The Hidden Logic of Cancer’s Evolution: Why Mutation Order Shapes Tumor Diversity

In a tumor, two cells with identical driver mutations can behave like strangers from different worlds—because the order in which they acquired those mutations determines not just their biology, but the very structure of the cancer ecosystem. According to a new modeling study by Chu, Sheng, Zhang, and Wang (2026), the sequence of genetic changes in cancer does more than alter cell fitness: it fundamentally reshapes intratumor heterogeneity, the mosaic of competing cell lineages that drives treatment resistance and disease progression. Their key finding? When a strong driver mutation occurs early in evolution, it amplifies diversity by spawning many independent malignant clones. But when that same mutation comes last, it suppresses diversity by allowing one lineage to dominate. This dual role produces a non-monotonic relationship between selective advantage and clonal concentration—one of the most counterintuitive results in cancer evolutionary dynamics to date.

This isn’t just theoretical nuance. It means that two tumors with identical mutational profiles could respond very differently to therapy depending on when each mutation arose. A JAK2 mutation in myeloproliferative neoplasms may lead to aggressive disease if it follows TET2, but milder outcomes if it comes first (Ortmann et al., 2015). The model explains why: mutation order alters the supply of founding clones and their relative growth advantages, shaping the tumor’s evolutionary trajectory long before treatment begins.

The Science

The researchers developed a multitype branching-process model to simulate how tumors evolve through multiple mutational pathways. Starting from a single wild-type cell, populations grow, divide, die, and acquire driver mutations stochastically. The core innovation lies in distinguishing malignant cells not only by their final genotype but also by the order in which they accumulated mutations—and crucially, by the independent transformation events that founded each clone.

They focused on a minimal yet revealing scenario: malignant transformation requires exactly two driver mutations, labeled mutation 1 and mutation 2. Two distinct evolutionary pathways emerge:

• Pathway (1→2): Wild-type → mutation 1 → mutation 2 (malignant)
• Pathway (2→1): Wild-type → mutation 2 → mutation 1 (malignant)

Even though both pathways end in cells carrying both mutations, they are treated as separate types—$s_{12}$ and $s_{21}$—to capture differences in intermediate fitness, mutation rates, and clonal genealogy.

Each cell type has its own birth rate $a_x$, death rate $b_x$, and net growth rate $\lambda_x = a_x - b_x$. Mutations occur during division at specified rates: $\mu_{\emptyset,1}$ for acquiring mutation 1 in wild-type, $\mu_{\emptyset,2}$ for mutation 2, and so on. The model allows these parameters to differ across pathways, enabling exploration of epistasis, state-dependent mutation rates, and asymmetric fitness effects.

To analyze long-term behavior, the authors applied the successive exponential approximation—a technique pioneered by Durrett and colleagues (Durrett et al., 2011)—which replaces stochastic population trajectories with deterministic exponentials modulated by random amplitudes. This captures early randomness while simplifying later dynamics, making analytical progress possible.

Their main metric is the Simpson’s index, $R_t$, defined as the probability that two randomly sampled malignant cells descend from the same founder clone. If all malignant cells come from one lineage, $R_t \approx 1$; if there are many equally sized clones, $R_t \approx 0$. Thus, higher values indicate lower clonal diversity—greater concentration.

Critically, they decomposed $R_t$ into contributions from pathway representation and within-pathway clonal structure:

$$R_t=R_{(12),t}{\pi }_{12}(t{)}^2+R_{(21),t}{\pi }_{21}(t{)}^2$$

where $\pi_{12}(t)$ and $\pi_{21}(t)$ are the fractions of malignant cells from each pathway, and $R_{(12),t}$, $R_{(21),t}$ are the within-pathway Simpson indices. This separation reveals two levers controlling overall heterogeneity: (1) which pathway dominates, and (2) how clonally concentrated each pathway is internally.

What They Found

Under conditions where both pathways produce malignant cells with equal net growth rates ($\lambda_{s_{12}} = \lambda_{s_{21}}$), the expected Simpson’s index converges to a closed-form limit that separates into two components:

1. Effective pathway weights: determined by mutation rates and growth dynamics up to the penultimate stage.
2. Within-pathway concentration terms: governed by the ratio of intermediate-to-malignant growth rates.

This decomposition exposes a fundamental asymmetry: the impact of a driver mutation depends entirely on when it acts.

When a strongly beneficial mutation occurs early—say, in the transition from wild-type to single mutant—it accelerates the expansion of the intermediate lineage. A larger pool of intermediate cells means more opportunities for the final malignant transformation, increasing the number of independently founded malignant clones. More founders = more diversity = lower Simpson’s index.

But when the same strong driver occurs late, in the final step to malignancy, it magnifies the advantage of clones founded earlier. Early-born malignant clones get a compounded head start, outcompeting later arrivals and leading to dominance by a few large lineages. Fewer dominant clones = higher Simpson’s index = greater clonal concentration.

These opposing forces create a surprising outcome: increasing the selective advantage of a driver mutation does not monotonically increase clonal concentration. Instead, depending on its position in the sequence, it can either promote or suppress diversity.

Simpson's Index Under Deterministic Mutation Order

Label	Value
f₁ = 0.01	0.45
f₁ = 0.02	0.58
f₁ = 0.03	0.62
f₁ = 0.04	0.59
f₁ = 0.05	0.51

Non-monotonic trend shows peak clonal concentration at intermediate early-driver strength. Data from {{image:0}}.

Figure 1 shows simulation results under the deterministic-order model (only one pathway active). As the fitness gain from the first mutation increases (from 0.01 to 0.05), the median Simpson’s index initially rises—indicating increased concentration—but then declines at higher values. This non-monotonic trend reflects the shift from age-dominated to supply-dominated regimes: weak early drivers allow late founders to catch up; strong early drivers flood the system with transformation events, boosting diversity.

