The Darwinian Cancer Paradigm The prevailing model of tumour evolution holds that cancer cells evolve like organisms in nature - through Darwinian selection in which beneficial mutations are positively selected, drive clonal expansion, and thereby determine intratumour genetic diversity. Under this model, genetic diversity within a tumour should be relatively low and structured by selection sweeps.
An Untested Assumption Despite its widespread acceptance, the Darwinian model of intratumour evolution had never been rigorously tested against quantitative predictions. Prior studies examined fewer than 10 samples from a single tumour - far too few to detect or reject selection patterns with statistical confidence. This study was designed to provide the statistical power needed for a definitive test.
Non-Darwinian Evolution as the Null Model The authors proposed non-Darwinian (neutral) evolution as the null hypothesis: that most mutations arise and drift randomly, with population size (N), mutation rate (u), and growth parameters determining diversity. Under this model, tumours with billions of cells should harbour hundreds of millions of distinct mutations, most at very low frequency.
HCC Study Design From a single hepatocellular carcinoma (HCC) tumour, 286 spatially distinct tumor samples were collected via honeycomb microdissection. Of these, 23 underwent whole-exome sequencing (WES) and all 286 were genotyped for discovered mutations. This coverage - to our knowledge the highest ever applied to a single tumour - enabled statistically rigorous testing of evolutionary models.
Honeycomb Microdissection A 1 mm-thick slice was cut through the middle of a 3.5 cm HCC tumour. 286 cylindrical samples (0.5 mm diameter, 1 mm height, approximately 20,000 cells each) were evenly distributed across four quadrants (A-D). 23 samples were selected for WES at average 74-fold read depth; the remaining 263 were genotyped for mutations identified in the sequencing.
Mutation Calling and Validation 269 somatic SNVs were identified in coding regions and splice sites. Shared SNVs were validated by cross-reference across multiple samples. All singleton SNVs (found in only one sample) were individually confirmed by Sequenom genotyping and Sanger sequencing, ensuring zero false positives. Copy number alterations were also characterized but the main analysis focused on SNVs.
Population Genetic Framework The data were analyzed using the infinite-site and infinite-allele models from population genetics. Clone sizes were represented as a mutation frequency spectrum (the number of clones appearing i times in n samples), and compared to predictions from the non-Darwinian model using standard neutral population genetics formulas. Simpson's diversity index H was calculated.
MALL Estimation Four independent methods were used to estimate MALL (the total number of coding mutations in the entire tumor): Mmin (minimum growth model), Meq (equilibrium model), M3D (3D growth model), and Mexp (exponential growth model). These estimates were then extrapolated to the full tumour size of greater than 10^9 cells.
20 Distinct Clones in 23 Samples The 35 validated polymorphic SNVs delineated 20 unique cell clones among the 23 sequenced samples. Simpson's diversity index H = 0.941 means that two randomly chosen samples would be genetically different with 94% probability - extremely high clonal diversity for a solid tumour.
Frequency Spectrum Fits Non-Darwinian Model The observed mutation frequency spectrum was [xi = 26, 7, 1, 1, 0, 0...], closely matching the expected neutral spectrum from non-Darwinian theory (chi-squared p = 0.865). No clones were found at unusually high frequency, which would be expected if positive Darwinian selection had driven any clone to expand. The test had sufficient power to detect even weak selection.
Sectoring Pattern of Spatial Growth Spatial mapping of the 286 samples revealed that clones grew outward in sectors from the ancestral clone. Derived subclones were consistently observed at the outer flanks of parent clones, consistent with outward growth without selective sweeps. Neighboring cells almost always differed by one to two coding mutations.
MALL Greater Than 100 Million All four estimation methods agreed that when extrapolated to the full tumour size of more than 10^9 cells, MALL (total coding mutations in the entire tumour) exceeded 100 million. Under Darwinian models, MALL would be orders of magnitude smaller. At even 10^6 cells (below imaging threshold), all four methods estimated approximately 10^5 coding mutations.
Low-Frequency Mutations Evade Selection Under non-Darwinian evolution, approximately 99% of tumour mutations are found in fewer than 100 cells. Such low-frequency mutations experience strong random drift, making selection nearly impossible to act upon. A mutation must reach a high enough frequency before it can be subject to meaningful selective pressure.
Spatial Constraints in Solid Tumours In solid tumours, cells do not migrate freely. When an advantageous mutation arises, the cells carrying it compete primarily with their direct neighbors - often cells carrying the same mutation. This spatial constraint blunts the competitive advantage that would allow a Darwinian sweep to occur. The situation differs from hematological cancers like leukemia, where cells circulate freely.
Ka/Ks Ratios Near 1 The observation that the ratio of nonsynonymous to synonymous mutations (Ka/Ks) is near 1 across hundreds of cancer genomes in TCGA data is consistent with non-Darwinian evolution and confirms that protein-altering mutations in tumours are not subject to the strong purifying selection seen between species (where Ka/Ks is typically less than 0.3).
Drug Treatment as a Selection Amplifier Drug treatment may be the physiological event that 'loosens up' the structured tumour cell population and allows competitive dynamics to emerge. Under this view, therapy converts non-Darwinian intratumour evolution into Darwinian selection - explaining why drug resistance can emerge rapidly from pre-existing rare variants that exist in the highly diverse non-Darwinian background.
Drug Resistance Is Likely Pre-Existing If a tumour of 10^9 cells harbours more than 100 million distinct coding mutations, then for virtually any drug target, resistant variants already exist within the tumour before treatment begins. This has profound implications for therapy: drug resistance may not be an evolutionary response to treatment, but rather selection of pre-existing rare clones.
Even Microscopic Tumours Harbor Diversity At the scale of 10^6 cells (below clinical detection), the model predicts approximately 10^5 distinct coding mutations. This means that by the time a tumour is detected, it has already accumulated extensive genetic diversity that can fuel resistance to single-agent treatments.
HCC-Specific Findings Six putative driver genes were identified among the fixed mutations in this HCC (mutations present in all cancer cells): CCAR1, CPXM2, DNAH7, TMPRSS13, TP53, and TSC1. Importantly, none of the 35 polymorphic mutations (those defining subclones) mapped to known driver genes, consistent with neutral evolution of polymorphic diversity.
Implications for Combination Therapy The finding that virtually all possible resistance variants exist before treatment argues strongly for upfront combination therapy targeting multiple distinct mechanisms simultaneously, rather than sequential monotherapy. Microscopic tumours may be the optimal treatment window before diversity expands further.
Testing in Other Solid Tumours This study was performed in a single HCC tumour. Future work should apply the same high-density sampling and population genetic framework to other solid tumour types - colorectal, breast, lung, and pancreatic cancers - to determine whether non-Darwinian evolution is a general principle across solid tumours or specific to certain cancer types.
Copy Number Alterations This analysis focused on SNVs because the mutation rate for copy number alterations (CNAs) is not well characterized. Future studies should develop comparable frameworks for assessing whether CNAs also evolve under non-Darwinian dynamics, or whether structural alterations are subject to different selective pressures.
Timing of Driver Mutations The six fixed driver mutations in HCC-15 (including TP53 and TSC1) arose early and were present in all cells. The timing of driver mutation acquisition relative to non-Darwinian diversification remains to be characterized. Single-cell sequencing approaches could resolve the temporal order of driver acquisition and neutral diversification.
Therapeutic Implications of Diversity The proposed relationship between intratumour genetic diversity and poor patient survival warrants systematic investigation. If high diversity predicts therapeutic resistance, diversity measures from multi-region biopsies or liquid biopsies could guide clinical decision-making about treatment intensity and combination strategy.