Tumor initiation and early tumorigenesis: molecular mechanisms and interventional targets

It is generally believed that tumorigenesis is a multi-stage process, wherein the initial step is the occurrence of an oncogenic mutation in a single somatic cell. The mutation endows cells with clonal advantages, allowing the mutant clone to expand and accumulate additional genetic and epigenetic alterations, ultimately resulting in an irreversible, highly heterogeneous, and invasive lesion¹ (Fig. 1). Mutations that confer growth competitiveness and promote cancer evolution are referred to as cancer driver mutations. Identifying driver mutations and revealing their roles in tumors represent key areas of focus in cancer genome research. Recent advancements in sampling and sequencing technologies facilitate the detection of somatic mutations and clonal expansion in normal tissues. It is surprising that even though driver mutations harbored by positively selected clones overlap to a great extent with cancer driver mutations and are pervasive in morphologically normal tissues, only a low annual incidence rate of cancer is diagnosed in populations. It is suggested that mutations alone are insufficient for tumor formation, and other prerequisite molecular events need to be identified. Additionally, humans have evolved various strategies to maintain homeostasis and defend oncogenic transformation. However, environmental insults and aging often disrupt the balance and increase the risk of cancer formation.^2,3 Although the mechanisms of these risk factors contributing to cancer progression have been widely explored, how they are involved in early tumorigenesis and interact with specific oncogenic mutations are still not completely understood. The non-genetic effects of external signaling may explain the paradox of genetic mutation and tumorigenesis. Epigenetic rewiring can serve as another impetus to release uncontrollable growth and survival potential.

Multistage tumorigenesis. In normal tissue, somatic mutations sporadically arise and either are eliminated by tumor-suppressive mechanisms or gain proliferative advantages to form clones. The mutant clones can still maintain homeostasis until they are exposed to additional stimulus. Their proliferation becomes uncontrolled, and malignant transformation initiates, progressing from premalignant lesions to advanced tumors. During this process, the transformed cells gradually accumulate additional genetic mutations and epigenetic alterations, exhibiting increasingly malignant traits such as immune evasion, structural disruption, and invasion. Simultaneously, the microenvironment of these cells evolves from being tumor-suppressive to supportive of malignancy. This includes dysfunctional immunosurveillance, the emergence of tumor-promotive inflammation, gradual transformation of fibroblasts to CAFs, as well as stiffening of the ECM. CAF cancer associated fibroblast, TAM tumor associated macrophages, MDSC myeloid-derived suppressor cell, ECM extracellular matrix. Created with BioRender.com

Full size image

Cells capable of forming a neoplastic phenotype after acquiring genetic and epigenetic alterations will henceforth be referred to as “transformed cells”. Their clonal evolution is the result of a balance between intrinsic competency and extrinsic selective pressures, which is influenced by neighboring competitors, the microenvironment, and the cooperative tissue architecture. It used to be difficult to detect the rare precursors of tumors, while being armed with innovative technology, the identities of transformed cells and their interactions with the environment are being elucidated. In this review, we explore the driver events that enhance the transforming competency of a cell into full-fledged tumors, and examine the key transitions underlying tumor initiation and early tumorigenesis driven by these events. In addition, given that numerous interventional strategies for advanced tumors are limited by their heterogeneity, premalignant stage is regarded as a promising timing for intervention.⁴ Therefore, we also summarize how the molecular processes can be utilized to predict patients who are at high-risk of developing consequential cancer, and to develop preventive strategies that intercept malignant transformation.

The earliest explanation for the origin of cancer can be dated back to the early 1900s, cell-free extracts of a diseased animal were able to transmit tumors to healthy animal, suggesting that tumors originate from a unit smaller than a cell⁵ (Fig. 2). In 1914, Theodor Boveri proposed the somatic mutation theory after observing chromosomal abnormalities in tumor cells.⁶ Subsequent studies validated DNA as the genetic material and revealed that tumorigenesis requires the accumulation of approximately six or seven mutations.^7,8 The term “oncogene” was introduced in 1960s when genetic material of certain viruses was verified to contribute to malignant transformation.⁹ The first specific tumor gene was identified in 1976 by Michael Bishop and Harold Varmus, that part of the DNA of avian sarcoma virus hybridized in the genomes of birds transforming normal cells to tumor cells, and named it as SRC.¹⁰ This indicated that the genetic material in our genome is capable of transforming normal cells. Subsequently, the first proto-oncogene, RAS, and tumor suppressor gene, RB1, were cloned in the early 1980s.^11,12,13 Following this, a significant number of these two classes of cancer genes were identified, accompanied by discovery of other forms of variations, including copy number alterations, translocations and promoter hypermethylation.¹⁴ In the middle of 2000s, benefiting from next-generation sequencing, cancer genomics flourished and promoted the launch of large-scale tumor sequencing initiatives, such as The Cancer Genome Atlas (TCGA) in 2006 and the International Cancer Genome Consortium (ICGC) in 2007.¹⁵ The TCGA consortium published its Pan-Cancer Analysis of Whole Genomes (PCAWG) data in 2020, which contained the whole genomic sequencing data of 38 tumor types from more than 2800 patients, largely expanding our understanding of cancer genomics.¹⁶ According to the influence in cancer development, mutations can be categorized as driver mutations and passenger mutations. The driver mutations confer fitness advantage for clone expansion while other preexisting mutations, lacking positive selection properties, are referred to as passenger mutations,¹⁷ and over 3,000 cancer driver genes have been identified experimentally or computationally to date.¹⁸ Notably, in the last decade, deep sequencing from low-input samples has helped to identify somatic mutations in normal tissues, which are highly concordant with the tumor driver mutations.¹⁹ It reveals a limitation of the somatic mutation theory, that is the mere presence of mutations is insufficient for tumorigenesis, suggesting that there are other driver events.

Research history of tumor initiation and early tumorigenesis. The upper section emphasizes the role of somatic mutations in tumorigenesis, while the lower section demonstrates the evidence of the driver events beyond genetic events. ICGC the International Cancer Genome Consortium, TCGA the Cancer Genome Atlas, PCAWG the Pan-Cancer Analysis of Whole Genomes, HTAN the Human Tumor Atlas Network

Full size image

On the other hand, Victor A. Triolo first proposed that cancer is a tissue-based disease in 1965.²⁰ Following studies have verified that the capability of mutated malignant cells to induce tumors is context-dependent. Injecting tumor cells into normal mouse blastocysts can result in the development of normal embryos, indicating that malignant cells alone do not necessarily lead to tumors.²¹ The role of tissue injury in Rous sarcoma virus-mediated tumorigenesis,^22,23 and tumors induced by carcinogen-treated extracellular matrices^24,25 both further confirmed that extrinsic factors influence the outcome of tumorigenesis. Accordingly, tissue organization field theory was proposed in 2011.²⁶ The theory posits that aberrant tissue organization and cell-cell interactions contribute to tumorigenesis, with carcinogens targeting the entire tissue. In 2018, the Human Tumor Atlas Network (HTAN) was launched,²⁷ aiming at setting three dimensional atlases at crucial transitions of multiple tumors, including tumor initiation and local expansion, based on single-cell and spatial methods, and elucidating complex interactions between cells and their dynamic tumor ecosystem. It is expected to help us better understand how microenvironmental factors and transformed cells cooperatively promote the early transformation. Furthermore, the pan-cancer analysis of epigenome, transcriptome, proteome, and post-translational modification were recently published,^{28,29,30,31,32,33,34} providing multidimensional information of the tumor biology and possibly giving insights for the research of tumorigenesis.

Genetic alterations

Single nucleotide variants

Single nucleotide variants continuously accumulate through lifespan, originating from errors during DNA replication and repair processes, resulting from both endogenous factors (e.g., cellular metabolites, reactive oxygen species, nitrogen species, and transposable elements) and exogenous factors (e.g., radiation, tobacco, alcohol, and other chemical mutagens). Spontaneous chemical modifications can also serve as mutagens.^35,36 Somatic mutations in morphologically normal tissues can establish a baseline for studying cancer genome evolution and for identifying key drivers of malignant transformation. In recent years, a series of studies have analyzed the mutational landscape across nonmalignant tissues, shedding light on tissue-specific mutational burdens, mutational signatures, and the spectrum and frequency of driver mutations and their clonal expansions (Table 1), which can be influenced by stem cell dynamics, tissue turnover patterns, and environmental exposures.^3,19 Mutational signatures, developed to depict various DNA damage and repair processes, offer insights into mutagenic mechanisms.³⁵ It shows that age-related signatures, such as single base substitution signature 1 (SBS1) and SBS5, are prevalent across phenotypically normal tissues, although their contributions vary.^37,38 These signatures are the primary mutagenic factors in most types of tissues, especially those with high rates of cellular proliferation, such as the intestines.^37,38 In contrast, exogenous mutational signatures often play a relatively minor role. However, there are some exceptions, such as the SBS22 mutational signature associated with aristolochic acid, which is common in the liver samples³⁷ and is also significantly enriched in the urothelial samples from Chinese donors.³⁹

To explore intra-individual heterogeneity, our laboratory analyzed 9 normal organs from the same donors, and found that the liver exhibited the highest mutational burden, significantly surpassing that of other epithelial tissues, whereas the pancreas had the lowest level of mutation burden.³⁷ In addition, we compared the mutational signatures across organs and found that aging induced mutagenesis was the most prevalent, although it varied significantly among different tissues. Certain organs, such as livers, were largely influenced by exogenous mutagens. We also spatially reconstructed clonal architecture at sub millimeter resolution, and revealed how clone expansions associate with tissue microstructures, harbored mutations, and environmental factors.³⁷ Similar phenomena have also been observed in other studies.^38,40

Similar to driver mutations in cancer, mutations conferring fitness are positively selected and promote clonal expansion in nonmalignant tissues. Intriguingly, although most driver mutations are classical cancer mutations, they can maintain homeostasis in normal tissues, and exert opposite effects on tumorigenesis.¹⁹ Furthermore, some mutations are less common in tumors than in normal tissues and have been validated to play a tumor-suppressive role through outcompeting oncogenic clones, exemplified by NOTCH1 loss of function (LOF) in the esophagus.⁴¹ In contrast, the frequency of some mutations increases in tumors, like TP53 in skin, esophageal and endometrial cancers and PTEN in endometrial cancer, indicating their contribution to tumor development.³ Given that these mutations are generally tolerable in normal tissues, there should exist other factors to further promote their proliferative potential and initiate malignant evolution. To accurately identify additional driver events and the timing they emerge, multiple sampling is required. We recently revealed more detailed genomic changes throughout the entire process of esophageal squamous cell carcinoma (ESCC) formation, using multistep tumorigenesis samples ranging from normal tissue, through low-grade and high-grade intraepithelial neoplasia, to tumors from the same individuals.⁴² We also reconstructed their temporospatial evolutionary dynamics and confirmed that biallelic loss of TP53 in low-grade intraepithelial neoplasia is one of the earliest steps in initiating malignant transformation, serving as a prerequisite for copy number alterations (CNAs) in oncogenic genes involved in the cell cycle, DNA repair, and apoptosis pathways.⁴² It was also verified in mouse models of esophageal and pancreatic tumorigenesis that Trp53 loss of heterozygosity (LOH) is a critical step for genomic instability and malignant transformation. Meanwhile, heterozygous Trp53 mutation can maintain clonality only to a limited extent in normal tissues.^43,44

Copy number alterations and structural variations

Large-scale chromosomal alterations are another widespread form of genetic mutations, encompassing numerical and structural variations and constituting 80–90% of cancer genomes.^45,46 CNAs comprise aneuploidy, whole-genome duplications (WGDs), and extrachromosomal DNA (ecDNA), while structural variations include genomic catastrophes such as chromothripsis, chromoplexy, and breakage-fusion-bridge cycles. The complex genomic rearrangements have a reciprocal causation with chromosomal instability (CIN), an ongoing state in which cells accelerate the production of aneuploidy, and both of which can converge onto initial chromosome segregation errors.^47,48 It is speculated that chromosomal alterations occur very early in the evolution of specific cancer types, suggesting their potentially pivotal roles in tumor initiation.^16,49,50,51 Indeed, while CNAs and aneuploidy are rarely observed in normal tissues,^{19,52,53,54,55} they can be detected in precancerous lesions, albeit at much lower levels than that in fully formed tumors.^{56,57,58,59,60,61} Furthermore, the levels of CNAs and CIN in precancerous lesions they indicate can serve as indicators of malignant progression.^60,62,63 ecDNA, a unique form of CNAs, consists of double-stranded circular chromatids, and may serve as a robust driver of tumor genome evolution due to the absence of centromeric sequences and uneven distribution in daughter cells during mitosis.⁶⁴ Notably, ecDNA has been detected early in the progression from high-grade dysplasia in Barrett’s esophagus to esophageal adenocarcinoma (EAC).^64,65 Their copy number and structural complexity increased along the tumor evolutionary trajectory. Patients who progressed to EAC exhibited higher levels of ecDNA compared to those who did not.⁶⁵

Benefiting from multi-region sampling and single-cell sequencing, ongoing CIN and complex evolutionary processes of CNAs and structural variations can be depicted accurately.^66,67,68,69 Through multi-region sampling of Barrett’s esophagus concurrently containing different states of dysplasia and microscopic EAC foci, it has been reported that the evolution of CNAs during EAC tumorigenesis can be launched ahead of the development of dysplasia. Multigenerational CIN was initiated by mitotic errors and subsequent genomic catastrophes, including WGD, and inactivation of TP53 played an enabling role in the propagation of CIN, aggravating the accumulation of CNAs.⁶⁹ Recently, signatures of CNAs and CIN have been summarized from pan-cancer studies, encompassing numerous structural and copy number-related biological phenomena, such as WGD, aneuploidy, LOH, homologous recombination deficiency, chromothripsis, and haploidization.^46,70 It is expected to facilitate integrated analysis of CNAs and structural variations, so as to better elucidate mutational processes and genomic complexity.

