Jamal-Hanjani, M., Wilson, G. A., McGranahan, N., et al. (2017). Tracking the evolution of non–small-cell lung cancer. New England Journal of Medicine, 376(22), 2109–2121.

Jamal-Hanjani et al. (2017) — Tracking the Evolution of Non–Small-Cell Lung Cancer (TRACERx)

Bibliographic Reference

Jamal-Hanjani, M., Wilson, G. A., McGranahan, N., Birkbak, N. J., Watkins, T. B. K., Veeriah, S., Shafi, S., Johnson, D. H., Mitter, R., Rosenthal, R., Salm, M., Horswell, S., Escudero, M., Matthews, N., Rowan, A., Chambers, T., Moore, D. A., Turajlic, S., Xu, H., … Swanton, C., for the TRACERx Consortium. (2017). Tracking the evolution of non–small-cell lung cancer. New England Journal of Medicine, 376(22), 2109–2121. https://doi.org/10.1056/NEJMoa1616288

Core Argument

Intratumor heterogeneity (ITH) in NSCLC is not merely a descriptive feature — it is a clinically significant, quantifiable prognostic biomarker driven by specific mutational processes and chromosomal instability. Through prospective multiregion whole-exome sequencing of 327 tumor regions from the first 100 patients in the TRACERx cohort, the authors demonstrate that: (1) branched evolution and ITH are near-universal in early-stage NSCLC; (2) copy-number heterogeneity — not mutational heterogeneity — predicts recurrence-free survival (HR 4.9, P = 4.4×10⁻⁴); (3) driver events show a consistent temporal ordering: targetable drivers (EGFR, MET, BRAF) are early and clonal, while chromatin modifiers and DNA damage response genes are frequently subclonal; (4) genome doubling is an early clonal event in 76% of tumors that enables subsequent chromosomal instability; and (5) APOBEC mutagenesis drives late subclonal diversification. The paper establishes multiregion sequencing as essential for accurately classifying driver clonality — without it, 76% of subclonal mutations would erroneously appear clonal.

Methods

Prospective cohort study (TRACERx, NCT01888601) — 100 patients with stage IA–IIIA NSCLC, treatment-naive, resected before systemic therapy. Multiregion whole-exome sequencing: 327 tumor regions (323 primary + 4 lymph node metastases) and 100 matched germline samples. Median 3 regions per tumor (range 2–8), median depth 426×. Somatic mutations classified as clonal (present in all cancer cells) or subclonal (subset). Copy-number alterations measured as percentage of genome affected. Phylogenetic trees constructed from mutation clustering by cellular prevalence (525 clusters identified, median 5 per tumor). Mutational signature analysis (Alexandrov et al. 2013, 2015). dN/dS selection analysis (Martincorena et al. 2015) with trinucleotide context correction. Mirrored subclonal allelic imbalance detection using germline heterozygous SNPs. Orthogonal validation performed. Data deposited in European Genome–Phenome Archive (EGAS00001002247).

Key Findings

1. Copy-number heterogeneity — not mutational heterogeneity — predicts survival. Patients with high subclonal CNA proportion (≥48%, the cohort median) had significantly worse relapse-free survival than those with low proportion: HR 4.9 (95% CI 1.8–13.1, P = 4.4×10⁻⁴). Median time to recurrence/death: 24.4 months vs. not reached. This remained significant in multivariate analysis (HR 3.70, P = 0.01) after adjusting for age, smoking, histology, adjuvant therapy, and stage. 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 state, that drives poor outcome.

2. Branched evolution and ITH are near-universal. Median 30% of somatic mutations were subclonal (range 0.5–93%); median 48% of copy-number alterations were subclonal (range 0.3–88%). 86% of tumor regions carried subclones from only a single branch of the phylogenetic tree. 525 mutation clusters identified (median 5 per tumor; range 2–15). Without multiregion sequencing, 76% of subclonal mutations would appear clonal, and 65% of branched subclone clusters would erroneously appear clonal.

