Rethinking cancer initiation: The role of large‐scale mutational events

Cancer initiation is revisited in light of recent discoveries in cancer pathogenesis. Of note is the detection of mutated cancer genes in benign conditions. More significantly, somatic clones, which harbor mutations in cancer genes, arise in normal tissues from early development through adulthood, but seldom do they transform into cancer. Further, clustered mutational events—kataegis, chromothripsis and chromoplexy—are widespread in cancer, generating point mutations and chromosomal rearrangements in a single cellular catastrophe. These observations are contrary to the prevailing somatic mutation theory, which states that a cancer is caused by the gradual accumulation of mutations over time. A different perspective is proposed within the framework of Waddington's epigenetic landscape wherein tumorigenesis is viewed primarily as a disruption of cell development. Cell types are defined by their specific gene‐expression profiles, determined by the gene regulatory network, and can be regarded as attractor states of the network dynamics: they represent specific, self‐stabilizing patterns of gene activities across the genome. However, large‐scale mutational events reshape the landscape topology, creating abnormal “unphysiological” attractors. This is the crux of the process of initiation. Initiation primes the cell for conversion into a tumor phenotype by oncogenes and tumor suppressor genes, which drive cell proliferation and clonal diversification. This view of tumorigenesis calls for a different approach to therapy.


| CANCER INITIATION: A BRIEF HISTORICAL PERSPECTIVE
2][3][4] Brief exposure of the skin of animals to a carcinogen, like tar, followed by repeated applications of a non-carcinogen, like croton oil, produced tumors.
The initiating event was viewed as a sudden, irreversible change in a small minority of cells in the treated area, giving rise to isolated, latent tumor cells, which are morphologically indistinguishable from the surrounding non-neoplastic cells.An additional stimulus, such as repeated exposure to a non-mutagenic tumor promoter, is required for the conversion of the initiated cells into tumors (papillomas).Promotion may be reversible if there is insufficient exposure to the promoter.Furthermore, the sequence is crucial insofar as initial application of the promoter followed by the carcinogen fails to produce tumors.Foulds later put forward the notion of a third step, progression, to describe the independent growth of tumor cells. 5,6

| THE SOMATIC MUTATION THEORY OF CANCER
The somatic mutation theory states that a cancer is caused by mutations in genes that normally control the cell cycle and cell proliferation. 7,8The mutations give the cell a selective growth advantage: an oncogene mutation causes a "gain of function" that promotes cell growth, while a loss of a tumor suppressor gene represents a "loss-offunction" mutation that turns off its inhibitory growth action. 9,10In addition, epigenetic changes, such as DNA methylation, histone modification and chromatin openness, contribute to cancer development. 11er time, the sequential accumulation of four to five mutations in a cell leads to its transformation into a tumor cell. 124][15][16][17] Histopathological evidence of a stepwise process of cancer formation is illustrated by the adenoma-carcinoma progression of colorectal cancer, in which the cancer develops in a seemingly linear manner from a small benign polyp (adenoma) to a larger adenoma to invasive carcinoma to metastatic cancer. 8,18,19large number of mutated genes are identified in individual human cancers.These can be divided into two functional categories: driver genes and passenger genes.Alterations of driver genes account for the selective growth advantage of the cancer cells and their consequent clonal expansion.1][32] The somatic mutation model, with its stepwise acquisition of gene mutations, cannot readily account for the presence of these genetic programs in cancer cells.
Second, several alternative genetic pathways can give rise to a particular cancer, suggesting that a number of different growth-controlling genes are involved.In colorectal cancer, for example, apart from mutation of the APC tumor suppressor gene, a negative regulator of the WNT pathway, in the earliest stages of tumor growth, there is not a consistent sequence of genetic changes during its development. 33It is possible that the important event is the alteration in the expression of a gene class rather than a specific gene mutation; hence, similar or complementary pathways would result in similar biological outputs.
Nonetheless, the absence of consistent driver mutations in specific cancers is noteworthy.Third, cancer traits do not all appear sequentially, and some are already encoded in mutant alleles that are acquired early.A cogent example of this is the ability of cancer to metastasize, which is regarded as a late clinical event.The genetic alterations responsible for metastasis are not unique, but are manifestations of well-known oncogenes and tumor suppressor genes, including RAS and MYC. 34,35In the early stages of cancer, the mutated genes promote cell proliferation, and their role in metastasis is expressed only much later, likely in concert with other genetic changes acquired by descendant cells.Fourth, it might be expected that putative founder mutations would be conserved as the cancer progresses from the premalignant to the malignant state.However, there are noteworthy exceptions.For example, the BRAF mutation, a known driver of malignant melanoma, occurs in about half of all cases, but is more common (62%) in dysplastic nevus, a premalignant lesion. 36,37Along the same lines, there is a paradoxical disconnect in the frequency of mutation of key genes in the continuum from the early to late stages of cancer.9][40] These findings could be explained by a loss of the mutation during progression of the cancer or the emergence of a dominant subclone without the particular mutation in the first place.
Nevertheless, the variable presence of specific driver genes as a cancer develops is counterintuitive.
Cancer genes have also been detected in several benign conditions, raising questions about their precise function.The aforementioned BRAF mutation, which is present in malignant melanoma, is also found in the majority (81%) of benign melanocytic nevi. 41,427][48][49] Further, germline inheritance of FGFR3 alterations causes dwarfism syndromes, but does not increase the cancer risk in affected individuals. 50The observation that driver gene alterations identified in cancer are also evident in benign conditions, sometimes at relatively high frequencies, is not readily explained.
5][56][57][58] Similarly, somatic mutations that drive clonal expansion of blood cells have been detected in 10% of healthy individuals by the age of 70 years. 59,60 summary, the discovery of gene-expression signatures of embryonic cells in cancer, the incongruent sequence of gene alterations in cancer development, the unexpected drop in frequency of driver genes during cancer progression, the surprising association of cancer genes with benign conditions, and, critically, the unexplained pervasiveness of cancer genes in healthy tissues raise questions about the role of the cancer genes in tumorigenesis, benign diseases and, indeed, normal aging.While there is no debate that oncogenes and tumor suppressor genes contribute to cancer formation, it seems that their functional impact might be influenced by various other factors, such as the nature of the genomic landscape, the tissue and cellular setting, the presence of other genomic co-drivers or suppressors, or immune surveillance. 27,51

