Clonal evolution of a tumor mutation tree — from a single founding clone to Darwinian selection of a pre-existing resistant subclone under therapy
Cancer begins, in nearly every case, from a single cell. That cell acquires an initiating "driver" mutation — an alteration in a gene controlling proliferation, survival, or genome stability — that grants it a selective growth advantage over its neighbors. All descendants of that one cell inherit the mutation, making it the most recent common ancestor (MRCA) of the tumor and the root node of what oncologists now reconstruct as a mutational phylogenetic tree, exactly analogous to a species phylogeny.
In 1976, pathologist Peter Nowell proposed that tumors arise and progress through an iterative Darwinian process: a single cell acquires a heritable change conferring a proliferative advantage, expands into a clone, and that clone becomes the substrate for further mutation and selection. This single insight reframed cancer not as a static disease state but as a population genetics problem — a lineage under continuous mutation, drift, and selection, just like any evolving species.
The founding mutation is usually a "driver": a change in a proto-oncogene (activating, e.g. KRAS G12D), a tumor suppressor (inactivating, e.g. TP53 loss), or a genome-stability gene (e.g. mismatch repair loss) that increases the mutation-generation rate itself. Everything downstream — every subclone, every resistant lineage — is built on this first genetic event.
Clinically, the founding clone is inferred rather than observed directly. Whole-genome or whole-exome sequencing of a tumor sample yields a list of somatic mutations and their variant allele frequencies (VAFs). Mutations present in essentially all tumor cells (clonal mutations, VAF near the tumor cell fraction) are inferred to have arisen early, in or near the founding clone. Mutations present in only a fraction of reads (subclonal mutations, lower VAF) arose later, after the population had already begun to diversify.
Computational tools (e.g. PyClone, SciClone, dN/dS phylogenetic reconstruction) cluster mutations by VAF and use them to build a tree rooted at the founding clone — the same logic used to build the tree of life from shared vs. derived genetic characters.
As the founding clone proliferates into millions and then billions of cells, DNA replication is imperfect and genomically unstable cancer cells accumulate mutations at an accelerated rate. Different cells acquire different additional mutations independently, and because these events are largely irreversible, the descendant lineages diverge — creating a branching tree of genetically distinct subclonal populations coexisting within a single tumor mass. This is intratumoral heterogeneity (ITH).
Not every new mutation matters equally. "Passenger" mutations are neutral hitchhikers carried along because they occurred in a cell that expanded for other reasons; they mark a lineage without changing its fitness. "Driver" mutations, occurring later in a subclone rather than in the founding cell, confer an additional selective advantage on top of the ancestral background — accelerating that particular branch's growth relative to its siblings.
Because different regions of a bulk tumor can be dominated by different subclones, a single biopsy taken from one location can badly under-sample the tumor's true genetic diversity — a phenomenon oncologists now call the "regional sampling bias" problem, directly motivating multi-region and liquid biopsy approaches.
The TRACERx (TRAcking Cancer Evolution through therapy, Rx) studies sequenced multiple spatially separated regions of the same primary tumor, revealing that the great majority of lung and renal cancers evolve through branched, not linear, phylogenies: a trunk of clonal (universally shared) mutations gives way to multiple divergent branches, each private to a tumor region or subclone.
Higher measured heterogeneity — more branches, more subclonal diversity — has been repeatedly associated with worse relapse-free and overall survival, because a more heterogeneous population carries a larger reservoir of pre-existing variants, including ones that might already tolerate a future therapy.
Branching evolution means a tumor is best understood not as one disease but as an evolving ecosystem of competing lineages — some cooperating, some competing for the same nutrients and space, all subject to selection.
A critical and often counter-intuitive insight from tumor evolutionary biology is that mutations conferring resistance to a specific therapy do not need to be caused by that therapy. Because mutation is ongoing and essentially random with respect to future selective pressures, a subclone carrying a resistance-conferring variant can already exist, at low frequency, within the tumor before treatment is ever given — mirroring the classic Luria–Delbrück fluctuation test in bacteria.
In 1943, Salvador Luria and Max Delbrück demonstrated with bacteriophage-resistant E. coli that resistance mutations arise randomly during growth, before exposure to the selective agent — not as a directed response to it. Fluctuation in the number of resistant colonies across replicate cultures could only be explained by mutations occurring stochastically at various points during pre-treatment growth.
The same logic applies to tumors. Well before a patient ever receives an EGFR inhibitor, ALK inhibitor, or BRAF inhibitor, a rare subclone within the primary tumor may already carry the exact resistance mutation (e.g. EGFR T790M, ALK G1202R, BRAF amplification) that will later become clinically dominant — simply because the tumor has been generating genetic diversity all along.