Chromosomal abnormalities promote tumorigenesis through their effects on abnormal gene expression, including disruption or loss of tumor suppressors, oncogene amplification, and formation of oncogenic fusion genes.^47,64 Loss of the 3p arm, harboring tumor suppressor genes such as VHL, PBRM1, BAP1, and SETD2, can be an initiating event in clear-cell renal cell carcinoma. An increased frequency of LOH at 9p has been observed from precancerous lesions to cutaneous squamous cell carcinoma (CSCC), possibly driven by loss of tumor suppressive gene CDKN2A in this region.^71,72 Driver fusion genes such as EML4-ALK in non-smoker lung adenocarcinoma (LUAD) are speculated to be generated from complex chromosomal rearrangements, including chromothripsis and chromoplexy, and to arise in early years of life.⁷³ Specifically, ecDNA can both promote gene amplification and function as mobile enhancers regulating the expression of oncogenes.^74,75 Nevertheless, it is worth noting that the role of CNAs and structural variations in tumorigenesis are context-dependent.⁷⁶ Complex chromosomal aberrations are likely to exert deleterious cellular effects, inducing senescence, DNA damage, proteotoxicity, essential and toxic gene changes.⁷⁷ However, under specific conditions, aneuploid cells can be preserved, for instance, when WGD occurs ahead, providing extra copies of essential genes to alleviate deleterious alterations.⁷⁸ Furthermore, TP53 inactivation often occurs earlier to support the occurrence of WGD and clonal expansion.^79,80 There are also paradoxical immune activation and evasion induced by CIN. Chromosomal mis-segregation generates micronuclei, from which DNA leakage into the cytoplasm can activate the immune system, leading to the clearance of genomic unstable cells via cGAS-STING and type I interferon (IFN) pathway.⁸¹ At some points, tumor cells develop strategies to overcome the IFN signaling. Simultaneously, the secretome induced by CIN stimulate chronic inflammation and pro-tumorigenic effects.^77,82

Epigenetic alterations

The epigenome is another layer of information to encode cell identity and could be passed onto daughter cells. Upon development, natural aging and environmental exposure, there are dynamic changes in DNA and histone covalent modifications that remodel chromatin states and structures, and the heritable epigenetic marks, such as DNA methylation, being referred to as “epimutations”, serve as another important impetus of malignant evolution independent to genetic mutations.^83,84,85 Accumulating evidence suggests that clones with aberrantly rewired epigenetic programs show increased tumor susceptibility in morphologically normal tissues.^58,86 Particularly, age-induced DNA methylation changes are parallel to those seen in malignant states, including increased CpG island methylation and global hypomethylation.^84,87 During precancerous evolution, epigenomes undergo a stepwise progression, culminating in a high level of intra-tumor heterogeneity in invasive lesions. For instance, a gradual increase of methylation aberrations was observed transitioning from precursors to invasive LUAD.⁸⁸ Actinic keratosis, a precancerous lesion of CSCC, displayed classic cancerous methylome features, with two distinct methylation patterns suggesting different progression pathways to malignancy.⁸⁹ Precancerous colorectal adenomas have also already undergone genome-wide methylation changes and showed preliminary heterogeneity at the adenoma stage.⁹⁰ In specific tumors, such as ependymomas, it seems that epigenetic alterations play a decisive role, with only minimal genetic alterations detected.⁹¹

Tumor driver events induced by epigenetic reprogramming are presented as overly either restriction or permission states for gene expression, which can induce all hallmarks of cancer.⁸⁸ Highly repressive states induced by DNA hypermethylation lead to gene inactivation, often occurring in tumor suppressor gene related pathways, including DNA repair, cell cycle regulation, and p53 signaling.⁹² Additionally, hypermethylation of promoter CpG islands is frequently observed in lineage-specific transcription factor (TF) sequences that carry bivalent H3K4me3 and H3K27me3 modifications, transforming these previously poised sequences into inactive states that promote dedifferentiation and tumorigenesis.^93,94 We have confirmed this process in early esophageal tumorigenesis.⁹⁵ Overly permissive states, also known as epigenetic plasticity, can stochastically induce expression of pro-carcinogenic programs. For example, hypomethylation in enhancers and lineage-committed TF regions serves as an important mechanism in leukogenesis,^96,97 which has already been leveraged by DNMT3A^R822 clone in nonmalignant hematopoiesis, leading to chaotic transcriptional phenotypes and increased tumor risks.^96,97,98 Another way to induce permissive states and enhance cellular plasticity involves the suppression of Polycomb repressors, such as through the inactivation of histone methyltransferases, as exemplified by early lung tumorigenesis induced by KMT2D inactivation.^99,100 Although DNA hypermethylation mainly induces suppressive states, they can also promote gene expression through dysfunctional chromosomal topology.¹⁰¹ Abnormal hypermethylation at cohesin and CCCTC-binding factor (CTCF)-binding sites reduces the binding of insulator protein and formation of insulators, thereby promoting aberrant regulatory interactions like the activation of a constitutive enhancer for the tyrosine kinase gene PDGFRA to upregulate its expression.¹⁰¹ An integrative multi-omics atlas of 11 major cancer types indicated that tumor-specific and concurrent epigenetic driver events are associated with cancer transition, with enhancer accessibility playing a more specific role in transition from normal to different types of tumors.²⁸ The evidence above suggests that roles of distal regulatory regions and chromatin topology in tumorigenesis warrant further exploration.

Epigenetic alterations and genetic mutations have complex interactions in promoting tumor initiation, with genetic mutations possibly serving as primers to induce epigenetic changes, or epigenetic reprogramming potentiating oncogenic competence of genetic mutations.^102,103 Genes that encode epigenetic modifiers are common driver mutations in specific cancers and can occur in precancerous stage, such as TET2, DNMT3A and ASXL1 in hematologic malignancies,^104,105 and SWI/SNF chromatin remodeling complexes in solid tumors.^106,107 Recurrent tumor driver mutations also have capabilities to mediate epigenetic remodeling. For instance, one of the tumor-suppressive roles of p53 is to safeguard epigenetically regulated lineage commitment, and its role in limiting cell fate reprogramming has been proven in several cell types.^{108,109,110,111} The oncogenic effects of Kras mutations mediated by chromatin remodeling have also been documented.¹¹² Conversely, epigenetic priming might precede genetic mutations, rendering cells more susceptible to oncogenic signals, exemplified by aging-related DNA methylation which can activate the Wnt pathway to be more sensitive to Braf mutation induced colon transformation.^113,114 Furthermore, epigenetic abnormalities play a role in accumulating mutations, such as through spontaneous deamination of DNA methylation,¹¹⁵ and DNA hypomethylation induced CIN.⁶³ In addition, methylated promoters of DNA repair genes underlie a field wherein the colorectal cancer (CRC) with higher rate of mutations arise.¹¹⁶ Multi-region single-gland genome, epigenome, and transcriptome profiling of concomitant colorectal adenomas and tumors demonstrated that genetic and epigenetic mutations mutually promoted accumulation of each other. Mutational signature showed that the epigenome alterations induced DNA mutation, while driver mutations were also found in chromatin modifier genes.¹¹⁷ However, the functions of chromatin accessible driver genes and genetic driver mutations were independent. Some accessible drivers were devoid of mutations.¹¹⁷ Parallel evolution of methylome and genome was also observed in lung tumorigenesis, where global hypomethylation was associated with high mutation burden, CNAs and allelic imbalance, as well as immune infiltration.⁸⁸ Beyond genetic and epigenetic interactions, it is recently reported that chromatin accessibility could also be modified by RNA modification, another regulatory layer for gene expression, being known as epitranscriptome. N⁶-methyladenosine (m⁶A) modifications of RNA are the most common form of mRNA modification, and their roles in regulating transcript stability, translation and localization have been proven to be intricately involved in tumorigenesis.¹¹⁸ Recently, specific crosstalk between RNA m⁶A and epigenetic marks, such as histone modifications and DNA methylations, is being unveiled.^{119,120,121,122,123} Our work indicated that m⁶A in super-enhancer RNA is capable of activating YTHDC2 and recruiting H3K4 methyltransferase MLL1 for co-transcriptionally directing H3K4me3 demethylation as well as being accessible to oncogene transcription.¹²⁴ In addition, we also found that m⁶A in RNA could be the cause of DNA demethylation in nearby genomic loci in both normal and cancer cells, which is mediated by RNA m⁶A modification reader FXR1 to recruit DNA dioxygenase TET1.¹²⁵ Altogether, different aspects of chromatin regulation are integrated to regulate cell fate and function. A deeper understanding is warranted to explore their roles and causal relationships.

Environmental factors

There are diverse environmental and systemic factors that have been epidemiologically confirmed as tumor risk factors, encompassing chemical and radical insults, unhealthy metabolic behaviors, specific pathogen infections, as well as aging. They induce versatile alternations in whole or at local positions, including both induction of genetic and epigenetic alterations in transformed cells and profound impacts on microenvironmental components that predispose to tumor initiation (Table 2). Since inflammation is a convergent response to various environmental alterations, we discuss its role in this part at first, which is followed by context-specific mechanisms of other risk factors.

Inflammation

Inflammation is a conserved response to potential insults, being involved in tissue repair, regeneration, and homeostasis regulation by stimulating cytokine production and mobilizing innate and adaptive immune systems to remove insults and protect the integrity of the tissue.^126,127 While acute inflammation aims to solve damage and has tumor-suppressive effects, chronic inflammation caused by unresolved and persistent damage is a well-known tumor risk factor and is considered an enabling hallmark of cancer.¹²⁸ It can be triggered by numerous external stimuli associated with tumors, including chemical carcinogens, radiation, and infections.¹²⁹ Additionally, aberrant autoimmune reactions, such as reflux esophagitis, inflammatory bowel disease, and atrophic gastritis, as well as systemic and subclinical inflammation related to ageing and obesity, can trigger similar pro-tumorigenic effects.^129,130 The mechanisms by which inflammation is involved in early tumorigenesis include not only oxidative stress and DNA damage, but also priming or releasing the expansion and transformative potential of cells harboring oncogenic mutations.^131,132 This process can be exemplified by inflammation-stimulated TP53 mutation clone expansion in colonic and leukemic transformations.^133,134,135 Notably, expanding mutant clones in inflammation can play roles independent of tumorigenesis, such as the regeneration role of ARID1A, KMT2D and PKD1 in liver injury¹³⁶ and tumor-suppressive NFKBIZ mutation in colitis.¹³⁷

It has been widely confirmed that cytokines and growth factors in chronic inflammation play pro-tumorigenic roles, such as interleukin 1 (IL-1), IL-6, transforming growth factor beta (TGF-β), IL-17A, and IL-22, and their functions, which regulate cell survival, proliferation and cell fate determination can be hijacked by cells harboring mutations, activating mitogen-activated protein kinases (MAPK), phosphatidylinositol-3-kinase (PI3K) -AKT, Janus kinase (JAK) -STAT and NF-κB pathways to increase the risk of tumors.^131,132 Specifically, inflammatory mediators can play a decisive role in early malignant evolution. For example, the cooperation between Sox2 overexpression and inflammation activated STAT3 is capable of inducing ESCC, while in the absence of environmental stimuli, mutations alone may only enhance proliferation without progressing towards tumors.¹³⁸ Liver injury induced dedifferentiation is also a promoter of tumorigenesis, where both mature hepatocytes and cholangiocytes have the potential to give rise to different type of primary liver cancers, comprised of hepatocellular carcinoma and intrahepatic cholangiocarcinoma.¹³⁹ The lineage commitment is dependent on both mutation backgrounds and epigenetic regulations of the injury signaling.^140,141 Hepatocytes harboring oncogenic mutations induced intrahepatic cholangiocarcinoma upon stimulation of damage-associated molecular patterns (DAMP)-associated cytokines induced by liver cell necroptosis.¹⁴¹ By contrast, apoptotic microenvironment promotes transformation of hepatocytes with the same mutation background to hepatocellular carcinoma.¹⁴¹ In hematological system, since chronic inflammation leads to stem cell differentiation and exhaustion, mutations conferring resistance to inflammation stress, such as TET2 and DNMT3A, can be positively selected and form clonal hematopoiesis of indeterminate potentials (CHIPs). Tet2 LOF hematopoietic stem/progenitor cells (HSPCs) upregulated TLR-TRAF6 in response to inflammation, resulting in a shift from the canonical NF-κB pathway to the noncanonical NF-κB pathway, thereby avoiding inflammatory damage to mutated stem cells, and facilitating the Tet2 mutation-induced progression of myelodysplastic syndrome.¹⁴² Dnmt3A LOF CHIP could also prevent hematopoietic stem cells from terminal differentiation through increasing methylation of IFNγ signaling pathways.¹⁴³

