Clonal Evolution

Therapy Resistance

Properties1
tagsclinical

12 min read

Therapy Resistance

Summary

Therapy resistance is the emergence of cancer cell populations that survive and proliferate despite treatment — the direct clinical consequence of clonal evolution under therapeutic selective pressure (Nowell 1976; Greaves & Maley 2012). Resistant clones may pre-exist as minor subpopulations before treatment or arise de novo through mutation during therapy (Bozic et al. 2013; Turajlic et al. 2019). Resistance is the primary cause of cancer mortality: metastatic disease is rarely curable because the tumor’s genetic heterogeneity ensures that some subclone survives any single-agent or combination therapy (McGranahan & Swanton 2017). Clinical monitoring via ctDNA can detect emerging resistance clones before radiographic progression (Wander et al. 2026).

Definition

Therapy resistance is the emergence of cancer cell populations that survive and proliferate despite treatment, driven by the selective pressure that therapy exerts on genetically heterogeneous tumor cell populations. It is the direct clinical consequence of clonal-evolution under therapeutic positive-selection.

Pre-existing vs. De Novo Resistance

Resistance can arise through two evolutionary routes (Turajlic et al., 2019):

Pre-existing resistance. Resistant mutations are present as minor subclones before treatment begins. Under the selective pressure of therapy, the sensitive population diminishes and the resistant clone expands — a clonal-sweep. Modeling suggests that detectable metastatic lesions can harbor ten or more resistant subclones (Bozic & Nowak, 2014, cited in Turajlic et al., 2019).

De novo resistance. A resistance mutation arises during treatment from the surviving cell population and expands. This takes longer to emerge and produces a monoclonal resistance pattern.

Both patterns are documented. In chronic lymphocytic leukemia treated with ibrutinib, resistance mutations in BTK and/or PLCG2 were detected up to 15 months before clinical progression (Ahn et al., 2017, cited in Turajlic et al., 2019). In other cases, polyclonal resistance with parallel expansion of distinct resistance mechanisms occurs, as in EGFR-mutant NSCLC treated with EGFR TKIs.

ctDNA Monitoring of Therapy Resistance

CDK4/6 inhibitor resistance in breast cancer. Wander et al. (2026) analyzed the GuardantINFORM real-world ctDNA database (1,473 pre-treatment, 4,944 post-treatment, 207 paired samples) to characterize CDK4/6 inhibitor resistance in HR+ breast cancer. Post-therapy samples showed significant enrichment of ESR1 mutations (16% → 32%) and RB1 mutations (3% → 7%), with ESR1 prevalence increasing monotonically with time on therapy (31% at <12 months → 59% at ≥48 months). A composite resistance panel including ESR1, RB1, PIK3CA, and TP53 was a significant predictor of overall survival (HR = 1.74). Paired pre/post samples showed polyclonal resistance in 20% of patients (ESR1), confirming that therapy creates selective bottlenecks favoring pre-existing resistant subclones. See wander2026-ctdna-cdk46-breast-cancer.

ctDNA as a real-time clonal monitor. Khatami & Tavangar (2018) reviewed ctDNA’s role in personalized cancer therapy, framing it as a “screenshot” of systemic clonal heterogeneity compared to tissue biopsy’s localized “snapshot.” The short half-life of ctDNA (minutes to hours) enables real-time tracking of resistance-conferring subclones under selective pressure, monitoring of minimal residual disease, and detection of drug-resistant minor clones months before clinical progression. Combined with the Stejskal et al. (2023) review of ctDNA biology, these sources establish ctDNA as a clinically validated modality for tracking clonal evolution during therapy. See khatami2018-ctdna-personalized-medicine and stejskal2023-ctdna-biology-review.

Mathematical Architecture

The Bozic et al. (2013) model rests on a continuous-time multitype branching-process-model. The mathematical structure connecting raw parameters to clinical predictions involves several intermediate quantities.

The compound mutation rate μ. The effective mutation rate for pre-existing resistance is not the raw point mutation rate u ≈ 10⁻⁹ but:

μ = u × log(Ms) / s

where M is the detection size (number of cells) and s = 1 − d/b is the survival probability of a single mutant lineage (~0.07 for typical parameters). This ~250× amplification occurs because: (i) log(Ms) counts the total number of cell divisions during growth from 1 to M cells — each an opportunity for mutation; (ii) 1/s accounts for the amplification of surviving clones by branching process dynamics. For M = 10⁹ and typical parameters, μ ≈ 2.5 × 10⁻⁷.

