Intratumor Heterogeneity

Summary

Intratumor heterogeneity (ITH) is the genetic and phenotypic diversity among cells within a single tumor — both the product of past clonal evolution and the substrate for future adaptation, including therapy resistance (Nowell 1976; Greaves & Maley 2012; McGranahan & Swanton 2017). ITH spans multiple components: point mutations (SNVs), copy number alterations (CNAs), structural variants (SVs), epigenetic states, and transcriptional programs, each requiring different measurement methods (Turajlic et al. 2019; PCAWG 2020). The relationship between ITH and clinical outcome is non-linear: intermediate ITH may predict poor prognosis (enough diversity to adapt), while both very low ITH (too few variants to select from) and very high ITH (fitness burden of maintaining many subclones) may predict better outcomes — the compression-entrenchment hypothesis. ITH is measured through the variant allele fraction (VAF) spectrum, corrected for purity and copy number to estimate cancer cell fractions (CCFs), which are then clustered into subclones and arranged into phylogenetic trees (Tarabichi et al. 2021).

Intratumor heterogeneity (ITH) is the genetic and phenotypic diversity present among cells within a single tumor. It is both the product of past evolution — the branching architecture left by the accumulation of heritable variation — and the substrate for future adaptation, providing the raw material upon which selection acts (Nowell, 1976; Greaves & Maley, 2012; McGranahan & Swanton, 2017; Turajlic et al., 2019).

ITH is distinct from inter-tumor heterogeneity (differences between tumors in different patients with the same cancer type) and inter-patient heterogeneity (differences across patients, including germline genetics, environmental exposures, and immune context). These three levels form a nested hierarchy: inter-patient variation arises from different germline backgrounds and exposures; inter-tumor variation captures differences between neoplasms sharing a tissue of origin; ITH captures variation within a single neoplasm’s cell population. ITH is the level at which Darwinian selection operates within a patient — the variation on which clonal evolution depends (Greaves & Maley, 2012).

The relationship between ITH and clonal-evolution is recursive: ITH arises from clonal evolution (mutation generates diversity, selection prunes it, drift generates neutral variation) and simultaneously enables further evolution (diverse populations adapt more readily to new selective pressures, including therapy). A tumor with zero ITH is evolutionarily frozen — it cannot respond to environmental change. A tumor with maximal ITH is maximally adaptable but may be burdened by the fitness costs of maintaining large populations of poorly adapted subclones. This tension — between adaptability and fitness — is the foundation of the compression-entrenchment hypothesis (see ITH as a Clinical Biomarker).

1. Components of ITH

1.1 Genetic ITH

Genetic ITH arises from all forms of somatic genomic alteration that create differences between cells within the same tumor. Each alteration type contributes a different layer of diversity and requires different measurement methods.

Point mutations (SNVs). Single-nucleotide variants are the most commonly measured dimension of ITH. Their allele frequency distribution in bulk sequencing — the variant-allele-fraction (VAF) spectrum — encodes information about clonal architecture. Mutations present at VAF consistent with the tumor’s estimated purity (after correction for copy number) are classified as clonal (present in all tumor cells); those at lower VAFs are subclonal (present in a subset). The ratio of subclonal to total mutations — the subclonal mutation fraction (SMF) — is the most widely used single-sample ITH metric (Turajlic et al., 2019; see Measurement and Quantification). The neutral theory predicts a characteristic $1/f^2$ distribution of subclonal VAFs under drift and rapid growth; deviations from this pattern may indicate ongoing selection (Graham & Sottoriva, 2017, as cited in Turajlic et al., 2019).

Copy number alterations (CNAs). Gains and losses of chromosomal segments contribute an independent dimension of ITH that is often greater in magnitude than SNV-based ITH. CNA-based ITH is measured through logR (read depth ratios) and B-allele frequency (BAF) signals from sequencing data. Subclonal CNAs — present in only a fraction of tumor cells — produce intermediate logR values and skewed BAF patterns that are challenging to distinguish from noise at low subclonal fractions (Tarabichi et al., 2021). In TRACERx Renal, the degree of subclonal copy-number complexity distinguished metastasis-competent from metastasis-incompetent clones more powerfully than SNV-based ITH (Turajlic et al., 2019).

Structural variants (SVs). Rearrangements — deletions, duplications, inversions, translocations, and complex events like chromothripsis and chromoplexy — create large-scale genomic differences between cells. Chromothripsis, present in 22.3% of cancers in the PCAWG cohort (PCAWG Consortium, 2020), is predominantly clonal (early), meaning it creates ITH at the level of the founding clone’s descendants rather than ongoing subclonal diversification. However, the structural rearrangements it produces can generate further subclonal diversity through genomic instability in the shattered regions.

Subclonal architecture. The phylogenetic structure of clones — the tree relating all detectable subpopulations — is the “shape” of ITH. Linear architectures (sequential sweeps) produce low diversity at any snapshot. Branching architectures (coexisting subclones from a common ancestor) produce moderate-to-high diversity. Neutral architectures (many small subclones, no dominant lineage) produce high diversity without strong fitness differentials. Punctuated architectures (early catastrophe, then stasis) produce low subclonal diversity despite high clonal aneuploidy (Turajlic et al., 2019). The relationship between these architectural types and measured ITH is not one-to-one: a branching tumor with one dominant subclone and several tiny ones may have similar SMF to a neutral tumor with even subclone sizes across the same total number — distinguishing them requires examining the clone frequency distribution, not just the clonal/subclonal binary classification.

1.2 Non-Genetic ITH

Genetic ITH is only one dimension of tumor diversity. Non-genetic heterogeneity — variation in chromatin state, gene expression, cellular phenotype, and microenvironment — contributes to ITH and may, in some contexts, compensate for low genetic diversity.

