Towards understanding cancer dormancy over strategic hitching up mechanisms to technologies

Abstract

Delving into cancer dormancy has been an inherent task that may drive the lethal recurrence of cancer after primary tumor relief. Cells in quiescence can survive for a short or long term in silence, may undergo genetic or epigenetic changes, and can initiate relapse through certain contextual cues. The state of dormancy can be induced by multiple conditions including cancer drug treatment, in turn, undergoes a life cycle that generally occurs through dissemination, invasion, intravasation, circulation, immune evasion, extravasation, and colonization. Throughout this cascade, a cellular machinery governs the fate of individual cells, largely affected by gene regulation. Despite its significance, a precise view of cancer dormancy is yet hampered. Revolutionizing advanced single cell and long read sequencing through analysis methodologies and artificial intelligence, the most recent stage in the research tool progress, is expected to provide a holistic view of the diverse aspects of cancer dormancy.

Introduction

Although chemotherapeutic agents and drugs targeting oncogenic driver genes have been developed and despite drugs and physical surgery being effective in reducing primary tumor burden, a large proportion of patients experience recurrence and metastatic outbreak [1, 2]. The majority of patients with advanced metastatic relapse have a severe prognosis and cannot be treated with current treatment strategies, due to resistance to pharmacological treatment, additional metastatic colonies, or functional organ damage [3,4,5]. Residual dormant cancer cells in a quiescent state after primary treatments, such as chemotherapy and radiation are important causes of metastasis and recurrence. Cancer cells reversibly alter their signaling, resulting in phenotypes such as proliferative versus dormant after primary treatment, can be dynamically altered by epigenetic changes [6, 7] and transcriptional regulation [8], and exhibit plasticity [9] and heterogeneity [10]. Although various intrinsic and extrinsic factors have been reported to control dormancy and reactivation of dormant cancer cells, it remains necessary to develop therapeutic strategies and treatment methods that efficiently target dormant cancer cells in clinic.

To characterize cancer cells, particularly those in dormancy, various sequencing has been used to dissect their genetic properties. Targeting certain genes to be turned on or off can decide the fate of cancer cells, however, the accessible techniques are still insufficient to decipher genetic features thoroughly. Single cell sequencing is used at the forefront to distinguish dormant cancer cells individually on their heterogeneity, nevertheless, long read sequencing that sheds light on in-depth information and is widely used in cancer biology has scarcely been reported. Boosting up the translation of genetic features into clinical decisions, artificial intelligence (AI)-assistant could be a promising prediction tool for the activity of thousands of dormancy-associated genes that are useful to understanding cancer dormancy biology and even more in diagnosis. Now AI can hardly be used in therapeutics directly, which requires further validation to be approved, so data analysis for training the AI is important in supplementing data generation. Overall life cycle of dormant cancer cells, the hallmarks, and putative environmental impacts have been elaborated, and how these were used to be targeted were expertly pointed in previous research [11, 12]. In addition to the previous perspectives, we aim to suggest advanced strategies for targeting cancer dormancy to achieve the ultimate goal of eradicating cancer.

Identifying dormant cancer cells and in depth understanding of properties are pivotal gaps to be addressed to compensate the barriers from current knowledge to translation into clinics. Herein, we discuss recent advances in describing the characteristics of mechanisms underlying cancer dormancy, and treatment strategies of dormant cancer cells. Attempting to elaborate on how to catch up on these gaps, we summarized interdisciplinary approaches that cover current edge sequencing technologies and the emergence of AI, which could be harnessed to the field of dormancy. Proposed combining multiple advanced technologies and analytical methods are expected to dissect modalities of cancer dormancy.

Definition of cancer dormancy

Cancer dormancy state can be classified into two distinct types: tumor mass dormancy and cellular dormancy [2, 13, 14]. In tumor mass dormancy, the tumor mass continues to divide until it is physically restricted by its size, immune system response, and blood supply. In this situation, cells are not fully inactive, but have limited expansion and a balance between proliferation and apoptosis is maintained and eventually net size of tumor is unaltered. Cellular dormancy is the process by which cells enter a reversible cell cycle arrest called quiescence [3]. Dormancy is a stage of cancer progression in which cells stop dividing but survive in quiescence while waiting for appropriate environmental conditions to resume proliferation. During quiescence, cells are not dividing, are extremely inactive and asymptomatic, and are arrested in the cell cycle in G₀-G₁. Within the tumor tissue, dormant cells arise by transition from proliferative cells. Dormant cancer cells are thought to exist in the successful treatment of primary tumors, initial tumor progression, and in micrometastasis. Additionally, dormant cancer cells may also have other mechanisms to evade immune responses.

Life cycle of dormant cancer cell and disseminated tumor cell

During evolution of dormant cancer cells from slow cycling cells to macrometastases in various organs involves a series of steps in life cycle of dormant cancer cells (Fig. 1A).

Fig. 1

Life cycle of dormant cancer cell. A Given a stress situation such as chemotherapy, proliferative cancer cells enter the dormant state. Dormant cancer cells become more drug resistant and disseminated, and when they circulate in the blood, they adapt to their surroundings and survive. The surviving dormant cancer cells escape from the blood vessels into the surrounding tissues and colonize them. After micrometastasis, dormant cancer cells reactivate, re-enter the proliferative state, and relapse cancer. B Factors governing dormancy and reactivation. Dormant and proliferative cancer cells exhibit distinct characteristics due to epigenetic regulators, transcription factors, biochemical factors, environmental factors, and others. Factors written in the red box indicate activated factors during reactivation and in proliferative cancer cells. In contrast, factors in the blue box represent activated factors during dormant state and in dormant cancer cells

Drug resistance

Dormant state has been identified as a key survival feature of cells initially exposed to anti-proliferative cancer drugs. Dormant cancer cells often survive primary cancer treatment; they can survive selectively after primary treatment and drug resistance can be induced during chemotherapy. Dormant cancer cells switch on stress-induced signaling such as the p38 MAPK pathway, PRKR-like endoplasmic reticulum kinase (PERK), and hypoxia-inducible factors (HIFs), which facilitate drug resistance [15,16,17]. Several studies have been identified that drug-tolerant persister cells are originally present in primary tumors and drug therapy may induce the selection of these dormant cells [7]. Conversely, other studies have suggested that drug treatment may promote the transition from proliferative to dormant cancer cells [18].

Dissemination and invasion

Dormant cancer cells that acquire drug-resistant phenotypes disseminate to distant organs. The dissemination of cancer cells corresponds to the earliest stage of the metastatic cascade [3], and is triggered by chromosomal instability caused by continuous errors in chromosome separation during mitosis [19].

Epithelial cells are difficult to move to other sites and are tightly connected to each other or the neighboring extracellular matrix (ECM) [20]. Epithelial-mesenchymal transition (EMT) refers to the process by which epithelial cells lose their cellular polarity and cell-cell interaction abilities, gaining metastatic potential to become mesenchymal stem cells [21]. EMT orchestrates reversible biological changes that allow certain epithelial cells to acquire a mesenchymal phenotype and regulate epithelial plasticity, which is important for the dissemination and invasion of cancer cells [22]. Dormant cancer cells undergo EMT to achieve cellular stemness and invasive potential, leading to dissemination. In primary tumors that undergo EMT, they have been shown to develop migratory and invasive phenotypes. Disseminated cancer cells from metastatic organs encounter inhibitory signals that result in cell cycle arrest and maintenance of dormant state [23]. A study on the function of genes related to cancer metastasis in a mouse model of breast cancer found that asparagine synthetase, a metabolic enzyme, has a strong positive correlation with the potential for metastasis. Lowering the level of asparagine in vivo resulted in a significant decrease in cancer cell metastasis, suggesting that asparagine availability is closely related to EMT [24]. It is now widely understood that EMT programs represent a wide range of transitions between epithelial and mesenchymal states [25]. Tumors containing a mixture of cells with epithelial and mesenchymal characteristics are more efficient in dissemination, invasion, colonization, and metastatic development [26]. These reversible transitions are dynamically regulated by multiple growth factors and complex signaling pathways [27, 28]. The regulatory mechanisms of various cell characteristics that govern gene expression signatures and chromatin landscapes, such as transcription, chromatin structure, and epigenetics, have been identified. Moreover, different EMT stages are regulated by different microenvironments and stromal cells. For example, cells with high metastatic potential of mesenchymal phenotype rapidly proliferate near endothelial and immune cells [25, 29]. Additionally, diverse metabolites, hypoxic stress [30], and matrix strength [21] induce and maintain the EMT status in various cancer cells.