| CANCER STEM CELLS AND THEIR NICHES
Tissues in the body are continuously renewed by tissue-specific stem cells or adult stem cells. 61,62In the classic model of hematopoietic cell hierarchy, the stem cell pool is small and quiescent.Through asymmetric division, the stem cell gives rise to a new stem cell, which replenishes the pool, and a transient amplifying cell, which divides rapidly and differentiates into short-lived cells that subserve tissuespecific functions.The updated model holds that in adult tissues specialized niche cells provide stem and progenitor cells with paracrine signals necessary for their maintenance or expansion.Adult stem cells are more abundant in their niches and actively divide throughout life. 63,64Further, adult stem cells are now recognized to divide in such a way as to produce one, two or no daughter stem cell.[70][71][72][73] Analogous to the renewal of healthy tissues, a cancer is maintained by a subpopulation of tumor-initiating cells or cancer stem cells with self-renewal and pluripotency properties. 74,75The cancer stem cell model differs somewhat from the somatic mutation model, which suggests that any cell can give rise to a cancer as consequence of a succession of time-dependent mutations.In the stem cell model, dysregulation of self-renewal pathways is the critical event that causes expansion of the stem cell population.Further, cells within a cancer are capable of undergoing phenotypic transitions, and they can stochastically enter into stem cell-like states in the absence of normal niche microenvironments. 76As cancer stem cells acquire mutations, they lose their dependency on niche factors and become increasingly autonomous.The accumulating mutations also interfere with the process of differentiation, leading to a shallow cellular hierarchy.Thus, as the cancer progresses, its cell composition shifts to an increasing proportion of cancer stem cells relative to non-stem cells. 64,77,78

| WADDINGTON'S EPIGENETIC LANDSCAPE
All cells in multicellular organisms arise from a single cell, the zygote, through the process of differentiation, in which cells undergo transitions into distinct cell fates or specialized cell types.In 1942, Waddington introduced the concept of an "epigenetic landscape" as a metaphor to capture this concept. 79,80(It is salient to note here that the term "epigenetic landscape" in this context refers to a systemlevel stable state of genetic interactions rather than the more common concept of "epigenetic trait" as a heritable phenotype resulting from changes in a chromosome without alterations in the DNA sequence.)In Waddington's epigenetic landscape, a cell is imagined as a pebble which rolls down from the top of the hill through its valleys to the bottom where it comes to rest (Figure 1A).moves into a valley, its subsequent route is restricted, which means its fate also becomes constraint.In this schema, as a progenitor cell moves down the hill, it multiplies and its progeny independently go one way or the other through the various valleys and sub-valleys.The descendant cells eventually come to rest in the lower regions as differentiated cells.The developmental paths or trajectories from immature to mature regions are determined by the landscape's structure or topology. 81,82Waddington was prescient in his intuition that the landscape's structure is underpinned by genes, which control the depths and positions of the valleys as well as the heights and positions of the intervening ridges (Figure 1B).