Epigenetic heterogeneity. Chromatin state — DNA methylation, histone modifications, chromatin accessibility, 3D genome organization — varies across cells within a tumor independently of the DNA sequence. IDH1/2 mutations produce genome-wide DNA hypermethylation, creating a uniform epigenetic state from a genetic driver (dual-regime-evolution). By contrast, stochastic methylation drift, microenvironmental signals, and therapy-induced stress generate cell-to-cell epigenetic variation without sequence changes. Epigenetic states are inherited through cell division by reader-writer complexes (DNMT1 for methylation, Polycomb/Trithorax for histone marks), making them heritable in somatic lineages — a Lamarckian dimension of ITH (dual-regime-evolution).

Transcriptional heterogeneity. Gene expression programs vary across cells due to both underlying genetic variation (allele-specific expression, cis-regulatory mutations) and epigenetic plasticity (chromatin state fluctuations, transcription factor availability, signaling pathway activity). Mikutenaite et al. (2025) demonstrated that transcriptional plasticity enables metastatic adaptation in prostate cancer without new driver mutations — distinct expression programs arose in different metastases from the same genetic subclone, with convergence on shared pathways (WNT, JAK-STAT, AR independence) across anatomically distinct sites (dual-regime-evolution). Transcriptional ITH can therefore be a source of adaptive variation even when genetic ITH is low.

Phenotypic heterogeneity. Cellular phenotypes — proliferative capacity, drug tolerance, EMT state, metabolic profile, differentiation status — vary across cells within a tumor. The Walens et al. (2020) cellular barcoding study in a HER2/neu breast cancer model demonstrated that ~50% of recurrent tumors arise via polyclonal recurrence with thousands of coexisting subclones, dependent on an autocrine IL-6-Jak/Stat3 signaling pathway that activates surviving residual cells without requiring a genetic selective sweep. This is phenotypic ITH as an adaptive strategy: the tumor survives therapy through coordinated phenotypic change (EMT induction, cytokine signaling) rather than through one clone outcompeting the rest. The other ~50% of recurrences in the same model used clonal dominance via Met amplification — a purely genetic ITH strategy (Walens et al., 2020; population-bottleneck).

Ecological cooperation-based ITH. Li & Thirumalai (2019) proposed a fundamentally different mechanism for ITH maintenance that does not rely on mutation, selection, or drift. Using replicator dynamics with evolving population size, they showed that stable coexistence of producer and non-producer cell subpopulations emerges from unequal allocation of diffusible paracrine growth factors (“public goods”) — a rule they call “distribution according to work.” This mechanism has three requirements: (a) nonlinear fitness functions of the producer fraction (linear fitness yields only unstable equilibria), (b) unequal sharing where producers retain a larger share of the public good than non-producers (b/a < 1), and (c) resource-limited conditions — below a critical exogenous resource concentration, cooperation and heterogeneity prevail; above it, competition eliminates diversity. The model quantitatively explained in vivo GBM (wtEGFR/ΔEGFR) experimental data without free parameters (parameters determined from three growth curves, predictions tested on two independent conditions), and fit in vitro pancreatic cancer (IGF-II producer/non-producer) data with two free parameters (a and p_0) (Li & Thirumalai, 2019; li2018-ith-mechanism). This mechanism is orthogonal to the four-mode evolutionary taxonomy (§4): it explains how two populations stably coexist once established, not how new branches arise. It makes a sharp, testable prediction that distinguishes it from neutral drift — ITH should collapse when exogenous resources exceed a critical threshold — and carries the counterintuitive therapeutic implication that nutrient supplementation (promoting competition) may be more effective than nutrient deprivation (which promotes cooperation). Hypoxic regions select for p53-loss clones; immune-rich regions select for clones with HLA LOH or PD-L1 upregulation; nutrient-poor regions select for metabolic reprogramming. Microenvironmental heterogeneity is not ITH itself, but its interaction with cellular diversity generates the spatially structured selection landscape that shapes ITH dynamics (Turajlic et al., 2019; clonal-evolution).

flowchart TD
    subgraph Sources["Sources of ITH"]
        MUT["Ongoing Mutagenesis\n(APOBEC, clock-like, therapy-induced)"]
        CIN["Chromosomal Instability\n(ongoing CNA generation)"]
        CAT["Catastrophic Events\n(chromothripsis, WGD)"]
        EPI["Epigenetic Plasticity\n(chromatin state variation)"]
        MICRO["Microenvironmental\nHeterogeneity"]
        THER["Therapy\n(bottleneck + re-diversification)"]
    end

    subgraph Types["Types of ITH"]
        GEN["Genetic ITH\nSNVs | CNAs | SVs"]
        NON["Non-Genetic ITH\nEpigenetic | Transcriptional\nPhenotypic | Microenvironmental"]
    end

    subgraph Measurement["Measurement"]
        SMF["Subclonal Mutation Fraction\nn_subclonal / n_total"]
        ENTROPY["Shannon Entropy of\nClone Frequency Dist."]
        NCLONES["Number of Detectable\nSubclones (GMM)"]
        VAF["VAF Distribution Shape\n(1/f² neutral null)"]
    end

    subgraph Significance["Clinical Significance"]
        MONO["Very Low ITH = Monoclonal\nCompleted compression\nEntrenched → worse outcome?"]
        OLIGO["Moderate ITH = Oligoclonal\nActive exploration\nVulnerable → better outcome?"]
        POLY["Very High ITH = Polyclonal\nCompression failure\nEntropic → worse outcome"]
    end

    Sources -->|"generate"| Types
    Types -->|"measured by"| Measurement
    Measurement -->|"predicts"| Significance

    MUT --> GEN
    CIN --> GEN
    CAT --> GEN
    EPI --> NON
    MICRO --> NON
    THER --> GEN
    THER -.-> NON

    MONO -.-|"Compression-Entrenchment Hypothesis"| OLIGO
    OLIGO -.-|"U-shaped ITH-outcome relationship"| POLY

