in breast CSCs, with Let-7a inactivating the NF-κB pathway in a Ras-dependent manner.⁴⁴⁵ Overexpression of S100 Calcium-Binding Protein A4 (S100A4) activates the Inhibitor Of Kappa B Kinase (IKK)/NF-κB signaling pathway, contributing to the stemness maintenance of bladder CSCs.⁴⁴⁶

The activated NF-κB pathway in CSCs is intimately linked to tumorigenesis and metastasis. The transition from a proneural to mesenchymal phenotype (PMT) characterizes the conversion of less aggressive proneural GSCs into highly aggressive mesenchymal GSCs.^447,448 Fos-Like Antigen 1 (FOSL1) has been identified as a key regulator of PMT, upregulating Ubiquitin-Conjugating Enzyme (UBC9) to enhance the SUMOylation of Cylindromatosis (CYLD). This process activates the NF-κB pathway, supporting the PMT program of GSCs.⁴⁴⁹ Similarly, Mixed Lineage Kinase 4 (MLK4) binds to phosphorylated IKKa, activating the NF-κB pathway and facilitating the transformation of GSCs into the mesenchymal phenotype.⁴⁵⁰ Upregulated BMI-1 in CD133⁺ liver CSCs enhances NF-κB activation and nuclear translocation, promoting CSC stemness and metastatic potential while inhibiting apoptosis.⁴⁵¹ The estrogen metabolite 2-methoxy estradiol (2-ME2) disrupts the NF-κB/HIF-1 axis, reversing the EMT program and abolishing the metastatic potential of nasopharyngeal carcinoma stem cells.⁴⁵² Stromal Cell-Derived Factor-1 (SDF-1) overexpression in breast cancer induces stemness and EMT phenotypes by activating the NF-κB pathway.⁴⁵³ Additionally, miR-221/222 inhibits Phosphatase And Tensin Homolog (PTEN), leading to NF-κB activation and enhanced stem cell characteristics, tumorigenesis, and metastasis in breast cancer cells.⁴⁵⁴ A positive feedback loop involving DiGeorge Syndrome Critical Region 8 (DGCR8)/circKPNB1/SPI1/DGCR8 promotes stemness in GSCs, with SPI1 upregulating the NF-κB pathway in a TNF-α-dependent manner, thereby promoting tumorigenesis.⁴⁵⁵

JAK/STAT pathway

The JAK/STAT pathway consists of three main components: tyrosine kinase-related receptors that receive signals, tyrosine kinase JAK that transmits signals, and transcription factors STAT.⁴⁵⁶ Upon binding of various stimulatory factors to the receptor, JAK is phosphorylated and activated, subsequently recruiting and phosphorylating the transcription factor STAT. This phosphorylated STAT then forms dimers and is translocated to the nucleus, where it binds to target genes, regulating downstream gene expression.⁴⁵⁷

Regulation of the JAK/STAT pathway is closely linked to the maintenance of stemness. Misra et al. demonstrated that selective inhibition of STAT3 significantly reduced the expression of stemness-related genes in breast CSCs.⁴⁵⁸ Alpha-casein acts as a STAT pathway antagonist, inhibiting the STAT3/HIF-1α axis and impairing the function of breast CSCs.⁴⁵⁹ Moreover, activation of the lipid metabolism-related STAT3/CPT1B/fatty acid β-oxidation (FAO) axis in breast CSCs correlates positively with stemness maintenance.⁴⁶⁰ Similarly, STAT pathway activation contributes to stemness maintenance in osteosarcoma, liposarcoma, and thyroid cancer.^461,462,463 Immunity may play a significant role in JAK/STAT pathway regulation in CSCs. IL-17E/IL-25 secreted by non-CSCs binds to IL-17 Receptor B (IL-17RB) on CSCs, activating the JAK/STAT3 pathways to regulate liver CSCs’ stemness.⁴⁶⁴ Additionally, IL-6 is secreted by regulatory T cells, which upregulates the STAT3 pathway in glioma cells, maintaining the stemness-associated phenotype.⁴⁶⁵

Regulation of the JAK/STAT pathway in CSCs is intricately linked to tumorigenesis, metastasis, and metabolic reprogramming. Kanno et al. demonstrated that Von Hippel-Lindau (VHL) inhibits the JAK2/STAT3 signaling pathway, thereby reducing the tumorigenic ability of GSCs.⁴⁶⁶ In prostate CSCs, IL-6-mediated activation of the JAK/STAT pathway is a crucial event in tumorigenesis, and its inhibition eliminates tumor initiation.⁴⁶⁷ LIM Domain Only 2 (LMO2) acts as an endogenous agonist of the JAK/STAT pathway by forming a complex with LIM Domain-Binding 1 (LDB1) that phosphorylates STAT3, promoting the expression of ID1 and thereby upregulating the stemness and metastatic potential of GSCs.⁴⁶⁸ Leptin, an adipocyte-derived hormone, activates the JAK/STAT pathway in gastric cancer cells, maintaining their stemness and metastatic potential.⁴⁶⁹ Interferon-Induced Transmembrane Protein 3 (IFITM3), derived from GSCs, activates the JAK/STAT3 pathway to upregulate Basic Fibroblast Growth Factor (bFGF) expression, promoting angiogenesis in glioblastoma, a critical step in metastasis.⁴⁷⁰ Contrary to the STAT3/CPT1B/FAO axis, which is activated to maintain the stemness of breast CSCs, viperin overexpression in CSCs partially inhibits FAO through the JAK/STAT pathway, thereby reprogramming metabolism to promote tumor progression.^460,471

TGF-β pathway

The TGF-β family ligands form a complex with receptors on the membrane, and the activated receptor kinase recruits and activates downstream Smad proteins, thereby inducing nuclear transfer of Smad proteins and exerting transcriptional regulation.⁴⁷² The TGF-β pathway plays a pivotal role in embryonic development, immune surveillance, and maintenance of homeostasis.⁴⁷³ Dysregulation of the TGF-β pathway in CSCs is closely associated with the occurrence and progression of tumors.⁴⁷⁴ Nakano et al. demonstrated that stimulation of the TGF-β pathway triggers the conversion of CD44⁻ non-colorectal CSCs into CD44⁺ colorectal CSCs. Moreover, sustained activation of the TGF-β pathway is crucial for colorectal CSCs to maintain an undifferentiated state.¹⁴¹ Similarly, activation of the TGF-β signaling pathway has been observed during breast CSC generation.⁴⁷⁵

Activation of the TGF-β pathway in CSCs is closely linked to tumorigenesis and stemness maintenance. The U2 Auxiliary Factor 65 (U2AF65)/circNCAPG/Ras-Responsive Element-Binding Protein 1 (RREB1) positive feedback loop was identified in GSCs, where U2AF65 binds to and stabilizes circNCAPG, thereby stabilizing RREB1 and promoting its nuclear translocation. Accumulated RREB1 activates the TGF-β1 pathway to maintain the stemness of GSCs and promote tumorigenesis.⁴⁷⁶ Similarly, Heat Shock Protein 47 (HSP47) induces the stemness and tumorigenesis of GSCs by activating the TGF-β pathway.⁴⁷⁷ Lymphoid Enhancer-Binding Factor 1 (LEF1) directly binds to and upregulates the expression of ID1, triggering the TGF-β pathway, which in turn promotes the stemness-associated phenotype and tumorigenicity of esophageal squamous cell carcinoma.⁴⁷⁸ Wang et al. discovered that CD51, a marker of colorectal CSCs, activates the TGF-β/Smad signaling pathway to support tumorigenesis.⁴⁷⁹ The Hematological And Neurological Expressed 1-Like (HN1L) overexpression triggers the TGF-β pathway by upregulating FOXP2, ultimately maintaining stemness and promoting tumorigenesis of prostate cancer.⁴⁸⁰