| CELLS AS ATTRACTOR STATES
Cell types are defined by their patterns of gene expression.A crucial issue is how cell development is orchestrated to produce reliably the gene-expression profiles of individual cell types from the myriad genes in the genome.This is mediated through an hierarchical, scalefree gene regulatory network (GRN), made up of a small subset of highly linked nodes (genes) and a large number of others that are sparsely connected.The most highly connected nodes form hubs, which control the activity of the other nodes and determine the network's behavior.Interference of nodes of high connectivity is more likely to affect the network performance compared with those of low connectivity whose removal leaves the network largely unchanged. 83,84e nexus between genes and phenotypic states was initially proposed by Delbruck in 1949 and later by Monod and Jacob in 1961, who showed that mutually regulating genes can generate a number of small gene regulatory circuits that settle down into more than one stable equilibrium state.These stable patterns of gene expression account for differentiation into different phenotypes. 85,86An appreciation of the dynamics of regulatory circuits can be gained by considering a simple, hypothetical network consisting of three genes (A, B, and C), each of which receives signals from the other two.For this exercise, we assume that the genes are either "on" or "off" at any given time, and we can specify the rules that govern their behavior: Gene A becomes active if both Genes B and C are active; Genes B and C become active if either of the other genes is active.Thus, the state of each gene is influenced by the others, and depending on the combination of inputs received, a gene will either be on or off at the next moment in time in accordance with the specific rule; this allows us to predict the long-term behavior of the system. 87The network has eight potential states, but settles into one of three equilibrium states with a recurrent pattern of activity (Figure 2).In State I, all the genes are off and the system remains in the off position.In State II, with either B or C in the on position, the system cycles between two states.With any F I G U R E 2 A simple network of three interconnected genes (A, B, and C).The state of individual genes at time (t) determines the state of the other genes at the next moment in time (t + 1) according to specific rules.In this example, Gene A becomes active if both Genes B and C are active; Genes B and C become active if either of the other genes is active.The system eventually settles down into three state cycles.other network state, the system eventually reaches a steady state with all the genes in the on position (State III). 88om an analysis of the statistical properties of large ensembles of Boolean networks, Kauffman, in 1969, found that complex networks of many thousands of interacting regulatory genes settle down into just hundreds of stable equilibrium states, called "attractors." Each attractor state corresponds to the gene-expression profile of a distinct cell type. 89,90The GRN represents a state space that contains all theoretically possible gene-expression profiles.Each geographic position in the state space corresponds to an attractor state or cell type: they have distinct patterns of gene activities across the genome as a result of the constraints placed by the gene-gene regulatory interactions of the GRN.Individual cell states display two critical features: selforganization and self-stabilization.Through self-organization, specific gene-expression profiles reliably form specific cell types during differentiation.Further, small perturbations due to changes in the expression levels of individual genes are typically insufficient to destabilize a cell, and several paths are available within its "basin of attraction" to restore it to its original state; this gives the system the property of robustness. 81,91The presence of attractor states in complex gene regulatory systems is supported by gene-expression profiling experiments.[94] In Waddington's model, the attractor state represents a stable equilibrium state within its valleys whereas the ridges that separate the valleys are unstable states. 82There are two ways by which a change in cell type can occur. 95,968][99] While most perturbations are typically small and incapable of displacing a cell from its attractor state, few are occasionally of sufficient force to cause it to "jump" over a ridge in the landscape, generating new phenotypes across cell generations. 1002][103] Second, extrinsic inputs or signaling, such as from growth factors associated with cell fate decisions, can cause a cell to cross over a ridge and switch to another state. 104