Sources and types of intratumor heterogeneity, their measurement, and clinical significance. Sources of ITH span genetic (mutagenesis, CIN, catastrophic events) and non-genetic (epigenetic plasticity, microenvironmental heterogeneity, therapy) domains. They generate two broad types of heterogeneity — genetic and non-genetic — which are measured through different metrics (SMF, Shannon entropy, subclone count, VAF distribution shape). The clinical significance is predicted by the compression-entrenchment hypothesis to follow a U-shaped relationship: both very low ITH (monoclonal, entrenched) and very high ITH (polyclonal, entropic) are associated with worse outcomes, while moderate ITH (oligoclonal, active exploration) is associated with better outcomes. Synthesized from Nowell (1976), Greaves & Maley (2012), McGranahan & Swanton (2017), Turajlic et al. (2019), PCAWG Consortium (2020), Walens et al. (2020), and the compression-entrenchment hypothesis (compression-progress-evolution; docs/superpowers/specs/2026-07-05-ith-outcome-test-design.md).

2. Sources of ITH

ITH is not a static property of a tumor but a dynamic outcome of ongoing processes that generate variation and of selective and neutral forces that shape its distribution.

Ongoing mutagenesis. APOBEC-family cytidine deaminases, clock-like mutational processes (SBS1, SBS5), and therapy-induced mutagenesis (platinum-based chemotherapy, PARP inhibitors) continue to generate point mutations throughout the tumor’s lifetime (McGranahan & Swanton, 2017). APOBEC activity can persist late in tumor evolution, producing subclonal mutational diversity that fuels adaptation. The PCAWG Consortium (2020) found evidence of mutational signature shifts in ~40% of tumors, reflecting changing mutational processes over time — including late APOBEC activation, HRD emergence, and therapy-induced signatures.

Chromosomal instability (CIN). Ongoing errors in chromosome segregation produce continuous copy-number heterogeneity. CIN generates large-effect variation — entire chromosome arms gained or lost per cell division — creating diversity at a scale far larger than point mutations. Turajlic et al. (2019) found that in TRACERx Renal, metastasis-competent clones were distinguished by the degree of aneuploidy and chromosome complexity, and that moderate CIN (25–75% aneuploidy) was associated with decreased survival in pan-cancer analysis, while excessive CIN (>75%) conferred improved prognosis — suggesting a fitness cost of extreme aneuploidy that constrains ITH. Jamal-Hanjani et al. (2017) introduced mirrored subclonal allelic imbalance — when the maternal allele is gained/lost in one subclone and the paternal allele in another — as a direct readout of ongoing dynamic CIN, detected in 62% of 92 evaluable TRACERx NSCLC tumors (375 events total). This phenomenon produces parallel evolution of driver copy-number alterations (CDK4, FOXA1, BCL11A) through distinct allelic routes converging on the same genes, and is significantly enriched in genome-doubled tumors (P = 0.004) — establishing genome doubling as an early permissive event that enables subsequent CIN-driven diversification.

Catastrophic events. chromothripsis (22.3% of cancers, predominantly clonal; PCAWG Consortium, 2020), whole-genome-duplication (WGD), and chromoplexy create large-scale genomic rearrangements in single catastrophic events. These events generate instant ITH at the time of occurrence — a single cell division produces a genome radically different from its neighbors. Because they are predominantly early events, they produce clonal (not subclonal) alterations: the heterogeneity they generate is between the initial clones (the “hopeful monsters” that survive the catastrophe) rather than ongoing diversification. However, the instability they create can seed subsequent subclonal variation.

Epigenetic plasticity. Chromatin state variation — methylation drift, histone mark redistribution, chromatin accessibility changes — generates phenotypic heterogeneity without sequence change. The dual-regime-evolution framework formalizes this as a non-Darwinian evolutionary regime: epigenetic states change in response to microenvironmental signals, are inherited through mitosis, and are not subject to the Weismann barrier. Epigenetic plasticity can substitute for genetic diversity: Geng et al. (2016) demonstrated that a clonal invasive plant with near-zero genetic diversity (94% of individuals sharing a single genotype) achieved full niche occupancy through phenotypic plasticity (population-bottleneck), and Mikutenaite et al. (2025) showed that transcriptional plasticity enables metastatic adaptation in prostate cancer without new driver mutations (dual-regime-evolution).

Microenvironmental heterogeneity. Spatial variation in oxygen, nutrients, immune cells, and ECM creates consistent regional selective pressures that maintain spatial ITH. A clone that is fit in the hypoxic core may be unfit in the normoxic periphery, and vice versa. This prevents any single clone from sweeping to fixation across the entire tumor, preserving diversity through spatial niche partitioning (Turajlic et al., 2019). Single-biopsy sampling systematically undersamples this spatial ITH: a core biopsy may show a p53-loss-dominated population while the periphery shows a p53-wild-type-dominated population.

Therapy. Treatment creates a selective bottleneck that reshapes ITH. The relationship between therapy and ITH is paradoxical (the bottleneck paradox): deeper responses (CR/vgPR) can produce more diverse relapses than shallower responses (PR). In the Myeloma XI trial, patients achieving CR/vgPR underwent a clonal bottleneck leading to branching clonal architecture at relapse, while incomplete responders maintained linear or stable patterns (Miething, 2019; population-bottleneck). This is consistent with the compression-progress framework: therapy is a forced decompression event; shallow decompression leaves the dominant clone repairable; deep decompression destroys the existing compression entirely, forcing surviving clones to re-explore the fitness landscape — generating renewed ITH (compression-progress-evolution).

3. Measurement and Quantification

3.1 From Sequencing Data

The standard pipeline for measuring ITH from bulk sequencing data involves several steps.