Activation of the TGF-β pathway in CSCs plays a pivotal role in tumor metastasis. Wen et al. demonstrated that targeted inhibition of the TGF-β/Smad pathway effectively eliminated the EMT program and metastatic potential of ovarian CSCs.⁴⁸¹ FZD7 activates the TGF-β1/Smad3 pathway to confer stemness to pancreatic cancer cells. Further evidence suggests that upregulation of FZD7/TGF-β1/Smad3 promotes the EMT program to support pancreatic cancer liver metastasis.³⁹⁹ Similarly, Epithelial Membrane Protein 3 (EMP3) in lung CSCs interacts with TGF-β Receptor Type 2 (TGFBR2) to activate the TGF-β/Smad pathway, subsequently upregulating stemness and promoting the EMT program.⁴⁸² Activation of the SIX Homeobox 1 (Six1)/Eyes Absent (EYA)/TGF-β pathway mediates CSC characteristics and EMT programs in breast cancer.⁴⁸³ Additionally, the interaction between miRNAs and the TGF-β pathway is a critical factor affecting tumor metastasis. MiR-495, identified as a stemness suppressor in oral squamous cell carcinoma, inhibits Homeobox C6 (HOXC6), thereby inhibiting the TGF-β pathway to prevent stemness characteristics and the EMT program of CSCs and induce their apoptosis.⁴⁸⁴ MiR-106b attenuates the expression of the inhibitory Smad protein Smad7 to trigger the TGF-β pathway and promote the EMT program.⁴⁸⁵ Angiogenesis is a crucial aspect of the metastatic cascade.⁴⁸⁶ Chen et al. identified Paired-Related Homeobox 1 (Prrx1) as a non-GSC stemness-promoting factor and a GSC stemness-maintaining factor in glioma. Prrx1 directly binds to the TGF-β1 promoter region to activate the TGF-β/Smad pathway, which in turn upregulates stemness and promotes vascularization in the tumor microenvironment.⁴⁸⁷

PI3K/AKT pathway

As a pivotal factor in the PI3K/AKT pathway, AKT undergoes structural changes and activation by PI3K, which subsequently modulates a cascade of downstream substrates to regulate various cellular behaviors.⁴⁸⁸ The mTOR is a classic downstream target of the PI3K/AKT pathway, while PTEN acts as a negative regulator by dephosphorylating AKT to suppress downstream signaling.⁴⁸⁹ The involvement of the PI3K/AKT pathway in driving the differentiation of normal stem cells into CSCs has been confirmed.^490,491 Moreover, this pathway is closely associated with the maintenance of stemness in CSCs. Madsen et al. demonstrated a positive correlation between PI3K/AKT/mTOR pathway activation and breast cancer stemness score.⁴⁹² Additionally, activation of the insulin/insulin-like growth factor signaling (IIS) pathway in breast CSCs further potentiates the PI3K/AKT pathway to sustain MYC expression, thereby enhancing the stemness traits of breast CSCs.⁴⁹³ Furthermore, PD-L1 contributes to the establishment of a suppressive immune microenvironment.⁴⁹⁴ Almozyan et al. revealed that the continuously activated PI3K/AKT pathway by PD-L1 is pivotal in maintaining the stemness of breast CSCs.⁴⁹⁵

Activation of the PI3K/AKT pathway in CSCs is intricately linked to tumorigenicity. Activation of the PI3K/AKT pathway and the MAPK/ERK pathway respectively promote and inhibit the stemness signatures and tumorigenic potential of lung cancer.⁴⁹⁶ The liver cancer tumor suppressor Connexin 32 (Cx32) attenuates the activity of the PI3K/AKT pathway, thereby suppressing stemness and tumorigenicity.⁴⁹⁷ The tumor suppressor miR-30a binds to and inhibits 5’-Nucleotidase Ecto (NT5E), thus downregulating the activity of the NT5E-mediated PI3K/AKT pathway, thereby impeding the stemness and tumorigenicity of GSCs.⁴⁹⁸ Non-coding RNAs also play a regulatory role in the PI3K/AKT pathway. Tumor suppressors miR-873 and miR-30a bind and inhibit patterns of Pleckstrin-2 (PLEK2) and NT5E respectively, leading to downregulation of the PLEK2 or NT5E-mediated PI3K/AKT pathway, thus hindering stemness and tumorigenicity of pancreatic CSCs and GSCs.^498,499 Similarly, miR-3187-3p, which can be sponged by circ_0000745, inhibits Erb-B2 Receptor Tyrosine Kinase 4 (ERBB4), thereby attenuating the activity of the PI3K/AKT pathway, exerting a suppressive effect on the tumorigenicity and stemness of ovarian cancer.⁵⁰⁰

The activation of the PI3K/AKT pathway in CSCs is intricately linked to metastasis. AKT-mediated phosphorylation of Testis-Specific Y-Like Protein 5 (TSPYL5), a factor involved in stemness maintenance, impedes its ubiquitination and degradation. Phosphorylated TSPYL5 further inhibits negative regulators of the PI3K/AKT pathway, forming an AKT/TSPYL5/PTEN positive feedback loop that sustains the expression of stemness-related genes and promotes EMT programs.⁵⁰¹ In head and neck squamous cell carcinoma, activation of the PI3K/AKT/mTOR pathway upregulates SOX2, promoting the maintenance of the stemness phenotype and the E-cadherin-mediated EMT program.⁵⁰² Stress-Induced Phosphoprotein 1 (STIP1) in osteosarcoma enhances MMP-2 and MMP-9 by activating the PI3K/AKT and ERK1/2 pathways, ultimately promoting osteosarcoma CSC metastasis.⁵⁰³ In breast CSCs, Transmembrane And Coiled-Coil Domain Family 3 (TMCC3) binds AKT to activate the PI3K/AKT pathway, thereby supporting tumorigenesis and metastasis.⁵⁰⁴ CAFs are pivotal in supporting metastasis. CAFs upregulate TNF Receptor Superfamily Member 19 (TNFRSF19/TROY), a marker of liver CSCs, which activates the PI3K/AKT/T-Box Transcription Factor 3 (TBX3) pathway by promoting polyubiquitination of the PI3K inhibitory subunit p85α. Accumulated TBX3 maintains stemness and promotes metastasis.⁵⁰⁵ Similarly, CAFs-secreted periostin induces the phosphorylation of FAK to activate AKT, enriching CSCs in the gastric cancer cell population.²¹⁶ Moreover, Liang et al. demonstrated that inhibition of the PI3K/AKT pathway attenuates the stemness characteristics and angiogenesis of endometrial cancer, which are closely associated with distant metastasis.^506,507

PPAR pathway

PPARs, belonging to the nuclear hormone receptor family, are ligand-activated receptors that regulate various metabolic processes like fat and glucose metabolism. There are three main subtypes: PPARα, PPARδ/β (PPARD), and PPARγ (PPARG).⁵⁰⁸

The activation of the PPAR pathway plays a pivotal role in maintaining the stemness of CSCs. In liver CSCs, activation of the PPARα pathway and the enrichment of its downstream factor, Stearoyl-CoA Desaturase 1 (SCD1), contribute to the stemness characteristics.⁵⁰⁹ Similarly, increased PPARγ activity has been observed in melanoma stem cells.⁵¹⁰ Co-culturing MSCs with gastric cancer cells leads to the enrichment of lncRNA Histocompatibility Leukocyte Antigen Complex P5 (HCP5) in MSC-stimulated gastric cancer cells, which sponges miR-3619-5p to promote the expression of PPARG Coactivator 1 Alpha (PPARGC1A). PPARGC1A accumulation triggers the PPAR Coactivator-1α (PGC1α)/CCAAT Enhancer Binding Protein Beta (CEBPB)/CPT1 axis, inducing FAO and stemness characteristics.²³⁰

Conversely, inhibition of the PPAR pathway has also been associated with maintaining CSC stemness. Activation of PPARγ induced by inhibiting TRAF2- and NCK-Interacting Protein Kinase (TNIK) correlates with the reduction of osteosarcoma cell stemness and their differentiation into adipocytes.⁵¹¹ Downregulation of PPARD in the acidic microenvironment of colorectal cancer inhibits Vitamin D Receptor (VDR) expression, promoting the emergence of a CSC phenotype.⁵¹² Similarly, PPARγ activation effectively inhibits the stem cell phenotype of bladder cancer.⁵¹³ Moreover, besides fat metabolism, the PPAR pathway also regulates CSC characteristics through glucose metabolism. Low expression of PPARα in AML CSCs inversely correlates with their stemness characteristics. PPARα binds to HIF1α, inhibiting the expression of its downstream Phosphoglycerate Kinase 1 (PGK1) gene, ultimately weakening glucose metabolism activity and inhibiting stemness.⁵¹⁴