Intravasation and circulation

Intravasation, the spread of cancer cells to organs through the lumen of the vasculature, is either actively or passively mediated [20], and influenced by cancer type, tumor microenvironment and vasculature [31]. In a three-dimensional in vitro microfluidic model, endothelial cells were important for obstructing tumor cell invasion and were regulated by factors present within the tumor microenvironment [32]. Using a tissue-engineered tumor microvascular system, a novel mechanism by which tumor cells located around blood vessels destroy vascular endothelial cells through mitosis-mediated regulation and enter the circulatory system has been elucidated [33]. Moreover, the structural constraints of tissue are known to play a critical role in exerting pressure on tumor cell invasion into the blood vessels [34]. Cell adhesion molecules are a subset of cell surface proteins involved in the attachment of cells to other cells or to the extracellular matrix in the cell adhesion [35]. Integrins are key cellular adhesion molecules that are associated with nearly every stage of cancer progression from primary tumorigenesis to metastasis. Altered expression of integrin is found in a variety of carcinomas, and it is known to be essential for the robust maintenance of oncogenic growth factor receptor signaling, cancer cell migration, and invasion-related phenotypes [36].

For most invasive cancer cells, the environmental conditions in the circulatory system are very harsh. Interactions between circulating tumor cells (CTCs) and the microenvironmental elements of the circulation are important factors for the ability of CTCs to survive and metastasize to distant sites [37, 38]. Most CTCs travel in the circulatory system as single cell types; however, circulating clusters composed of several cell types, including stromal cells and immune components from primary tumors, have a high potential to form metastasis [39,40,41]. Neutrophils play an important role in CTC cluster formation, and increase CTC viability by regulating leukocyte activation [42]. Moreover, interactions between CTCs and platelets inhibit the detection of CTCs by immune cells and provide the necessary structures to tolerate the physical stress of circulation [43, 44].

Immune evasion

Disseminated tumor cells (DTCs) may interact with the immune system, with both intrinsic and extrinsic mechanisms protecting them from immune attacks. Pancreatic DTCs spread to the liver of mice [45] and bone marrow DTCs isolated from patients [46] downregulate the expression of MHC class I, which is critical for CD8 + T cell recognition. To minimize the ability of the immune system to detect dormant cancer cells, these cells reduce the expression of molecules that immune cells can recognize DTCs [4]. Recently, circulating dormant cancer cells have been shown to evade immune responses by expressing programmed death ligand 1 (PD-L1) [47, 48]. Additionally, the microenvironment of DTCs may contribute to immune cloaking. For example, a subpopulation of immunosuppressive T regulatory cells in the bone marrow regulates hematopoietic stem cell engraftment and the cell cycle through adenosine-mediated preservation from oxidative stress and immune killing [49]. The perivascular niche can repress immune function through the expression of PD-L1 and production of immunosuppressive cytokines such as IL-6 [50,51,52,53]. This allows dormant cancer cells to evade the immune response until metastasis begins. Therefore, their survival can persist for an extended period.

Extravasation and colonization

An important factor for cancer metastasis is the ability of CTCs to attach and extravasate via vascular endothelial cells to colonize the metastatic sites. When CTCs stop in the capillaries, they either leak out via transendothelial migration or grow intravascularly [54,55,56]. Because organs such as the liver and bones are composed of extremely porous sinusoidal vessels, CTCs have a high metastatic potential in these organs [20]. Additionally, extravasating cells encounter rigid barriers and basement membranes that require genetic and molecular mediation to pass through. CTCs can survive in a lack of nutrients and oxygen in the bloodstream and be maintained under immune destruction. Dormant cancer cells flow through the blood to distant organs and penetrate blood vessels to reach metastatic organs. Dormant cancer cells arrive at distant organs, they can convert into a proliferative state through reactivation. Dormant cancer cells undergo mesenchymal to epithelial transition to form metastatic outgrowth in target organs by acquiring epithelial characteristics [57]. Extravasation is a complicated process involving various factors such as ligand-receptor interactions, cytokines, chemokines, and surrounding non-carcinoma cells [58, 59]. Integrins play an important role in governing extravasation by regulating anchorage-independent survival of CTCs [36]. In recent years, it has been shown that DTCs activate the programmed necrosis of endothelial cells similar to leukocyte transendothelial migration, leading to the extravasation of metastatic cells [60]. Additionally, the RIPK family of serine/threonine protein kinase plays an important role in epithelial cell necroptosis and DTC extravasation [61].

DTCs that extravasate into target organs face severe environments that hinder survival [29]. Various cancer cell-derived factors and bone marrow-derived cells mediate the formation of a premetastatic niche in which DTCs can settle and grow [62, 63]. Recent studies have demonstrated that exosomes regulate the metastatic niche composed of bone marrow progenitor cells and play an important role in the formation of a premetastatic niche in the liver [63, 64]. Moreover, communication between DTCs and host cells is critical for cancer cell colonization. Hepatocytes play a role in increasing the sensitivity of the liver to metastatic colonization by regulating the activity of myeloid cell and fibrotic potential in the liver. They are also closely associated with the activation of IL-6/signal transducer and activator of transcription 3 (STAT3) signaling and increased secretion of serum amyloid A1 and A2 (SAA). Successful inhibition of the IL-6/STAT3/SAA axis inhibits the formation of a premetastatic niche and metastasis of DTCs in the liver [65]. Recent studies have highlighted the importance of the vascular system for proper metastatic colonization. In a mouse model of breast tumor heterogeneity, overexpression of SERPINE2 and SLPI affected the ability of distant metastasis of breast cancer cells [66].

Factors determining dormancy and reactivation

During tumor progression, the transition between dormant and proliferative cancer cells occurs dynamically through epigenetic, transcriptional, biochemical, and environmental factors, which help cells acquire the distinct characteristics of each state (Fig. 1B). Chemotherapeutic agents targeting fast-growing cancer cells can induce reversible changes in several cellular signaling pathways, causing the cells to enter a dormant state.

Epigenetic regulators

Melanoma cells with high expression of H3K4 demethylase JARID1B (KDM5B) have been shown to sustain a delayed cell cycle via regulating Jagged1/Notch1 signaling [67]. In fast-growing glioblastoma state, H3K27 methyltransferase Enhancer of Zeste 2 Polycomb Repressive Complex 2 Subunit (EZH2) was highly expressed. In contrast, H3K27 demethylases KDM6A/B were upregulated in dormant glioblastoma after receptor tyrosine kinase (RTK) inhibition and essential for the maintenance of slow-cycling persistent glioblastoma stem cells [68]. Methylcytosine dioxygenase Ten-Eleven Translocation 2 (TET2), known for its role in regulating 5-hydroxymethylcytosine (5hmC) in DNA and tumor necrosis factor α (TNFα) signaling, was found to be highly expressed in chemoresistant dormant cancer cells of colon cancer, melanoma, and glioblastoma [6].

Transcription factors

In head and neck squamous cell carcinoma (HNSCC), c-JUN promotes G₁-S phase transition and FoxM1, which promotes the G₁-S phase/G2-M phase transition, and is highly expressed in fast-growing cancer cells [69]. Transcription factors SOX2, SOX9, NANOG and RARB are highly expressed in dormant HNSCC, prostate cancer, and breast cancer. NANOG promotes the production of acetyl-CoA by mediating fatty acid oxidation (FAO)-ATP-citrate lyase (ACLY) signaling and transcription of p21 and p27 to maintain dormant state [70]. Treatment of 5-Aza (5-Azacytidine) and atRA (All-trans retinoic acid) promotes TGFβ2 and SMAD4 expression and inhibits proliferation to induce dormant state [71]. In ER + breast cancer, ZEB1/ZEB2, VIM, and FN1 are upregulated by transcription factors that regulate EMT and maintain a dormant state, whereas E-cadherin and Ki-67 are increased in proliferative cancer cells and mediate the reactivation of dormant cancer cells in metastatic organs [28, 72].