Step 1: VAF computation. For each somatic mutation, compute the variant allele fraction — the proportion of sequencing reads supporting the variant allele:

$VAF=n_{variant}/n_{total}$

where n_variant is the number of reads carrying the variant allele and n_total is the total read depth at that locus.

Step 2: CCF correction. Convert VAF to cancer cell fraction (CCF) — the proportion of tumor cells carrying the mutation — by correcting for tumor purity and local copy number:

$CCF=\frac{VAF\times (\rho \times C{N}_t+(1-\rho )\times 2)}{\rho \times C{N}_m}$

where ρ = tumor purity, CN_t = average tumor copy number at the locus, and CN_m = mutant allele copy number (Tarabichi et al., 2021). CN_m is typically assumed to be 1 (heterozygous mutation on one copy) in the absence of SNP-based phasing, but this assumption becomes uncertain in amplified regions.

Step 3: Clonal vs. subclonal classification. Mutations are classified as clonal if CCF >= 0.85 (accounting for measurement uncertainty around the true CCF = 1.0) and subclonal if CCF < 0.85 (Turajlic et al., 2019; Tarabichi et al., 2021). This threshold is a field convention, not a biologically derived quantity — the primary signal is the distribution of CCFs, not the binary classification.

Step 4: Subclonal Mutation Fraction (SMF). The primary ITH metric:

$SMF=n_{subclonal}/n_{total}$

SMF is the most widely used single-sample ITH metric. It ranges from 0 (all mutations are clonal — monoclonal architecture) to 1 (all mutations are subclonal — maximally diverse). Intermediate values (0.20–0.60) indicate oligoclonal architecture with multiple co-dominant subclones (see docs/superpowers/specs/2026-07-05-ith-outcome-test-design.md).

SMF has important limitations. It conflates the number of subclones with the number of subclonal mutations per subclone: a tumor with one subclone bearing 5 subclonal mutations (SMF = 0.50 if 10 total) is biologically different from a tumor with 5 subclones each bearing 1 mutation (also SMF = 0.50). The framework’s prediction (see ITH as a Clinical Biomarker) is about clonal architecture — number of competing clones — not subclonal mutation count. Secondary metrics address this limitation:

• Shannon entropy of clone frequency distribution: $H=-{\sum }_i{p}_i\ln p_i$, where p_i is the fraction of tumor cells belonging to clone i (Walens et al., 2020). This metric captures both the number of clones and the evenness of their distribution.
• Number of detectable subclones: Cluster mutations by CCF using Gaussian mixture models (1–5 components, BIC-based selection) and count clusters with mean CCF < 0.85.
• VAF distribution shape: Under neutral evolution in an exponentially growing tumor, the cumulative VAF distribution follows a $1/f^2$ power law (Graham & Sottoriva, 2017, as cited in Turajlic et al., 2019). Deviations from this pattern — excess high-frequency subclonal mutations — may indicate ongoing selection.

3.2 Detection Limits

ITH measurement from bulk sequencing is constrained by fundamental detection limits.

Subclone detection floor. The minimum detectable subclone CCF scales with sequencing depth:

$CC{F}_{\text{min}}\approx \frac{3}{\sqrt{depth}}$

At standard 100x coverage, CCF_min ≈ 0.30 (a subclone comprising ~30% of tumor cells). At 500x, CCF_min ≈ 0.13. At 1000x, CCF_min ≈ 0.09 (Tarabichi et al., 2021). Subclones below this threshold are invisible in bulk data, though they may be clinically significant.

Bulk sequencing time bias. Mutations that arose fewer than ~7 doublings before sampling fall below detection at standard 100x depth. This creates a systematic blind spot: recent subclonal expansions are invisible, and neutral evolution may be overestimated because the low-frequency tail of the VAF distribution — where ongoing sweeps would appear — is censored by the detection floor (Turajlic et al., 2019).

Panel vs. exome vs. WGS. Targeted gene panels (e.g., 163-gene panels) measure ITH only in the assayed genes. A tumor classified as monoclonal by panel sequencing may be highly subclonal outside the panel. Exome sequencing (~1% of the genome) offers broader coverage but still undersamples non-coding regions. Whole-genome sequencing (WGS) provides the most complete ITH measurement but at higher cost and computational burden. The compression-entrenchment test design (docs/superpowers/specs/2026-07-05-ith-outcome-test-design.md) acknowledges that panel-limited ITH measurement biases against finding the predicted U-shaped relationship — a positive result survives this bias; a null result could reflect measurement noise rather than absence of the phenomenon.

Multi-region sequencing. The gold standard for ITH measurement is multi-region sequencing (e.g., TRACERx protocol), which captures spatial ITH by sampling multiple distinct regions of the same tumor. Single-region sequencing systematically underestimates ITH: a core biopsy may contain only one subclone while the periphery contains several others. In the first 100 TRACERx NSCLC patients (Jamal-Hanjani et al., 2017), multiregion sequencing of 327 tumor regions revealed a median of 30% of somatic mutations were subclonal (range 0.5–93%) and a median of 48% of copy-number alterations were subclonal (range 0.3–88%). Without multiregion sequencing, 76% of subclonal mutations would erroneously appear clonal, and 65% of branched subclone clusters would appear clonal — single-biopsy analysis systematically misclassifies ITH. Multiregion data enable application of the crossing-rule: if clone A has higher CCF than clone B in one region but lower in another, they are sibling clones in a branching phylogeny, not an ancestor-descendant pair (Tarabichi et al., 2021). However, multi-region sequencing is rarely available in clinical cohorts, and the compression-entrenchment test design is designed for single-biopsy data with all appropriate caveats.