| Cancer cells as discrete attractors
Kauffman postulated that cancer attractor states exist in the genome, but are not normally expressed.Over evolutionary time, cell differentiation trajectories from immature to mature phenotypes have become streamlined and the smooth "canalization" of the epigenetic landscape makes these attractors inaccessible. 105However, genetic mutations can reshape the contours of the landscape, such as tilting its slope or lowering the height of a separating ridge.This facilitates state transitions, allowing cells to veer from regular differentiation pathways and enter unused attractor states, among which are gene-expression profiles that encode a neoplastic phenotype.[108]

| LARGE-SCALE MUTATIONAL EVENTS IN CANCER
In 2020, the Pan-Cancer Analysis of Whole Genomes (PCAWG) Consortium published a comprehensive meta-analysis of whole-genome sequences from 2658 tumors across 38 cancer types.Three clustered mutational events in cancers are described. 12First, kataegis, a focal hypermutation process detected in about 61% of cancers, is caused mainly by the APOBEC family of cytidine deaminases.3][114][115] This generates multiple gene amplifications/deletions, and accounts for 3.6% of all drivers and about 7% of copynumber drivers. 12,116Third, in chromoplexy, identified in about 18% of cancers, different chromosomes are broken and the multiple strands of double-stranded DNA are ligated to each other in a new configuration.The scrambled rearrangements result in multiple changes, such as generation of oncogenic fusion genes as well as disruption or deletion of genes located near rearrangement breakpoints. 12,117,118The discovery of these large-scale mutational events in cancer is not easily reconciled with the somatic mutation model of tumorigenesis, which espouses a time-dependent, multistep accumulation of mutations.
It is important to recognize that while most cancers harbor multiple mutations, a small number across different cancer types may have infrequent or no identified driver mutations. 12,119This may be due to technical difficulty in detecting the mutation or it may reflect other biological processes, such as chromosomal changes, copy number variations or epigenetic modifications.For example, in cases of chromophobe renal cell cancer and pancreatic neuroendocrine tumor with no driver mutation, chromosomal aneuploidy is consistently observed, suggesting that the underlying pathogenetic mechanism resides in the chromosomal gains and losses. 12It is also likely that there are cancerassociated genes not yet discovered or that undescribed mutations in the noncoding and regulatory regions of the genome may be sufficient to give rise to a cancer in the absence of more targeted driver events.These exceptions highlight the complexity of oncogenesis, which generally involves mutations within the coding sequences of known cancer genes and specific noncoding elements, like promoters, but may also encompass alterations in gene regulatory mechanisms.

| DISCUSSION
The somatic mutation theory, which has influenced our thinking for decades, states that a cancer arises from a succession of genetic and epigenetic changes, which give rise to the various neoplastic traits.An alternative view is that the transformation of normal adult stem cells into cancer stem cells, brought about by the disruption of self-renewal pathways, is the cause of cancer.While the changes that initiate the process are incompletely understood, it is widely assumed that mutations in stem cells are the proximate event, although chronic tissue inflammation has been put forward as the possible inciting cause. 64,120 extension of the stem cell model is that the occasional mutation in non-stem cells in the niche could give them tumor initiating capability. 78ile these models are widely accepted, there is the confounding issue that cancer driver mutations are commonly acquired in normal tissues from early development through adulthood.In fact, positively selected clones in non-cancerous tissues grow in number and size over a person's lifetime, possibly as a remodeling process of aging as well as in response to chronic inflammation and environmental insults. 52The overlap in driver mutations between normal and corresponding neoplastic tissues has been taken as evidence that cancer arises from one of these phenotypically normal somatic clones.However, somatic clones do not necessarily become cancerous, and this conundrum poses crucial questions about what triggers a cancer.