Distinguishing cells from lineages. The compound μ appears in formulas for the expected number of resistant cells (X ≈ M n₁₂ μ). The raw u appears in formulas for the probability that zero resistant lineages were founded (p₁↑ = exp(−M u n₁₂)). These count different things: X counts cells, weighted by clonal amplification; p₁ counts founder events, unweighted. The wiki uses both — understanding the distinction prevents conflation.

The extinction filter. A single resistant mutant cell has probability d/b ≈ 0.93 of going extinct before establishing a detectable lineage (for typical parameters). Even when the expected number of resistant cells is 1, the probability that zero survive is exp(−E[lineages]) ≈ exp(−M u n₁₂ s), which is substantially higher than naive Poisson (exp(−1) ≈ 0.37) would suggest. The branching process introduces a ~7% survival filter that the qualitative “lost by genetic drift” description masks — every single mutant faces a 93% chance of immediate extinction regardless of its selective advantage.

Why the four probabilities multiply. The cure probability is p_erad = p₁↑ p₁↓ p₂↑ p₂↓. These four terms correspond to independent stochastic events: (↑) pre-treatment vs (↓) during-treatment, (1) one-step vs (2) two-step resistance. They are approximately independent because they involve different cell populations at different times — pre-treatment sensitive cells, during-treatment sensitive cells (with altered birth/death rates), and intermediate resistant types (10, 01). The independence is not exact (the populations share ancestry) but is an excellent approximation.

The Cross-Resistance Barrier

Bozic et al. (2013) formalized combination therapy resistance in a continuous-time multitype branching process model. The critical determinant of dual therapy success is whether any single point mutation can confer resistance to both drugs simultaneously. Let n1 and n2 be the number of point mutations that confer resistance to drug 1 and drug 2 individually, and n12 the number that confer cross-resistance to both.

When cross-resistance exists (n12 ≥ 1): The expected number of dual-resistant cells at treatment start is X ≈ M × n12 × μ, where M is the lesion size and μ = u × log(Ms)/s (u ≈ 10^−9 per base pair per division, s = 1 − d/b the survival probability). Critically, X is independent of n1 and n2 — a single cross-resistance mutation dominates because it requires only one mutational step. For a lesion of 10^9 cells with n12 = 1, X ≈ 10^9 × 10^−9 ≈ 1 cell expected (orders of magnitude vary by parameter values).

Even with a single expected resistant cell, the probability that resistance is present at detection is substantial. In the clinical cohort of 22 patients (pancreatic, colorectal, melanoma; total burden 8.5 × 10^8 to 2.6 × 10^11 cells), none were predicted to be cured by dual therapy when n12 ≥ 1. Stochastic extinction can still occur: small numbers of resistant cells may be lost by genetic drift during treatment, enabling ~26% cure probability even with cross-resistance for typical lesions.

When no cross-resistance exists (n12 = 0): X ≈ M × n1 × n2 × μ² — resistance requires two independent mutations. Since μ ≈ 10^−9, μ² ≈ 10^−18, making dual-resistant cells orders of magnitude rarer. For the same 22-patient cohort with n12 = 0, eight patients with smallest tumor burden had >95% predicted cure probability. Even patients with the largest burden had >20% recurrence risk, and for tumors with rapid cell turnover (1-day interdivision time), failure risk rose to 37%.

Triple therapy and beyond. For k drugs with no cross-resistance, X ≈ M × n1 × n2 × … × nk × μ^k — resistance becomes exponentially rarer as k increases. However, if n123 ≥ 1 (a mutation conferring resistance to all three drugs), triple therapy also fails. This generalizes: cross-resistance to all k drugs negates the benefit of adding more agents.

Simultaneous vs Sequential Therapy

Bozic et al. (2013) delivered a stark mathematical proof:

ScenarioSequential therapySimultaneous therapy
n12 ≥ 1 (cross-resistance)Fails in 100% of lesionsCures ~26% of lesions
n12 = 0 (no cross-resistance)Fails in 100% of lesionsCures >99% of lesions

Sequential therapy fails even without cross-resistance because treatment with drug 1 alone allows the tumor population to generate and expand cells resistant to drug 2 before drug 2 is ever administered. With cross-resistance, ~74% of sequential failures are due to dual-resistant cells that pre-existed before any treatment; ~26% arise during the first drug’s administration window.

The clinical implication is direct: sequential administration precludes any chance for cure. Simultaneous administration should be the default for combination targeted therapy, and drugs should be developed as combinations from the outset rather than added sequentially after resistance to the first agent emerges.

Role of Cancer Stem Cells

The effective population size for resistance depends on the cancer stem cell fraction. Bozic et al. (2013) noted that if cancer stem cells represent only 0.1% of tumor cells (as in CML), resistance is ~0.1% as likely — explaining imatinib’s remarkable success in chronic-phase CML. In solid tumors, stem cell fractions are typically >5% and sometimes near 100%, making resistance far more probable. This explains why monotherapy for solid tumors almost invariably fails while CML can be controlled for years with a single agent: the target population for resistance mutations is vastly smaller in CML.