ctDNA-based ITH measurement. Circulating tumor DNA offers a complementary approach that partially addresses the spatial sampling problem. Because ctDNA is shed from multiple tumor regions into the bloodstream, it captures systemic clonal heterogeneity: subclones present in one spatial region but absent in another may both be represented in plasma. Abbosh et al. (2017) demonstrated phylogenetic ctDNA profiling in the first 100 TRACERx patients: tumor-specific phylogenetic trees enabled bespoke ctDNA panels that detected relapse with 93% sensitivity and a median 70-day lead time over imaging. Subclone volume was the primary determinant of ctDNA detection — larger subclones shed more DNA — meaning ctDNA-based ITH estimates are biased toward larger clones but capture spatial diversity that single biopsies miss. Stejskal et al. (2023) reviewed the biological basis: ctDNA fragments (~150 bp, reflecting nucleosomal protection during apoptosis) have a short half-life (minutes to 2 hours), providing a real-time snapshot of clonal composition rather than the historical record captured by FFPE tissue. For the ITH empirical test design, ctDNA-based SMF estimation introduces sample-type-specific confounders: detection sensitivity varies by cancer type (<50% in brain, renal, prostate, thyroid cancers), clonal hematopoiesis variants contaminate plasma without matched normal filtering, and ctDNA fraction can be <1% of total cfDNA in early-stage disease. See abbosh2017-ctdna-tracerx and stejskal2023-ctdna-biology-review.

3.3 Confounders

ITH measurement is confounded by multiple factors that must be accounted for:

• Tumor mutation burden (TMB). TMB affects SMF measurement: more total mutations means more opportunities to detect subclonal mutations, biasing SMF upward in high-TMB tumors. SMF-outcome analyses must include TMB as a confounder to ensure SMF is not a TMB proxy (docs/superpowers/specs/2026-07-05-ith-outcome-test-design.md §4.7).
• Sample purity. Low-purity samples reduce the dynamic range of VAF measurements: a mutation present in 30% of tumor cells at 50% purity has VAF ~0.15, while the same mutation at 20% purity has VAF ~0.06 — potentially falling below detection. CCF correction propagates this uncertainty.
• Copy number. Amplified regions make CCF correction ambiguous: if CN_t > 2 and CN_m is unknown (all copies may carry the mutation, or only one), the same VAF is consistent with a range of CCF values. The standard assumption CN_m = 1 provides a lower bound on CCF.
• Sequencing depth. Lower depth leads to a higher detection floor, fewer detectable subclonal mutations, and therefore lower SMF. Depth must be included as a covariate in SMF-outcome analyses.

4. ITH and Evolutionary Modes

Turajlic et al. (2019) formalized a taxonomy of four evolutionary modes, each associated with a characteristic ITH profile. These modes are not mutually exclusive — a single tumor may transition between modes at different stages of its evolution — and all arise from different combinations of the same fundamental processes (mutation, drift, selection).

Mode	Typical ITH	Mechanism	Clinical association
Linear	Low (SMF < 0.20)	Sequential selective sweeps purge diversity. Each new driver expands to fixation, eliminating competing lineages.	Nowell’s (1976) original model. Dominant in early tumors (N ~ 10^3–10^5). May reflect strong selection coefficients.
Branching	Moderate (SMF 0.20–0.60)	Multiple subclones coexist, each bearing distinct private mutations, with no single clone achieving fixation. Clonal interference — competition between adaptive lineages — is the hallmark.	Most common pattern in solid tumors by multi-region sequencing. Arises when driver advantages are modest (~0.4%; Greaves & Maley, 2012) and mutation rates sustain continuous new variants before any single sweep completes.
Neutral	High (SMF > 0.60)	No strong selection differentials. Mutation and drift determine clone frequencies. The VAF distribution follows the characteristic 1/f^2 power law.	May dominate between selection events (Turajlic et al., 2019). Bulk sequencing can overestimate neutral prevalence because recent sweeps fall below detection (~7 doublings at 100x).
Punctuated	Low subclonal ITH despite high clonal aneuploidy	A single early catastrophic event (chromothripsis, WGD) generates large-scale genomic change, followed by relative stasis. High clonal aneuploidy, low subclonal diversity.	Associated with aggressive clinical behavior: fast growth, widespread metastasis, monophyletic metastatic seeding (Turajlic et al., 2019). PCAWG Consortium (2020): chromothripsis in 22.3% of cancers, predominantly clonal.

The relationship between mode and ITH is not deterministic. A branching tumor with one dominant subclone may have SMF similar to a linear tumor nearing a sweep — the difference lies in the shape of the clone frequency distribution (whether multiple clones are present at moderate frequencies vs. one dominant clone with many tiny ones), not the binary clonal/subclonal split. This is why the compression-entrenchment test design uses both continuous SMF and subclone count as ITH metrics, and emphasizes the shape of the SMF-outcome relationship (spline-based U-consistency test) over categorical classification (docs/superpowers/specs/2026-07-05-ith-outcome-test-design.md §4.5).

5. ITH as a Clinical Biomarker

5.1 The Compression-Entrenchment Prediction

The compression-entrenchment hypothesis, derived from the compression-evolution isomorphism (compression-progress-evolution), predicts a non-monotonic U-shaped relationship between ITH and clinical outcome:

• Very low ITH (monoclonal, SMF < 0.20): The tumor represents a completed compression — one clone’s genomic program encodes the microenvironment more efficiently than any competitor. The dominant clone is entrenched at a fitness peak. Standard therapy cannot easily destabilize it because small perturbations do not generate sufficient compression progress — the clone is well-adapted to its current environment and resistant to displacement. Predicted outcome: worse survival than oligoclonal tumors.

• Moderate ITH (oligoclonal, SMF 0.20–0.60): No single clone has achieved complete compression. Multiple clones are actively competing, exploring the fitness landscape. The tumor remains in a transitional, vulnerable state — it is “interesting” in Schmidhuber’s (2009) sense (steep learning curve, high compression progress potential). Therapeutic perturbation can exploit this vulnerability. Predicted outcome: better survival than monoclonal or polyclonal tumors.