Activation of the PPAR pathway in CSCs is closely associated with tumorigenicity, differentiation, and metastasis. GSCs exhibit overexpression of PPARα compared to normal neural stem cells. Knockdown of PPARα significantly reduces the expression of stemness-related genes and fat metabolism-related genes in GSCs, leading to a decrease in tumorigenicity.⁵¹⁵ In hepatic CSCs, fatty acid 4-phenylbutyric acid (4-PBA) upregulates the expression of PPARα, preventing its degradation, thereby promoting the initiation and tumorigenicity of hepatic CSCs.⁵¹⁶ N1-methyladenosine methylation-driven expression of PPARδ in hepatic CSCs activates the PPAR pathway, regulating cholesterol metabolism to maintain stemness and enhance tumorigenicity.⁴¹³ Activation of PPARδ has been implicated in colorectal cancer liver metastasis induced by a high-fat diet, where it increases Nanog transcription.⁵¹⁷ Moreover, PPARα activation is positively correlated with the invasive and stemness phenotypes of GSCs⁵¹⁸. Conversely, stimulation of PPARγ may inhibit the migration ability of GSCs.⁵¹⁹ Activation of PPARγ has also been shown to downregulate the stemness of brain CSCs and induce the expression of differentiation-related genes such as Collagen Type II Alpha 1 (COL2A1) and Motor Neuron And Pancreas Homeobox 1 (HLXB9).⁵²⁰ Similarly, PPARγ activation reduces the activity of SOX2 and YAP1 genes, inhibiting the stemness of osteosarcoma stem cells and promoting their differentiation.⁵²¹

Molecular crosstalks

The formation and maintenance of CSCs involve complex interactions between multiple signaling pathways. For instance, SCD1 has been identified as a target for colorectal CSCs, inhibiting both WNT and Notch signaling pathways simultaneously to maintain the stemness-associated phenotype.⁵²² Similarly, NK6 Homeobox 1 (NKX6-1) in leiomyosarcoma upregulates stemness by activating Notch and SHH pathways.⁵²³ Tumors with high expression of Notch and hedgehog signaling pathways exhibit stronger stemness, often associated with a hypoxic microenvironment and activation of regulatory T cells.⁵²⁴ Protein kinase CK2 activates AKT, NF-κB, and STAT3 pathways to maintain the stemness of AML cells.⁵²⁵ Breast CSCs overexpressing Cyclooxygenase-2 (COX-2) activate PI3K/AKT, Notch, and WNT pathways via E-type Prostaglandin Receptor 4 (EP4), contributing to breast cancer metastasis.⁵²⁶ Moreover, Frizzled10 (FZD10) activation in liver CSCs through N6-methyladenosine methylation mediated by METTL3 stimulates the WNT and Hippo pathways, critical for hepatic CSC self-renewal.⁵²⁷

The formation and the stemness of CSCs are supported by crosstalk between multiple pathways. IL-6 and NO secreted by MDSCs activate STAT3 and Notch signaling in breast cancer cells, collectively inducing CSC formation.²⁸⁸ Notch signaling can drive NF-κB pathway-related gene expression in skin CSCs (Fig. 4a).⁵²⁸ Notably, NF-κB pathway upregulation also activates the Notch pathway to support breast CSC expansion (Fig. 4b).⁵²⁹ Additionally, activated PPARγ inhibits the STAT5 pathway, downregulating HIF2α and Cbp/P300 Interacting transactivator with Glu/Asp-Rich Carboxy-Terminal Domain 2 (CITED2) expression, which are protectors of CML CSCs (Fig. 4c).⁵³⁰ Lastly, Breast Cancer Susceptibility Gene 1-Associated Protein (BRAP) inhibits the TGF-β/PI3K/AKT/mTOR pathway, weakening the stem cell properties of GSCs (Fig. 4d).⁵³¹ LncROPM exerts a direct binding effect on Phospholipase A And Acyltransferase 3 (PLA2G16), thereby augmenting its expression and facilitating phospholipid metabolism. This process subsequently activates the PI3K/AKT, WNT/β-Catenin, and Hippo/YAP pathways to maintain the characteristics of breast CSCs (Fig. 4e).⁵³²

Fig. 4

Crosstalk of signaling pathways in CSCs. a, b The Notch pathway can be activated by the NF-κB pathway while activating the NF-κB pathway. c PPARγ inhibits the STAT5 pathway to downregulate the expression of HIF2α and CITED2, ultimately attenuating the stemness characteristics of CSCs. d BRAP inhibits the TGF-β/PI3K/AKT/mTOR axis to reduce the stemness of CSCs. e LncROPM upregulates PLA2G16 expression to facilitate phospholipid metabolism, which subsequently activates the PI3K/AKT, WNT/β-Catenin, and Hippo/YAP pathways to maintain the stemness of CSCs. f Amplified miR-139 through the miR-139/PDE2A/Notch1 loop, inhibits the WNT pathway to attenuate the tumorigenicity of CSCs. g LINC00115, upregulated by the TGF-β pathway, sponges miR-200s to activate the ZNF596/EZH2/STAT3 axis to promote the stemness and tumorigenesis of CSCs. h Activation of the TLR4/NANOG axis subsequently upregulates the YAP1/SMAD3 and IGF2BP3/AKT/mTOR/SMAD3 pathways to inhibit the nuclear transfer and phosphorylation of SMAD3, ultimately attenuating the tumorigenicity of CSCs. i WNT/β-Catenin pathway downstream effector PROX1 inhibits each other with Notch1, thereby elevating the stemness of CSCs and hindering their differentiation. j PROX1, which can be activated by the WNT/β-catenin pathway, inhibits each other with Notch1, thereby enhancing the stemness of tumor cells and hindering their differentiation. k SOX9/PROM1 positive feedback loop in inhibits differentiation by activating the CSC program, which positively correlates with WNT pathway and negatively correlates with TGF-β pathway. l The PI3K/AKT pathway, activated by CBX7, further stimulating the NF-κB/miR-21 axis, and ultimately promoting the stemness characteristics and metastasis of tumor cells

The tumorigenicity, differentiation, and metastasis capabilities of CSCs are regulated by the intricate crosstalk among multiple signaling pathways. In GSCs, the miR-139/Phosphodiesterase 2 (PDE2A)/Notch1 loop inhibits stemness and tumorigenicity by suppressing WNT signaling (Fig. 4f).⁵³³ Chronic hypoxia-induced HIF-2α overexpression activates WNT and Notch pathways, leading to enhanced stemness-associated phenotype and tumorigenesis in breast CSCs.⁵³⁴ Additionally, LINC00115, activated by the TGF-β pathway, upregulates ZEB1 and Zinc Finger Protein 596 (ZNF596) expression to activate ZNF596/Enhancer Of Zeste Homolog 2 (EZH2)/STAT3, promoting the stemness and tumorigenesis of GSCs (Fig. 4g).⁵³⁵ Activation of the AKT and YAP pathways in CSCs may inversely correlate with tumorigenesis. In liver cancer-initiating stem cells, the Toll-like receptor 4 (TLR4)/Nanog/YAP1/insulin-like growth factor 2 mRNA-binding protein 3 (IGF2BP3) axis inhibits TGF-β pathway activity and tumor-initiating ability, which can be counteracted by TGF-β pathway activation (Fig. 4h).⁵³⁶ The interaction between the WNT pathway and hedgehog, Notch, and TGF-β pathways influences the differentiation of colorectal CSCs. Hedgehog signaling negatively regulates WNT signaling, while PTCH1-dependent non-canonical hedgehog signaling positively regulates WNT signaling, contributing to CSCs’ differentiation (Fig. 4i).⁵³⁷ PROX1, a downstream effector gene of the WNT/β-Catenin signaling pathway, can inhibit each other with Notch1, thereby enhancing the stemness of colorectal cancer cells and hindering their differentiation (Fig. 4j).⁴²⁶ Moreover, the SOX9/PROM1 positive feedback loop in colorectal cancer inhibits differentiation by activating the stem cell program, positively correlating with WNT pathway activation and negatively correlating with TGF-β pathway activation (Fig. 4k).⁵³⁸ In lung CSCs, the HIF-1ɑ/miR-1275 axis co-activates WNT/β-Catenin and Notch pathways, enhancing stemness and metastatic potential.⁵³⁹ In gastric cancer, Chromobox Protein Homolog 7 (CBX7) upregulates the PI3K/AKT pathway, activating the NF-κB pathway to promote miR-21 expression, enhancing the CSC-associated phenotype and metastasis (Fig. 4l).⁵⁴⁰

Clinical predictive values of CSCs

Using CSC markers, investigators have revealed a negative correlation between the presence of CSCs and patients’ survival in various types of cancers.⁵⁴¹ Moreover, given the close correlation between CSCs and multi-drug resistance, it is also reasonable to use CSC as a parameter to predict patients’ prognosis after a specific type of anti-cancer treatment.