MYC is highly expressed in fast-growing cancer cells and is known to mediate cancer cell proliferation and survival. Among MYC transcription factors, MYCN has been shown to determine the fate of cancer cells after treatment with anti-proliferative drugs. MYCN-low cells enter senescence, while MYCN-high cancer cells showed accelerated growth after chemotherapy [73].

Biochemical factors

The opposite functions of kinases p38 and ERK signaling have been largely investigated as regulators of the dormancy-reactivation transition. p38 is a negative regulator of the cell cycle; activated p38 induces a dormant state in cancer cells by activating the transcription factors p53 and nuclear receptor subfamily 2 group f member 1 (NR2F1) [14, 74, 75]. Upregulated p53 and NR2F1 in dormant cancer cells promote the CDK inhibitors p16 and p27, which induce G₀/G₁ cell cycle arrest [76]. However, in proliferative cancer cells, p38 activity is low and ERK is highly activated. When ERK is activated in the proliferative state, uPAR (plasminogen activator, urokinase receptor) expression is promoted and activated uPAR facilitates ERK signaling reflecting a positive feedback loop [75]. In Pancreatic ductal adenocarcinoma (PDAC), highly expressed uPAR induced by KRAS mutations promotes ERK signaling and the cell cycle [77]. Then, ERK signaling activates Focal adhesion kinase (FAK) and PI3K/Akt signaling in proliferative cancer cells [78].

F-Box and WD repeat domain containing 7 (Fbxw7) is a subunit of E3 ubiquitin ligase that is highly expressed in dormant cancer cells and promotes proteasomal degradation of CyclinE, c-Myc, Notch, and c-JUN to arrest the cell cycle in dormant cancer cells. S-phase kinase associated protein 2 (Skp2) is highly expressed in proliferative cancer cells and promotes proteasomal degradation of p27, p21, p57, and p130 to sustain proliferation [79].

Environmental factors

In recent studies, cancer dormancy is also affected by the extracellular microenvironment, including cell-cell and cell-ECM interactions. Aged lung fibroblasts have been reported to induce the reactivation and metastatic outgrowth of melanoma cells by secreting the WNT antagonist sFRP1. Secreted sFRP1 inhibits dormancy activator Wnt family member 5a (WNT5A) in melanoma cells [80]. In contrast, aged dermal fibroblast secretes sFRP2 that induces dormancy and drug-resistance in melanoma cells. Bone metastatic cells have been exhibited reduction of quiescence in the aged bone marrow microenvironment via secreted factors [81]. Additionally, dormant cancer cells were found to generate a type III collagen-rich ECM niche, which maintain the slow growth state of cancer cells [82].

In breast cancer, members of the IL-6 cytokine family are highly expressed and promote cell growth. In contrast, the leukemia inhibitory factor receptor (LIFR) induces a dormant state by activating STAT3 and SOCS3, which, in turn, inhibit the expression of cytokines, including IL-6. The induction of LIFR in disseminated breast cancer cells promotes a dormant phenotype in the bone marrow, where LIFR activation triggers the JAK/STAT signaling pathway to suppress cell proliferation and induce quiescence, allowing the cells to persist in a non-proliferative state [83, 84]. The Stimulator of Interferon Response CGAMP Interactor 1 (STING) pathway has also been identified as a key regulator of immune reactivity during the exit from dormancy in metastatic cancer cells. STING activity is upregulated in metastatic progenitors that re-enter the cell cycle but is suppressed by hypermethylation and chromatin repression during dormancy re-entry, particularly in response to TGFβ. Treatment with STING agonists in mice eliminates dormant metastases and prevents spontaneous outbreaks in a T cell- and natural killer cell-dependent manner [85].

Additionally, soluble factors secreted by hepatic stellate cells are known to influence the characteristics of cancer cells. Inflammatory cytokines IL-8 and monocyte chemoattractant protein-1 (MCP-1), secreted by activated hepatic stellate cells, drive the emergence of dormant breast cancer cells in the liver. IL-8, in particular, counteracts serum-starvation-induced growth arrest in MDA-MB231 cells and promotes cancer cell proliferation both in vitro and ex vivo. Blocking the IL-8 receptor, C-X-C motif chemokine receptor 2 (CXCR2), reduces IL-8-induced cancer growth, suggesting that IL-8 and MCP-1 play crucial roles in regulating cancer dormancy and recurrence in the liver [86]. A recent study identifies Coco, a secreted antagonist of TGFβ ligands, as a key regulator of metastatic dormancy and reactivation in breast cancer cells, specifically in the lung. Coco promotes the reactivation of dormant cancer cells by blocking lung-derived BMP ligands, which normally suppress traits associated with cancer stem cells. Additionally, Coco induces a gene expression signature linked to metastatic relapse in the lung, revealing that metastasis-initiating cells must overcome organ-specific antimetastatic signals, such as BMP signaling, to reactivate in specific niches [38].

Others

A recent study identifies Centrosome-associated protein chromosome 4 open reading frame 47 (C4orf47) as a key regulator of cancer dormancy in pancreatic cancer under hypoxic conditions. Upregulated by HIF-1α and C4orf47 suppresses the cell cycle and inhibits proliferation by increasing cell cycle repressors (Fbxw-7, p27, and p57) and downregulating promoters (c-myc, cyclinD1, and cyclinC). Additionally, C4orf47 induces epithelial-mesenchymal transition, enhances cell plasticity and invasiveness, and regulates the p-ERK/p-p38 ratio, suggesting its involvement in pancreatic cancer dormancy and its potential as a prognostic biomarker [87].

Genetic properties of cancer dormancy

Cancer cells metastasize to various tissues via the vascular or lymphatic systems. Metastatic cancer may exist in tissues in an inactivated, dormant state. These dormant cancer cells can be prompted to reactivate because of stimulation by various factors [88]. Dormancy scores are a method used to evaluate dormancy-related genes. The gene expression signature is used to measure dormancy score as a standard unit to calculate the potential for dormancy or late recurrence. The score is determined by normalizing the expression levels of dormancy-associated genes that are up- or downregulated in various human cell lines and patient’ samples, both in vitro and in vivo, and by measuring the sum of this gene expression. High scores indicates a greater likelihood of dormancy and late relapse [89]. Several factors, including the microenvironment, lead to cancer cell dormancy. For example, breast cancer cells within the ECM surroundings highly express IGFBP5, which affects tumor suppression [90]. Herein, the effects of gene expression by categorizing genes into coding genes and non-coding genes, considering how genes influence cancer cells through gene expression regulation, epigenetic regulation, and microenvironment. The representative hallmarks of cancer dormancy-associated genes include NR2F1, NANOG, C-X-C motif chemokine ligand 12 (CXCL12), Wnt5A, interferon regulatory factor 7 (IRF7), growth arrest specific 6 (GAS6), and basic helix-loop-helix family member E41 (BHLHE41), also, several noncoding genes includes NR2F1-AS1, and ELEANOR (Table 1) (Fig. 2). These genes act on one or more types of cancers by interacting with regulatory elements through various mechanisms (Fig. 3A).