The epigenetic plasticity conferred by inflammation lowers the barriers for malignant transformation. A typical example is pancreatic tumorigenesis initiated from Kras mutant acinar cells and promoted by injury and pancreatitis. Kras mutation is insufficient for transformation, and injury-induced inflammation is indispensable in the development of pancreatic intraepithelial neoplasms (PanIN) and pancreatic ductal adenocarcinoma (PDAC).^144,145,146 Inflammation induces transdifferentiation of acinar cells to ductal cells, which is a reversible process termed as acinar-to-ductal metaplasia (ADM), and can be resolved as tissue regenerates.^145,146 However, the reprogramming can be co-opted by Kras mutations to irreversibly transform the ADM program to PanIN and PDAC programs.^147,148,149 Distinct chromatin states between normal regeneration and Kras induced tumorigenesis could be mediated by a chromatin reader, bromodomain and extra-terminal family member reader, BRD4. The divergence was initiated as early as 48 hours after pancreatic injury induced by caerulein in mouse models.¹⁴⁸ Besides, another study identified that a precancerous cell subset with ductal identities and oncogenic potential had emerged in ADM, and Kras mutation maintained the pro-oncogenic programs, ultimately resulting in PDAC.¹⁴⁹ It is because inflammation activated AP-1 to dominate the pro-oncogenic transcriptional program, and its key components Junb and Fosl1 could be stabilized by Kras mutation.¹⁴⁹ Similar cooperation between gene and environment was depicted in oncogenic epidermal wound repair, where stress-induced TFs, such as AP-1, ETS2 and STAT3, induced transient lineage infidelity between epidermal stem cells and hair follicle stem cells. In tumorigenesis, stress-TFs were enhanced, resulting in a permanent lineage infidelity and newly activated oncogenic enhancers for malignant transformation, which were divergent from normal regeneration.¹⁵⁰

An emerging field of study of inflammation-induced epigenetic rewiring is tissue memory, which is an adaptation to recurrent stress and has been identified in various tissues, including skin, lung, intestine and pancreas.^{151,152,153,154,155,156,157} In parallel to the immune memory, epithelial cells set long-term memory based on epigenetic modifications they have adopted during injury, which can be partially maintained after the resolution, enabling a more rapid response to a next similar damage.¹⁵⁸ However, there is a trade-off between tissue long-term adaptation and tumorigenesis that the persistent abnormal epigenetic program primes a field permissive for tumorigenesis. For instance, pancreatic epithelium develops tissue memory of ADM to rapidly instigate a protective program for a secondary pancreatic injury and reduce tissue damage,¹⁵⁷ which can be enhanced by Kras mutations via MAPK constitutive signaling to increase fitness. Nevertheless, Kras mutations induces an irreversible ADM reprogramming and increase tumor risk simutaneously.¹⁵⁷ Similarly, in wound-priming epidermis, there are memory stem cells located in distal intact areas, which are prepared both to respond to another damage adaptively, and to give rise to tumors detrimentally. This is achieved through epigenetic and transcriptional reprogramming and mediated by a long-lasting loss of histone repressive mark H2AK119ub.¹⁵⁹

Chemical and radical insults

Environmental carcinogens are prevalent in nature, derived from air pollution, cigarettes, alcohol, ultraviolet (UV) radiation, etc. These carcinogens promote tumor progression through various mechanisms, including genotoxicity, epigenetic modification, chronic inflammation, immune suppression, oxidative stress, and activation of receptor-mediated signaling pathways.^160,161

In the tumor initiating stage, chemical and radical carcinogens not only induce mutations and contribute to specific mutational signatures,^54,162 but also promote clonal expansion of specific mutations. For example, driver mutations of CSCCs, such as NOTCH, TP53, FAT1 and FGFR3, are more prevalent in chronically UV exposed skin than in unexposed healthy skin.^52,163 Similarly, smoking promotes clonal expansion in the blood, including ASXL1, DNMT3A, and TET2 CHIPs.^105,164 Intriguingly, the landscape of clone expansion is likely to be reversible. The high mutational burden and driver mutation frequency in the bronchial epithelium decrease after smoking cessation, likely due to the rescue effect of quiescent cell expansion, which was previously protected from tobacco mutagenic insults.¹⁶²

The positively selected mutant clones are expected to exhibit resistance to stress. In sun-exposed skin, plasmacytoid dendritic cells with Tet2 LOF are protected from UV-induced cell death, providing a reservoir for the accumulation of more oncogenic mutations and subsequent malignant transformation.¹⁶⁵ Mouse esophageal stem cells harboring Trp53 mutations are less vulnerable to radiation-induced oxidative stress and replace differentiated wild-type cells for clone expansion.¹⁶⁶ HSPCs with Trp53 mutation were also insensitive to radiation-induced differentiation. Mutant p53 bound to enhancer of Zeste homolog 2 (EZH2), a catalytic subunit of Polycomb repressive complex 2 that is responsible for trimethylation of Lys-27 in histone 3 (H3K27me3), thereby promoting the expression of self-renewal program in Trp53-mutant CHIP.¹⁶⁷

In addition to providing a hostile environment, multiple insults can directly activate epithelial cells to induce epigenetic and transcriptional changes, or they can act on immune cells to trigger inflammatory responses, indirectly promoting tumor development. Nicotine activates the AKT-extracellular-regulated kinase (ERK)-MYC pathway via the nicotinic acetylcholine receptor and inhibits the Gata6 promoter, a key regulator of acinar cell differentiation.¹⁶⁸ This leads to the dedifferentiation of acinar cells and further promotes the activation of Kras mutation, thereby facilitating the transformation of Kras-mutant ADM and PanIN.¹⁶⁸ Chronic exposure to cigarette smoke has also proven to induce time-dependent epigenetic changes, which makes bronchial epithelial cells more susceptible to single Kras mutation induced tumorigenesis.¹⁶⁹ Alterations in transformed cells, such as epithelial-to-mesenchymal transition (EMT), anchorage-independent growth, and RAS/MAPK signaling upregulation, are closely associated with gene silencing induced by hypermethylation.¹⁶⁹ The Epidermal growth factor receptor (EGFR) gene mutation is identified as a common driver mutation in healthy lung tissue exposed to environmental particulate matter measuring ≤2.5 μm (PM2.5), and is associated with a higher incidence of LUAD.¹⁷⁰ Hill et al. showed that PM2.5 induced lung macrophage infiltration and secretion of IL-1β, which mediated the reprogramming of alveolar type (AT) II cells into a progenitor-like state.¹⁷⁰

Metabolic factors

Cellular metabolism is regulated by both intrinsic metabolic properties of the cell and the intake of external nutrients. Tumors modify their metabolic patterns to evade nutrient restraints and fulfill their heightened demands for aberrant growth and proliferation. Alternatively, tumors produce oncogenic metabolites that regulate gene and protein expression to promote tumor progression.^171,172 Recent findings suggest that metabolic remodeling begins earlier at precancerous stages. In early precancerous lesions of lung squamous cell carcinoma, activities such as fatty acid metabolism, oxidative phosphorylation, and the citric acid cycle are enhanced.^173,174 These early metabolic changes in tumorigenesis might play a role in driving tumor initiation by interacting with predisposed mutations.¹⁷⁵ There are two primary mechanisms. One is that mutations drive early metabolic alterations and adaptations. The other is that abnormal metabolic environment facilitates transformation of mutated cells. Classical oncogenic mutations, such as PIK3CA, TP53, RAS, and MYC, are all implicated in metabolic regulation by influencing the activity and localization of metabolic enzymes at transcriptional and post-transcriptional levels.¹⁷⁶ Specifically, they have the potential to recapitulate epigenetic modifications through upregulating expression of metabolic effectors. In the early stage of pancreatic tumorigenesis, mutant Kras and loss of Trp53 enhance acetyl coenzyme A and α-ketoglutarate synthesis, respectively. The metabolites, in turn, epigenetically promote dedifferentiation and PanIN formation.^177,178 Additionally, Kras mutations promote metabolic remodeling via post-translational modification of metabolic enzymes. They suppress ubiquitylation and degradation of branched-chain amino acid transaminase 2, an enzyme essential for the catabolism of branched-chain amino acids and mitochondrial respiration, thereby contributing to the progression of PanINs.¹⁷⁹ Apart from recurrent cancer mutations, mutations in genes encoding metabolic enzymes, including succinate dehydrogenase, fumarate hydratase, and isocitrate dehydrogenase 1 or 2, have the capability to accumulate oncometabolites, disrupting dioxygenases and their epigenetic regulatory functions.¹⁷¹ The isocitrate dehydrogenase mutation induced oncometabolite, (R)-2-hydroxyglutarate, was confirmed to promote the early tumorigenesis of acute myeloid leukemia (AML) and gliomas through the inhibition of histone lysine demethylases 5.¹⁸⁰

In addition to mutation-driven metabolic remodeling, unhealthy systemic metabolic status, including high-fat and high-carbohydrate diets, and metabolic diseases they induce, such as obesity and type 2 diabetes mellitus, can increase the risk of tumors.^181,182 Obesity triggers several pathological processes associated with tumor development, including hyperglycemia-related insulin resistance, abnormal hormone secretion, inflammation and dysregulation of lipid metabolism. Under physical conditions, insulin signaling systematically senses blood glucose levels and promotes proliferation and anabolic metabolism. In the presence of obesity, insulin resistance in metabolic tissues leads to hyperglycemia and hyperinsulinemia, while tumor cells develop strategies to maintain their sensitivity to insulin-induced proliferative signaling.¹⁸¹ Transformed mutant cells can adopt similar strategies, utilizing the proliferative signaling and gaining competitive advantages.^183,184 Furthermore, hyperglycemia induced by both glucose and fructose consumption enhances tumorigenesis by accelerating glycolysis and de novo lipogenesis.¹⁸⁵ Recently, glucose was reported to act as a signaling molecule, directly binding to and activating NSUN2, thereby activating NSUN2-TREX2 signaling. This led to inhibition of dsDNA accumulation, subsequent cGAS/STING pathway activation, and immune activation.¹⁸⁶ In terms of the high-fat diet (HFD), Sasak et al. found that it disrupted cell competition outcomes by enhancing lipid metabolism.¹⁸⁷ In normal epithelium, the apical extrusion of Ras^V12 transformed cells could be mediated by Warburg-like effects and damage to mitochondrial membrane potential. However, HFD increased the levels of free fatty acids and promoted their metabolic transformation to acetyl coenzyme A, which played a role in restoring the mitochondrial membrane potential and inhibited the clearance of Ras^V12 cells.^187,188 Furthermore, inflammation and immune responses play crucial roles in linking the pathological processes of obesity to tumorigenesis.¹⁸⁹ Pancreatic Kras mutation can downregulate peroxisome proliferator-activated receptor (PPAR)-γ, exacerbate inflammation and further promote the formation of PanIN, mediated by fibroblast growth factor 21, which is an endocrine regulator for metabolic homeostasis.¹⁸⁹ HFD also promotes the activation of PPAR-δ and the secretion of CCL2 in Kras-mutant pancreatic cells. Consequently, immunosuppressive cells are recruited, promoting the transformation from PanIN to PDAC.¹⁹⁰ Additionally, the mechanisms by which HFD promotes tumorigenesis are also related to microbial dysbiosis. Alterations in gut microbiota and metabolites are crucial for HFD-associated colorectal tumorigenesis, inducing cell proliferation, impairing gut barriers, and promoting oncogenic gene expression.¹⁹¹

Microbiome

The human body harbors diverse microbiome communities that interact with the host in complex ways.¹⁹² Dysbiosis has been implicated in the development of numerous diseases, including cancer.¹²⁸ The tumorigenic effects of specific microorganisms have been well-established across several types of tumors. The World Health Organization has classified several microorganisms as Group 1 carcinogens, including Helicobacter pylori (H. pylori), Epstein-Barr virus (EBV), human papillomavirus, hepatitis B virus (HBV), and hepatitis C virus (HCV).¹⁹³ Excepted for the classic oncogenic pathogens, many other microorganisms have also been discovered to be associated with tumors,¹⁹⁴ and several large-scale pan-cancer studies have revealed the presence of microbes in almost all types of tumors.^195,196,197 For example, Fusobacterium nucleatum, polyketide synthase-positive(pks⁺) Escherichia coli (E. coli), and enterotoxigenic Bacteroides fragilis (ETBF) are associated with the occurrence of CRC.^198,199,200 Furthermore, it has recently been discovered that microbiome alterations emerge in early precancerous stages of CRC, indicating their promoting role in early tumorigenesis.^201,202,203 In other tumor types, there are also some cues that microbiota is involved in tumor formation, such as Streptococcus anginosus (S. anginosus) in gastric cancer,²⁰⁴ Acidovorax species in lung squamous cell carcinoma,²⁰⁵ and Bacteroides fragilis in breast cancer.²⁰⁶

The microbiome plays a crucial role in tumorigenesis through various mechanisms, including physical binding or secretion of metabolites and toxins, which lead to genotoxicity and epigenomic abnormalities, activation of signaling pathways, and modulation of the immune system and inflammatory responses.^207,208 One of the most well-known examples of genotoxicity is the integration of the HBV genome into the host liver cell genome, which results in genetic mutations and chromosomal abnormalities that promote liver cancer.^209,210 HBV DNA most commonly integrates into the telomerase reverse transcriptase (TERT) promoter region, disrupting the tight suppression of TERT transcription and leading to abnormal liver cell proliferation.^211,212,213 Furthermore, when HBV inserts into the human cyclin A gene, it generates novel tumor-specific chimeric proteins with oncogenic functions.^214,215,216 Pathogenic E. coli also promotes tumorigenesis through genomic alterations. The toxin colibactin, secreted by pks⁺ E. coli, causes interchain crosslinking and double-strand DNA breaks, leading to gene mutations and tumorigenesis.^217,218