3. Temporal ordering of driver events. Targetable drivers (EGFR, MET, BRAF) and TP53 were almost always clonal and occurred before genome doubling — they are initiation events. In adenocarcinomas, these included TERT amplification, 8p loss, and 5p gain. In squamous-cell carcinomas, NOTCH1 mutations, FGFR1 amplification, 3q gain (SOX2, PIK3CA), and 3p/5q/17p loss were early clonal events. By contrast, >75% of tumors carried at least one subclonal driver alteration. PIK3CA, NF1, KRAS, TP53, and NOTCH family members were frequently subclonal. 51% of driver alterations in chromatin remodeling, histone methylation, or DNA damage response/repair genes were subclonal or late.

4. Genome doubling is early, clonal, and permissive for CIN. Genome doubling was detected in 76% of tumors and was clonal in all but 3. In adenocarcinomas, genome doubling was significantly associated with both subclonal mutation burden (P = 0.02) and subclonal CNA burden (P = 0.003). Mirrored subclonal allelic imbalance — evidence of ongoing dynamic CIN — was significantly enriched in genome-doubled tumors (P = 0.004). This establishes genome doubling as an early permissive event that enables the chromosomal instability driving subsequent diversification.

5. Mirrored subclonal allelic imbalance reveals parallel evolution through CIN. Detected in 62% of 92 evaluable tumors (375 events total, 1–43% of affected genomes). When the maternal allele is gained/lost in one subclone and the paternal allele in another, it produces mirrored B-allele frequency profiles — direct evidence of ongoing CIN producing parallel evolution. Five tumors showed focal amplifications of different parental alleles converging on the same cancer genes (CDK4, FOXA1, BCL11A, MUC1, NKX2-1). At the chromosome-arm level, parallel evolution was observed in 13 tumors, with most events in regions previously classified as significantly gained/lost in NSCLC — consistent with positive selection.

6. Mutational processes are temporally partitioned. Smoking signature 4 dominated early/truncal mutations: Spearman r = 0.90 with early mutation burden in adenocarcinomas (P < 1.1×10⁻¹⁶). In 7 of 12 long-term former smokers (>20 years since last exposure), signature 4 persisted in late mutations — suggesting a long latency period before clinical presentation. APOBEC signatures 2/13 dominated late/subclonal mutations and were not correlated with signature 4. Clock-like signatures 1A and 5 correlated with subclonal burden. 19 tumors had subclonal driver mutations in an APOBEC context — APOBEC activity directly generates subclonal driver events. One lifelong nonsmoker with squamous-cell carcinoma had >1,000 clonal signature 4 mutations — occupational exposure to arsenic, benzene, and coal tar mimicked tobacco’s mutagenic effects.

7. Selection is ongoing and temporally structured. dN/dS > 1 for all exonic missense mutations considered together. When temporally dissected: significant positive selection for late but not early mutations. Nonsense mutations depleted early (dN/dS < 1) but not late — early constraints relax as tumors evolve. Early driver mutations were histologic-subtype-specific; late/subclonal drivers affected a broader range of pan-cancer genes. Squamous-cell carcinomas had greater early constraint (more nonsense depletion) than adenocarcinomas. Adenocarcinomas had a significantly higher rate of clonal driver acquisition (P = 0.001).

8. Clinical implications for targeted therapy. 86 of 100 tumors had alterations under investigation in genomically profiled drug trials (NLMT, MATCH). Of these, 17 (20%) had subclonal targetable mutations/CNAs. In 12 of 17 (71%), both a clonal AND a subclonal targetable alteration were present — clonal targets should be prioritized. Multiregion sequencing identified significantly more driver alterations than single-sample analysis (P = 0.004). Single-biopsy-based target identification systematically misses subclonal drivers and overestimates the clonality of detected drivers.