CSC markers can be used to predict response and survival after chemotherapy. In 47 patients with esophageal squamous cell carcinoma who receive neoadjuvant chemotherapy followed by radical esophagectomy, those with a high pre-chemotherapy expression of CD133 have significantly shorter survival compared to those with low CD133 expression, while the difference in survival is not significant between patients with high CD44 expression and low CD44 expression.⁵⁴² This study also reveals that CD44^high/CD133^high expression is associated with significantly poorer survival compared to those with CD44^low or CD44^high/CD133^low, suggesting that the combined use of CSC markers can provide better predictive values.⁵⁴² In 112 patients with advanced NSCLC treated with platinum-based chemotherapy, high Nanog levels were independently associated with shorter PFS (hazard ratio (HR) = 3.09, 95% confidence interval (CI) 2.01−4.76) and OS (HR = 3.00, 95% CI 1.98−4.54).⁵⁴³ Likewise, overexpression of CXCR4 correlates with poorer PFS and OS in 124 patients with epithelial ovarian cancer receiving cisplatin-based chemotherapy.⁵⁴⁴ Some studies also investigated the predictive value of single-nucleotide polymorphisms (SNPs) of CSC markers, which can influence the transcription, translation, and splicing of these proteins.⁵⁴⁵ For instance, LGR5 rs17109924 is associated with prolonged time to recurrence (TTR) (HR 0.38, 95%CI 0.19−0.79; P = 0.006) based on data from 391patients with colon cancer treated with adjuvant 5-fluorouracil-based chemotherapy.⁵⁴⁶ However, this correlation does not occur in patients treated with surgery alone, indicating that the correlation is mainly attributed to the impact of LGR5 rs17109924 on adjuvant chemotherapy.

CSC markers or stemness-related gene signatures also correlate with response to radiation and can be used to predict patients’ prognosis after radiotherapy. A study suggests that the CSC marker CD44 expression can be used to predict local recurrence of larynx cancer based on data from 19 patients.^547,548 Patients with rectal cancer and high expressions of CSC markers, CD133, OCT4, and SOX2, are prone to develop distant recurrence compared to those with low expressions of the genes.⁵⁴⁹ A systemic review identifies a series of CSC markers, including CD133, CD44, ALDH1, LGR5, and G9a, as indicators for the prognosis of patients with rectal cancer receiving radiotherapy.⁵⁵⁰ Using a machine learning method and data from the TCGA database, a model based on five tumor stemness and immune-related signatures, including Carbamoyl Phosphate Synthetase I (CPS1), CCR2, NT5E, Anillin (ANLN), and ATP-Binding Cassette Sub-Family C Member 2 (ABCC2), demonstrates favorable predictive values predicting radiotherapy responses.⁵⁵¹ A study suggests that P16^INK4A expression is negatively associated with CSC markers and predicts poor survival of patients with cervical cancer after radiotherapy.⁵⁵² GSC markers, CD133 and O6-methylguanine-DNA methyltransferase, are also associated with patients’ responses to radiotherapy.⁵⁵³

Some studies attempt to predict patients’ response to or survival after targeted therapy and immunotherapy using CSC markers or stemness-related gene signatures. For instance, a study shows that head and neck squamous cell carcinoma patients with low CD44, a CSC marker, have a significantly better HR for OS than those with high CD44 expression when comparing nimotuzumab plus cisplatin-radiation (NCRT) with cisplatin-radiation (CRT), suggesting that CD44 might be a favorable reference for whether to use nimotuzumab.⁵⁵⁴ Using a five tumor stemness and immune‐specific‐gene (CPS1, CCR2, NT5E, ANLN, and ABCC2) signature, a study constructs a machine-learning model that can predict therapeutic responses in melanoma patients receiving adoptive T cell therapy (area under ROC curve (AUC) = 0.717) and immune checkpoint blockades (AUC = 0.703).⁵⁵¹ The AUCs of this signature were higher than those of PD-1 (0.687 for adoptive T cell therapy and 0.505 for immune checkpoint blockades).⁵⁵¹

Cancer stem cells and tumor chemotherapy resistance

The clinical significance of chemotherapy resistance

Chemotherapy remains the cornerstone of current clinical oncology, offering significant reductions in tumor burden and enhanced patient survival, thereby securing its widespread clinical application.⁵⁵⁵ However, patients initially responsive to chemotherapy inevitably evolve into drug resistance. This phenomenon, termed acquired resistance, poses a significant challenge to both clinicians and researchers. Research has progressively revealed that acquired resistance is intricately linked to intratumoral heterogeneity.³⁰⁰ Chemotherapy selectively eliminates sensitive subpopulations, allowing resistant cells to prevail and drive disease progression. The contribution of CSCs to intratumoral heterogeneity and acquired resistance has been a focal point of research for decades. This chapter explores the mechanisms through which CSCs mediate resistance and their ramifications for chemotherapy strategies.

Mechanism of resistance of CSCs to chemotherapy

In cancer biology, CSCs exhibit dynamic states of proliferation and quiescence. During dormancy, CSCs reduce their metabolic activity, enabling prolonged survival in a quiescent state. Upon extracellular stimulation, however, CSCs may re-enter the cell cycle, regaining proliferative capacity.^556,557 This duality poses significant challenges for chemotherapy, as quiescent CSCs exhibit resistance to such treatments, and often develop more resistant phenotypes.^558,559,560 This resistance is largely attributed to the mechanism of conventional chemotherapy, which targets rapidly dividing cells and acts in a cell cycle-specific manner (Fig. 5).⁵⁶¹ However, CSCs, with their slow division rates, often residing in the G1 or S phase, exhibit resistance to a variety of chemotherapeutic agents including cisplatin, taxol, and doxorubicin.^562,563 For instance, the overexpression of Zinc Finger E-Box-Binding Homeobox 2 (ZEB2) increases the proportion of colorectal CSCs in G0/G1 phase, leading to platinum resistance.⁵⁶⁴ Distinguishing quiescent CSCs from proliferative CSCs remains a challenge due to the lack of specific surface markers and common genotypic and phenotypic characteristics.⁵⁶⁵ CD13 has been proposed as a marker for quiescent hepatic CSCs, which have been proven capable of neutralizing chemotherapy-induced ROS and DNA damage.³¹¹ Moreover, epigenetic modifications also play crucial roles in regulating the quiescent state of CSCs. For example, SET Domain-Containing Protein 4 (SETD4) promotes breast CSC dormancy through the trimethylation of histone H4 lysine 20, facilitating heterochromatin formation.⁵⁶⁶ Elevated levels of miR-135a reduce the methylation at the CG5 site of the Nanog promoter by directly targeting DNA Methyltransferases 1 (DNMT1). Then, the combination of SET And MYND Domain Containing 4 (SMYD4) and unmethylated Nanog promoter will activate the expression of Nanog in those Nanog-negative tumor cells, thus promoting the switch of CSCs.⁵⁶⁷ Endothelial cells, by expressing miR-126, can induce dormancy in CML stem cells. Concretely, by targeting the PI3K/AKT/mTOR signaling pathway, miR-126 blocks the cell cycle progression of CSCs.^568,569 Soluble growth factor/receptor pathways, such as CXCL1/CXCL12, Bone Morphogenetic Protein-4 (BMP4), and LIF, have also been shown to regulate the quiescent state of activated CSCs. For instance, CXCL1 induces liver CSC quiescence via mTORC1 kinase activation, while knocking out CXCL12 downregulates quiescence-associated genes, such as TGF-β and STAT3, facilitating the exit of leukemia stem cells from dormancy.^570,571 BMP4 directly regulates the quiescent state of CML leukemia stem cells through a JAK/STAT3 pathway dependent on BMPR1B kinase activity, and the LIF Receptor (LIFR) correlates with the expression of quiescence-associated genes in CSCs, such as TGFβ2 and Notch1. Knockout of LIFR promotes the proliferation of breast CSCs and enhances their capacity for bone destruction.^572,573 More and more evidence suggest the involvement of extracellular vesicles (EVs) in CSC quiescence regulation. EVs from CAFs of hormone therapy-resistant breast cancer patients promote estrogen receptor-independent oxidative phosphorylation and hormone therapy resistance.⁵⁷⁴ Furthermore, CAFs create a resistant niche through close interactions with CSCs, secreting factors like IL-6 and IL-8 that support CSC survival.⁵⁷⁵ In colorectal cancer, CAF-derived EVs trigger resistance to 5-fluorouracil in CSCs, which is the standard of care.⁵⁷⁶ Endothelial cells can promote resistance in GSCs through the secretion of NO, enhancing Notch signaling, or by releasing CD44 ligands.