In addition to genetic mutations, it has been reported that bacteria significantly contribute to epigenetic alterations.^219,220 The human commensal bacterium ETBF could promote distal colonic tumorigenesis in the Apc^MinΔ716/+ mouse model. When another Braf^V600E was induced, new tumors emerged in the midproximal colon, which exhibited similar phenotypes to human BRAF-mutant serrated-like tumors. The colonization of ETBF and Braf mutation synergistically increased the levels of CpG islands DNA methylation and induced characteristic immunophenotypic alterations, including IFN pathway activation, and myeloid-derived suppressor cells and CD8⁺ T cell infiltration.²¹⁹ Furthermore, the microbiota can exert an epigenetic modulation role by influencing the oncogenic effects of mutant proteins. Trp53 mutation plays context-specific roles in intestinal tumorigenesis, promoting tumorigenesis in the distal gut while suppressing tumors in the proximal gut.²²⁰ The tumor suppressive effect was achieved through disrupting the binding of T cell factor 4 to chromatin and repression of the WNT signaling. A high density of microorganisms in the distal gut, along with their metabolite gallic acid, has the potential to reverse the protective role of mutant p53 and activate the oncogenic WNT pathway. The administration of antibiotics effectively reduced WNT activation and cell proliferation.²²⁰ Furthermore, Fu et al. recently discovered that S. anginosus promoted the tumorigenesis of H. pylori-negative gastric cancer through direct interactions.²²¹ The surface protein of S. anginosus, TMPC, could activate gastric epithelial cell receptor ANXA2, enabling colonization of S. anginosus in gastric mucosa and activation of MAPK pathway.²²¹ As a result, S. anginosus damaged the gastric barrier function, promoted cell proliferation, and inhibited apoptosis of epithelial cells, and ultimately induced gastric cancers.²²¹

Microbes can also play a pro-tumoral role by regulating the immune microenvironment. The microbiome in pancreatic cancer selectively activates Toll-like receptors in monocytes, which in turn drives immune suppression by inducing T-cell anergy, ultimately fostering tumorigenesis.²²² Fungi migrating from the intestine to the pancreas also experience fungal dysbiosis. They activate the mannose-binding lectin-complement cascade reaction to accelerate PDAC formation.²²³ On the contrary, some microorganisms play roles in inhibiting immunosuppression and tumor formation.^224,225 Ruminococcus gnavus and Blautia producta, belonging to Lachnospiraceae family, could inhibit the growth of colon tumors by degrading dissolved glycerophospholipids, suppressing their immunosuppressive function, and maintaining the immune surveillance function of CD8 T cells.²²⁴ Similarly, during the occurrence of CRC, the urea cycle is activated because of the absence of beneficial bacteria with ureolytic capacity. The accumulation of urea could induce macrophages to polarize towards a pro-tumorigenic phenotype, characterized by polyamine accumulation, thereby promoting the tumorigenesis of CRC.²²⁵ Altogether, the complex crosstalk between the microbiome and their hosts in tumorigenesis involves both tumor cells and their microenvironmental cells, inducing changes at genetic, epigenetic, transcriptional, and metabolic levels, which warrants further exploration.

Aging

Aging is considered the primary risk factor for tumorigenesis.²²⁶ There are systemic and local changes that overlap with that in tumors, including genomic instability, epigenetic alterations, inflammatory responses, and dysbiosis,²²⁷ which may have already played a role as early as in tumor initiating stages. Abnormal epigenetic alterations associated with aging underlie mutation-induced tumorigenesis. In mouse-derived organoids, aging-like spontaneous methylation of DNA promoter CpG-island induced colon more susceptible to the Braf^V600E-driven proximal colon tumorigenesis by activating Wnt signaling.¹¹⁴ However, there are some aging hallmarks, including telomere attrition, decreased stem cell plasticity, and cellular senescence-associated cell cycle arrest, possessing tumor-suppressive properties.²²⁷ Tumor-initiating cells always evade these tumor-suppressive mechanisms through mutations, such as inactivating mutations in TP53, CDKN2A, and CIP1.²²⁸ Moreover, mutations in the promoter of TERT, which allow for the maintenance of telomeres, are one of the most common driver mutations in a variety of tumors, and can be detected even in cirrhotic regenerative nodules, preventing cellular senescence and cell-cycle arrest, and thereby enhancing the proliferative potential of the transformed cells.^213,229

In addition to transformed cells, various microenvironmental cells, including fibroblasts, immune cells, and endothelial cells, generally exhibit an increased rate of senescence.^230,231 This is accompanied by the secretion of a large quantity of senescence-associated secretory phenotype, including various cytokines, growth factors, enzymes, and extracellular matrix (ECM). Although senescence-associated secretory phenotypes promote the clearance of senescent cells by activating the immune system in youth, it exerts immunosuppressive, pro-inflammatory, and pro-fibrotic effects in aging and chronic inflammation, contributing to tumorigenesis by directly targeting tumor cells or indirectly remodeling the microenvironment.²³² In cell competition, hepatocyte growth factor, a component of the senescence-associated secretory phenotype secreted by fibroblasts, was confirmed to inhibit Ras^V12 cell elimination by inducing their EMT and transformation from apical to basal extrusion.²³³ Furthermore, the senescence and dysfunction of immune cells can lead to immunosuppression, possibly further increasing the risk of cancer.²³⁴ Clearance of senescent macrophages was shown to reduce tumor burden and intercept non-small cell lung cancer at early and intermediate tumor stages by promoting immune surveillance in a Kras-driven lung cancer model.²³⁵

The identities of transformed cells are the result of the combined influence of intrinsic genetic and epigenetic profiles and external signaling. These factors collectively activate oncogenic pathways and remodel the microenvironment (Fig. 3). Consequently, there are not only cell-autonomous alterations that override cellular quality control mechanisms, enabling the gradual acquisition of hallmarks of cancer, but also adaptations to the extrinsic stress from their surrounding healthy counterparts, microenvironmental components, and tissue architecture. In addition, transformed cells actively reshape the external factors to be tailored for their oncogenic identities.

Interactions between oncogenic driver events. a In addition to genotoxicity, chemical and radical insults can induce cell injury, differentiation, and apoptosis. Oncogenic mutations that can confer resistance to such injuries provide proliferative advantages. On the other hand, the insults stimulate proliferative and self-renewal pathways by transcriptional and epigenetic regulation. Immune cells can also be activated to regulate transformed cell fate and promote tumorigenesis. b Unhealthy diet patterns induce hyperglycemia and hyperinsulinemia, and further cause differential response to insulin signals, which can facilitate cells harboring Src or Ras mutation in gaining competitive advantages and promote tumorigenesis. High levels of fatty acids also promote retention of Ras-mutant cells in cell competition by metabolism remodeling and mitochondrial membrane potential restoration. In addition, fatty acid and glucose participate in tumorigenesis as signaling molecules by modulating immune response and inflammation. c Microbiota interacts with transformed cells to affect host DNA methylation, transcription, metabolism and immune microenvironment to have an influence on malignant transformation. d Aging induces senescent stromal cells to secrete SASPs, which can reverse the outcome of cell competition and promote EMT of the mutant cell. Aging also cause spontaneous methylation, further promoting mutation-driven tumorigenesis. e The pathological processes mentioned above can converge at inflammation, which releases tumorigenic potential of expansive clones by activating oncogenic pathways and increases epigenetic plasticity. For instance, in pancreatic inflammation induced plastic state, ADM, Kras-mutant cells are more likely to transform to malignant status, while in the absence of inflammation, Kras can only induce PanIN without progression to PDAC. EMT epithelial-to-mesenchyma transition, ROS reactive oxygen species, nAChR nicotinic acetylcholine receptor, MDSC myeloid-derived suppressor cell, PDK pyruvate dehydrogenase kinase, TCF4 T cell factor 4, HGF hepatocyte growth factor, TET2 tet methylcytosine dioxygenase 2, EGFR Epidermal growth factor receptor, UV ultraviolet, Gata6 GATA Binding Protein 6, EZH2 enhancer of zeste homolog 2, PPAR-δ peroxisome proliferator-activated receptor-delta, FGF21 fibroblast growth factor 21, CCL2 PDK4, pyruvate dehydrogenase kinase 4, ADM acinar-to-ductal metaplasia, PanIN pancreatic intraepithelial neoplasms, PDAC pancreatic ductal adenocarcinoma, ETBF enterotoxigenic Bacteroides fragilis, Created with BioRender.com

Full size image

Cell-autonomous processes

Cells in normal tissues are hierarchically organized to restrain tumorigenesis. The initiating transformed cells must reprogram their cell fates, so as to gain uncontrollable self-renewal abilities and aberrant differentiation.² There are mainly three ways, encompassing activation of unlimited proliferative potential in stem cells, dedifferentiation of lineage-committed and differentiating cells, as well as leveraging intermediate states during trans-differentiation as the precursor of cancer (Fig. 4).

Cell-autonomous processes in tumorigenesis. After acquiring genetic and epigenetic mutations, transformed cells enter a malignant continuum where they reprogram their developmental pathways, allowing gradually gains of uncontrollable self-renewal capabilities and aberrant differentiation potential, primarily through three mechanisms: originating from stem cells, dedifferentiating from lineage-committed cells, and hijacking intermediate states during trans-differentiation. Created with BioRender.com

Full size image

It is believed that stem or early progenitor cells are more likely to achieve malignant transformation, based on their inherent self-renewal capacity and longevity.^236,237,238 On one hand, stem cells can accumulate more genetic mutations and epigenetic alterations necessary for tumor formation.²³⁹ On the other hand, stem cells and early progenitor cells exhibit high levels of cellular plasticity and are highly susceptible to fate transition.²⁴⁰ In the developmental hierarchy of melanocytes, progenitor stages, including neural crest and melanoblasts, are susceptible to transformation by BRAF^V600E and additional mutations, while differentiated melanocytes resist these cancerous signals.²⁴¹ The difference is induced by ATPase family AAA domain-containing 2 (ATAD2) in neural crest and melanoblasts, which regulates chromatin accessibility.²⁴¹ This enables TFs including SOX10 and MYC to form complexes with ATAD2, initiating the expression of downstream neural crest genes and oncogenic MAPK pathway genes, respectively.²⁴⁰

In addition to stem cells, there is accumulating evidence suggesting that committed cells are also able to give rise to cancer, specifically after undergoing dedifferentiation into stem-like cells upon oncogenic mutations or environmental stimulation.^242,243,244 For instance, melanoma induced by Braf^V600E and Pten loss can be originated from mature, pigment-producing melanocytes located in the interfollicular regions of mouse tails, which experienced transcriptional reprogramming and dedifferentiation prior to invasion.²⁴⁵ Consistently, Kaufman et al. identified the fate change during melanoma initiation in a Braf^V600E and Tp53 loss zebrafish model. Re-expression of neural crest progenitor program in melanoma, characterized by embryonically expressed gene Crestin, was driven by neural crest progenitor transcriptional factors, such as SOX10.²⁴⁶ Another example where dedifferentiation is implicated in tumor initiation is observed in mammary epithelium. Pik3ca mutation in lineage-restricted mammary basal and luminal cells can both induce multipotent stem-like cells, which is followed by development of tumor heterogeneity and multilineage mammary tumors.²⁴⁷ Luminal progenitor cells derived from BRCA1 basal-like breast cancers have also been confirmed to undergo dedifferentiation.^248,249 Mechanically, MYC plays a central role in the reprogramming of the lineage-specific cells. It inhibits mammary luminal-specific TFs, leading to the decommissioning of enhancers that disrupts their original transcriptional program. Additionally, MYC activates de novo enhancers and activates oncogenic pathways, such as the WNT pathway, which supports stem cell features and predisposes luminal epithelial cells to tumor initiation.²⁵⁰

Trans-differentiation is a common physiological response to injury, converting cells that are initially committed to one differentiation fate into an entirely different direction, either directly or through a stem or progenitor cell intermediate. The process can be implicated in tumor initiation, as exemplified in lung tumorigenesis that hijacks repair and regeneration programs. Many types of lung epithelial cells are highly plastic, and are capable of abandoning their cell fidelity to differentiate into each other upon injury, such as the transformation of club cells into AT2 cells, and AT2 cells into AT1 cells, which can be leveraged to promote tumorigenesis.^{111,251,252,253,254} Specifically, the intermediary state during these transformations is likely to be the key progenitor giving rise to tumors. For instance, KRT8 intermediate cells, which transition between AT2 and AT1 cells, have been identified in normal lung tissues adjacent to LUAD lesions.²⁵³ The KRT8 cells expand in precancerous and cancerous stages and are implicated in tobacco-associated KRAS-mutant LUAD,²⁵³ marked as reduced differentiation, enhanced plasticity and harboring KRAS driver mutations.²⁵³ The high-plasticity cells can also play a role in later progression and development of tumor heterogeneity. In a mouse model of LUAD tumorigenesis originating from Kras^G12D mutation and Trp53 loss in AT2 cells, a subset of transitional and high-plasticity cells emerging from adenomas was computationally predicted to drive cellular heterogeneity.²⁵⁴Although they are distinct from stem cells, they exhibit high growth and differentiation potential and play a transitional role in giving rise to the most heterogeneous cancer cell identities, which are indispensable for LUAD progression²⁵⁴ Other classic cases include pancreatic and epidemic injury, where lineage infidelity and epigenetic reprogramming at intermediate stages can be exploited by oncogenic mutations to activate malignant programs.^{147,150,255,256} Specifically, it is suggested that EMT is a drastic state of plasticity, and its intermediate state also exists, which endows cells with the highest capacity of invasion and metastasis.^257,258 Recent evidence indicates that the EMT can occur at a very early stage of tumorigenesis.^259,260 In squamous cell carcinomas induced by FAT1 LOF, the mutation triggers both a mesenchymal state mediated by YAP1-ZEB1 and a sustained epithelium state through EZH2 inactivation and SOX2 expression, illustrating a hybrid EMT phenotype with enhanced stemness and increased metastatic potential.²⁶⁰

