Clonal Evolution

Population Bottleneck

Properties1
tagsfoundations, selection

9 min read

Population Bottleneck

Definition

A population bottleneck is a severe, transient reduction in population size that collapses genetic diversity, leaving only a small subset of individuals — the survivors — to repopulate. In clonal evolution, bottlenecks are most commonly therapy-induced: effective treatment eliminates the majority of clones, and the tumor repopulates from the resistant or dormant survivors. The bottleneck’s severity, not just its occurrence, determines the subsequent evolutionary trajectory: shallow bottlenecks preserve clonal architecture (linear or stable relapse); deep bottlenecks force re-diversification (branching relapse).

The Bottleneck Paradox

The central empirical puzzle is the bottleneck paradox: deeper therapeutic responses — those that eliminate more tumor cells and more subclones — can produce more clonal diversity at relapse than shallower responses.

In the Myeloma XI trial, Jones et al. (2019) demonstrated this pattern through WES of diagnosis/relapse pairs (N = 56). Patients achieving a complete response (CR) or very good partial remission (vgPR) underwent a clonal bottleneck leading to branched clonal architecture upon relapse — new clones, new mutations, sometimes completely new sets of genetic aberrations. Patients with partial responses (PR) maintained linear evolution or stable clonal patterns, with roughly the same clones present at diagnosis and relapse (Miething, 2019). Maintenance lenalidomide therapy did not alter this pattern: the bottleneck severity, not the maintenance drug, determined clonal architecture at relapse.

This is paradoxical under standard population-genetic theory: a more effective therapy should leave fewer survivors, reducing diversity, which should produce less diverse relapses. The bottleneck paradox inverts this expectation.

Resolution via Compression-Progress

The compression-progress framework (compression-progress-evolution) resolves the paradox. Therapy is a forced decompression event — it destroys the tumor’s compressed genomic program. The severity of decompression determines what happens next:

Bottleneck severityCompression stateEvolutionary outcome
Shallow (PR, incomplete response)Partial decompression. The dominant clone’s compression is disrupted but not destroyed.Linear/stable relapse. The same clone repairs its compression and re-expands.
Deep (CR/vgPR)Complete decompression. No surviving clone has a compression adequate to the post-therapy microenvironment.Branching relapse. Survivors re-enter exploration phase, generating diverse new compressions.

In Schmidhuber’s terms: a shallow bottleneck leaves the system on the same learning curve — the existing compression is damaged but repairable. A deep bottleneck destroys the learning curve entirely — survivors must find a new one, and the exploration process generates diverse candidate compressions. The branching architecture at relapse IS the exploration phase made visible.

This reframing has a therapeutic corollary: maximizing tumor kill may be counterproductive if it maximizes re-diversification. The optimal bottleneck may be one that is deep enough to eliminate the dominant clone but shallow enough to preserve competitive dynamics among survivors — preventing any single clone from completing a new compression. This is the evolutionary logic of adaptive therapy, now grounded in the compression framework rather than empirical observation.

Cross-Domain Evidence: Ecological Invasion

Geng et al. (2016) provide independent-domain evidence for a parallel dynamic in ecological invasions. The invasive clonal plant Alternanthera philoxeroides (alligator weed) experienced different bottleneck severities in different introduced ranges:

  • China: A severe founder bottleneck. 94% of sampled individuals shared a single multi-locus genotype — near-complete genetic uniformity.
  • USA: Multiple introductions. Each sampled individual had a unique multi-locus genotype, with diversity comparable to the native range (Argentina).

Despite this dramatic difference in genetic diversity, both introduced ranges achieved full bioclimatic niche occupancy. The mechanism: phenotypic plasticity. Under common garden conditions, Chinese clones showed the same magnitude of plastic response to water availability as genetically diverse USA and Argentine clones. Plasticity, not genetic diversity, drove invasion success.

The structural parallel to cancer is precise:

flowchart LR
    subgraph Ecology["Ecology (Geng et al. 2016)"]
        E1["Founder bottleneck<br/>(single introduction)"] --> E2["Low genetic diversity<br/>(94% identical)"]
        E2 --> E3["High phenotypic plasticity"]
        E3 --> E4["Full niche occupancy<br/>(invasion success)"]
    end

    subgraph Cancer["Cancer (Miething 2019)"]
        C1["Therapy bottleneck<br/>(CR/vgPR)"] --> C2["Low clonal diversity<br/>(few survivors)"]
        C2 --> C3["Epigenetic plasticity<br/>(re-diversification)"]
        C3 --> C4["Branching relapse<br/>(therapeutic failure)"]
    end

    E1 -.-"bottleneck".-> C1
    E2 -.-> C2
    E3 -.-> C3
    E4 -.-> C4

Cross-domain bottleneck dynamics. Left: ecological invasion (Geng et al., 2016). Right: cancer evolution (Miething, 2019). Both systems exhibit the same structure — bottleneck → low diversity → plasticity-driven adaptation → successful repopulation. The dashed lines mark the functorial mapping between domains. Synthesized from Miething (2019) and Geng et al. (2016).

Category-Theoretic Structure

The cross-domain mapping can be formalized using the olog framework of Giesa, Spivak, & Buehler (2011) (buehler2011-reoccurring-patterns). Define a Bottleneck olog with objects:

  • Population — a set of individuals (cells or plants) with heritable variation
  • Bottleneck event — a function bottleneck: Population → Population that maps a population to a subset of survivors
  • Diversity — a measure diversity: Population → ℝ⁺ (genetic diversity or clonal heterogeneity)
  • Plasticity — a function plastic_response: Population × Environment → Phenotype mapping survivors to their phenotypic range
  • Adaptive success — a predicate succeeds: Population → {true, false}

The commutativity condition: for any population P, the path P → bottleneck → diversity → adaptive_outcome must commute with P → bottleneck → plasticity → adaptive_outcome. That is: after a bottleneck, adaptive success is determined by plasticity, not by residual diversity.