When both pathways coexist, additional dynamics emerge. If fitness effects are commutative—meaning the total fitness of double mutants is the same regardless of acquisition order—the limiting Simpson’s index depends only on the effective weights and within-pathway ratios. But if fitness is non-additive, such as in threshold-like epistasis (e.g., second mutation unlocks full malignancy only after the first), the system can generate highly concentrated populations even with symmetric mutation rates.

Clonal Concentration Under Non-Additive Fitness

Label	Value
g₁ = 0.05	0.38
g₁ = 0.04	0.51
g₁ = 0.03	0.63
g₁ = 0.02	0.74
g₁ = 0.01	0.82

Larger late-stage fitness gains lead to higher clonal concentration. Data from {{image:3}}.

Figure 3 illustrates this effect. As the single-mutant fitness advantage $g_1$ decreases while double-mutant advantage $g_2$ remains fixed, the gap $g_2 - g_1$ widens. Both total ($R_t$) and within-pathway ($R_{(12),t}$) Simpson indices increase—showing that a larger jump in fitness at the final transition concentrates the population. This mirrors clinical observations where late-emerging subclones take over after therapy: they’re not necessarily more resistant, but they grew faster from the moment they appeared.

Finally, when terminal fitness differs by pathway—what the authors call non-commutative fitness—the faster-growing pathway asymptotically dominates, regardless of initial accessibility. Even a rare route can prevail if its malignant cells have a slight edge in net growth rate.

Why This Changes Things

For decades, oncologists have treated driver mutations as interchangeable units of risk. Find KRAS? Bad prognosis. Detect TP53? Brace for resistance. But this model shows that context—specifically, temporal context—is everything.

Consider two patients with colorectal cancer, both harboring APC and KRAS mutations. If APC was lost first, the resulting field of hyperproliferative cells creates fertile ground for multiple KRAS+ clones to emerge independently. The tumor becomes a patchwork of semi-independent lineages—a “big bang” model of evolution (Sottoriva et al., 2015). Such tumors may be harder to eradicate completely, as diverse subclones harbor different vulnerabilities.

But if KRAS mutated first, fewer intermediate cells exist to undergo APC loss. Malignant transformation happens less frequently, yielding fewer founding clones. One lineage might dominate due to early establishment and rapid growth. While seemingly simpler, this tumor could still relapse—if that dominant clone evolves resistance.

This reframes how we interpret genomic data. Bulk sequencing sees only the endpoint: “APC−/−, KRASG12D.” Single-cell phylogenetics can reconstruct order—but rarely connects it to clonal architecture. The current study provides the missing link: a quantitative framework predicting how mutation order shapes the distribution of clone sizes.

It also challenges assumptions in precision medicine. Targeted therapies often assume uniform sensitivity across cells with a given mutation. But if mutation order affects epigenetic state, microenvironmental adaptation, or metabolic programming—even without changing the target antigen—then cells with identical genotypes may respond differently based on ancestral history.

From an evolutionary perspective, the findings reveal cancer as a multi-layered selection process. Selection acts not just on mutations, but on trajectories. Pathways that maximize founder supply may win in stable environments; those that maximize growth speed may dominate under stress. And because mutation order influences both, tumors navigate a complex fitness landscape shaped by historical contingency.

Clinically, this suggests that measuring clonal diversity—via Simpson’s index or similar metrics—could improve risk stratification. In Barrett’s esophagus, high clonal diversity predicts progression to adenocarcinoma (Maley et al., 2006). This model offers a mechanistic explanation: diverse premalignant fields arise when early drivers expand intermediate pools, increasing transformation flux. Monitoring diversity over time might detect shifts toward high-risk evolutionary paths before malignancy emerges.

Moreover, the non-monotonic relationship between driver strength and clonal concentration implies that moderately strong early drivers could paradoxically reduce relapse risk by maximizing diversity—making complete eradication less likely, but competitive suppression among clones more probable. Conversely, weak early drivers followed by strong late ones may set the stage for sudden clonal sweeps, akin to post-treatment relapses.

What's Next

While elegant, the model has limitations. It assumes infinite population size, no spatial constraints, and neutral passenger mutations. Real tumors face physical limits, immune surveillance, and ecological interactions—all of which can alter clonal dynamics. Future work should integrate spatial structure (e.g., using agent-based models) and microenvironmental feedback.

Another frontier is validating predictions in human data. Longitudinal single-cell sequencing from pre-cancer to invasive stages—such as in lung adenocarcinoma or cervical dysplasia—could test whether early versus late driver acquisition correlates with observed clonal architectures. Projects like TRACERx are already generating such datasets (Jamal-Hanjani et al., 2017).

The model also opens new questions about therapeutic design. Should we aim to increase clonal diversity to prevent takeovers by aggressive subclones? Could “adaptive” therapies exploit competition between lineages, as proposed in prostate cancer (Zhang et al., 2017)? Or might drugs that delay the final malignant transition—thereby extending the window for immune clearance—be more effective than those targeting proliferation?

Perhaps most provocatively, the work suggests that cancer evolution is not merely stochastic but structured by developmental logic. Just as embryogenesis follows reproducible sequences despite cellular noise, tumorigenesis may follow probabilistic yet predictable paths dictated by mutation order, fitness landscapes, and chance.

We are beginning to see tumors not as chaotic masses, but as evolving ecosystems with genealogical memory. And in that memory, written in the order of mutations, lies a blueprint of resilience—and perhaps, a roadmap to defeat it.