Recently, a series of studies analyzing different stages of precancerous samples across various tumor types at single-cell resolution have demonstrated dynamic evolutionary trajectories preceding tumor formation, revealing a continuum of changes that lead to acquisition of hallmarks of cancers, including cell cycle, cell fate regulation, and metabolic reprogramming¹²⁸ (Table 3). For example, through single-cell multi-omics analysis of HSPCs from patients with myeloproliferative neoplasm, a convergent genomic evolutionary pattern of a double-hit TP53 mutation in hematopoietic stem cells was identified, and based on this trajectory, pre-leukemia stem cells ultimately progressing to secondary AML were found to undergo differentiation arrest prior to TP53 mutation occurrence, and the subsequent P53 mutant clones could be selected by inflammation, leading to clonal expansion.¹³⁵ Conventional colon adenomas can be traced back to originating from colonic stem cell (CSC). Throughout the progression from normal stem cells to adenomas and then to colon cancer, there is a gradual change in gene expression and chromatin accessibility, including upregulation of stem-like programs and increased antioxidative stress capability.^261,262 On the other hand, premalignant phenotypes induced by intrinsic and environmental drivers have been explicitly depicted in preclinical models. In mouse models and organoids, gastric premalignancies resulting from Trp53 mutations and exposures relevant to the disease have demonstrated the acquisition of renewal properties, activation of the WNT pathway independent of exogenous WNT ligands, and the abilities to overcome cell cycle distress and DNA damage stress.²⁶³ Similarly, progenitors of pancreatic tumorigenesis, induced by Kras mutations and inflammation, are characterized as gaining proliferative potential, with activation of cell cycle genes and other pathways.¹⁵⁰ Furthermore, in colorectal cancer originating from CSCs, CSCs are fixed predominantly on a highly proliferative phenoscape, whereas there is a continuous differentiation phenoscape that spans revival CSCs to proliferative CSCs under normal condition.²⁶⁴ YAP signaling regulates polarization of revival stem cells, which can be activated by fibroblast derived TGF-β, while APC loss and KRAS^G12D mutation collaboratively activate MAPK-PI3K signaling, trapping CSCs in the cancerous proliferative fate.²⁶⁴ Compared to ECM signaling, the intrinsic mutations exert a more dominant effect in regulating the stem cell fates.²⁶⁴ The evidence above also suggests that the regulation of malignant transformation involves the interplay between intrinsic cellular factors and microenvironmental factors, which needs to be evaluated in a tissue-specific context.

Clonal expansion by cell competition

Multicellular organisms develop surveillance mechanisms that compare cellular fitness with neighboring cells to preserve the most robust populations in environments with limited space and nutrients, a process termed ‘cell competition’. In epithelial tissues, mutant cells that alter fitness often become the losers and are eliminated by neighboring wild-type cells. Therefore, the process is an important tumor-suppressive mechanism, referred to as ‘epithelial defense against cancer’.²⁶⁵ However, in some cases, mutations can endow cells with ‘winner’ properties, allowing them to eliminate surrounding normal cells and gaining space for clonal expansion and tumor development, which is called ‘supercompetitor’.²⁶⁶

The molecular mechanisms to elicit cell competition include mechanical force, cell-cell contact, and secretory signaling, and losers can be eliminated through various forms, including extrusion, apoptosis, differentiation, necroptosis, and entosis, which are quite different from one tissue to another.²⁶⁷ For instance, in mouse pancreas and intestinal epithelium, apical extrusion of living cells was employed to eliminate Ras-mutant cells, through intercellular communications and alterations in cytoskeleton^188,268 (Fig. 5a). On the other hand, in self-renewing tissues, stem cell fate is a decisive factor for cell competition. Stem cells compete to occupy stem cell niche, and the winners have persistent self-renewal properties, while the differentiated cells would be removed from the stem cell niche. The structure of stem cell niches varies across tissue, which may be the cause for various clone sizes and structures in different tissues.⁷ In the intestinal glandular epithelium, the stem cell niche is located at the bottom of the crypt. Accordingly, competitions are confined to a single crypt and clones rarely expand to other crypts. Under normal conditions, intestinal stem cells (ISCs) stochastically differentiate and migrate upward along the crypt, shedding at the top. Otherwise, they maintain self-renewal and occupy the entire niche to form a monoclonal crypt, a phenomenon referred to as ‘crypt fixation’²⁶⁹ (Fig. 5b). Oncogenic mutations, such as KRAS, APC, and PIK3CA, have the potential to disrupt the neutral drift and tend to achieve crypt fixation.^270,271,272 The scenario is different in stratified epithelium, where stem/progenitor cells are distributed throughout the entire basal layer without interference from microstructures. Therefore, fitter stem cells have the potential to expand across the entire structure theoretically, until they encounter cells with the same fitness and end the competition (Fig. 5c).

Cell competition across tissues. a Live cells can be extruded from simple intestinal epithelium by intercellular communications and cytoskeleton rearrangement. b Intestinal stem cells compete for dominance within the stem cell niche located at the bottom of the intestinal crypt. Mutant supercompetitors are more likely to maintain stemness, replace wild-type counterparts, occupy the ISC niche, and subsequently take over the entire crypt. The displaced wild-type cells, referred to as “losers,” differentiate, migrate upward along the crypt, and are ultimately shed at the top. The fate of stem cells can be regulated by secretory signals that come directly from supercompetitors and indirectly from stromal cells surrounding the crypts, stimulated by the supercompetitors. Stemness inhibitory signals, including BMP activators and NOTUM, differentially affect wide-type cells and supercompetitors by preventing wild-type cells from maintaining stemness, while having less effects on supercompetitors. c In stratified epithelium, the outcome of stem cell competition is also regulated by cell fate decisions. However, it is not limited to specific microstructure as the crypt, the winner clone has the potential to expanding to a large area. WT wild-type cells, BMP bone morphogenic protein, ISCs intestinal stem cells. Created with BioRender.com

Full size image

Recent studies indicated that oncogene-mutation supercompetitors have the ability to outcompete their wild-type counterparts by both rising their own fitness and decreasing their competitors fitness.^270,271,272 Apc^–/– ISCs secrete notum palmitoleoyl-protein carboxylesterase, an antagonist of WNT signaling to inhibit wild-type ISC proliferation and to facilitate Apc-mutant clones towards crypt fixation, ultimately contributing to adenoma formation.^270,271 Analogously, ISCs carrying Pik3ca or Kras mutation enhanced secretion of BMP ligand, mediating the differentiation of wild-type ISCs.²⁷² Super-competition has also been observed in Asxl1 CHIP, where mutant HSPCs generate mature offspring with elevated expression of pro-inflammatory genes.²⁷³ The inflammatory environment induced differentiation of wild-type cells, while the mutant HSPCs upregulated genes that suppress inflammation to protect themselves from differentiation.²⁷³

Interactions with microenvironmental components

The microenvironment is composed of diverse immune cells, fibroblasts, and ECM,²⁷⁴ which have sophisticated interactions with transformed cells. On one hand, the healthy microenvironment plays a tumor-suppressive role and exerts the selective pressure to sculp clonal landscape. On the other hand, the transformed cells can remodel the surrounding niche to support their fitness, and accumulating work has identified early transformation of the microenvironment during tumorigenesis (Table 2). In this part, we aim to illustrate the interplays and co-evolutionary dynamics between mutant clones and their microenvironment during tumorigenesis.

Immune cells

The immune system possesses the capacity to suppress and shape tumors. Immune surveillance can be stimulated by mutation-induced neoantigens. Accordingly, immunogenic pressure selects for transformed cells that can evade immune recognition and killing, as well as those with the capability to sculp an immunosuppressive landscape.

A convergent immune identity is present in almost all established tumors, including varying extents of suppression of cytotoxic T lymphocytes, natural killer cells, and dendritic cells, increases in regulatory T (Treg) cells and other suppressive cells, activation of pro-inflammatory cells, as well as transformation of myeloid cells into pro-oncogenic phenotypes^{275,276,277,278} (Fig. 6a). There is a continuum of immune evolution accompanying the transformation of cells from pre-cancerous stages (Table 3). For example, a stepwise process of CRC tumorigenesis was shown to be accompanied by a shift from pro-inflammatory to immune-suppressive macrophage populations, along with upregulation of ‘don’t eat me’ CD47-SIRPα signaling.²⁷⁹ Moreover, during the progression of preneoplasia to invasive LUAD, the immune system evolves with downregulation of immune-activation pathways, such as dendritic cell maturation and the acute phase reaction pathway, and upregulation of immunosuppressive pathways including T cell exhaustion signaling and poly adenosine diphosphate-ribose polymerase (PARP) signaling pathways.²⁸⁰ More importantly, the immune transformation may play a decisive role in the transition from precancerous lesions to tumors. Lung carcinoma in situ only progresses to cancer if immune evasion occurs while lesions with an active immune response and higher infiltration of CD8⁺ T cells would regress.²⁸¹

Interactions of transformed cell and microenvironmental components. a Abnormal genetic, epigenetic and transcriptional signals in transformed cells can paradoxically induce immune activation, while simultaneously developing strategies to achieve immune evasion. Their crosstalk is primarily mediated by direct cell-cell interaction signals and paracrine signals, such as chemokines, cytokines and growth factors and direct cell-cell interaction signals. As a positive feedback, tumor supportive immune cells, such as TAMs, which can produce IL-1β signals to further promote malignant evolution. b Transformed cells, along with environmental stress and genetic alterations, can activate fibroblasts through both secretory and contact signals, transforming them into CAFs with diverse tumor-promoting properties. In turn, fibroblasts secrete stemness signals to differentially regulate mutant and wild-type cells during cell competition. c Environmental signals can induce ECM remodeling, and a single transformed cell with ECM adhesion loss can also produce ECM to support its survival. In turn, abnormal mechanical signals in the ECM, including stiffness and viscoelasticity, under pathological conditions such as inflammation, aging, wound repair, and T2DM, predispose mutant cells to malignant progression through the activation of the YAP/TAZ pathway. The pro-tumorigenic effects can be aggravated by mutations in the RTK-Ras pathway. Additionally, a stiff ECM inhibits filamin from translocating from perinuclear areas to the interface of wild-type and mutant cells, further inhibiting the extrusion of mutant cells. TAMs tumor associated macrophages, MDSC myeloid-derived suppressor cell, cGAS cyclic GMP-AMP synthase, STING stimulator of interferon genes, IRF3 IFN regulatory factor 3, T2DM type 2 diabetes mellitus, Yap yes-associated protein, TAZ transcriptional co-activator with PDZ-binding motif, LPAR4 lysophosphatidic Acid Receptor 4. Created with BioRender.com

Full size image

As mentioned beforehand, many environmental factors change the immune landscape, stimulating chronic inflammation and increasing tumor susceptibility. In addition, the transformed cells can be a key driver of immune remodeling. Mechanically, tumor cells are able to regulate immune cell activation, chemotaxis, and polarization through paracrine secretion of cytokines, chemokines, and growth factors, or through direct cell-cell interaction signals, such as tumor antigens presented by major histocompatibility complex class l (MHC-I), programmed death ligand 1 (PD-L1), and CD47.²⁸² In turn, a remodeled immune ecosystem supports further malignant progression. The crosstalk between transforming cells and the immune microenvironment is complicated and synergistically promotes the co-evolution. Caronni et al. found that transformed cells secreted high-level prostaglandin E2 and tumor necrosis factor (TNF) and thus promoted infiltration of IL-1β expressing tumor-associated macrophages (TAMs), which drove inflammatory reprogramming of neighboring transformed cells, resulting in a positive feedback loop to aggravate inflammation and tumor progression.²⁸³ Another case at this point is in Hras-mutant benign cutaneous papilloma. Upregulation of TGF-β pathway induced transcriptional reprogramming of cancer stem cells, resulting in upregulated expression of leptin receptors in cancer stem cells and angiogenesis.²⁸⁴ As a result, benign tumor cells enhanced sensing and responding to circulatory leptin levels, and activated downstream PI3K-AKT-mammalian target of rapamycin (mTOR) pathway, leading to malignant transformation.²⁸⁴