Fig. 5

Mechanism of resistance of CSCs to chemotherapy. CSCs possess the ability to maintain a quiescent state and reduce metabolic activity, thereby exhibiting resistance to chemotherapy. Furthermore, CSCs are capable of metabolic reprogramming, utilization of ABC transport proteins, and activation of DNA repair pathways, which allows them to evade chemotherapy. Additionally, the microenvironment plays a crucial role in supporting CSC survival. The balance between ROS and anti-apoptotic versus pro-apoptotic signals, along with exosomes secreted by tumor-associated fibroblasts, dynamically regulates CSCs

The HIF pathway emerges as one of the most pivotal regulators of the quiescent state in CSCs.⁵⁷⁷ With the identification of quiescent CSCs in MM, the expression levels of TRIM44 were elevated. This E3 ubiquitin ligase facilitates the deubiquitination and stabilization of HIF-1α under both normoxic and hypoxic conditions, underscoring the intricacy of oxygen sensing in tumorigenesis.^578,579 Furthermore, the significance of HIF2α in the stability and transformation of CSCs in glioblastoma also highlights the critical role of oxygen levels in CSC biology.⁵⁸⁰ The markers of CSCs such as OCT4, Nanog, SOX2, Krüppel-Like Factor 4 (KLF4), c-Myc, and miR-302 are induced in hypoxic environments further supporting the adaptive responses of CSCs to oxygen deprivation.⁵⁸¹ Notably, hypoxia not only modulates cell plasticity but also stimulates the proliferation and expansion of pre-existing CSC pools, suggesting a dynamic interplay between CSC quiescence and activation.^582,583,584 In conclusion, CSCs exhibit long-term stability and a quiescent phenotype under hypoxia, characterized by low metabolism and reduced oxidative phosphorylation. Conversely, the presence of oxygen triggers the activation of tricarboxylic acid (TCA) cycle enzymes and oxidative phosphorylation, transitioning CSCs into a proliferative state.^585,586

Metabolic reprogramming stands as one of the hallmarks of the bioenergetics of CSCs.⁵⁸⁷ ALDH enzymes serve as potential inducers of metabolic reprogramming, thereby promoting chemotherapy resistance. For example, ALDH enzymes mitigate aldehyde accumulation by converting them into less toxic carboxylic acids, which play a crucial role in detoxification within CSCs.⁵⁸⁸ Moreover, ALDH enzymes maintain a low level of ROS by consuming these reactive aldehydes induced by ROS.⁵⁸⁹ The increased activity of ALDH1A1 and ALDH3A1 subtypes enables CSCs to metabolize cyclophosphamide and its analogs, such as 4-hydroperoxycyclophosphamide, ifosfamide, and etoposide, and detoxify their intermediate products aldophosphamide into carboxyphosphoramide.^590,591 ALDH also contributes to the synthesis of retinoic acid and neurotransmitter γ-aminobutyric acid (GABA), essential for the homeostasis and differentiation of CSCs.^330,592 Inhibition of ALDH activity using all-trans retinoic acid (ATRA) significantly improves prognosis in leukemia, highlighting the enzyme’s involvement in cell differentiation and survival pathways including Notch, mTOR, and PI3K/AKT.^593,594,595 The mediating role of ALDH activity in therapy resistance has been established across various cancers, including breast, pancreatic, lung, Ewing’s sarcoma, stomach, glioblastoma, head and neck, ovarian, and colorectal cancers in recent years. Types of chemotherapy agents covered include doxorubicin, paclitaxel, gemcitabine, gefitinib, temozolomide, doxorubicin, and platinum, implicating it as a key marker of CSC drug resistance.^{596,597,598,599,600,601,602,603,604} Among the 19 ALDH family members, ALDH1 is considered most closely associated with CSCs.^605,606 However, the complex role of ALDH in CSC biology and therapy resistance warrants further investigation.

The ABC transporter superfamily, encoded within the human genome, represents the largest group of transmembrane proteins. These transporters are categorized into seven subfamilies, ABC-A to ABC-G, based on the similarity or disparity of their domain structures.⁶⁰⁷ Recent studies have revealed a significant upregulation of ABC transporters in CSCs, highlighting their pivotal role in mediating chemotherapeutic resistance by extruding harmful toxins and xenobiotic compounds from cells, thereby reducing intracellular drug concentrations.^83,608 ABCB1, also known as Multidrug Resistance Protein 1 (MDR1) or P-glycoprotein, was the first member of the ABC transporter family identified in humans.⁶⁰⁹ Wright et al. demonstrated that ABCB1 expression serves as a crucial marker for doxorubicin resistance in breast CSCs.⁶¹⁰ Various members of the ABC transporter family, especially ABCB1, ABCC1, and ABCG2, are recognized for their heightened expression in CSCs and their involvement in MDR mechanisms.^611,612,613 Certain cancers, such as melanoma, might exhibit specific ABC transporter profiles, with ABCB5 playing a significant role.^614,615 From the structure and properties of ABC transporter, ABCB1, characterized by two ATP-binding sites, exhibits enhanced drug transport capabilities.⁶¹⁶ Distinct ABC transporters have been implicated in various chemotherapeutic resistances. For example, ABCC1 is primarily associated with resistance to anthracycline drugs, whereas ABCG2 exhibits the broadest spectrum of drug resistance.^617,618,619 New roles for ABC transporters in CSCs have been uncovered. ABCB5 possesses the ability to regulate IL-8-dependent CSC maintenance in melanoma and promote the invasion of tumor cells in colorectal cancer. ABCG2 also plays a role in the enhancement of CSC tumorigenic potential.^83,620 It should be noted that the ABC transporter family is intimately linked with signaling pathways. The ABCB1 gene promoter contains multiple targets for the β-Catenin complex, suggesting a reciprocal relationship where the WNT/β-Catenin signaling pathway targets ABCB1 activity.⁶²¹ Activation of the WNT/β-Catenin pathway can induce ABCB1 expression, facilitating chemotherapy resistance in CSCs.^622,623 ABCG2 is also involved in the WNT/β-Catenin signaling cascade.^613,624 And its expression can be regulated by the Notch pathway as well.⁶²⁵ Furthermore, the Hippo pathway effector YAP1 promotes the drug resistance of CSCs through ABCG2.^626,627 The PI3K/AKT pathway regulates ABCG2 in GSCs at the plasma membrane, instead of the mTOR pathway.⁶²⁸ Inhibition of the PI3K/AKT pathway results in the downregulation of ABCG2 in CML cells.⁶²⁹ Despite these insights, clinical successes with specific ABC transporter inhibitors remain scarce, underscoring the ongoing need for mechanistic exploration.