Table 1 Dormancy inducible genes in cancer cell

Cancer dormancy in a state not growing and the cell cycle arrested go unnoticed for years. Recent studies have demonstrated that it can be induced by several cancer dormancy hallmark genes. These genes show their respective roles in diverse cancers through their associated mechanisms. When it comes to inducing dormancy in squamous cell carcinoma, NR2F1 was found to be highly expressed [69]. Knockdown of NR2F1 in dormant squamous cell carcinoma led to a decrease in p38α/β, p16, p27, and p15 levels that known dormancy-associated genes as tumor suppressing. Upregulated NR2F1 has also been observed in prostate cancer, which is known to have a long dormancy phase [107], and an epigenetic association by NR2F1 was related to DNA methylation and histone modification. Following treatment with the DNA methylation inhibitor 5-azadeoxycytidine (5-Aza-C), the expression of NR2F1 increased, and H3K4me3 and H3K27ac became abundant at the transcription start site, indicating transcriptional activation. The expression of SOX9 and RARβ were changed as a consequence of NR2F1 level alteration, indicating the role of NR2F1 in inducing dormancy (Fig. 3D) [91]. In addition, it was assessed the dormancy score with NR2F1 in breast cancer in a state of tumor cell dormancy or angiogenic failure. The dormancy score was highly evaluated with the ER + feature among breast cancer cell lines. The candidate NR2F1 within the upregulated genes were knocked down in the proliferative state of breast cancer cell MCF-7 of ER + luminal type. These cells were subcutaneously injected into the mammary fat pad of severely combined immunodeficient mice, and palpable cancer was detected in all mice after 12 days [89].

Fig. 2

Therapeutic strategies targeting dormancy-reactivation of cancer cells. A Proliferating cancer cells can be eliminated using traditional anti-proliferative chemotherapeutic drugs through a process of inducing reactivation of dormant cancer cells. B Dormant cancer cells after primary treatment for cancer can be maintained in dormant state for a long time. C After primary treatment for cancer, residual cancer cells can be completely eradicated. The size of the yellow circles refers the expected tumor burden at each step

The transcription factor NANOG also plays a role in inducing dormancy in DTCs. Maintaining pluripotency and self-renewal in stem cells by NANOG has been studied to understand how it induces dormancy in colorectal cancer cells. NANOG is known to promote the self-renewal and growth of colorectal cancer stem cells [108]. NANOG knockdown increased the number of colorectal cancer cells by promoting proliferation, whereby the cells entered the G₀ phase under serum-free conditions. The expression levels of cell cycle arrest-associated genes that encode for NR2F1, hDEC2, p21, and p27 were increased, and NANOG regulatory function was observed in the FAO pathway, which is utilized for energy by cancer cells [109]. FAO increases the expression of ACLY, and subsequently enhances the production of acetyl-CoA, which is necessary for the acetylation of histone residues to promote transcription. Therefore, increased H3K27ac in the promoter region of NANOG increases transcription to induce NANOG-mediated dormancy [92].

Because cancer cells can enter dormancy during chemotherapy, genes related to dormancy entry and exit can be regulated through chemokine signaling followed by treatment of glioblastoma with temozolomide (TMZ). TMZ treatment of GBM cell lines LN229 and T98G highly modulated differential gene expression levels, contributed by CXCL12 along with CXCR4 and CXCR7 axis, and CXCL16 and CX3CL1. Moreover, TMZ treatment regulated the genes associated with dormancy entry and exit in GBM cells. The genes related to entering the dormant state include chemokine C-C motif ligand 2 (CCL2) and SAA2, whereas the genes related to exiting the dormant state include follistatin-like 3 (FSTL3), thrombospondin type 1 domain containing 4 (THSD4), and vascular endothelial growth factor C (VEGFC). Defective CXCR7 reduces the expression of dormancy entry genes, which supports the CXCL12-CXCR4-CXCR7 balance as a crucial axis for regulating dormancy. Stimulation with CXCL16 and CX3CL1 decreased the expression of THSD4 and VEGFC, whereas the expression of FSTL3 was unaltered [93].

Bone metastasis from prostate cancer can maintain a dormant state through the Wnt/β-catenin signaling pathway, specifically including the WNT5A. WNT5A inhibits the proliferation of prostate cancer cells and promotes the expression of siah E3 ubiquitin protein ligase 2 (SIAH2) by the receptor tyrosine kinase-like orphan receptor 2 (ROR2)-dependent manner. The increased Siah2 levels subsequently inhibit Wnt/β-catenin signaling, indicating the associations of WNT5A with dormancy [94].

Breast cancer is nonlinear resulting in cells that survive chemotherapy and then become dormant [110]. The type I interferon (IFN) pathway-associated gene IRF7 also participates in breast cancer mechanisms, through its high expression level in the dormant state after chemotherapy. The presence of dendritic cells, T cells with CD4 + or CD8 + in breast cancer cells have also been identified after chemotherapy. Silencing IRF7 diminished the level of IFN-β, besides, blocking the IFN receptor led to CD4 + and CD8 + cells reduction, which weakened the suppressive power of tumor formation accordingly. Increased IRF7 promotes IFN-β signaling and increases the levels of CD4 + and CD8 + indicating that IRF7 putatively affects chemotherapy-mediated dormancy of breast cancer cells [95].

Protein kinase D (PKD) promotes tumor angiogenesis by enhancing the expression of mastocyte-mediated angiogenic factors in the prostate cancer microenvironment [111]. Osteoblastic PKD1 induces prostate cancer cell dormancy by increasing the ERK/p38 activity ratio, and prostate cancer cell proliferation increases upon PKD1 deficiency. PKD1 positively regulates cAMP response element binding protein 1 (CREB1), a transcription factor, accompanied by higher or lower levels of PKD1 and phosphorylation. GAS6 expression is higher during growth arrest [112]. The interaction of GAS6 with CREB1 was assessed in CREB1 deficient osteoblastic cells, where the upregulation of PKD1 blocked the GAS6 expression. In contrast, activated CREB1 promoted the expression and secretion of GAS6. Therefore, the survival and proliferation of osteoblastic cancer cells can be converted to dormancy through PKD1-derived GAS6 signaling regulation [96].

BHLHE41 was associated with p38 MAPK-mediated dormancy in breast cancer has been reported [69] and p38 knockdown resulted in the proliferation of breast cancer cell MDA-MB-231 in the endosteal bone niche. Genes that suppress tumor cell proliferation have been identified in the bone niche, including BHLHE41, ADRB2, CASR, CD63, CDC2l1, FLT1, HBP1, KEAP1, LSP1, NOB1, NRG1, P11, TTF1, and WNT3, which are associated with p38 signaling and metastasis. Downregulated BHLHE41 especially increased proliferation of breast cancer cells in the endosteal bone niche. Additionally, BHLHE41 affects tumor formation by interacting with TP53; therefore, BHLHE41 deficiency causes mutations and loss of TP53, leading to encode tumor suppressor proteins [97].

In addition to gene expression regulation, several factors have been found to induce dormant states, such as histone modification, hypoxic milieu, and the enrichment of laminin/collagen in ECM. Histone modifications activate multiple pathways associated with cancer dormancy, and hypoxia and laminin/collagen-enriched ECM promote molecules related to dormant cancer cells [113, 114]. In HNSCC, upregulated macroH2A variants, such as macroH2A1.1 and macroH2A1.2, induce dormancy in disseminated cancer cells. The expression of macroH2A1 is upregulated through the activation of the TGFβ2 and p38α/β pathways, and the macroH2A variants strongly inhibit tumor growth and reduce the nuclear signals of H3K9me3 and H3K27me3. Elevated macroH2A2 promotes TNFα and nuclear factor kappa B (NF-κB) cascades, by which macroH2A2 induces growth arrest of disseminated cancer cells and prevents metastasis in HNSCC [104].

Dormant cancer and epigenetics have been demonstrated to be interrelated with several epigenetic mechanisms in disseminated cancer cells that induce dormancy [113]. Histone deacetyltransferase inhibitors (HDACi) induce LIF regardless of the ER state, which suppresses the proliferation of breast cancer cells. LIFRs are known to impede breast cancer progression [115], accompanied by dormancy through STAT3, which is activated in breast cancer cells in response to stimulation of the LIFR cascade, allowing the transition of disseminated breast cancer cells into dormancy [84]. Treatment with HDACi sensitizes LIFR, resulting in STAT3 phosphorylation HDACi induces the acetylation of H3K9, leading to an increase in LIFR transcription, and may inhibit the proliferation of breast cancer cells [105].