The immunomodulatory roles of transformed cells can be induced by genetic and epigenetic mutations and aberrate signaling. The driver mutations may serve as a major source of heterogeneity in the immune landscape of early tumors. Early transformation of host immunity in lung tumorigenesis was verified to be strongly associated with the type of driver mutations.²⁸⁰ Mutant Kras induced stronger immune activation compared with that of EGFR mutations from normal and premalignant to cancerous states, including CD8⁺ T cell infiltration, a low ratio of CD4⁺/CD8⁺ T cells and Treg/CD8⁺ T cells, and higher T cell clonality.²⁸⁰ Indeed, the immunomodulatory roles of the two classic tumor driver mutations have been widely explored. Cells harboring Kras mutation acquire capability to activate STAT3, secrete IL-6 and other proinflammatory cytokines. They also activate NLRP3 inflammasome and release chemokines, such as CCL5 and CXCL3, mediating tumor-promoting inflammation and immune modulation, and further promoting tumor progression.²⁸⁵ Similarly, EGFR mutations have been reported to promote Treg infiltration by upregulating CCL22 through activation of JUN amino-terminal kinase (JNK)/cJUN, and impede CD8⁺ T cell recruitment through downregulation of IRF1 and CXCL10 pathway.²⁸⁶ Pten deletion promoted PI3Kβ-mediated immune evasion through activation of the AKT and BMX-STAT3 pathways with reduced GM-CSF production, inactivation of dendritic cells, downregulation of antigen presentation pathways, and attenuation of IFNγ-mediated anti-tumor responses. In addition, mutations in TP53, another classical tumor suppressor gene, can not only maintain chronic inflammation by secreting IL-8 through the NF-κB pathway,²⁸⁷ but also inhibit innate immune response by disturbing the cytosolic DNA activated STING-TBK1-IRF3 pathway.²⁸⁸

In addition to genetic mutations, epigenetic and transcriptional factors are also involved in shaping the immune microenvironment. Repression of CXCL9 and CXCL10 expression, as well as impairment of CD8⁺ T cell infiltration in tumors, can be induced by mutations in isocitrate dehydrogenase and global hypermethylation.²⁸⁹ Meanwhile, oncogenic pathways, such as WNT-β-catenin, TGF-β, NF-κB and HIF, have the capability to alter the immune landscape by affecting the communication network between immune cells and cancer cells.²⁹⁰ A genome-wide CRISPR screening for genes modulating immune evasion from cytotoxic T lymphocytes in mouse cancer cells identified those involved in regulating IFN-response and TNF-induced cytotoxicity.²⁹¹ Similarly, Martin et al. performed CRISPR screening in immunodeficient and normal mice, identifying multiple tumor suppressor genes that were positively selected by the adaptive immune system. These tumor suppressor genes are involved in various crucial cellular processes, such as chromatin interaction, antigen presentation, protein stability regulation, TGF-β signaling, and IFNα signaling.²⁹² Although the evidence above is primarily based on research in established tumors, the effects of immune evasion are now being highlighted at the earliest stages of tumorigenesis. SOX17 deregulated IFNγ receptor expression and further lowered the expression of MHC-I and CXCL10, as well as CD8⁺ T cell infiltration. These changes played indispensable roles in the in vivo adaptation of genetically engineered naïve colon cancer organoids.²⁹³ In ESCC tumorigenesis, pathological overexpression of SOX2 activated endogenous retroviral elements and promoted double-stranded RNA formation, which should have activated immune surveillance.²⁹⁴ However, parallel upregulation of ADAR1 in turn attenuated the IFN signaling and contributed to immune escape.²⁹⁴ Interestingly, metabolic identities of tumor cells and immune components can also play a role in their interactions, forming competitive or dependent relationships with each other. On one hand, metabolites of tumor cells promote immunosuppressive effects,^295,296 and in turn, phagocytosis of TAMs facilitates nutrient accumulation to meet energy requirement of tumor cells.²⁹⁷ On the other hand, there is nutritional competition between immune cells and tumor cells.²⁹⁸ mTORC1 signaling in TAMs plays a role in regulating the competition.²⁹⁹ Under normal protein diet conditions, the mTORC1 pathway is weakened in TAMs and thereby be enhanced in Myc-overexpressing tumor cells, resulting in a competitive advantage of tumor cells. Conversely, under low-protein diet conditions, activation of the GTPase-activating proteins GATOR in TAMs leads to TFEB/TFE3 nuclear translocation and mTORC1 activation in TAMs. As a consequence, TAMs gain an advantage over tumor cells in metabolic competition, exerting tumor-suppressive effects.²⁹⁹ Whether the mechanism is involved in early tumorigenesis warrants further exploration.

There are some arguments for the timing of immune activation and evasion. It is believed that there is an immune ignorance at the earliest cancerous stage where only a few transformed cells are present, and low levels of neoantigens they produced are deficient to activate immune clearance.^275,300 The immune surveillance may not be a decisive factor for the initial clonal expansion.⁴¹ A mathematical model was developed to separate the fitness of driver mutations based on positive oncogenicity and negative immunogenicity. It revealed that TP53 mutations in non-cancerous tissues were primarily selected for their pro-oncogenic proliferative advantage rather than negatively selected by immunogenicity.³⁰¹ When progressing to advanced tumors, the pro-tumoral evolutionary force shifted into powering immune evasion. The shift could also explain the reason for different TP53 hotspot mutations between cancer and normal tissues.³⁰¹ The timing of switch from immune ignorance to activation and subsequent evasion need further exploration. High-resolution multiregional spatial and single-cell multi-omics sequencing are well suited to assess this issue. For instance, Cody et al. constructed a pseudo-temporal trajectory of colorectal tumorigenesis based on CIN and hypermutated pathways in their spatial multi-omic atlas, and mapped immune state changes along progression pseudotime, thereby facilitating prediction of immune exclusion.³⁰²

Fibroblasts

Fibroblasts constitute the primary stromal cellular components and serve major roles in ECM production, tissue structure maintenance, regulation of stem cells, interactions with immune cells, and participation in wound repair. Their role in regulating cell fate through paracrine orchestration can be hijacked by transformed cells to promote tumorigenesis^272,303 (Fig. 6b). In the ISC niche, prostaglandin E2 secreted by a rare population of PTGS2-expressing fibroblasts can act on Sca-1⁺ ISCs and activate Cox2-Yap signaling for Apc^Min/+ stem cell expansion and colon tumorigenesis.³⁰³ The stem cell niche signals produced by stromal cells also participate in the competition between oncogenic-mutant and wild-type cells. Pik3ca mutant ISCs showed an expansion advantage, partially by inhibiting stromal WNT signaling and creating a detrimental condition for the survival of wild-type ISCs.²⁷²

Alternatively, it is well-documented that cancer-associated fibroblasts (CAFs) are an important component in the TME. They can be activated by various stimuli in cancerous tissues, including TGF-β, inflammatory factors such as IL-1, IL-6, and TNF-α, physiological and genomic stress, ECM changes, and contact signals.^304,305,306 In addition, CAFs promote tumor growth by remodeling the ECM, inducing immune evasion, and directly interacting with tumor cells.³⁰⁴ It has been confirmed that they emerge and contribute to the earliest stage of tumorigenesis (Table 3). The transformed cells are the main driver for CAF transformation.^307,308,309 We identified a reciprocal mechanism between fibroblasts and epithelial cells, evolving synchronously in the multistep ESCC tumorigenesis.³⁰⁷ In the early stage of tumorigenesis, epithelial cells gradually downregulated ANXA1 expression due to the suppression of transcription factor KLF4. Subsequently, the formyl peptide receptor type 2, an ANXA1 receptor on fibroblasts responsible for fibroblast homeostasis, was dysregulated and drove the transformation of CAFs. This process was accompanied by TGF-β secretion from transformed cells, further accelerating CAF transformation.³⁰⁷ Similarly, the epithelial-stromal interactions mediated by JAG1 on ductal carcinoma in situ cells and NOTCH2 on fibroblasts play a role in CAF transformation and mammary tumorigenesis.

Apart from transformed cells, other abnormal signals can prime pro-tumorigenic identity of fibroblasts before transformation. Mutations in fibroblasts, such as BRCA1 and NOTCH1 can also be regarded as the prerequisite of tumorigenesis. In addition, external stimuli, including aging, dense microenvironment and exposure to oncogenic insults can all promote transformation.^{233,310,311,312} Dermal fibroblasts under UV exposure induce suppression of NOTCH and its effector CSL, and promote the production of inflammatory cytokines, growth factors and matrix metalloproteinases, contributing to precancerous actinic keratosis lesions and CSCC formation.³¹³

Extracellular matrix

The ECM is mainly composed of fibrous proteins and glycosaminoglycans, providing mechanical support, cellular anchoring, and storage for water and various bioactive molecules.³¹⁴ Additionally, the ECM communicates with cells through local adhesions, converting chemical and mechanical signals into biological signals, regulating key cellular processes such as proliferation, apoptosis, fate decision, and migration. This process is known as mechanosensing and mechanotransduction.³¹⁵ During tumorigenesis, the ECM experiences remodeling mainly driven by CAFs, tumor cells, and macrophages, leading to increased deposition, cross-linking and stiffness. As a result, the changes promote malignant progression by transducing abnormal biomechanical signals to transformed cells, as well as regulating immune recruitment and activation.^314,316

The abnormal ECM has been profoundly investigated in established tumors, however, their earlier roles in regulating clone evolution before cancer formation and the accurate timing for oncogenic disorganization are unclear. Recently, Wu et al. reported the role of ECM remodeling in tumor initiation, where a solitary transformed cell at the very beginning of tumorigenesis met much more stress in a normal microenvironment than that in established tumors, including loss of cell-cell contact between tumor cells and pro-tumor ECM³¹⁷ (Fig. 6c). The individual pancreatic cancer cell enhanced production of ECM and adapted to isolated stress by increasing expression of the stress-responsive gene lysophosphatidic acid receptor 4 (LPAR4) and promoted the production of fibronectin-rich ECM, which could compensate for the absence of stromal-derived factors and help tumor initiation.³¹⁸ Furthermore, the ECM could also support neighboring cells without upregulated expression of LPAR4 through integrins α5β1 or αVβ3.³¹⁸

Ahead of the transformed cell driven ECM remodeling, many pathological conditions, such as chronic inflammation, aging, and tissue injury, can increase stiffness of ECM and prime a tumor-vulnerable state.³¹⁹ At the initial stage, stiffness influences the epithelial defense of oncogenic mutation. Filamin, an actin filament cross-linking protein located at the interface of wild-type and mutant cells, facilitates the extrusion of mutant cells under normal physiological conditions. However, when ECM is stiff, filamin relocates to the perinuclei and leads to the failure of epithelial defense against cancer and causes tumorigenesis.³²⁰ Furthermore, stiff signaling plays a role in cell fate regulation, and further regulates the susceptibility to oncogenic transformation.^321,322 In the condition of SmoM2 induced basal cell carcinoma, the back skin, which has a denser collagen I network compared with the skin of the ear, was not susceptible to the mutation-induced progenitor state reprogramming and tumor initiation.³²¹ Chronic UV exposure and aging can decrease the expression of collagen, overcoming the natural resistance and increasing the risk of tumorigenesis.³²¹ Orthogonally, oncogenic mutations render tumor-initiating cells to be more sensitive to signals of ECM stiffness. Even slight changes in ECM rigidity can trigger abnormal responses in cells harboring mutated oncogenes in the RTK-Ras pathway, such as human epidermal growth factor receptor 2 (Her2) and Kras.³²² Stiffness and the mutations synergistically activated the YAP/transcriptional co-activator with PDZ-binding motif (TAZ) pathway, subsequently promoting the transformation of precancerous states.³²² In addition to stiffness, viscoelasticity is another pro-tumorigenic mechanical property of ECM, which can be induced by advanced glycation end-products (AGEs) accumulation by type 2 diabetes mellitus. It is characterized by decreased interconnectivity of collagen matrix, shorter fiber length and greater heterogeneity, activating integrin-β1-tensin-1-YAP pathway and promoting cancer progression.³²³ Notably, YAP/TAZ serves as a molecular hub for mechanosensing and mechanotransduction, which is activated by mechanical signals transmitted by cytoskeletons, and is followed by the nucleus translocation and gene expression.³²⁴ Since abnormal YAP/TAZ pathway is strongly associated with various tumors,³²⁴ it is suggested that its oncogenic mechanotransductive signaling may be a general trait implicated in early malignant transformation. The interaction between mutations and mechanical signaling during tumorigenesis warrants further investigations.