Resistance in the Vemurafenib Era

Bozic et al. (2013) analyzed 68 index lesions from 20 melanoma patients receiving the BRAF inhibitor vemurafenib. Responses ranged from complete remission to stable disease to mixed partial responses. The net growth rate of untreated lesions was 0.01/day; median tumor decline during treatment was −0.03/day (10th–90th percentile: −0.01 to −0.07/day). Smallest lesions were most likely to become undetectable. The data confirmed the model’s core prediction: monotherapy always fails in detectable lesions because resistant cells are pre-existing.

Fitness Cost of Resistance

Resistance often carries a fitness cost. KRAS mutations conferring resistance to EGFR inhibition in colorectal cancer were detectable in cell-free DNA during treatment but became undetectable upon treatment withdrawal — “they require ongoing therapy for their maintenance and that resistance comes at a cost” (Turajlic et al., 2019, p. 414). This fitness cost creates therapeutic opportunities: intermittent dosing schedules can exploit the fitness disadvantage of resistant clones.

Bozic et al. (2013) quantified the limits of this principle. When cross-resistance exists (n12 ≥ 1), a 10% fitness cost per resistance mutation only marginally improves cure probability — the pre-existing resistant population still dominates. When n12 = 0 with large lesions and high cell turnover, costly resistance becomes meaningful: eradication probability rises from 47% to 68% for a 10^11-cell lesion with 1-day turnover. Costly resistance is most beneficial precisely when other factors (no cross-resistance, moderate tumor burden) already favor treatment success.

Nowell’s Prescience

Nowell (1976) anticipated the centrality of therapy resistance to clonal evolution: “With variants being continually produced, and even increasing in frequency with tumor progression, the neoplasm possesses a marked capacity for generating mutant sublines, resistant to whatever therapeutic modality the physician introduces” (p. 27). This insight — that the same evolutionary process that creates the tumor also undermines its treatment — remains the central challenge of cancer therapy.

Clinical Benefit Without Clonal Selection

Not all therapeutic effects operate through clonal selection. Jones et al. (2019) demonstrated a striking dissociation in the Myeloma XI trial (jones2019-clonal-evolution-myeloma): lenalidomide maintenance significantly improved progression-free survival (a clear clinical benefit), yet had no detectable impact on clonal evolution patterns at relapse. The mutational profile, copy number changes, and evolutionary patterns (branching vs. linear vs. stable) were indistinguishable between the lenalidomide and observation arms. No enrichment of CRBN-pathway resistance mutations was observed — only one patient developed a CRBN mutation, likely stochastic.

This dissociation has important implications:

  • Clinical benefit ≠ clonal selection pressure. Lenalidomide’s mechanism (immune modulation, cereblon-mediated degradation of IKZF1/3) may suppress tumor growth without exerting strong direct selective pressure on clonal composition. The clinical benefit operates through an immune-mediated or microenvironmental mechanism, not through eliminating specific clones.

  • Bottleneck severity, not drug target, determines evolutionary outcome. The depth of response (CR vs. PR) was the dominant determinant of clonal architecture at relapse — not which maintenance drug was used. This suggests that the evolutionary trajectory is shaped more by the severity of the population reduction than by the specific selective pressure applied.

  • Relevance to adaptive therapy. If some therapies suppress tumor growth without selecting for resistance, they may be suitable candidates for long-term maintenance — providing clinical benefit without driving the evolutionary escalation that produces increasingly resistant clones. This hypothesis requires prospective testing: Jones et al. (2019) demonstrated the pattern in one drug (lenalidomide) in one cancer type (myeloma); whether the dissociation between clinical benefit and clonal selection pressure generalizes to other immunomodulatory agents remains unknown.

This finding complements the pre-existing vs. de novo resistance framework: lenalidomide maintenance may represent a third category — therapy that provides clinical benefit without resistance evolution, because it operates through mechanisms that do not create strong clonal selection gradients (Jones et al., 2019).

Resistance to Immunotherapy

Immune checkpoint blockade is also vulnerable to evolutionary escape. Resistance mechanisms include:

  • Loss of clonal neo-antigens through deletion of the encoding chromosomal region
  • Outgrowth of alternative subclones lacking subclonal neo-antigens (immune editing)
  • HLA loss of heterozygosity preventing antigen presentation
  • Mutations in JAK1, JAK2, and B2M disrupting interferon signaling and antigen presentation (Turajlic et al., 2019)

Graph View

Backlinks