• Very high ITH (polyclonal, SMF > 0.60): Compression failure. No clone has found an adequate compression of the microenvironment. The tumor is entropic, generating vast diversity that provides substrate for adaptation to any selective pressure. Predicted outcome: worse survival (consistent with standard population-genetic theory).

This U-shaped prediction discriminates between the compression-entrenchment hypothesis and standard population-genetic theory. Standard theory predicts a monotonic relationship: more genetic diversity leads to more adaptation substrate, which leads to worse outcome. Both theories agree on the polyclonal end (high ITH leads to worse outcome). They disagree on the monoclonal-to-oligoclonal range: standard theory has no reason to predict that monoclonal tumors are systematically worse than oligoclonal ones, while the compression-entrenchment hypothesis predicts exactly this.

Formal hypothesis (from docs/superpowers/specs/2026-07-05-ith-outcome-test-design.md):

$\log h(t)=\log h_0(t)+{\beta }_1\cdot \text{SMF}+{\beta }_2\cdot {\text{SMF}}^2+\gamma \cdot X$

where X is a vector of confounders. The primary test is H_0: β_2 ≤ 0 vs. H_1: β_2 > 0. If β_2 > 0 (significant), the relationship is U-shaped — supporting the compression-entrenchment hypothesis. If β_2 ≤ 0, the prediction is falsified.

Immunotherapy caveat. The prediction applies to tumors treated with standard chemotherapy, targeted therapy, or surgery. In immunotherapy, high ITH leads to increased neoantigen burden, which leads to better response — a well-established mechanism that operates in the opposite direction. Immunotherapy-treated patients are excluded from the primary analysis and analyzed separately as an exploratory cohort (docs/superpowers/specs/2026-07-05-ith-outcome-test-design.md §2.3).

5.2 Existing Evidence

Miething (2019) — bottleneck paradox. In the Myeloma XI trial (N = 56 diagnosis/relapse pairs), patients achieving CR/vgPR underwent a clonal bottleneck followed by branching relapse (high ITH), while patients with partial responses showed linear evolution (low ITH). This is consistent with the compression-entrenchment framework: deep therapy forced decompression led to renewed exploration, which led to high ITH at relapse. However, this describes post-treatment ITH dynamics, not the pre-treatment ITH-outcome relationship that the compression-entrenchment hypothesis primarily addresses (CN-017 distinguishes these).

Walens et al. (2020) — dual-route recurrence. Cellular barcoding in a HER2/neu breast cancer model revealed two recurrent routes: ~50% clonal dominance (Met amplification, low ITH, genetic compression breakthrough) and ~50% polyclonal recurrence (Jak/Stat pathway, high ITH, plasticity-driven). The clonal dominance route is the compression-entrenchment pattern: a clone achieves a superior compression (Met amplification) and sweeps to dominance, producing a monoclonal recurrent tumor. The polyclonal route is the compression-failure pattern: many clones survive without any achieving dominance.

Turajlic et al. (2019) — punctuated evolution and clinical phenotype. Tumors with early clonal aneuploidy (punctuated evolution, low subclonal ITH despite high clonal aneuploidy) grow fast, metastasize widely, and seed metastases monophyletically. This is consistent with the entrenchment pattern: the early catastrophic event produces a genome that is well-adapted across diverse microenvironments, and the resulting clone is difficult to destabilize.

Jamal-Hanjani et al. (2017) — CNA heterogeneity, not mutation heterogeneity, predicts survival. In the first 100 TRACERx NSCLC patients (327 tumor regions), high subclonal copy-number alteration proportion (≥48%, the cohort median) was associated with significantly worse relapse-free survival: HR 4.9 (95% CI 1.8–13.1, P = 4.4×10⁻⁴). This remained significant in multivariate analysis (HR 3.70, P = 0.01). By contrast, subclonal mutation proportion showed no association with survival (HR 0.86, P = 0.70). A static measure of chromosome disruption (mean aberrant genome fraction) was also not prognostic — it is the dynamic ongoing chromosomal instability, not the accumulated genomic state, that drives poor outcome. This finding has two implications for the compression-entrenchment framework: (a) CNA-level ITH and mutation-level ITH are not interchangeable — they have different clinical significance and should not be conflated in ITH-outcome analyses; (b) the U-shaped prediction is about clonal architecture (number and structure of competing clones), not mutation count heterogeneity, and CNA heterogeneity may be a more direct readout of clonal architecture than mutation-based SMF.

TRACERx — subclonal diversity predicts progression. Multi-region sequencing in NSCLC within the TRACERx study has shown that higher subclonal diversity predicts disease progression (Al Bakir et al., 2023). This is consistent with the high-ITH arm of the U-curve but does not address the monoclonal-to-oligoclonal range where the compression-entrenchment prediction deviates from standard theory.

Weng et al. (2026) — recurrent ITH is a regulated hallmark. In 34 metastatic lesions from 9 mCRPC patients, Weng et al. (2026) used snRNA-seq + snATAC-seq + WGS to show that ITH is not random — metastases converge on predictable proportions of six transcriptional archetype modules, regardless of clonal background, organ site, or local microenvironment. Three findings are directly relevant to the wiki’s framework: (a) transcriptional ITH develops predominantly independently of CN subclones (1.6–28.2% variance explained by genetics), providing further evidence for the dual-regime model (dual-regime-evolution); (b) the recurrent, predictable nature of ITH validates the compression-entrenchment hypothesis’s assumption that ITH reflects a stable tumor property — if ITH were purely stochastic, the U-shaped outcome prediction would be untestable; (c) therapeutic target expression (PSMA, STEAP1/2, B7-H3) is linked to transcriptional ITH, not genetic changes — meaning ITH directly modulates drug target availability independently of clonal architecture. See weng2026-ith-prostate-cancer for the full source summary.