Because pre-existing resistant subclones are typically present at very low variant allele frequencies (often well under 1%), detecting them requires ultrasensitive assays: digital droplet PCR (ddPCR), duplex/error-corrected next-generation sequencing, or deep amplicon sequencing of plasma-derived circulating tumor DNA (ctDNA).
Studies using such assays have detected resistance mutations like EGFR T790M in a meaningful subset of treatment-naive tumors, at frequencies far below what standard clinical genotyping can see. This has direct clinical implications: the size of the pre-existing resistant reservoir at treatment start is one of the strongest predictors of how quickly resistance will become clinically apparent.
Resistance is frequently not "acquired" during therapy in the sense of being newly created by the drug — it is often selected from pre-existing diversity that was already there, waiting.
Once therapy begins, it acts as an extremely strong selective force. Susceptible subclones — including the bulk of the tumor, often even the founding lineage — are killed or suppressed. Any pre-existing resistant subclone, unaffected or minimally affected by the drug, is largely spared. With competing lineages removed, the resistant subclone gains access to space, nutrients, and growth signals it previously had to share — competitive release — and begins to expand relative to the rest of the population, exactly as antibiotic-resistant bacteria expand under antibiotic pressure.
In population genetics terms, a targeted therapy imposes a strongly negative selection coefficient on susceptible clones (their effective growth rate becomes negative — the population shrinks) while leaving the selection coefficient of a resistant clone largely unchanged, or even improved via "competitive release" as competing lineages vacate the tumor microenvironment.
The net effect is that even a resistant subclone present at under 1% of the tumor at treatment start can, given the same intrinsic proliferation rate as before, come to dominate the tumor within months, purely because its competitors have been selectively removed — not because the resistant clone itself grew unusually fast.
This dynamic is mechanistically identical to how antibiotic resistance emerges in bacterial infections: a large susceptible population is present, a small resistant subpopulation pre-exists (or emerges) by chance, the antibiotic (or targeted cancer drug) kills susceptible organisms, and the resistant subpopulation — now facing far less competition — expands to fill the vacated niche. In both settings, the drug does not "create" resistance from scratch nearly as often as it reveals and amplifies resistance that was already present.
This reframing has motivated adaptive therapy strategies in oncology: rather than dosing to maximum tolerated levels (maximizing selection pressure for resistance), some protocols deliberately leave susceptible tumor burden intact to continue suppressing the resistant clone through ordinary competition — trading short-term tumor shrinkage for longer-term control.
Serial biopsies and ctDNA monitoring during therapy frequently show a resistant clone's allele frequency rising from a barely detectable trace to tumor-dominant levels over just weeks to months of treatment.
Over the full treatment course, the once-rare resistant subclone can progressively displace the susceptible bulk tumor to become the dominant tumor population. Clinically, this manifests as disease progression on imaging or rising biomarkers despite the drug's initial, often dramatic, effectiveness — a pattern oncologists recognize as acquired resistance, even though the underlying resistant genotype may have predated treatment entirely.
Because resistant-clone expansion is a gradual, exponential process, molecular signals often precede clinical or radiographic evidence of relapse. Rising ctDNA levels, or a rising fraction of a known resistance mutation in serial liquid biopsies, can detect the resistant subclone's dominance weeks to months before a scan shows measurable tumor regrowth — creating a window for early intervention or therapy switching before the disease burden becomes clinically severe.
By the time progression is clinically obvious, the resistant lineage may already represent the overwhelming majority of tumor cells, meaning that continuing the original therapy provides little additional benefit — the selective pressure keeps the susceptible remnant suppressed, but it can no longer meaningfully act on a population it was never designed to target.
Because the resistant clone typically has its own specific vulnerability (a second driver mutation, a bypass signaling pathway, a changed drug-binding pocket), identifying the resistance mechanism at progression allows a rationally chosen next-line therapy — for example, a third-generation inhibitor designed specifically to overcome a common gatekeeper resistance mutation.
However, each new therapy imposes a new round of selective pressure, and the same clonal evolution logic applies again: a small subclone resistant to the second-line drug may already be present, ready to be selected in turn. This has driven interest in upfront combination therapies (attacking multiple pathways simultaneously, so no single mutation confers resistance to the whole regimen) and adaptive dosing strategies that manage tumor burden as an evolving ecosystem rather than a target to be eradicated in one strike.
The central clinical lesson of clonal evolution: an excellent initial response to therapy is not the same as cure. As long as any genetically diverse tumor population survives treatment, Darwinian selection continues to operate on it.