Several studies have elucidated that the mechanisms of DNA-damaging chemotherapeutic agents underlying tumor cells, which include DNA crosslinkers (cisplatin, carboplatin, oxaliplatin), DNA synthesis inhibitors (methotrexate), and topoisomerase inhibitors (doxorubicin, daunorubicin).⁶³⁰ These agents predominantly target the S phase of tumor cells, where DNA replication occurs, exploiting the diminished DNA repair capacity in tumor cells, which culminates in genomic instability and subsequent apoptosis. Notably, in CSCs, DNA damage checkpoints are activated, facilitating repair mechanisms that enhance cell survival. Sequencing data reveal an upregulation in the majority of DNA damage response and repair genes within CSCs, indicating a superior DNA repair efficiency.^23,631 The p53 signaling pathway and apoptosis are pivotal to DNA damage repair. Upon detrimental DNA damage, the ATM and Ataxia Telangiectasia And Rad3-Related (ATR) kinase complex with Poly ADP-Ribose Polymerase 1 (PARP-1) and BRCA1, phosphorylate CHK1 and CHK2, thereby activating p53, which leads to cell cycle arrest, DNA repair, or the execution of apoptosis.⁶³² Remarkably, genes like p53, which induce cell death, often harbor mutations or are dysregulated in CSCs, and inhibiting p53 aggregation can restore sensitivity to platinum-based treatments.⁶³³ DNA repair proteins directly or indirectly linked to CSCs’ drug resistance include CHK1, CHK2, ATR, MSI1, RAD50, and RAD51, with RAD51 playing a significant role in resistance to PARP inhibitors.^{634,635,636,637}

Furthermore, CSCs can prevent DNA damage through effective ROS clearance.⁶³⁸ A highly compatible ROS scavenging system has evolved in the CSCs of some tumors to maintain low ROS concentrations. Antioxidant enzymes like superoxide dismutase, glutathione peroxidase, and catalase are markedly active in CSCs.^639,640,641 NRF2, a transcription factor, mediates CSC drug resistance by not only regulating the expression of genes involved in the cellular antioxidant response, but also by stimulating drug efflux through raising ATP Binding Cassette Subfamily F Member 2 (ABCF2) expression among other functions.^642,643 ROS overload or elevated ROS levels induced by chemotherapy have been implicated in the activation of HIFs, which can trigger the activation of pro-survival and developmental pathways such as Notch, WNT, and hedgehog. These pathways in turn contribute to the sustenance of CSCs’ survival.⁶⁴⁴ Additionally, studies have identified a negative feedback loop between ROS and COX-2 within CSCs, where elevated ROS levels induce COX-2 expression, which in turn mitigates ROS accumulation, fostering CSC enrichment and metastasis.^645,646 Autophagy, a critical biological process for cellular homeostasis, has been recognized as a pivotal resistance mechanism in metastatic prostate CSCs, significantly contributing to ROS scavenging.⁶⁴⁷ To sum up, CSCs exhibit heightened sensitivity to any alteration in the oxidant/antioxidant balance, acquiring resistance under both low and elevated ROS levels.

One of the primary approaches of chemotherapeutic agents is the induction of apoptosis.⁶⁴⁸ The balance between pro-apoptotic (BCL2-Associated X Protein (BAX), BCL2 Antagonist/Killer (BAK), BCL2 Asociated Death Promoter (BAD)) and anti-apoptotic (BCL2, BCL-XL, MCL1) proteins constitutes a focal point of cellular response to apoptosis.⁶⁴⁹ In CSCs with chemotherapy resistance, the balance tips towards anti-apoptotic proteins. It has been shown that compared to tumor cells, CSCs exhibit higher levels of anti-apoptotic gene expression (such as BCL2, BCL-XL).^650,651 Knockdown of the biomarkers of CSCs, such as CD44, increases apoptosis, evidenced by elevated expression of pro-apoptotic proteins BAX and caspases-3, -8, and -9, while the levels of anti-apoptotic proteins BCL2 and BCL-XL decrease.⁶⁵² Further analysis by Konopleva et al. demonstrated that the overexpression of anti-apoptotic genes BCL-XL and BCL2 could also induce a quiescent state in CSCs.⁶⁵³ Moreover, the upregulation of specific cell surface receptors (such as EGFR, Fibroblast Growth Factor Receptor (FGFR), HER2R) in CSCs can inhibit apoptosis by downregulating the pro-apoptotic protein BAD.⁶⁵⁴ Additionally, CSCs can evade apoptosis by prolonging the G2/M phase in the cell cycle through upregulation of G2/M checkpoint proteins CHK1 and CHK2.⁶⁵⁰ Some CSCs overcome apoptosis by upregulating the expression of Inhibitors Of Apoptosis Proteins (IAPs).⁶⁵⁵ The expression of Cellular FLICE-Like Inhibitory Protein (C-FLIP) also affects the apoptosis receptor initiation pathway, inhibiting caspase activation and thereby hindering the apoptotic process. Studies have shown that different splice variants of c-FLIP are associated with resistance to chemotherapeutic drugs.^656,657 CSCs employ various indirect or direct mechanisms to evade apoptosis, such as endoplasmic reticulum stress. It has been discovered that several components directly involved in endoplasmic reticulum protein processing are dysregulated in CSCs.⁶⁵⁸ In vitro CSC models observed the inactivation of IRE1 (XBP-1 splicing) and the activation of the PERK (elF2α phosphorylation) pathway, both key conduits of the endoplasmic reticulum stress response.⁶⁵⁹ Mitochondrial integrity is crucial for the survival and maintenance of CSCs, with its dysregulation having profound effects on autophagy and apoptosis.⁶⁶⁰ Studies indicate mitochondrial alterations in CSCs of CML compared to normal stem cells. Resistant CSC subpopulations can be identified by higher mitochondrial mass and increased endopeptidase activity.⁶⁶¹

In conclusion, the chemoresistance mechanisms of CSCs constitute an interactive network (Fig. 5). Targeting individual components may not eliminate the resistance posed by CSCs. To devise accurate CSC-targeted treatments that enhance sensitivity, further exploration in the domain of resistance is warranted.

Clinical trials targeting CSCs combined with chemotherapy

With a profound understanding of the pivotal role CSCs play in chemotherapy resistance, researchers have initiated a series of targeted clinical trials aimed at exploring potential therapeutic strategies for CSCs (Table 4). These trials broadly fall into three categories. The first category, guided by the ChemoID assay, identifies subsequent treatment regimens using patient biopsy samples before treatment to enrich stem cells and test their response to chemotherapy drugs. Results from a phase III clinical trial, exemplified by NCT03632135, demonstrated a significant reduction in patient mortality risk in the ChemoID assay-guided group, suggesting that the ChemoID assay could become a routine diagnostic and treatment method akin to genetic testing in the future. The second category involves the development of specific inhibitors targeting mechanisms by which CSCs contribute to chemotherapy resistance. This category encompasses most clinical trials, such as those using vismodegib or PF-04449913 to inhibit the hedgehog pathway, LGK974 targeting the WNT pathway, and OMP-52M51 against DLL4 in the Notch pathway. Although γ-secretase inhibitors are the largest group of drugs targeting the Notch pathway, trial outcomes have not been disclosed yet. Additionally, the development of the drugs RO4929097 and PF-03084014 has been halted for various reasons. NCT04137627 evaluated melatonin as an antioxidant in combination with neoadjuvant chemotherapy for changes in tumor stemness expression in oral squamous cell carcinoma, but results showed no statistical difference despite a reduction in miR-210 and CD44 expression, implying the tumor microenvironment might play a role in CSC resistance mechanisms but may not be the dominant factor. In theory, PARP inhibitors involved in DNA repair could also be effective against CSCs, but current clinical trials have not measured changes in CSCs or biomarkers before and after treatment, necessitating further exploration of their effect on CSCs. The third category targets markers specific to CSCs for treatment, such as CD44v6. Bivatuzumab mertansine, an antibody-drug conjugate (ADC) targeting CD44v6, enhances the specificity of chemotherapy and has shown promising results in various cancer treatments. However, its effectiveness against CSCs, as indicated by NCT02254005, remains inconclusive. These results indicate that we cannot yet conclusively determine whether targeting CSCs can reverse chemotherapy resistance. We look forward to the anticipated outcomes of these clinical trials in the coming years.

Table 4 Clinical studies on combination of chemotherapy and CSC-targeting therapies

Cancer stem cells and tumor immunotherapy resistance

CSCs and immune evasion

Immune cells within the tumor microenvironment play a pivotal role throughout the oncogenesis and progression of tumors. Unlike their counterparts in normal tissues, these immune cells often exhibit attenuated inflammatory responses or enhanced suppressive functions, thereby facilitating tumor immune evasion. This section aims to provide an overview of the principal roles played by TAMs, MDSCs, NK cells, T cells, and B cells in mediating immune escape of CSCs. Their complex interplay and the mechanisms through which they contribute to the immunological cloak that shields CSCs from the host’s immune defense are critical to understanding and developing novel therapeutic strategies.