5.3 The Empirical Test

The ITH-outcome test design (docs/superpowers/specs/2026-07-05-ith-outcome-test-design.md) is a pre-registered statistical protocol for testing the U-shaped prediction. It specifies:

• Primary endpoint: Overall survival (OS). Secondary: progression-free survival (PFS).
• Primary analysis: Cox model with SMF + SMF^2 + confounders (stage, age, sex, ECOG, purity, depth, TMB, cancer type, sample type).
• Primary test: β_2 > 0 (one-sided, α = 0.05 — the directional prediction justifies one-sided testing).
• Shape test (three-step): (1) M1 (quadratic) must beat M0 (linear) by LRT; (2) M2 (5-knot spline) must not beat M1 — or if it does, must still be U-shaped (both arms significant); (3) from M1, compute hazard ratio between SMF=0 and the fitted minimum, with bootstrap CI for minimum location.
• Treatment interaction: The U-curve should be steeper under systemic therapy than surgery-only — entrenchment matters most when therapy is applied.
• Positive controls: Higher stage predicts worse OS; older age predicts worse OS. If these fail, data are unreliable.
• Negative control: In R0-resected patients (tumor removed), SMF must not predict survival (the evolutionary contest is over).
• Sensitivity analyses: Purity tertiles, CCF threshold ladder, SMF threshold ladder, bootstrap SMF, mutation count ladder, landmark analyses, dfbeta outlier exclusion.

The design specifies eight possible outcomes in an interpretation matrix (strong success, weak success, partial, null, data failure, confounded, fragile) with pre-registered thresholds for each.

6. ITH in the Olog

In the cancer-evolution-olog, ITH is formalized as the object IntratumorHeterogeneity (Turajlic et al., 2019; McGranahan & Swanton, 2017):

IntratumorHeterogeneity — The set of diversity measures over TumorCellPopulation. An element ι ∈ IntratumorHeterogeneity is a scalar (Shannon diversity index, number of subclones, VAF distribution width, SMF) capturing the degree of clonal diversity. Examples: Shannon index H = 0 (monoclonal); H = 2.3 (highly polyclonal); number of subclones detected = 7.

Key arrows involving ITH:

Arrow	Domain leads to Codomain	Meaning
diversity	TumorCellPopulation leads to IntratumorHeterogeneity	Computes a diversity measure from clonal composition
hasITH	TumorCellPopulation leads to IntratumorHeterogeneity	Alternative name for the diversity arrow
constrains	IntratumorHeterogeneity leads to Relapse	Low ITH (entrenchment) constrains the tumor’s ability to adapt — paradoxically worsens outcome under standard therapy?
enables	IntratumorHeterogeneity leads to AdaptationSubstrate	High ITH provides substrate for adaptation (consistent across both compression and standard frameworks)

Cross-domain functor mappings (from cross-domain-functors):

• Functor F: EcologyOlog leads to CancerOlog maps GeneticDiversity (ecology) to IntratumorHeterogeneity (cancer). The commutativity condition C4 (Niche Occupancy Independence) asserts that low ITH does NOT predict low relapse risk — the diversity-only path gives the wrong answer, just as the ecology condition asserts that low genetic diversity does NOT predict low niche occupancy. This is the formal encoding of the compression-entrenchment prediction in category-theoretic terms (cross-domain-functors §F-C4).

• Functor G: CompressionOlog leads to CancerOlog maps CompressorDiversity (multiple candidate compressions) to IntratumorHeterogeneity (cancer). Under G, the compression-entrenchment gradient — very low, moderate, very high ITH — maps onto the compression states of successful compression, active exploration, and compression failure, respectively (compression-progress-evolution §Intratumor heterogeneity as compression fragmentation).

The ITH-plasticity trade-off (from cross-domain-functors §F-C1). The functor F’s central commutativity condition states that adaptive success after a bottleneck is determined by plasticity, not by residual genetic diversity. This predicts a trade-off: a monoclonal tumor with high epigenetic plasticity may be as adaptable as a polyclonal tumor with high genetic diversity. The empirical implication is that measuring only genetic ITH may systematically underestimate the adaptive capacity of tumors — a challenge for ITH-based prognostication that the compression-entrenchment test design partially addresses through its treatment-interaction analysis.

Commutativity condition CC2 (Bottleneck-Diversity) from the cancer olog. The post-bottleneck subclonal architecture obtained by “apply therapy to surviving cells and reconstruct architecture” must equal the architecture obtained by “compute pre-treatment architecture and filter through bottleneck.” The bottleneck function on populations must commute with the architecture function. The empirical anchor: Myeloma XI data (Miething, 2019) confirms that diagnosis-clonal relationships are preserved at relapse — the bottleneck prunes but does not reorder the phylogeny. Walens et al. (2020) confirms that recurrent clones are present in the pre-treatment population (cancer-evolution-olog §3.2).

7. Limitations

Single-biopsy captures one spatial sample. A core needle biopsy samples a tiny fraction of a tumor’s volume. ITH measured from a single biopsy can differ dramatically from the full spatial ITH, especially in tumors with strong spatial niche partitioning (hypoxic core vs. well-vascularized periphery). Multi-region studies (e.g., TRACERx) show that single-biopsy ITH systematically underestimates true diversity and can misclassify a spatially structured branching tumor as monoclonal. This biases against finding the predicted U-shaped ITH-outcome relationship — a positive result survives this bias, but a null result cannot distinguish “no effect” from “measurement too noisy” (Turajlic et al., 2019; docs/superpowers/specs/2026-07-05-ith-outcome-test-design.md §9.4).

Panel-based ITH measures only assayed genes. Most large clinical cohorts use targeted gene panels covering 163–500 genes. ITH measured in these regions may not reflect genome-wide ITH. A tumor classified as monoclonal by panel could be highly subclonal in non-coding regions, regulatory elements, or genes not on the panel. This also biases against finding the predicted effect — a challenge the test design acknowledges as a limitation that favors accepting the null (conservative bias; docs/superpowers/specs/2026-07-05-ith-outcome-test-design.md §9.2).

