Bibliographic Reference
Bozic, I., Reiter, J. G., Allen, B., Antal, T., Chatterjee, K., Shah, P., Moon, Y. S., Yaqubie, A., Kelly, N., Le, D. T., Lipson, E. J., Chapman, P. B., Diaz, L. A. Jr., Vogelstein, B., & Nowak, M. A. (2013). Evolutionary dynamics of cancer in response to targeted combination therapy. eLife, 2, e00747. https://doi.org/10.7554/eLife.00747
Core Argument
The major barrier to curing solid tumors with targeted therapy is the pre-existence of resistant cells at the start of treatment. Whether combination therapy succeeds depends critically on one factor: whether any single point mutation can confer resistance to both drugs simultaneously (cross-resistance). If such a mutation exists (n12 ≥ 1), dual therapy will fail in most patients with advanced disease; if it does not (n12 = 0), dual therapy can cure the majority of patients with typical tumor burdens. Simultaneous administration is vastly superior to sequential administration, which always fails. For patients with the largest tumor burdens, triple therapy without cross-resistance may be required.
Methods
Continuous-time multitype branching process model with four cell types (00 sensitive to both drugs, 10 resistant to drug 1, 01 resistant to drug 2, 11 resistant to both). Two phases: pretreatment (exponential growth from a single sensitive cell, birth rate b, death rate d) and treatment (death rate increased to d′ for sensitive cells). Parameters: point mutation rate u ≈ 10^−9 per base pair per cell division; ~50 potential resistance mutations per drug (n1 = n2 = 50); cross-resistance mutations n12 ∈ {0, 1}. Birth rate b = 0.14/day (7-day interdivision time), death rate d = 0.13/day (from vemurafenib dataset: 21 lesions, net growth 0.01/day). Treatment death rate d′ from 68 lesions: median decline −0.03/day. Model validated against 20 melanoma patients on vemurafenib monotherapy. Multi-lesion analysis: 22 patients (7 pancreatic, 11 colorectal, 6 melanoma) with total tumor burden quantified (8.5 × 10^8 to 2.6 × 10^11 cells).
Key Findings
-
Cross-resistance is the critical barrier. If even one of the 6.6 billion base pairs in a diploid genome can mutate to confer cross-resistance to both drugs (n12 ≥ 1), dual therapy will not achieve sustained control for most patients with advanced disease. The expected number of fully resistant cells at treatment start is X ≈ M n12 μ — independent of n1 and n2. A single cross-resistance mutation has orders-of-magnitude more impact than sequential acquisition of two separate resistance mutations (X ≈ M n1 n2 μ²). Triple therapy faces the same constraint: if n123 ≥ 1, it too will fail.
-
Dual therapy without cross-resistance can cure. For n12 = 0, the expected number of dual-resistant cells scales as M n1 n2 μ² — orders of magnitude fewer. In a 22-patient cohort, 8 patients with smallest tumor burden would have >95% cure probability. Those with largest burden still had >20% recurrence risk. Extraordinary sensitivity to tumor size: for M = 10^8, eradication probability ~99.9%; for M = 10^11, ~88% with typical parameters.
-
Simultaneous therapy is vastly superior to sequential therapy. When n12 ≥ 1, sequential therapy fails in 100% of cases (~74% due to pre-existing dual-resistant cells, ~26% due to resistance arising during first-drug treatment). Simultaneous therapy can cure ~26% of lesions even with cross-resistance. When n12 = 0, sequential still fails in 100% of cases, while simultaneous succeeds in >99%. The clinical implication: sequential administration “precludes any chance for cure — even when there are no possible mutations that can confer cross-resistance.”
-
Stochastic effects can rescue small lesions. Even with cross-resistance mutations present, small numbers of dual-resistant cells can be lost by genetic drift during treatment, enabling complete eradication. This explains why tumors can recur after long remission without invoking cancer stem cell dormancy, angiogenesis, or immune escape.
-
Resistance pre-exists at detectable tumor sizes. For monotherapy with ~50 potential resistance mutations, resistant cells are always present in lesions detectable by conventional imaging (M ≥ 10^8 cells). Monotherapy will eventually fail in all such lesions — resistance is a “fait accompli” for single agents.
-
Cancer stem cell fraction determines resistance probability. If cancer stem cells represent only 0.1% of tumor cells, resistance likelihood is ~0.1% as likely — explaining imatinib’s remarkable success in CML. In solid tumors, stem cell fractions are typically >5% and sometimes near 100%, making resistance far more probable.
-
Fitness costs of resistance have limited impact. When cross-resistance exists (n12 ≥ 1), a 10% fitness cost per resistance mutation only marginally improves cure probability. 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.
-
Vemurafenib clinical data. Among 20 melanoma patients: complete responses in some, stable disease in others, partial responses with mixed lesion responses in most. Smallest lesions were most likely to become undetectable. The median tumor decline rate during treatment was −0.03/day.
Concepts Introduced or Used
therapy-resistance, cross-resistance, combination-therapy, targeted-therapy, branching-process, multitype-branching-process, sequential-therapy, simultaneous-therapy, drug-resistance, cancer-stem-cell, pre-existing-resistance, de-novo-resistance, stochastic-extinction, fitness-cost, tumor-burden, vemurafenib, BRAF-inhibitor
Entities Referenced
- Drugs: vemurafenib (BRAF inhibitor), imatinib (BCR-ABL inhibitor), gefitinib/erlotinib (EGFR inhibitors), panitumumab/cetuximab (EGFR mAbs), crizotinib (ALK inhibitor)
- Genes: BRAF V600E, EGFR, KRAS, EML4-ALK, BCR-ABL, BTK, PLCG2, ABC transporters
- Cancer types: melanoma, colorectal adenocarcinoma, pancreatic ductal adenocarcinoma, NSCLC, CML
- Clinical cohorts: 20 melanoma patients (vemurafenib), 22 patients (metastatic burden quantification), FAP cohort (from Bozic 2010)
Limitations (as stated by authors)
- The model applies to small-molecule targeted therapies, not immunotherapies (CTLA-4, PD-1, CAR-T) — the immune system can replicate and evolve, and the factors underlying immunotherapy success/failure are not sufficiently understood for similar modeling.
- Parameter estimates derived from melanoma; other cancers may have different birth/death rates, though results are qualitatively robust across wide parameter ranges.
- The cancer stem cell fraction is poorly quantified in solid tumors and may be plastic (non-stem cells converting to stem cells), complicating effective population size estimates.
- The model assumes constant birth and death rates and neglects spatial constraints, metabolic gradients, and clonal cooperation/interference.
Relevance to Clonal Evolution
This paper operationalizes clonal evolution theory into clinically actionable predictions. It demonstrates that the evolutionary principle Nowell articulated in 1976 — pre-existing variation enables therapeutic escape — can be formalized mathematically to predict which drug combinations will succeed and which will fail. The central finding that cross-resistance (n12) is the critical determinant reframes drug development: identifying compounds with no shared resistance mutations is more important than increasing potency. The simultaneous-over-sequential proof challenges current clinical practice and has direct implications for trial design. This paper bridges the gap between the qualitative Darwinian framework of cancer evolution and the quantitative predictions needed for precision oncology.