Framed another way, what are the conditions present when a normal cell changes into a cancer cell?
There are differences in the repertoire of mutational changes between somatic clones and cancer clones. 51Although the vast majority of drivers in somatic clones are known cancer drivers, usually only few are present; in contrast, multiple drivers are found in cancer clones.The common mutations in the somatic clones are point mutations and small indels, mainly of tumor suppressor genes.Another difference is the extent of chromosomal instability.Somatic clones have a diploid genome and only rarely acquire copy number variations and chromosomal rearrangements.In contrast, aneuploidy is a hallmark of the malignant state. 121,122The observation of multiple mutations and aneuploidy in cancer is consonant with the large-scale mutational events that have been detected in cancers: kataegis produces numerous mutations, while chromothripsis and chromoplexy generate copy number alterations and genomic rearrangements.Therefore, it seems plausible that encoded in these genetic changes is the key to cancer initiation.
We can formulate a new proposition based on the premise that cells are attractors with stable gene-expression profiles and that tumor cells represent abnormal unphysiological attractors within the state space of the cellular genome, which are by-passed by normal developmental trajectories. 106Cancer initiation can be viewed as a disruption of the cell developmental pathways due to genomic modifications that significantly alter the topology of the epigenetic landscape.Individual mutations exert a limited effect, since, for the most part, the scale-free nature of the GRN protects the network from major upheavals; at most, the attractor basins might be distorted. 123,124However, mutational events like kataegis, chromothripsis and chromoplexy affect the GRN in more profound ways.
It is likely that in most cases the genetic and chromosomal damage caused by these events is incompatible with cell survival, and the cell undergoes apoptosis.But when the damage falls below the apoptotic threshold, the mutations rewire the cell's genetic circuitry, thus reshaping its epigenetic landscape and creating new pathways to typically unoccupied attractors that lead to a neoplastic endpoint.This is the essence of cancer initiation.A study of the dynamics of a cancer attractor at single-cell resolution reveals that among the heterogenous cells within the attractor is a small group at the outer edge of the basin-"edge cells"-that appear poised to shift to another attractor state. 94Such transitions would result in the emergence of a new phenotype.Edge cells display certain unusual characteristics: distinct mRNA expression patterns involving numerous genes, nontypical marker expression at low frequency, low proliferation rate, and high apoptosis rate.Could initiated cells be akin to edge cells?
The viewpoint put forward here proposes that major genomic modifications reconfigure the cell's GRN and are responsible for switching its fate, favoring a neoplastic state.The full-blown malignant state is attained through additional mutations in the well-known cancer genes, which drive uncontrolled cell proliferation and clonal diversification.These operate in the conducive milieu created during the process of initiation.

| The path forward
While the role of somatic mutations in protein-coding genes is well described in tumorigenesis, the impact of mutations in the regulatory regions and other functional elements of the noncoding cancer genome, which makes up about 98% of the genome, is incompletely understood.Establishing whole-genome sequencing databases of the diverse mutation patterns that shape the cancer genome will provide a means to probe the spectrum of the different biological processes influenced by tumor-specific regulatory regions. 125,126A potential area of inquiry is the elucidation of the contribution of the threedimensional (3D) organization of the genome to tumorigenesis, primarily with respect to topologically associating domains (TADs) and their boundaries. 127,1280][131][132] These disruptions can affect gene expression, and a link has been demonstrated between TAD border weakening and cancer development. 129,133It can be postulated that somatic mutations within TAD boundaries may act as a head start to tumorigenesis.
Of tantamount importance is the computational modeling of the dynamic properties of the GRN that controls the behavior of the cell.The inhomogeneous connectivity distribution of the GRN, with a few highly connected nodes and most with low connectivity, introduces a topological vulnerability to the system. 134moval of a few nodes with a key role in maintaining the network's connectivity has a serious effect on the ability of the remaining nodes to communicate with each other. 124This network feature permits us to contemplate a different therapeutic strategy.
The present approach of designing drugs that target mutated genes and their abnormal proteins, many of which are often downstream, generally produces a limited and usually transitory benefit to patients because the tumor cell can bypass their growthinhibiting action.It seems more advantageous to select and block aberrant pathways that feed into the network hubs.This would go a long way in realizing the goal of precision therapy.

CONFLICT OF INTEREST STATEMENT
The author reports no conflict of interest.
As the cell travels down the landscape, it encounters a series of branching points, where it stochastically enters one of the two adjacent valleys.The path it takes determines its ultimate downhill location or "fate."Once a cell F I G U R E 1 Waddington's epigenetic landscape.(A) A cell, represented as a pebble, starts at the top of a hill and rolls down the landscape through a series of branching points that represent decision events.(B) The landscape is underpinned by the activity of genes, represented as pegs underneath the hills and valleys.The modeling of the landscape is controlled by the pull of the numerous guy-ropes, which are anchored to the genes.Credit: Waddington CH.The Strategy of the Genes: A Discussion of Some Aspects of Theoretical Biology by C.H. Waddington.(Allen & Unwin, 1957).Wellcome Collection.Attribution-Non-Commercial 4.0 International (CC BY-NC 4.0).