Detection floor masks low-frequency subclones. At standard sequencing depths, subclones below ~0.10–0.15 CCF are invisible. These subclones may be clinically significant — they can harbor pre-existing resistance mutations that expand under therapy. The detection floor also creates a systematic bias in SMF: low-depth samples have fewer detectable subclonal mutations and therefore lower SMF, independent of true ITH. Depth must be included as a covariate (Tarabichi et al., 2021; Turajlic et al., 2019).

ITH is dynamic — a single time point is insufficient. ITH changes over time as the tumor evolves. A tumor that is monoclonal at diagnosis may be highly polyclonal at relapse, and vice versa. The compression-entrenchment test design uses pre-treatment SMF as the predictor, but a single pre-treatment measurement may not capture the evolutionary trajectory. Repeated sampling over time — through therapy, progression, and metastasis — would provide a more complete picture but is logistically prohibitive at scale (McGranahan & Swanton, 2017).

Distinguishing clonal from subclonal requires purity plus CN correction. The CCF = VAF x (purity x CN_t + (1-purity) x 2) / (purity x CN_m) formula depends on accurate purity estimation and copy-number calls. Purity is typically estimated from sequencing data (e.g., ABSOLUTE, ASCAT) with considerable uncertainty. Copy-number calls in subclonal regions are especially error-prone. These errors propagate into CCF and SMF, adding measurement noise that biases against finding the predicted relationship (Tarabichi et al., 2021).

ITH measurement is confounded by TMB, purity, sequencing depth, and sample type. The compression-entrenchment test design addresses these through pre-specified confounder inclusion and sensitivity analyses, but residual confounding cannot be ruled out in observational data. The negative control (R0-resected patients) provides the strongest guard against SMF being a proxy for general tumor biology rather than capturing evolutionary dynamics (docs/superpowers/specs/2026-07-05-ith-outcome-test-design.md §5.2).

Cannot distinguish mechanism from the U-shape alone. Even if a U-shaped ITH-outcome relationship is confirmed, it does not uniquely establish the compression-entrenchment mechanism. Alternative explanations include: immune coldness of monoclonal tumors (low neoantigen diversity leads to poor immunotherapy response, but the immunotherapy exclusion partially addresses this); catastrophic-sweep biology (early chromothripsis/WGD producing both low ITH and aggressive biology); or a general non-linearity in the relationship between diversity and adaptive capacity that has nothing to do with compression. Mechanistic discrimination requires orthogonal evidence: immune infiltration data, longitudinal sampling, experimental perturbation, or matched epigenetic measurements (docs/superpowers/specs/2026-07-05-ith-outcome-test-design.md §9.1).

The SMF thresholds are field conventions, not derived quantities. The CCF threshold of 0.85 for clonal/subclonal classification and the SMF categorical cutoffs (0.20, 0.60) are drawn from field practice, not derived from the compression framework. The continuous SMF analysis (primary) avoids thresholds entirely, and the sensitivity ladder tests threshold robustness, but the dependence on conventions is a genuine limitation (docs/superpowers/specs/2026-07-05-ith-outcome-test-design.md §9.9).

Monoclonal classification is biologically heterogeneous. A tumor classified as monoclonal may be: (a) a recently post-sweep tumor that eliminated competing clones (entrenched, consistent with the framework), (b) an indolent tumor that never developed subclonal diversity (not entrenched — simply never explored genotype space), or (c) a tumor where the biopsy missed a spatially separated subclone (sampling artifact). These have opposite prognoses. Single-biopsy, single-timepoint data cannot distinguish them (docs/superpowers/specs/2026-07-05-ith-outcome-test-design.md §9.7).

Revision history

• 2026-07-28 — Added ecological cooperation-based ITH mechanism. Li & Thirumalai (2019) public goods game with unequal allocation: stable producer-nonproducer coexistence via “distribution according to work.” Orthogonal to mutation-driven ITH — explains maintenance, not origin. Testable prediction: ITH collapses above critical resource threshold. (li2018-ith-mechanism)
• 2026-07-16 — Added TRACERx primary cohort data. Integrated Jamal-Hanjani et al. (2017) — the first 100 TRACERx NSCLC patients (327 regions). Added quantitative ITH benchmarks (median 30% subclonal mutations, 48% subclonal CNAs), the key prognostic finding (CNA heterogeneity HR 4.9, mutation heterogeneity HR 0.86 P=0.70), mirrored subclonal allelic imbalance (62% of tumors, 375 events), and the dynamic vs. static CIN distinction. Updated multi-region sequencing section with TRACERx underestimation data (76% of subclonal mutations misclassified without multiregion sequencing). Added to sources and related frontmatter. (jamal-hanjani2017-trackingEvolution_nsclc)
• 2026-07-05 — Major revision. Expanded from skeleton (3 sources, ~300 words) to comprehensive concept page (8 sources, ~5,500 words). Core additions: (1) genetic and non-genetic component analysis across all alteration types; (2) complete CCF-to-SMF measurement pipeline with formulas and detection limits; (3) evolutionary mode table mapping Turajlic’s four modes to characteristic ITH levels; (4) compression-entrenchment hypothesis with formal statistical specification from the pre-registered test design; (5) formalization in the cancer evolution olog with cross-domain functor mappings; (6) Mermaid diagram synthesizing sources, types, measurement, and clinical prediction; (7) nine-item limitation section covering spatial sampling, panel bias, detection floor, dynamic nature, purity/CN correction, confounding, mechanistic non-identifiability, threshold conventions, and biological heterogeneity of monoclonal classification. This page is the definitive ITH reference for 14+ wiki pages linking via intratumor-heterogeneity.