Research across various tumor types has revealed that TAMs often constitute up to 50% of the immune cell population, positioning them at one of the forefront research fields of immune cells within the microenvironment.⁶⁶² The heterogeneity of TAMs emerges as a critical mechanism behind immunotherapy resistance. Engagement of damage-associated molecular patterns (DAMPs) with specific pattern recognition receptors on macrophages, such as TLR4, triggers pro-inflammatory signaling and polarization towards the M1 phenotype.⁶⁶³ M1-TAMs, characterized as classically activated macrophages, exhibit enhanced pathogen phagocytosis capabilities, thereby exerting anti-tumoral properties. However, the tumor microenvironment promotes polarization towards the M2-TAMs, which possess pro-tumoral potential, supporting and sustaining CSCs and therapy resistance through the secretion of chemokines and activation of stemness pathways, such as sonic hedgehog ligands. For instance, GSCs can recruit M2-TAMs by secreting periostin.⁶⁶⁴ CSCs may also elevate M2–TAM levels through chemokines like CCL2 and macrophage colony-stimulating factor 1 (CSF1).⁶⁶⁵ Drug-resistant lung CSCs activate the Interferon Regulatory Factor 5 (IRF5)/M-CSF pathway to promote the production of M2-TAMs from CD14⁺ monocytes.⁶⁶⁶ In turn, M2-TAMs secrete factors like milk fat globule-EGF factor 8 (MFG-E8), activating STAT3 and sonic hedgehog signaling in CSCs, thereby enhancing treatment resistance.⁶⁶⁷ Moreover, TAMs secrete substantial amounts of TGF-β1, maintaining CSC characteristics and promoting EMT.^257,668 Lu et al. demonstrated that CSCs undergoing EMT upregulate CD90/Thy1 and EphA4, crucial proteins mediating physical interactions between CSCs and TAMs. EphA4 receptor activation secretes Src and NF-κB, inducing CSCs to secrete various cytokines maintaining stem cell status.²⁸² ScRNA-seq has clarified the bidirectional feedback mechanisms between CSCs and TAMs. CSCs secrete S100A11 protein to promote TAM polarization towards the M2 phenotype, which in turn enhances CSC self-renewal and metastatic capabilities.²⁴⁷ The crosstalk between CSCs and macrophages is intricate, wherein CSCs not only polarize macrophages towards a tumorigenic state but also employ protective mechanisms to avoid macrophage phagocytosis. Elevated expression of CD47, observed in CSCs from both hematological malignancies like AML and solid tumors like pancreatic, liver, and lung cancers, interacts with Signal Regulatory Protein α (SIRPα) on TAMs, broadcasting a “don’t eat me” signal to protect them from macrophage engulfment.^{669,670,671,672,673,674} Recent studies have also identified TAMs as “iron donors” within the tumor microenvironment, fulfilling the high iron demand of CSCs and playing a crucial role in influencing iron homeostasis.⁶⁷⁵ Although the mediators involved in this crosstalk may vary with tumor pathology, such interactions may one day become potential therapeutic targets against CSCs.

Tumor-infiltrating myeloid cells represent a heterogeneous lineage that includes TAMs, MDSCs, and so on, the latter being a focal point of research due to their impact on limiting the efficacy of immunotherapy.⁶⁷⁶ MDSCs, immature myeloid cells derived from the bone marrow, are categorized into polymorphonuclear MDSCs (PMN-MDSCs) and monocytic MDSCs (M-MDSCs), expressing CD15 and CD14, respectively.^677,678 These cells exert immunosuppressive effects through distinct mechanisms, with the ratio of PMN-MDSCs to M-MDSCs in peripheral blood being crucial.⁶⁷⁹ In murine models of melanoma, prostate, and cervical cancers, CSCs promote PMN-MDSC infiltration by overexpressing G-CSF, CXCL5, and TGFβ.^680,681,682 In turn, PMN-MDSCs increase STAT3 phosphorylation, CD133 and CD44 expression, and sphere formation of colorectal CSCs in vitro by secreting S100A9 protein.²¹⁰ PMN-MDSCs also enhance the ratio of CSCs, spheroid-forming ability, and expression of stemness-related genes in myeloma cells by inducing piRNA-823 expression.²⁸⁶ Moreover, studies have identified M-MDSCs as primary drivers of the CSC phenotype in pancreatic and breast cancers.^288,290 In breast tumor models, M-MDSCs comprise the majority of tumor-infiltrating MDSCs. Mechanistic analysis has shown that NO produced by M-MDSCs promotes the CSC phenotype through activation of Notch signaling and sustained STAT3 phosphorylation in cancer.^288,677 The relationship between CSCs and MDSCs is bidirectional, as CSCs also recruit MDSCs to limit T cell activity, creating a favorable environment for tumor growth. MDSCs in peripheral lymphoid organs are predominantly PMN-MDSCs, which exhibit relatively mild immunosuppressive activity compared to M-MDSCs. PMN-MDSCs primarily produce high levels of ROS to exhibit immunosuppressive activity, which are unstable and transiently, requiring antigen-specific interactions with T cells to ultimately induce tumor-specific T cell tolerance.⁶⁸³ In contrast, M-MDSCs produce substantial amounts of NO, arginase 1, and immunosuppressive cytokines with longer half-lives, effectively inhibiting nonspecific T cell responses without the need for direct contact between MDSCs and T cells.⁶⁷⁹ It is noteworthy, however, that despite functional annotation and transcriptomic profiles widely recognizing PMN-MDSCs as distinct from inflammatory neutrophils, tumor-associated neutrophils, and PMN-MDSCs share overlaps in markers and suppressive functions, suggesting a close phenotypic and functional relationship.⁶⁸⁴

T cells are the most pivotal immune effector cells in the anti-tumor response, executing cytotoxic effects on tumor cells through classical pathways such as perforin/granzyme release, death receptor engagement, and induction of apoptosis.⁶⁸⁵ Studies have revealed that CSCs evade T cell-mediated immune rejection by downregulating key components of the antigen processing and presentation machinery and suppressing T cell anti-tumor functionality.⁶⁸⁶ Tumor antigens are broadly classified into two categories: (1) tumor-specific antigens (TSAs), encoded by mutated or rearranged genes, and (2) tumor-associated antigens (TAAs), encoded by genes specific to the normal cellular lineage.⁶⁸⁷ CSCs may selectively avoid expressing differentiation-related TAAs, thus resisting T cell-mediated rejection.^688,689 Another mechanism of immune evasion involves the downregulation or loss of MHC-I by CSCs.⁶⁹⁰ As MHC-I play a crucial role in immune recognition, their absence or reduced expression can limit T cell-mediated lysis of CSCs.^689,691 The induction of T cell tolerance by CSCs is another key strategy in escaping immune surveillance. A primary mechanism of tolerance induction involves the clonal deletion of antigen-reactive T cells through apoptosis or death, with the Factor-Related Apoptosis (Fas)/Fas-L pathway serving as a significant mediator.^692,693 CSCs may actively destroy T cells through the expression of Fas-L, moreover, the autocrine secretion of soluble Fas-L protects CSCs from cytotoxic T cell-mediated Fas killing.^693,694,695 Furthermore, CSCs can evade immune attack by downregulating Fas.⁶⁹³ The tumor-expressed ligand Receptor-Binding Cancer Antigen Expressed On SiSo Cells (RCAS1) has also been found to induce apoptosis in T, B, and NK cells expressing its receptor.⁶⁹⁶ Immunogenic tolerance can be achieved through non-deletional processes, such as the inactivation of antigen-reactive cells.⁶⁹⁷ The secretion of TGF-β underpins the inhibition of T cell proliferation mediated by MSCs.⁶⁹⁸ CSCs produce TGF-β and IL-10, directly suppressing T cells to avoid immune-mediated destruction.^699,700,701 The TGF-β signaling pathway is also specifically activated in CSCs, with secreted morphogens of the TGF-β superfamily and their receptors preferentially expressed by CSCs.^702,703,704 T cell activation in the immune response requires two signals.⁷⁰⁵ The first signal comes from the T Cell Receptor (TCR) recognizing the MHC/antigen peptide complex, conveying an antigen-specific recognition signal.⁷⁰⁶ The second signal is provided by co-stimulatory molecules of antigen-presenting cells (APCs), offering a non-specific synergistic co-stimulation signal.⁷⁰⁷ CSCs may reduce T cell responsiveness to tumor antigens by actively modulating the activation state of APCs and may express negative co-stimulatory molecules to disrupt anti-tumor immune responses.^708,709 PD-1/PD-L1-mediated negative co-stimulatory signal transduction is the most common way of inhibiting lymphocyte activation.^710,711 Other mechanisms include the induction or active recruitment of regulatory T (Treg) cells, which can effectively suppress the activation, proliferation, and cytokine production of other T cells, crucial for maintaining immune self-tolerance and homeostasis.^{712,713,714,715} In summary, CSCs employ a multitude of processes to drive tumor escape from immune-mediated rejection responses.