The functor F: EcologyCat → CancerCat maps:

  • A. philoxeroides population → tumor cell population
  • Founder bottleneck → therapy bottleneck
  • ISSR marker diversity → subclonal mutational heterogeneity
  • Phenotypic plasticity (morphological) → epigenetic plasticity (chromatin)
  • Water availability treatment → therapeutic pressure
  • Bioclimatic niche occupancy → metastatic colonization / relapse

This functor is structure-preserving: the bottleneck → low diversity → plasticity → success path commutes in both domains. The bottleneck paradox — that deeper bottlenecks produce more adaptive outcomes through plasticity-driven re-diversification — is a commutativity condition that holds across domains.

Direct Experimental Evidence: Dual-Route Recurrence

The strongest direct experimental evidence for bottleneck dynamics comes from Walens et al. (2020), who used cellular barcoding to track individual clones through tumor regression, residual disease, and recurrence in a HER2/neu mouse model of breast cancer. Clonal diversity decreased progressively during regression and residual disease — the expected signature of a therapy-induced bottleneck. The critical finding was that recurrent tumors arose via two distinct routes:

RouteFrequencyMechanismClonal architectureDrug sensitivity
Clonal dominance~50%Met gene amplificationFew subclones dominateSensitive to Met inhibition
Polyclonal recurrence~50%Jak/Stat pathway activationThousands of subclones (similar to primary)Sensitive to Jak/Stat inhibition

The clonal dominance route is a genetic compression breakthrough (compression-progress-evolution): a single clone acquires Met amplification, achieving a compression of the post-therapy microenvironment superior to all competitors, and sweeps to dominance. This is the entrenchment pattern — the Met-amplified clone has found a compression that works, and the tumor plateaus. The polyclonal route represents a plasticity-driven strategy (dual-regime-evolution): Jak/Stat pathway activation enables diverse subclones to survive without a single dominant compression, analogous to the phenotypic plasticity that drives invasion success in ecologically bottlenecked populations (geng2016-genetic-diversity-phenotypic-plasticity).

Both routes produce viable recurrent tumors, but they have different therapeutic vulnerabilities — a finding with direct clinical implications. If a recurrent tumor is Met-amplified and clonally dominant, targeting Met should be effective. If it is polyclonal and Jak/Stat-dependent, a different strategy is needed. The bottleneck does not just reduce diversity; it selects for specific adaptive strategies, and knowing which strategy a tumor has adopted may guide therapy at recurrence.

Falsifiable Predictions

The bottleneck framework generates testable predictions that distinguish it from standard population-genetic theory:

  1. Bottleneck-severity threshold. There exists a threshold of tumor cell kill beyond which relapse architecture switches from linear to branching. Below the threshold, surviving clones repair existing compressions; above it, they re-diversify. This threshold should be identifiable from dose-response data with paired diagnosis/relapse sequencing.

  2. Plasticity as a mediator. The relationship between bottleneck severity and relapse diversity should be mediated by epigenetic plasticity. Tumors with low epigenetic plasticity (e.g., DDR-proficient, low basal chromatin remodeling activity) should show less re-diversification after deep bottlenecks than tumors with high plasticity.

  3. Non-monotonicity of adaptive success. Intermediate bottlenecks — deep enough to eliminate the dominant clone, shallow enough to preserve competition — should produce worse adaptive outcomes (from the tumor’s perspective: less successful repopulation) than either very shallow or very deep bottlenecks. This is the therapeutic sweet spot.

  4. Bottleneck timing matters. A bottleneck applied before the tumor has achieved a stable compression (early in evolution, during active exploration) should be more effective than a bottleneck applied after entrenchment (late, monoclonal architecture). This is the compression-entrenchment hypothesis (see docs/superpowers/specs/2026-07-05-ith-outcome-test-design.md).

Limitations

  • Cross-domain evidence is analogical, not homologous. Geng et al. studied an ecological invasion; Miething commented on a myeloma trial. The structural parallel is precise but the mechanisms differ: plant phenotypic plasticity operates through developmental and physiological pathways; cancer epigenetic plasticity operates through chromatin remodeling and gene expression. The functorial mapping preserves structure, not mechanism.
  • Single-study evidence in each domain. The bottleneck paradox is documented in one myeloma trial (N = 56). The plasticity-compensates-for-diversity dynamic is documented in one plant species. Replication in other cancer types and other invasive species is needed.
  • Bottleneck severity is not directly measured. In the myeloma study, response depth (CR/vgPR vs. PR) is a clinical surrogate for bottleneck severity, not a direct measure of population reduction. The exact relationship between tumor cell kill and architectural outcome is unknown.
  • Maintenance therapy’s null effect may be drug-specific. Lenalidomide’s lack of effect on clonal patterns may reflect its immune-dependent mechanism (Miething, 2019). Cytotoxic or targeted therapies that directly select for resistant clones may produce different bottleneck dynamics.
  • Cannot distinguish bottleneck-driven from time-driven re-diversification. Branching at relapse could be driven by the bottleneck (survivors re-explore) or simply by the passage of time (more cell divisions → more mutations → more diversity). Longitudinal sampling with multiple time points is needed to distinguish these.

Graph View

Backlinks