Tissue architecture restraint

Tissue structure is shaped by collective mechanical characteristics of individual cells, as well as their interactions with neighboring counterparts, stromal cells and the ECM. The maintenance of three-dimensional structural balance relies on stable number and arrangement of cells, which is also an important tumor-suppressive mechanism. Since there are tight interconnections and limited space in solid tissues, cell proliferation and elimination generate mechanical stress by the resistance of surroundings, thereby providing feedback to regulate cell behaviors.³²⁵ When over-proliferative cells cause density increase and compression, dense responsive signals are activated to suppress proliferation and eliminate redundant cells.³²⁶ Differential sensitivity to mechanical signals triggers cell competition.³²⁷ The mutations that endow cells with insensitivity to compression would be preserved (Fig. 7a). For example, when subjected to compression, Scribble mutant Madin-Darby canine kidney cells tended to undergo apoptosis due to p53 activation by ROCK and p38 pathways.³²⁸ On the contrary, Ras^V12 mutant cells downregulated ERK in neighboring wild-type cells via competition, triggering apoptosis of wild-type cells.³²⁹

Tissue architecture restraint for clone expansion and alterations in tumorigenesis. a Cell density sensing mechanisms trigger apoptosis in cells sensitive to compaction, whether mutant or wild-type. b Mutations that disrupt the ECM and enable anchorage-independent survival allow cells to move to the lumen and expand. Additionally, the loss of cell-cell junctions can unleash the proliferative potential of mutations in situ instead of through translocation. c Mutant intestinal crypts are more likely to split rather than merge, increasing their number but still keeping overall balance through spreading and decreasing local crypt density. The Kras mutation speeds up this splitting, to a degree that cannot be counteracted by dispersal, leading to tumorigenesis. d When the homeostatic tissue architecture is disrupted, the mutant cells mediate a dysregulated tumor structure. This manifests as alterations in cell-cell junctions, cell-ECM adhesions and cytoskeleton rearrangement. The initial tissue curvature, as well as the stiffness of the basal membrane and suprabasal cells, all affect the nascent tumor morphogenesis. In addition, the EMT mediated by EFNB1-EPHB4 interactions in epithelial cells harboring TP53 mutations occurs alongside early malignant morphogenesis. MMP metalloprotease, EMT epithelial-to-mesenchymal transition. Created with BioRender.com

Full size image

Furthermore, cell-cell junctions and cell-ECM adhesions are other important factors to arrest oncogenic growth and maintain homeostasis^326,330(Fig. 7b). A well-organized acinar structure formed by a non-transformed human mammary epithelial cell line, MCF10A, remained quiescent in the presence of sporadic oncogenic mutations with proliferative potential until they expressed matrix metalloproteinases and disrupted cell-matrix adhesions. This disruption resulted in the translocation of mutant cells into the lumen, releasing more space for expansion.³³⁰ In addition, although detachment from the ECM alleviated the space limitation, the loss of the survival signal provided by the ECM would also lead to decreased fitness and apoptosis.^330,331 Only cells that achieve anchorage-independent survival could continue to expand.³³⁰ Furthermore, the extrusion of mutant cells is also regulated by cell-cell junctions. Disruption of cell-cell junctions leaded to the transformation of the proliferative cells from lumen translocation to proliferation in situ.³³⁰

Some tissues have microstructures, which impose another barrier to the expansion of mutant clones.^37,332,333 As mentioned earlier, the expansion of mutant stem cells is typically limited to a single intestinal crypt. Further expansion requires crypt fission, but it is a rare event for normal adult tissues, at approximately one fission every 27 years.³³⁴ Additionally, there are concurrent crypt fusions to maintain crypt density.³³⁵ Some mutations can break the balance and speed up crypt fission^333,336 (Fig. 7c). This may account for discrepant elevations in the frequency of crypt fission without a concurrent rise in crypt fusion.³³⁴ Alternatively, dispersal of intestinal crypts occurs to counteract rising crypt density. However, the rate of crypt fission in Kras mutant crypts is too fast to be accommodated through dispersal, resulting in an increase in the local density of crypts, which increases the risk of polyps and tumor formation.³³⁴

Alongside overriding the structural restrictions of normal tissues, early tumor morphogenesis is shaped by cell proliferation, abnormal mechanics of transformed cells, and their microenvironment^337,338 (Fig. 7d). Ras mutation induces MCF10A transformed cells to aggregate from two-dimensional (2D) to 3D structure through differential localization of E-cadherin at the top and bottom layers, reduction of adhesion to ECM, and redistribution of epithelial tension regulators. Neighbor structures of the lesion are also implicated. In tubular epithelia, whether lesion growth occurs outwards or inwards to the ductal lumen results from the balance between cellular tension of the lesion and the resistance of the tissue curvature.³³⁸ In stratified epithelium, the assembly of the basal membrane and the stiffness of superbasal layers also play a significant role in shaping tumors. Tumor budding is promoted by well-remodeled and soft basement membrane in SmoM2 induced basal cell carcinomas (BCCs). By contrast, in HRas^G12V induced squamous cell carcinoma, stiffness from basal membrane and superbasal stratification promotes a folding architecture, which is more likely to develop an invasive tumor.³³⁹ However, molecular mechanisms underlying the gradual oncogenic tissue disorganization are not well understood. Based on spatial transcriptomic technology, our laboratory recently deciphered spatiotemporal expression patterns and identified key molecules driving the stepwise tissue destruction in esophageal tumorigenesis. Transformed cells interacted with each other through EFNB1-EPHB4 and triggered cell proliferation and EMT by SRC/ERK/AKT signaling, which were possibly instigated by ΔNP63 overexpression due to a TP53 mutation.²⁵⁹

Molecule-based cancer risk prediction

A better understanding of molecular and phenotypic determinants of malignant transformation facilitates cancer prevention, while the first step is to conduct risk assessment. Traditionally, it relies mainly on histopathological identification of precancerous lesions and combined demographic risk factors to identify individuals at high-risk of developing cancer, whereas predictive values are generally low. Only a small proportion of pathologically identified precancerous lesions progress to invasive tumors, inducing overdiagnosis and unnecessary interventions.³⁴⁰ In addition, as we have discussed above, some precancerous molecular alterations can emerge precedent or independent of morphological abnormalities. Therefore, molecular drivers identified in early tumorigenesis can be exploited to improve the efficacy of risk stratification, and further improve targeted surveillance and early interception.

Detection of germline mutations to evaluate inherited cancer susceptibility is widely explored, as exemplified by BRCA1 and BRCA2 pathogenic variations for breast and ovarian cancers.³⁴¹ In the past few decades, large-scale case-control association studies across cancer types have facilitated the identification of cancer-risk loci and the development of polygenic risk scores for risk prediction.³⁴² The combinations of polygenic risk scores and other known risk factors, including family history, lifestyle and reproduction, have been shown to accurately predict life-long risks of breast cancer.³⁴³ On the other hand, the pervasive existence of cancer driver mutations in normal tissues not only provides opportunities but also places higher demands for somatic molecule-based risk prediction, requiring accurately distinguishing between those as normal background and those as a consequential cancer signal during tumorigenesis. For example, in Barrett’s esophagus, TP53 mutation and 17p LOH are relatively more specific predictors of progression to esophageal adenocarcinoma,^59,344 with the TP53 mutation even capable of predicting progression in samples with no dysplasia.³⁴⁵ Furthermore, a predictive panel of multiple driving mutations will have better performance than the TP53 mutation alone.^{346,347,348,349} Another case of point is in CHIP, where it has been used in combination with hematologic and biochemical indicators to develop three independent risk prediction models for progressions to different myeloid neoplasms, including AML, myelodysplastic syndromes, and myeloproliferative neoplasms in a cohort of 454,340 UK Biobank participants, enabling early prediction of tumor occurrence in normal individuals.³⁵⁰ Given that CIN and CNAs are already present in specific precancerous diseases and accumulate throughout malignant progression, such as in Barrett’s esophagus, CNAs may also be a potential strategy for predicting risk of cancer.^{351,352,353,354} A genomic instability-based model was reported to distinguish patients with Barrett’s esophagus at high-risk of progression, among which 50% patients in the high-risk group were predicted 8 years before transformation of high-grade dysplasia or cancer.³⁵¹ Furthermore, based on evolutionary measurement, genetic clonal diversity and clonal expansion are explored as predictive indicators of malignant evolution in colon, esophagus, and blood, potentially being a more universal method for various cancers.^{355,356,357,358,359} DNA methylation is also a promising type of risk prediction marker, demonstrating value in risk prediction for Barrett’s esophagus progression and gastric cancer formation.^{360,361,362,363} Liquid biopsy tests of circulating cell-free DNA fragments and/or their methylation patterns have gained widespread attention due to their non-invasiveness, low cost, and viable implementation. Tests for tumor DNA methylation have been validated in detecting multiple advanced cancers, whereas it appears to perform poorly in early-stage tumors.^364,365 Advances in technology and more precises predictive panels are required to enhance this promising testing tool for use in premalignant stages.

Rapid development of high-throughput omics technologies in recent years has facilitated explorations of numerous biomarkers, and predictive panels based on transcriptomics, proteomics and metabolomics have been developed for specific tumors (Table 4). Based on serum metabolomics, lung adenocarcinoma and its preneoplasia can be distinguished from benign lesions by a metabolic panel.¹⁷³ A gut microbiome-based panel has also shown efficacy in distinguishing CRC and adenoma from normal tissues, and further research is needed to verify its predictive role in disease progression.^202,203 Based on multiplexed ion beam imaging by time of flight and tissue transcriptomics, Risom et al. mapped a spatial cellular landscape of ductal carcinoma in situ (DCIS) and delineated spatial and functional coordinated changes in stromal components from DCIS to invasive breast cancer, including myoepithelium, fibroblasts, and immune cells. Based on the features, they developed a risk prediction model for breast cancer invasion, which is largely dependent on myoepithelium and stroma. Intriguingly, disruption of myoepithelium indicates low risk of progression, which is contrary to the traditional belief that an intact myoepithelial barrier protects from tumor invasion, and the mechanism has not yet been detected.³⁶⁶ Altogether, multidimensional molecular features in the transition of tumors could be utilized to develop predictive assays. Nevertheless, most studies to date are based on small cohorts and sometimes lack validation cohorts, requiring further validations before being introduced into clinic.

Intervention strategies

Chemoprevention

Chemoprevention refers to the use of synthetic or natural substances to reduce the risk of developing cancers (Table 5). The most popular chemoprevention strategy is endocrine therapy for breast cancer prevention. Indeed, endocrine therapies have been widely attempted for breast and prostate cancer prevention, by inhibiting binding of sex steroids and their receptors to block downstream gene regulation and tumor cell growth.³⁶⁷ Females with high-risk breast cancer are recommended to use selective estrogen receptor modulators, such as tamoxifen and raloxifene, or aromatase inhibitors, which inhibit aromatization of androgens and decrease the level of estrogens, but specific adverse events need considerations, including fracture, thrombosis, endometrial cancer, and cataract.^368,369 In placebo-controlled randomized trials, tamoxifen can reduce the incidence of breast cancer by 31%, while raloxifene, aromatase inhibitors, exemestane and anastrozole, reduce it by 56% and 55%, respectively.³⁶⁹ They may also be effective in preventing DCIS.³⁶⁹ Similarly, 5α-reductase inhibitors, such as dutasteride and finasteride, have been attempted for prostate cancer prevention by inhibiting the synthesis of dihydrotestosterone, the most potent endogenous androgen.^370,371,372 Although they have demonstrated an overall reduction in prostate cancer risk, the efficacy in high-grade prostatic tumor prevention requires further confirmation.^370,371,372

Given the important roles of inflammatory responses in tumorigenesis, anti-inflammatory regimens for cancer prevention are of great interest. Nonsteroidal anti-inflammatory drugs, especially aspirin, have shown preliminary efficacy in the prevention of various cancers, including those of the central nervous system, breast, esophagus, stomach, head and neck, liver, bile duct, colorectum, endometrium, lung, ovaries, prostate, and pancreas.^373,374 Evidence for aspirin in preventing CRC is the most definitive. However, due to its severe adverse event of gastrointestinal bleeding, it is currently only recommended for Lynch syndrome and patients with removed familial adenomatous polyposis but is not routinely recommended for healthy individuals.^375,376,377 Targeting key pro-carcinogenic inflammatory factors, such as IL-1, IL-6, and TNF-α, may enable more precise cancer prevention. In the cardiovascular CANTOS trial, the intervention arm using canakinumab, an IL-1β monoclonal antibody, significantly reduced lung cancer incidence.³⁷⁸ However, the costs and fatal adverse events of cytokine targeting therapy necessitate careful consideration for preventive applications.

Metformin is the first-line treatment for type 2 diabetes mellitus, primarily targeting molecules involved in energy metabolism, such as mitochondrial complex I, MAPK, and mTOR. It also plays a role in reducing insulin levels, enhancing insulin sensitivity and exerting effects on immune cells.^379,380 In cell competition models, metformin reverses insulin resistance or enhances aerobic glycolysis, eliminating the competitive advantage of mutant cells, suggesting its potential inhibitory effect on tumor initiation.^184,381 Since a preliminary retrospective case-control study in Scotland was reported in 2005, the preventive use of metformin for tumors has been supported by several observational studies.³⁸² A randomized controlled trial in Japan confirmed the protective role of metformin from adenoma and polyp recurrence³⁸³; however, there is a lack of further evidence from intervention trials for the reduced risk of various cancers with metformin use.³⁸⁴ It is hypothesized that personalized regimens of metformin may be necessary, in order to particularly target tumors that are dependent on oxidative phosphorylation, as metformin primarily targets mitochondrial respiration. Additionally, due to metabolic reprogramming of tumor cells after metformin treatment, combination therapy targeting metabolic pathways on which tumor cells depend may enhance metformin efficacy.³⁸⁰ Another focus of metabolic regulation is statins, a class of drugs used to treat lipid disorders. As inhibitors of 3-hydroxy-3-methylglutaryl coenzyme-A (HMG-CoA) reductase (HMGCR), statins are associated with reduced overall cancer risk, and liver, prostate, lymphoma, and CRC risks.^{385,386,387,388,389,390,391,392,393} Statins inhibit de novo cholesterol synthesis via the mevalonate pathway and promote the removal of plasma low-density lipoprotein cholesterol by acting on low-density lipoprotein receptors. Mechanistically, statins exert anti-tumor effects by promoting cell death, regulating angiogenesis, and modulating immunity.³⁹⁴ In early stages of colorectal tumorigenesis, they can also modulate gut microbiota.³⁹⁵ Treatment with statins increases microbial tryptophan availability in the gut, promoting the growth of Lactobacillus reuteri, which converts tryptophan into indole-3-lactic acid and regulates Th17 cells to inhibit tumor formation. More evidence is required to support their clinical use.