In human papillomavirus (HPV) positive cancer cells, the tumor genes E6 and E7 stimulate the degradation of E6-mediated p53 to prevent tumor suppression. Under hypoxic conditions however, the downregulation of E6 and E7 permits normal activation of p53, which inhibits the proliferation of cancer cells [98]. In addition to hypoxia, ECM can induce dormancy. In lung cancer cells, laminin/collagen IV-rich ECM decreases cell proliferation and induces G₀/G₁ cell cycle arrest. Lung cancer cells with ECM were more abundant in the G₀/G₁ phase, and the presence of ECM reduced lung cancer cell invasion and mobility (Fig. 3C). The expression of genes related to tumor progression was attenuated, and the level of the anti-apoptotic molecule Bcl-xL was also diminished. Additionally, the protein level of NF-κB increased, while the expression of p-ERK and p-Akt decreased inducing G₀/G₁ phase cell cycle arrest [99].

Associations have been demonstrated for microRNA and long noncoding RNA in cancer having been accumulated [116, 117], especially regarding the functional regulation of cancer cell dormancy. The long noncoding RNA NR2F1-AS1 was demonstrated to be upregulated in ER + breast cancer, promotes metastatic dormancy, and regulates the expression of NR2F1 with the involvement of polypyrimidine tract-binding protein 1 (PTBP1). NR2F1-AS1 binds to NR2F1, which recruits PTBP1, a protein that regulates transcription by binding to the promoter region [118]. NR2F1-AS1 also strengthens the interaction between PTBP1 and NR2F1, and enhances the function of the internal ribosome entry site (IRES) of NR2F1 5’-UTR, facilitating the translation of NR2F1. Downstream, promoted cancer cell dormancy were revealed to be influenced by NR2F1-AS1/NR2F1-mediated miR205 and ΔNp63. miR205 represses EMT [119], and ΔNp63 is a variant of lacking full-length transactivation domain tumor protein 63 (TP63), a miR205 translational regulator [120]. NR2F1 represses transcription by binding to the promoter of ΔNp63; consequently, downregulation of ΔNp63 enhances dormancy [100].

A cluster of long noncoding RNAs known as ELEANOR promotes dormancy in breast cancer. ELEANOR was found in ER-positive breast cancer cells and facilitates late recurrences. Additionally, ELEANOR plays a role in the metastatic progression of breast cancer by activating the ER signaling pathway and CD44 expression in ER + breast cancer cells. Specifically expressed CD44 in breast cancer stem cells is associated with cancer progression, metastasis, and dormancy. When ELEANOR is reduced, both the population of breast cancer stem cells and the expression of CD44 decrease concomitantly. ELEANOR enhances CD44 to maintain a dormancy (Fig. 3A) [101].

The roles of microRNAs in breast cancer have been accumulated, and their implication is emerging in relation to breast cancer-associated dormant conditions. When the breast cancer cell lines MDA-MB-231 and T47D were co-cultured with bone marrow stroma, both cell lines were arrested in the G₀ phase and microRNAs were detected to pass through gap junctional intercellular communication (GJIC) between breast cancer cells and the stroma. CXCL12 expression decreases after contact with the stroma in breast cancer cells [121], followed by bone to breast cancer metastasis. Four identified miRNAs (miR127, miR197, miR222, and miR223) were found to have homology within the 3’-UTR of CXCL12 under stromal and T47D cell co-culture conditions. In particular, miR197 inhibited CXCL12 secretion, in turn, the increased miR197 attenuated CXCL12 levels, leading to cancer cell metastasis [102]. An miR190, induced cancer dormancy due to its soaring expression levels [122]. The known cancer dormancy-associated genes CBLB, PDCD4, and SNTB1 were upregulated in T98G glioblastoma and KHOS osteosarcoma cells, where miR190 expression levels increased. The expression of miR190 within the cellular machinery drives cancer dormancy, which also alters the gene profiles of overlapping genes responsible for IFN signaling [103].

The tumor microenvironment induces dormancy as inevitable factor in disseminated cancer cells [123], and noncoding RNAs also participate here playing a role in inducing hypoxia-associated molecules [124]. In salivary adenoid cystic carcinoma (SACC), the miR922/DEC2 axis is activated during dormancy. While miR922 is highly expressed, the dormancy-associated gene DEC2 is downregulated in SACC. The inhibition of miR922 impairs cancer cell mobility and induces DEC2 expression. microRNAs have been reported to play a crucial role in lipid metabolism reprogramming; however, the mechanisms remain unclear [125, 126]. miR922 increases FAO and decreases triglyceride synthesis, implying that miR922 may influence the lipid metabolism through the synthesis of triglycerides and cholesteryl esters in SACC. Moreover, HIF-1α reduced miR922 levels in hypoxic environments, which in turn increases the expression of DEC2 to induce dormancy [106].

Interrogate cancer dormancy through integrative genome sequencing

Dormant cells can be characterized by genetic signatures, associated with their lifespan and properties as aforementioned. Advanced techniques involving sophisticated deep sequencing have allowed us to access veiled genetic dissections in a wide range of individual cells. Cancer studies using sequencing technologies have attempted to elucidate features attributed to metastasis, growth, habitat, and diagnosis. For almost two decades, RNA-sequencing has provided a genomic view; however, bulk- and short read technical challenges limit the acquisition of missing information. A more holistic window that captures a more complete genome structure is now achievable via single cell, long read, integrated sequencing and analysis.

Single cell sequencing is a prominently adopted method for dormancy research, which assesses cancer cell types including patients with cancer, in vivo cancer model, and in vivo cancer dormancy model. Since the cancer cells and their probable dormant state exhibit genomic instability, it is possible to infer the promotion of genetic differences between original and derivative tumor cells. Single cell sequencing can be used to distinguish dormant from proliferative cells, allowing for a higher resolution of cancer heterogeneity [127]. For cell tracing, strategic single cell barcoding enables the genetic or optimal detection of cancer heterogeneity. Barcoded clones are engrafted and traced spatiotemporally based on their unique biological properties from primary to metastatic sites [128]. Tens of thousands of cells can be separately identified in parallel with their dynamic characteristics using single cell profiling, and this was represented in several in vivo studies.

A tumor model was developed by transplanting Wnt/iFGFR^GFP transgenic mice-derived tumor chunk into GH mice, which were divided into four different stages: primary, dormant, long-term dormant, and recurrent. This model was designed to explain the minimal residual disease and its stages in dormancy, single cell analysis revealed microenvironmental signatures and signal transduction statuses. Following FGFR inhibitor (BGJ398) treatment, the off-drug treatment lasted for 14 days; then, tissues were collected and analyzed with single cell resolution. Gene expression markers, including Nr2f1, Cdkn1a, Zfp281, and Gas6 were similar to those in cancer dormancy, and the model shared gene expression signatures with residual cancer patients. The data also described that the immune cell population is uniquely distributed by comparing primary and relapsed dormancy and long-term dormancy. Immunosuppressive cells, including regulatory T cells and myeloid cells, were dominant in dormancy, comprising the cellular milieu where Notch signaling seemed to affect cancer recurrence [129]. In this study, no genetic associations were found with disease-free survival in breast cancer [129], whereas correlations with relapse were found in another study that profiled metastatic bone-derived analysis [130].

Research models were developed using a reporter-tagged breast cancer cell line (PyMT-B01) injected directly into the left cardiac ventricle and tibia. Proliferative and dormant gene expression signatures were analyzed using single cells isolated from the bone marrow and lungs. DiD dye-positive PyMT-B01 cells were identified as dormant cancer cells with less fluorescence diminishing. In the context of gene expression, several genes, such as Cfh, Gas6, Mme, and Ogn were upregulated in the dormant state in both the bone and lungs. These genes were also distinguishable in both dormant cancer cells and patients with a history of breast cancer, showing higher expression levels, suggesting the possibility of recurrence after disease-free survival [130]. In another breast cancer study, a cancer dormancy model was developed using triple-negative breast cancer (TNBC), which accounts for 15% of all breast cancer types. PKH26 labeled MDA-MB-231 cell lines were separated based on their fluorescence intensity, and bulk sequencing was performed and compared with single cell data from the aforementioned studies [129, 131]. ECM-associated gene expression was enriched as a dormancy feature. These genes specifically participated in several pathways, including MAPK, Notch, and Hedgehog, and the immune cell composition varied [131]. Compared to TNBC, ERα+, the most prevalent breast cancer subtype shows higher recurrence risk after 5 to 20 years [132]. A model of ER + or triple-negative cell lines grafted to the milk ducts of immune-compromised mice (MIND) were used to study ER + tumor cell dissemination and dormancy. Intraductally injected cells metastasized to multiple tissues upon cell types with different radiances, many of which were detected in a dormant state in secondary tissues. Six months after MCF-7 cell line xenografts, single cell analysis revealed the transcriptomic profiles in both primary and metastatic lungs. Cancer dormancy cells expressed ZEB2, VIM, FN1, and COL3A1, whereas proliferative cells expressed CDH1, KRT18, and EPCAM [72].