In the era of immune checkpoint inhibitors (ICIs) and adoptive T cell therapies, the pivotal role of T cells in anti-tumor immunity has become indisputable. However, these advancements have also exposed numerous limitations of T cells, underscoring the urgent need to unravel immunological mechanisms. With the advancement of scRNA-seq technology, the subpopulations and states of B cells within the tumor microenvironment are increasingly scrutinized. For instance, in melanoma, genes associated with early B cell stages are extensively expressed.⁷¹⁶ In breast cancer, B cells predominantly exist as naive B cells, memory B cells, with fewer plasma cells and germinal center cell clusters observed. Notably, compared to peripheral blood B cells, tumor-associated B cells exhibit higher levels of somatic mutations and greater clonal expansion.^717,718 Another distinctive function of B cells was observed in ovarian cancer, where B cells preferentially express IgA, while in breast cancer, B cells mainly express IgM and IgG. This IgA can target antigens and be internalized by tumor cells in an antigen-independent manner through Polymeric Immunoglobulin Receptor (PIGR), sensitizing tumors to T cell.⁷¹⁹ Current research indicates that the states of tumor tissue-associated B cells vary across different types of tumors, but largely remain in a pre-antibody class-switched state. Some studies suggest that in the presence of ongoing tumors, exhausted or dysfunctional CD8⁺ and CD4⁺ T cells seek the aid of B cells in the microenvironment, through the expression of CXCL13, to form tertiary lymphoid structures (TLS).^720,721,722 Research across multiple cancers demonstrates that the presence of TLS and B cells in tumor tissues correlates with better prognoses, and the anti-tumor efficacy of T cells is enhanced in the presence of B cells.^{723,724,725,726} Interestingly, TLS can also be exploited by tumor cells under certain conditions to promote lymphatic infiltration of tumor cells, leading to lymphatic metastasis.⁷²⁷ However, few studies have revealed a direct significant correlation between the CSC phenotype and both TLS and B cells.⁷²⁸ Tumors with low TLS infiltration may present higher CSC characteristics, with increased proliferation and metastatic potential.⁷²⁹ In summary, the presence of TLS is considered a crucial component of anti-tumor immunity. With the development of scRNA-seq and spatial transcriptomics, research into the functions of TLS within tumors and their relationship with CSCs is expected to mature and refine further.

NK cells represent a crucial component of the innate immune system, constituting the third major lymphocyte type, following T cells and B cells. They play a complementary role to T cells by eliminating reduced or absent MHC class I expression tumor cells which evade CD8⁺ T cell detection, and can also recruit dendritic cells to indirectly enhance T cell-mediated responses.⁷³⁰ Emerging evidence suggests that CSCs may be particularly susceptible to NK cell-mediated targeting. In colorectal cancer models, CSCs exhibit increased vulnerability to NK cell cytotoxicity, associated with the upregulation of natural cytotoxicity receptors, especially NKp30 and NKp44.⁷³¹ Intriguingly, GSCs demonstrate resistance to unstimulated NK cells but exhibit heightened sensitivity in co-culture models following pre-treatment with IL-2 and IL-15.⁷³² This preferential susceptibility might be mediated by increased expression of Natural-Killer Group 2 Member D (NKG2D) ligands UL16 Binding Protein 1 (ULBP1), ULBP2, and MHC Class I Chain-Related Protein A (MICA) on CSCs.⁷³³ Beyond their capacity to directly eliminate CSCs, NK cells can also induce their differentiation. In the presence of CSCs and IL-2, the cytotoxicity of NK cells is suppressed, and cytokine production is enhanced, a state referred to as “split energy”.⁷³⁴ These split anergic NK cells secrete high levels of Interferon-γ (IFN-γ), which induces the expression of MHC-I, differentiation receptors, and PD-L1 while reducing CD44 levels on CSCs. This induction of CSC differentiation subsequently leads to slowed tumor growth and decreased metastatic spread.⁷³⁵ Therefore, NK cells appear to counter tumor progression through a dual-step mechanism: initially eliminating a portion of CSCs and then, following a phase of split energy inducing cellular differentiation within the remaining CSC population.⁷³⁶ However, the local tumor microenvironment can directly inhibit NK cell effector mechanisms. Tregs suppress NK cell functions in a TGF-β-dependent manner, while CAFs inhibit NK cell functions through cell-cell communication and the release of PGE2.^737,738,739 CSCs can also impede NK immune responses through various inhibitory mechanisms. In metastatic melanoma, the expression of Indoleamine-2,3-Dioxygenase (IDO) and/or production of PGE2 can modulate the expression of NKp30, NKp44, and NKG2D.⁷⁴⁰ In neuroblastoma, TGF-β suppresses NK cell functions by regulating the expression of activation receptors and chemokine receptor repertoires, chiefly interfering with their migration and accumulation within tumor nests.^741,742 In ovarian tumors, the expression of Macrophage Migration Inhibitory Factor (MIF) and the glycoprotein MUC-16 can downregulate NKG2D and disrupt the formation of synapses between tumor cells and NK cells.^743,744 Additionally, CSCs evade immune surveillance and reduce NK cell-mediated killing by actively shedding MICA and MICB and recruiting Tregs.^745,746 Kryczek et al. observed that IL-22 promotes the CSC phenotype in preclinical and patient-derived models, with IL-22 being produced by NK and T cells.⁷⁴⁷

In summary, immune cells’ fight against tumors mainly goes through three stages: immune elimination, immune equilibrium, and immune evasion. In the initial phase, T cells and NK cells identify and eradicate proliferating CSCs before they develop into full-blown cancer. Consequently, CSCs with high immunogenicity are gradually eliminated by the immune system, such as high MHC-I or NKG2D, leaving behind those with low immunogenicity or those in a quiescent state to survive into the second phase. Ultimately, these selected CSCs expand uncontrollably with the help of immunosuppressive effectors, and the immune system becomes incapable of suppressing them.

Immunotherapy targets cancer stem cells

Tumor immunotherapy represents a therapeutic approach that harnesses the immune system to generate tumor-specific immune responses, aimed at suppressing and eliminating tumor cells. Based on the different mechanisms of the immune response against tumors, tumor immunotherapy can be categorized into “active immunotherapy” and “passive immunotherapy”.⁷⁴⁸

The core of passive immunotherapy hinges on administering immune effectors with antitumor activity, such as tumor-specific T cells and antibodies. This method offers rapid action but fails to elicit a lasting immune response.⁷⁴⁸ Tumor-specific monoclonal antibodies (mAbs) represent the most well-known form of immunotherapy, widely utilized in clinical practice.⁷⁴⁹ The mechanisms for mAbs primarily encompass (1) specific recognition of molecules expressed on tumor cell surfaces, leading to tumor cell death via phagocytosis, complement system activation, and antibody-dependent cell-mediated cytotoxicity (ADCC); (2) disruption of signaling pathways essential for tumor cell progression and survival, or inducing death signals by binding to surface receptors; and (3) conjugation with cytotoxic drugs or radioactive isotopes for specific delivery to tumors.⁷⁵⁰ Adoptive cell immunotherapy (ACI or AIT) involves infusing immune cells with anticancer activity back into the patient. This includes chimeric antigen receptor T-cell (CAR-T) therapy, tumor-infiltrating lymphocytes (TILs) therapy, NK cell therapy, and cytokine-induced killer (CIK) cell therapy.^751,752 The principle behind these therapies is the isolation of immune cells with cytotoxic potential from the patient. CAR-T therapy involves genetically engineering isolated T cells to bind tumor cell antigens;^753,754 CIK cells, expressing both CD3 and CD56 membrane proteins, are a novel type of immune cell known as NK-like T lymphocytes with potent anticancer activity.^755,