A deeper understanding of molecular events in early tumorigenesis offers insights into novel interventional targets. Accurately identifying the cell state at the root or pivotal transitional point along the pathway of malignant transformation is essential for searching for potential targets within these cells. For instance, compared with wild-type luminal epithelium, the mammary epithelium in individuals carrying BRCA2 mutations exhibits an increased number of ERBB3^lo luminal epithelial cells, which potentially serve as the cells of origin for both ER⁺ and ER^- breast cancers.³⁹⁶ mTORC1 signaling is significantly upregulated in these cells. Short-term treatment with a mTORC1 inhibitor substantially curtailed tumorigenesis in a preclinical model, thus uncovering a potential strategy for BRCA2 breast cancer prevention.³⁹⁶ In tobacco-induced LUAD, considering that the identified progenitor cell subset harbors KRAS mutation, it is logical to hypothesize that KRAS mutation inhibitor can play a role in intercepting the earliest tumorigenesis.²⁵³ Similarly, as TP53 mutations are widespread in normal tissues and are identified as the early key driver events in various tumor initiation, there are many explorations for TP53 targeting strategies. Methods such as blocking murine double minute 2 (MDM2), as well as restoring dysfunctional p53 are being attempted.³⁹⁷ It is expected to identify more promising targets and accordingly develop robust agents in the future.

Immunoprevention

Immunoprevention involves modulating the host immune system to elicit early anti-tumor immune responses, eliminating tumor cells at precancerous stages. The most classic example of immunoprevention is vaccination against carcinogenic pathogens, such as human papillomavirus for preventing and treating cervical intraepithelial neoplasia,^398,399 and HBV for preventing hepatocellular carcinoma.^{400,401,402,403} In addition to viral-based vaccines, vaccines targeting tumor antigens have been attached great attention in recent years. The targets can be tumor-associated antigens abnormally overexpressed in tumor cells compared to normal tissues, such as carcinoembryonic antigen (CEA) and HER2, or aberrantly post-translationally modified antigens, such as Mucin 1 (MUC1).⁴⁰⁴ Among them, HER2-based vaccines have preliminarily shown success in interception of DCIS in clinical trials.^405,406 In addition, since aberrant hypoglycosylation of MUC1 occurs in precancerous lesions of multiple epithelial cancers, there are many attempts of vaccines targeting MUC1 in clinical trials for various cancers.⁴⁰⁴ However, it was recently reported that there were limited effects of MUC1 vaccine in preventing colonic adenomas in a randomized controlled trial. Individuals with advanced adenoma received MUC1 peptide vaccine within 1 year after adenoma removal. Despite eliciting significant antigen-specific immune responses, adenoma recurrence did not significantly decrease.⁴⁰⁷ Therefore, further study is still required to improve vaccine efficacy for preventive usage. Another strategy is to target tumor-specific driver mutations that have already occurred in precancerous stages, such as KRAS and EGFR mutations for lung cancer prevention.^408,409,410 These tumor-specific antigens are likely to be more immunogenic with better responses, but clinical trials are required to verify their efficacy and safety. Clinical trials on the mutant KRAS-targeted long peptide vaccine for high-risk pancreatic cancer recipients and EGFR-targeted vaccine for high-risk lung cancer recipients are currently underway (NCT05013216, NCT04298606).

As described above, immunosuppressive microenvironment has emerged at an early stage of tumorigenesis; accordingly, immunomodulatory strategies are being attempted for tumor prevention. For precancerous actinic keratosis of the skin, a combination of calcipotriol and 5-fluorouracil was adopted in a randomized controlled trial, which can induce squamous cell expression of thymic stromal lymphopoietin, thereby mobilizing anti-tumor immunity. Compared to using 5-fluorouracil alone, the combination showed significant lesion reduction, accompanied by upregulation of thymic stromal lymphopoietin, HLA-II, natural killer cell group 2D ligand expression, as well as CD4 T cell infiltration.⁴¹¹ Long-term follow-up indicated that the effects of immunomodulation persisted three years later, with a decrease in the incidence of CSCC.⁴¹² PD-L1 and PD-1 upregulation has been observed in precancerous lesions of the oral cavity^413,414 and lung tumors,⁴¹⁵ suggesting that PD-1 monoclonal antibodies are an ideal early immunoprevention strategy. Currently, relevant clinical trials are underway (NCT03347838, NCT03603223). In a preliminary trial to evaluate the safety and clinical response of anti-PD-1 therapy among patients with high-risk proliferative verrucous leukoplakia, 12 patients (36%) (95% CI, 20.4%-54.8%) had a > 80% decrease in size and degree of dysplasia after receiving nivolumab, suggesting potential clinical activity for nivolumab in high-risk proliferative verrucous leukoplakia.⁴¹⁶

Lifestyle and dietary interventions

Lifestyle and dietary interventions are low-cost, low-risk, and accessible preventive strategies. There are various advocated healthy lifestyles against cancers, including avoiding and ceasing exposure to carcinogens such as tobacco, alcohol, and UV, as well as adopting healthy diets and engaging in regular exercise.^{417,418,419,420} In terms of dietary interventions, multiple healthy dietary patterns, such as the Mediterranean diet, vegan diets, and various healthy diet guideline indices have been proposed.⁴¹⁷ Their core tenets are avoiding carcinogenic dietary components, controlling total calories, and increasing proportions of beneficial constituents.⁴²¹ However, most evidence is based on epidemiological associations from population studies, and many confounding factors cannot be excluded.⁴²¹ Exploring the molecular mechanisms behind specific nutrients in healthy diets for cancer prevention can not only strengthen the evidence supporting existing dietary interventions, but also yield insights for developing novel and scientifically grounded strategies. Low calorie intake and various fasting regimens are confirmed to inhibit nutrient sensing pathways and activate nutrient scarcity sensors to regulate cellular stress responses, modulating tumor cell activity and anti-tumor immune response.⁴²² Another popular regimen is ketogenic diet, which means to intake low carbohydrates, high fat, and moderate protein to enhance ketone metabolism. Some clinical trials have confirmed its therapeutic effects in patients with breast cancer undergoing chemotherapy or radiotherapy.^423,424,425 Although there is a lack of clinical evidence to support the preventive usage of ketogenic diet, its benefits have been shown in preclinical models. Ketogenic diet induced-β-hydroxybutyrate could bind the Hcar2 receptor on intestinal stem cells and activate tumor suppressive TF Hopx to inhibit cell proliferation and exert anti-cancer effects.⁴²⁶ Furthermore, since oral supplement of β-hydroxybutyrate alone could achieve an anti-tumor effect, it may be served as an alternative regimen for the ketogenic diet, possibly addressing the issue of low compliance with strategies that change the overall dietary pattern.⁴²⁶ Specific diets may also act as prebiotics or probiotics.⁴²⁷ The most popular one is high-fiber diets, which are associated with a lower risk of multiple cancers.^{428,429,430,431,432,433} Dietary fiber can be fermented into short-chain fatty acids by microbes to regulate microbe composition and diversity, protect intestinal mucus barrier, and prevent bacterial translocation, thereby modulating systemic metabolism, immunity, and inflammation.^{434,435,436,437} A small randomized cross-over trial confirms that supplementing fermentable fiber inulin and inulin-propionate ester, which is aimed at delivering short-chain fatty acids to the colon, can modulate gut microbes, metabolism, and inflammation, thereby improving insulin resistance.⁴³⁸ Recently, the BE GONE trial confirmed similar findings through the supplementation of beans, a fiber-rich food. Soybeans can act as prebiotics, regulating gut microbes, inflammation, and metabolism, improving biomarkers of metabolic obesity and colon cancer.⁴³⁹ However, direct evidence from fiber intervention trials for cancer prevention is still lacking. Another study held an opposite conclusion. It found that high-dose soluble fiber could dysregulate gut microbiota and metabolites, leading to enrichment of potentially pathogenic bacteria and depletion of probiotics, and prompt colorectal tumorigenesis.⁴⁴⁰ Specific effects of high-fiber diets are still warranted to be further explored.

Apart from adjusting dietary structure and macronutrient intake, direct supplementation of specific anti-tumor nutrients and metabolites is a more implementable strategy. Given the epigenetically tumor-suppressive effects of α-ketoglutarate demonstrated in mouse models, dietary supplementation of its precursor molecule glutamine may be a potential preventive strategy.¹⁷⁷ Other dietary supplements, including marine omega-3 fatty acids sourced from fish and seafood,⁴⁴¹ as well as the plant-derived natural alkaloid berberine,⁴⁴² have demonstrated preliminary efficacy in preventing CRCs. Further large-scale clinical trials and long-term follow-up are required. On the other hand, various vitamins have been proposed for tumor prevention, and some of them have illustrated promising applications. Nicotinamide, which belongs to vitamin B3 family, plays a role in inhibiting oxidation and DNA damage.^443,444 A Phase III clinical trial showed that it can effectively reduce the risk of non-melanoma skin cancers and actinic keratoses in high-risk populations.⁴⁴⁵ However, it failed to show a preventive effect in immunocompromised individuals following organ transplantation, possibly due to DNA damage resulting from the use of immunosuppressive drugs.⁴⁴⁶ In addition, low-dose acitretin, a vitamin A derivative, has been demonstrated to have a preliminary preventive effect on skin cancer in organ-transplanted recipients. Renal transplanted patients with actinic keratosis received acitretin therapy (20 mg/d) for 1 year and there was an improvement of actinic keratosis in all patients.⁴⁴⁷ Mechanistically, acitretin exerts an anti-tumor effect by increasing the number of epidermal Langerhans’ cells and enhancing skin immune monitoring.⁴⁴⁷ Other examples of effective preventive strategies include vitamin D for DCIS and high-dose folic acid for recurrent colorectal adenoma.^448,449 However, these clinical trials are limited by their small scale and short-term follow-up. Apart from the examples mentioned above, there is a notable scarcity of successful cases in interventions using other vitamins and micronutrients.^450,451 This underscores the need for more high-quality retrospective and prospective studies to evaluate the potential impacts of micronutrients. Such studies should be conducted in conjunction with preclinical research that demonstrates molecular mechanisms, thereby facilitating the identification of compounds suitable for future dietary interventions.

Driven by genetic and epigenetic alterations along with environmental signaling, transformed cells not only acquire cell-intrinsic proliferative advantages, but also actively remodel their environment to support their aberrant behaviors during early tumorigenesis. Encouragingly, apart from mutagenesis, many determinants of tumorigenesis are reversible, and understanding the molecular mechanisms underlying early malignant evolution provides significant translational opportunities. Cancer prevention aims to identify high-risk individuals and implement early interventions with high efficacy, low adverse events, and the potential to cure. Since many targetable aberrative pathways in advanced tumors have also been found in the earliest stages, including those affecting the cell cycle, anti-apoptosis, metabolic remodeling and immune evasion, classic anti-tumor agents might be repurposed for earlier interventions. However, extensive clinical trials to verify their efficacy and safety are warranted, and the balance of expenses and benefits should be considered.

Tumors originate from individual cells, presenting significant challenges in capturing these rare cell subsets. Advances in next-generation sequencing, single-cell, and spatial omics have revolutionized the study paradigm of tumorigenesis. At an extremely high resolution, precursor clones of various tumors have been identified, and co-evolutionary dynamics of the transformed cells and their microenvironment are being depicted. Furthermore, integrative analyses of paired omics modalities, such as genome, epigenome, and transcriptome, have been preliminarily applied to map the early tumorigenesis events,^{117,135,302,452} offering insights into the ordering and interplays among multiple evolutionary drivers, as well as their roles in regulating cellular phenotype. As multiplexing spatial and single-cell multi-omics technologies continue to enhance their throughput, resolution, and accuracy, coupled with innovations in bioinformatics tools to analyze unprecedented high-dimensional data,⁴⁵³ it is anticipated that multi-omics approaches can be leveraged to achieve a more comprehensive understanding of the complex biological processes of tumorigenesis.

To date, many studies primarily infer evolutionary trajectories computationally from multisampling of cancer specimens. However, this approach is limited to capturing only the dominant malignant clones and their major driver events. There is a loss of information regarding dynamic precancerous clonal competition and selection, since other precancerous clones may have been swept out in advanced tumors. Therefore, the importance of employing multiple sampling strategies to cover various stages of the malignant continuum is being increasingly recognized. Specifically, acquiring both cancerous and non-cancerous clones with shared ancestors simultaneously can optimize phylogenetic analysis results, depicting both the malignant evolutionary dynamics and the fate of remaining non-cancerous clones with partially shared mutations. This approach highlights the additional changes necessary for evolution into a malignant phenotype and their sequence among various driver events.⁴⁵⁴ Given that there are some premalignant stages that do not progress to invasive tumors, it is emphasized that rational cohort design in longitudinal studies to distinguish premalignant lesions from regression to progression can indicate key mechanisms that ultimately drive tumorigenesis. Yet, significant challenges remain in ensuring patient compliance and completely removing precancerous lesions during initial sampling, which usually aims at prevention and may interrupt natural disease progression.⁴⁵⁵ Alternatively, inducing autochthonous tumors in animal models or organoids offers an alternative way to study the early evolutionary processes. By prospectively introducing driver events informed by prior knowledge, and integrating lineage tracing with in vivo imaging techniques, real-time clone dynamics and their temporal evolutionary trajectory are visible, further facilitating the study of biological functions of specific perturbations in early tumorigenesis. At this point, Yao et al. recently reported their protein level reporter system, which is capable of tracing mutant p53 protein accumulation, a cancer-specific event as well as a potential mark for early transformed cells.⁴⁵⁶ The system sensitively identified rare precancerous cells in noncancerous tissues, and further facilitated characterization of cellular phenotypes underlying transformation, as well as the identification of potential interventional targets.⁴⁵⁶ In the future, a deeper understanding of the ordering and interactions of the driver molecular events, and their dynamic evolution under varying local and systemic environmental pressures and during specific tumorigenic phases, will help us gain more insights into tumor prevention, diagnosis, and early intervention.