A syngeneic RM1 model was adopted for castration-resistant prostate cancer. PKH26 stained RM1 cells were injected intracardially, followed by single cell analysis after 16 days from the long bones. Actively proliferative or dormant cells were isolated, revealing cancer dormancy highly expressing interferon regulated genes highly. There were tripled levels of Irf7, which is retained during cancer dormancy, among three different cell clusters: apparent dormant state, less dormant state, and proliferative cells, all showing recognizably different gene profiles [133]. Equal sequencing data were further analyzed in another study using public clinical data from The Cancer Genome Atlas (TCGA). Differentially expressed genes were identified that prostate cancer dormancy was associated with MAD2L1 gene expression, suggesting the utilization of prognostic markers. A meta-analysis was performed within 10 sets of GEO datasets from various types of cancer, including breast and prostate cancers. This provided an integrative interconnected map regarding dormancy, with potentially related genes, including CLU, APP, NEU1, by protein-protein interaction, gene regulation, and transcription factor-free gene regulation [8].

Long read sequencing can outperform previous sequencing methods in the context of missed information. Cancer cells are highly diversified due to their genetic variation, even within an individual tissue mass, referred to as complex single-gene variance [134]. These can emanate from the accumulation of genetic variations, such as repetitive sequences and structural variants from whole transcripts, as well as epigenetic and base modifications [135]. Long reads obtained through advanced sequencing compensate for the drawbacks due to frequent changes in cancer cells [136, 137]. According to lung cancer chemotherapy, cancer cells exhibit defective alternative splicing and accompanying splicing factor gene expression. Using PacBio isoform sequencing and next generation sequencing, alternative splicing events and differentially spliced transcripts were identified in a cisplatin-treated non-small cell lung cancer (NSCLC) cell line. In a comparison between noncancerous control and dormant and reactivated cells, cisplatin treatment diminished isoform diversity such as exon inclusion/exclusion, and overall changed the expression of RNA helicase (DDX and EIF4A) [138]. Regarding cancer therapeutic resistance, ER + breast cancer relapse is observed in almost half of the patients receiving endocrine-therapy, which implicates some alterations that induce resistance to therapy. Patients who showed cellular adaptation against therapy did not exhibit novel mutations in driver genes upon recurrence. Long read sequencing using nanopore in addition to single cell technologies, has revealed that epigenetic reprograming could be a key mechanism to explain the circumstances of dormancy; for example, changes in chromatin accessibility with augmented H3K9me2, H3K27me3, and H4K20me3 [139]. To understand dormant cancer cells, tumor microenvironments were analyzed based on the combined calculation of nanopore and next generation sequencing, and the abundance of immune cells that are known to affect cancer were revealed. Although nanopore sequencing shows limited read depth for detection, it is advantageous to obtain chromosomal rearrangement reads accurately, such as MYB::NFIB and SS18::SSX2 fusion gene [140]. However, a lack of long read sequencing in dormant cancer research can be found at the current point of publications.

To distinguish and understand the characteristics of cancer dormancy, the followings are required (1) isolating dormant statues of cells from proliferative cells and (2) uncovering genetic and epigenetic cancer cell heterogeneity that used to be hidden. Here, coupling advanced sequencing technologies and single cell long read sequencing has a strong advantage for these considerations. Although the application of this method has not been reported in cancer dormancy, single cell long read sequencing has been harnessed in numerous aspects of cancer research. The framework revealed allelic imbalance and SNV-mediated isoforms in ovarian cancer [141] and cell type-dependent profiling of mutations and isoforms in the tumor microenvironment [142]. Furthermore, long read sequencing may enlarge the genomic window through long-range phasing to assess the assembly of each haplotype. It is expected to help demonstrating allelically differentially methylated regions (aDMRs), including promoter regions, as well as versatile onco-viral integration and extrachromosomal DNA (ecDNA) [143].

Clinical goals regarding dormant cancer cell

Despite clinical trials of primary treatments mostly targeting proliferative cancer cells, residual dormant cancer cells survive in slow-cycling and drug-resistant states, and are a major cause of metastasis and poor survival of cancer patients. Given the risk of dormant cancer cells relapse and metastasis, it is necessary to develop effective therapeutic strategies that target both dormant and proliferative tumor cells.

Approaches to investigate and modulate the dormancy-reactivation transition

To understand the biology of dormant cancer cells that may utilize clinically, multi-omics data, including genome, exome, transcriptome, and epigenome data, were obtained from in vitro cancer cells, in vivo xenograft tissues and patient tumor tissues (Table 2). Most studies related to multiomics data production and analysis have been conducted at the level of cultured cells divided into dormant and proliferative states, indicating serious limitations to their clinical application. Recently, sequencing of primary, metastatic, and recurrent tumors was conducted in a TracerX study of patient tumor samples [144].

Table 2 Sequencing data from dormant or recurrent or metastatic cancer

Based on omics data and previous references, the transition between the dormant and proliferative states of cancer cells can occur reversibly and dynamically through intrinsic pathways, extrinsic signaling, remodeling of ECM [147, 148]. Biological methods, including gene disruption and targeted therapies to maintain dormancy or induce reactivation, are summarized in Table 3. Most of these methods have been examined in in vitro or in vivo mouse models and have not yet been validated in clinical trials. Three therapeutic strategies (awakening, sleeping, and killing dormant cancer cells) have been proposed to modulate the dormancy-reactivation axis [149] (Fig. 2).

Table 3 Summary of dormancy or reactivation inducers

Fig. 3

Mechanisms of various dormancy inducers in cancer cells. A Noncoding RNAs, including ELEANORs and miR197, regulate dormancy-inducing gene expression in the nucleus and cytoplasm, respectively. The ELEANORs bind transcription factors to promote gene expression, and miR197 suppress target mRNA expression. B Various axis and signaling pathways that participate in regulatory mechanisms to induce dormancy. STAT1 is activated by chemotherapy, promoting IRF7 transcription, which stimulates IFN-β that activates CD4 + and CD8 + T cells. FAO pathway increases ACLY to elevate the levels of acetyl-CoA, which activates the transcription factor p300 to promote the transcription of NANOG and enhance the activation of factors that induce cancer cell dormancy. CCL2 and SAA2 transcribed under CXCL12 stimulation. WNT5A activates SIAH2 via ROR2, subsequently inhibits β-catenin signaling. TGFβ2 activates p38 to translocate to promote the transcription of NR2F1. The elevated NR2F1 leads to increased levels of SOX9 and RARβ. PKD1 promotes CREB1 phosphorylation to translocation, which activates the transcription of GAS6. Transcription of BHLHE41 is enhanced via p38 cascade, which increased p53 as a consequence. C Maintenance of cancer dormancy is influenced by hypoxic conditions and the ECM. Dormant or proliferative condition can be sustained under low or high oxygen levels, and ECM tightness surrounding cancer cells affect oxygen permeability. D 5-Aza-C induces histone modifications H3K4me3 and H3K27ac promotes the transcription of NR2F1 and leads to cancer cell dormancy

Therapeutic strategies i. awakening strategy

To improve the efficacy of anti-proliferative drugs, awakening strategies have been proposed to re-enter dormant cancer cells into the cell cycle. Following the reactivation of dormant cancer cells into fast-cycling cancer cells, the reactivated cancer cells can be eliminated using well-established anti-proliferative drugs [149]. Awakening of dormant cancer cells can be induced by various ways described in Table 3. Dormant cancer cells have been reactivated in Fbxw7 knockout mice model and the treatment of Paclitaxel, an anti-proliferative drug, could eliminate the reactivated cancer cells efficiently [156]. Disruption or overexpression of several genes including WNT5A, COCO, PFKFB3, and DEC2, and environmental factors including aging and blood flow, have been revealed to reactivate dormant cancer cells; however, further studies are needed to determine the effectiveness of combination therapy with anti-proliferative drugs. With this strategy, undetectable residual disease may persist, because not all cells respond to cell cycle reactivation. Although there is a risk when reactivated cancer cells are not properly eradicated and minimal residual disease can occur, this strategy has the advantage of being applicable in a combination with effective chemotherapeutic drugs. The combined effects of awakening inducers and chemotherapy have been demonstrated; however, the clinical efficacy of awakening strategy remains unclear. Inhibition of the TGFβ/p38 MAPK signaling is known to reactivate dormant cells and strongly promote multi-organ metastasis of HNSCC, suggesting that reactivated cells may become more aggressive depending on the context [168]. Therefore, unless awakening is followed by highly effective anticancer drug treatment, this strategy may have a detrimental effect on the patient’s prognosis.

Therapeutic strategies ii. Sleeping strategy

Since the reactivation of dormant cancer cells causes recurrence and metastasis, a method for maintaining dormancy that does not worsen cancer progression was suggested. It maintains the dormant state of cancer cells by suppressing proliferative signals or activating dormancy pathways. Inducers that promote dormancy or inhibit reactivation can be utilized to maintain the dormant state of cancer cells. However, continuous drug treatment may be essential because the drug is not persistent and can only induce temporary sleeping in dormant cells. Similar to the existing anti-proliferative therapies, minimal residual disease and existence of non-responder cells may remain. STING agonists induce dormancy in lung cancer cells and inhibit metastasis and relapse, supporting the possibility of a sleeping strategy [85]. Dormancy of cancer cells can also be stably maintained through dormancy factors such as NR2F1, which is induced by the activation of the retinoic acid signaling pathway. TGFβ2, a component of the metastatic dormant niche, is known to maintain the dormant state of tumor cells [169]. Moreover, glutaminase inhibitors [158], neutralizing antibody targeting sFRP1 [80], environmental factors including aged cells [165], and type III collagen [82] have been shown to affect dormancy maintenance. The disadvantages of this strategy is that it does not completely remove dormant cancer cells from the body, and that this life-long treatment incurs high clinical costs.

Therapeutic strategies iii. Killing strategy

One strategy to complement the methods of awakening dormant cancer cells is to develop drugs and methods that directly kill dormant cells. By specifically inducing the death of dormant cancer cells, both fast- and slow-growing cancer cells can be removed simultaneously using combinational treatment with traditional chemotherapeutic approaches. YAP/TEAD and ROCK inhibitors, which can induce apoptosis by acting only in the dormant state context, have been assessed, but further investigation is needed on how to induce the death of drug-resistant dormant cancer cells [166]. Fibronectin inhibition has also been reported to hinder the survival of dormant cancer cells [167]. However, a major disadvantage of strategies involving the elimination of dormant cancer cells is the uncertainty regarding their killing efficacy. Additionally, given that current diagnostic technologies are unable to detect individual dormant cells in patients, assessing the efficacy of killing approaches is challenging. If dormant cancer cells are not completely killed, dormant cells that survive a particular treatment are likely to become more aggressive and potentially result in a significantly worse clinical prognosis.

Despite various studies on the molecular mechanisms and targets of the dormancy-reactivation axis, the efficacy of three proposed strategies in completely eradicating both dormant and proliferative cancer cells remain unproven and controversial. The findings discussed herein may facilitate novel pharmacological cancer treatments combined with established cancer therapies.

Artificial intelligence applies on prediction for cancer dormancy

AI is increasingly extending its influence over application on the study of cancer and provides unprecedented insights all around cancer. By mimicking human cognitive behavior for processing data, computational power drives the prediction biological mechanisms that reflect input datasets. Thereafter, an AI algorithm is used to find patterns within large generated datasets [170]. AI has been able to facilitate a significant shift in recent years by: first, exponential growth of computing power and the development of training models for machine learning and deep learning, and second, the wide-reaching digitalization of research data, including genomic, clinical, and image data [171, 172]. In addition, massive amounts of biological big data have been democratized, allowing easily access for utilization and integration into novel prediction [173]. Consequently, AI reshapes the existing research structure and is expected to uncover various insights (Fig. 4).

Fig. 4

Toward precision oncology in response to cancer dormancy. Multiple data may facilitate the prediction of cancer dormancy, including information from macromolecules, big data, histological patterns from tissue biopsy, and genetic information from liquid biopsy. The data go through an AI framework to allow the prediction of cancer dormancy and years to recurrence, which overcomes current medical diagnostic challenges

Although the field of studying cancer dormancy has extended its knowledge, we should go through several challenges for the ultimate goal of preventing cancer relapse after a period of dormancy; in turn, AI can be harnessed to confront these challenges.

• Systemic identification of dormant cancer cells. Despite the need to identify dormant cancer cells prior to evaluation and monitoring, few methodologies exist to detect such cells. Given assistance by AI is expected to assess molecular patterns such as circulating tumor DNA (ctDNA) or CTC analysis that may reflect dormant cancer cells before recurrence [174, 175].

• Prediction of recursive risk of cancer. Metastasis and relapse are continuing concerns during clinical remission, and it is difficult to predict the recursive risk. AI may anticipate the signs of cancer relapse by evaluating histological images, molecular patterns, and previously collected big data [176]. Particularly, AI-generated prediction looks forward to reducing time for therapeutic plans through remote monitoring of systemic conditions as well as metastatic cancer [177].

• Development of therapeutic strategies considering critical aspects of cancer dormancy. Though the clinical goal of complete remission with eradication of cancer dormancy requires a greater understanding of cancer biology and ethical concerns limit clinical trials. To address such limits, AI has been broadly adopted in drug discovery and in diagnosis and prognosis. There are no FDA-approved standard therapies regarding cancer dormancy and no exact AI assistance addressing cancer dormancy is available, in turn, leveraging AI to establish models may give exploited pathological insight to develop therapeutic strategies [178, 179].

Each of these steps is required to translate research into clinical application, and use of AI-assistant is a possible time-saving method in such achievement. For a more precise and delicate care of patients with cancer, AI could help detection, classification, genomics, TME analysis, treatment, follow-up prognosis, and drug discovery. The prediction of tumor origin after metastasis using deep learning has been attempted through cytological image training and exhibits predictive concordance with long-term outcomes of patients with cancer of unknown primary origin [180]. However, predicting whether a primary tumor will metastasize, will become a risk factor for recurrence, and where it will reside after first diagnosis and treatment remains a critical issue that needed to be addressed by AI.

Using a convolutional neural network (CNN) image-trained AI, AI-based Lung Adenocarcinoma Recurrence Predictor (AILARP) was trained to predict recurrence through pulmonary adenocarcinoma histological images from. 124 patients (128 tissue specimen images), achieving an accuracy of 0.87 area under the curve (AUC) over patient-wise image analysis [181]. Similarly, CNN-based digitized histological image training was conducted to predict recurrence of colorectal cancer. Cohorts of 376 and 377 patients in stages I to III were used, and the prognostic prediction of recurrence-free survival (RFS) was significant for all patients (P < 0.0001) [182]. The prediction of prostate cancer recurrence was also demonstrated using histological image training in 204 independent patients. Relying on the microscopic decision of the Gleason’s pattern, a CNN-based AI predicted the years to recurrence even beyond the International Society of Urological Pathology (ISUP) grade that was used to predict prominence [183]. A meta-analysis evaluating AI-based recurrence prediction after the first-line treatment of hepatocellular carcinoma (HCC) indicated its applicability, supported by its sensitivity, specificity, and AUC measurement [184]. However, most AI-based research is limited by training dataset characteristics, such as small cohort number, sample sustainability depending on storage duration, and incomplete information of patients, and far more. A systematic review of published studies on ovarian cancer diagnosis using AI addressed that an ambiguous bias in all studies [185].

Since the properties of dormant cancer cells are in a slow cycling or arrested state before they relapse, the diagnosis of potentially being cancerous again is demanding to detect by using non-invasive radiological scan [186]. Therefore, molecular surveillance is required to monitor whether the cells can be defined as cancer dormant at primary or disseminated metastatic sites. With the advent of genomic characterization, surveillance using liquid biopsy has been a promising approach in precision oncology. Immune deficits serve as a metastatic outbreak after long-term dormancy, and 0.01–0.5% of cell-free DNA (cfDNA) is released from tumor cells, referred to as ctDNA, into the blood of patients with cancer. For minimal residual disease in melanoma, increased ctDNA analysis has been suggested to predict cancer before radiological scanning [187]. Similar results have been obtained in breast cancer [188], and bladder cancer [189]. Concomitant with the amount of ctDNA, the methylation and mutation of ctDNA are potential biomarkers for cancer detection and monitoring [190]. Following the accumulation of cfDNA and ctDNA understanding, ctDNA detection may address several scenarios including recurrence and subsequent treatment strategies [191]. Studies on some cancer types have started to use machine learning to decipher concealed non-genetic traits of ctDNA. The distinct pattern of DNA methylation in both coding and noncoding regions of CpG can be a putative diagnostic marker of lung cancer, and six AI platforms were used to identify CpG methylation in lung cancer. Deep Learning (DL) and support vector machine (SVM) scored 1.00 AUC, postulating a high accuracy in epigenetic alteration-associated cancer prediction [192]. In addition to DNA methylation, 30.5% of cohort with chronic obstructive pulmonary disease (COPD) had ctDNA mutations, which were detected in plasma using deep sequencing. Machine learning model using multiple COPD clinical criteria predicted ctDNA mutations with AUC at 0.774, which is opposite to other data, including the emphysema index [193]. The level of ctDNA decreases by approximately 90% after oral cancer treatment, and prognosis can be admitted to detect ctDNA arising and mutation appearing. Although unspecific to oral cancer, machine learning models have shown high accuracies regarding cancerous or noncancerous traits [194]. Moreover, for the clinical translation of the detected ctDNA, classification of the cancer type is crucial, in addition to knowing the presence of cancer. Ensemble modeling, using Light Gradient Boosting Machine (LGBM) was performed on patient blood data covering ctDNA and protein biomarkers, followed by classify cancer type at 97.81% AUC and identifying at 99.95% AUC using hyperparameters [195]. The recurrence risk of resectable colorectal cancer was assessed in a 2 to 49-month follow-up of 6,061 patients. This was conducted as part of the CIRCULATE-Japan GALAXY observational study, providing evidence for the necessity of monitoring ctDNA to assess recurrence risk [196]. Despite the cfDNA fragment length peaking at 167 bp, longer fragments were also detected depending on the harboring nucleosome [197, 198]. In addition to the next-generation sequencing, long read sequencing has revealed over 1 kb cfDNA in plasma, which contained a high number of CpG sites to analyze methylation patterns as well as a large portion of informative SNP sites [199]. An AI model using ctDNA to directly predict cancer dormancy has not yet been reported; however, the AI models for cancer dormancy-related risks, including metastatic tumors or recurrence in the short to long-term after primary tumor detection, advocate its feasibility.

Meanwhile, the accuracy of AI-derived information requires further validation and target identification for collaborative strategies, including experimental, multiomics, clinical, and computational approach. The cancer research infrastructure has grown owing to bioinformatics and big data, and is constantly revolutionizing technologies with AI-mediated reshaping. AI has established that the training of immense information, multifaceted aspects, and delicately curated data is essential for optimal prediction outcomes. Cancer heterogeneity exhibits numerous kinds of transcripts, not only count in linear coding genes but also in peculiar forms of genes such as ecDNA. As a driver oncogene, ecDNA ranging from 1 to 3 MB in size has been found in 17.1% of tumor samples from the 100 K Genome Project, and is known to contribute to cancer treatment resistance [200]. Patients with hematological malignancies carrying ecDNA have a high propensity of relapse after complete remission [201]. Accelerated cancer recurrence is the opposite of cancer cell dormancy, and ecDNA-related treatment resistance compared with treatment-induced dormant conditions is another questionable aspect. Encompassing diverse forms of genes in cancers, the applications of long read sequencing could broaden precision cancer genomics [202], allowing the construction of entire sequences and gene modifications. Subsequently, federated learning, which is yet to be trained simultaneously, may provide insights regarding cancer dormancy to facilitate precision oncology.

Perspectives and conclusions

Cancer dormancy at the single cell level indicates that the cell cycle is slow or arrested at the G₀ stage, which exhibits primary cancerous, metastatic, immunogenic, and treatment-induced features. Cancer is a great burden worldwide, and the tremendous efforts to combat it still face challenges. One of clinical problem is that the dormant cancer cells can escape current therapeutics and become awake as a relapsed cancer by certain contextual cues; awake-killing or maintenance in dormant strategies are prominent in preventing cancer recurrence. Thus, a comprehensive view of cancer dormancy is required to postulate therapeutic targets not limited to proliferative cancer cells.

Highly heterogeneous cancer cells needed to be analyzed with respect to structural variant, epigenetic modification, haplotype phasing, and complexity determination. Integrative genome sequencing that hybridized single cell and long read sequencing enables to resolve of these information [203]. Nonetheless, sample collection to run the single cell sequencing for cancer dormancy is challenging due to their uncertain residence [12]. The rare cell types require specialized analysis, for instance, cell identification as a significant step for further clustering analysis demands caution not to be defined as a dominant cluster [204]. Especially for hybrid approaches are cost-prohibitive, which limits research progress worldwide. Meanwhile, the optimal usage of AI requires massively generated and well-curated data for training. Although some aspects remain challenged, integrative sequencing technologies, in combination with molecular sequencing at exceptional depths, are expected to determine the most applicable targets to interpret cancer dormancy. Likewise, AI assistance may speed up the translation into clinical utilization from multiple model-driven data.

To overcome these technical hurdles, various downstream analysis methods have been considered for hierarchical cell types, subtypes, and states [205]. It is possible to deduce the future state of single independent cells using RNA velocity, which measures the underlying kinetics of transcripts. RNA velocity has been used to decipher multiple rates of cell tracing from various stimulations using cellDancer, a deep neural network [206]. Pseudo-temporal analysis or trajectory inference also allows the assembly of individual cells in a lineage-specific manner to infer dynamic cellular paths [207]. Breast cancer cells MCF-7 may undergo dormancy or proliferation due to changes in p21 and CDK2 levels, which are also affected by hypoxic stimulation. A stochastic model using cell cycle trajectories demonstrated cell fate heterogeneity upon energy exposure, such as in certain microenvironments [208]. Tumor metastasis to multiple organs in a mouse model of minimal residual disease was analyzed via trajectory analysis using single cell sequencing. Distinct cellular states, including dormancy, were related to the Jagged-1/Notch cascade during delayed recurrence [129]. Trajectory analysis has also been suggested for the deep learning models VITAE and scTour to enhance data accuracy and integrity [209, 210]. Moreover, perturbation modeling guides external stimulus-mediated single cell characterization, which involves drug treatment-driven cellular dynamics. Predicting the response to certain biological perturbations, which was advanced using CellOT, an input-convex neural networks, may allow a wider understanding of the molecular machinery [211].

Combined approaches of single cell long read sequencing to access a holistic view of the cancer dormancy genome, and training AI using histological images, ctDNA, and ecDNA are daunting and currently bear technical obstacles. Nevertheless, many methodologies to determine modalities are revolutionizing the currently limited approaches that fail to uncover relationships and traits. Recent TracerX projects have also focused on ctDNA in liquid biopsy to track metastases and relapse risk in patients with NSCLC [212]. The development of clinical strategies through various classes of therapy, such as chimeric antigen receptor T-cell (CAR-T), nanotechnology, and gene editing, is also expected to be improved [213,214,215]. Leveraging the unanswered questions and new knowledge will pave the way for seeking desired outcomes with immense potential.