Concepts Introduced or Used

• TRACERx (TRAcking Cancer Evolution through therapy (Rx)): Prospective multiregion sequencing cohort study; target enrollment 842 patients; this paper reports the first 100. The definitive prospective test of ITH-clinical outcome associations.
• Mirrored subclonal allelic imbalance: A genomic signature of ongoing dynamic chromosomal instability — when the maternal allele is gained/lost in one subclone and the paternal allele in another, producing mirrored B-allele frequency profiles across tumor regions. Direct evidence that CIN produces parallel evolution.
• Clonal vs. subclonal driver classification: Driver events classified by their presence in all tumor cells (clonal = truncal = early) vs. a subset (subclonal = branch/leaf = late). Critical for therapeutic target prioritization: clonal targets are present in every cell; subclonal targets will leave untargeted clones.
• Temporal dissection of selection: Using dN/dS separately for early vs. late mutations to determine when in tumor evolution selection operates. Reveals that positive selection is ongoing and that constraints relax over time.
• Dynamic vs. static chromosomal instability: The distinction between ongoing CIN (measured by inter-region CNA heterogeneity) and accumulated CIN (measured by mean aberrant genome fraction). Only the dynamic measure is prognostic — the process matters, not the accumulated state.
• Genome doubling as a permissive event: Early clonal WGD enables subsequent CIN by providing a buffer against the deleterious effects of chromosomal losses. Occurs in 76% of NSCLCs, almost always before subclonal diversification.
• Mutational signature timing: Partitioning mutations by evolutionary timing (early/truncal vs. late/subclonal) reveals that different mutational processes dominate at different evolutionary stages — smoking early, APOBEC late.

Entities Referenced

• TRACERx Consortium — Multicenter UK prospective cohort; this is the flagship paper
• EGFR, MET, BRAF — Targetable drivers; almost always clonal and early in adenocarcinomas
• TP53 — Predominantly clonal and early in both subtypes
• PIK3CA, NF1, KRAS, NOTCH1-3 — Frequently subclonal drivers
• CDK4, FOXA1, BCL11A, MUC1, NKX2-1 — Genes showing parallel evolution through focal amplification
• KMT2C, KMT2D, COL5A2, UBR5 — Chromatin modifiers and DDR genes; frequently subclonal/late
• APOBEC3A/3B — Cytidine deaminases driving late subclonal mutagenesis (signatures 2/13)
• Mutational signatures 1A, 2, 4, 5, 13 — Alexandrov et al. COSMIC signatures
• NLMT, MATCH trials — Genomically profiled drug studies
• Adenocarcinoma, squamous-cell carcinoma — The two major NSCLC subtypes; show distinct evolutionary patterns

Limitations (as stated by authors)

• First 100 patients only. Target enrollment is 842; findings require validation in the full cohort. The association between copy-number heterogeneity and survival “is being assessed in the next 742 patients enrolled.”
• Exome-only. Whole-exome sequencing misses non-coding mutations, structural variants, and epigenetic heterogeneity. The full extent of ITH is underestimated.
• Bulk sequencing. Cannot resolve the clonal composition within a single tumor region at single-cell resolution.
• Early-stage only. All patients had resectable stage IA–IIIA disease. Findings may not generalize to advanced/metastatic NSCLC.
• Short follow-up. Median follow-up time limited for recurrence-free survival analysis in a cohort with curative-intent resection.
• No post-treatment sampling. Pre-treatment resection only. Evolutionary dynamics under therapy (a primary TRACERx objective) are not addressed here — this is the baseline evolutionary state.

Relevance to Clonal Evolution

This is a Tier 1 paper — arguably the single most important empirical study in the wiki’s NSCLC corpus, along with Gerlinger et al. (2012) for renal carcinoma. Its contributions span the full breadth of the clonal evolution framework:

Empirical foundation for key concepts

• intratumor-heterogeneity: The definitive prospective quantification of ITH prevalence in NSCLC. Establishes that ITH is near-universal (median 30% subclonal mutations, 48% subclonal CNAs) and clinically significant (CNA heterogeneity → HR 4.9). Provides the empirical anchor for the wiki’s ITH concept page: subclone counts, mutation clusters, spatial distribution metrics, and the critical finding that mutational and copy-number heterogeneity have different prognostic significance.

• branching-evolution: Phylogenetic trees for 100 tumors. 86% of regions carry subclones from a single branch — direct evidence that branched evolution produces spatially segregated subclones. Extends Gerlinger 2012’s renal carcinoma findings to NSCLC at scale (100 vs. 4 patients).

• subclonal-architecture: 525 mutation clusters, median 5 per tumor. Comprehensive phylogenetic reconstruction across a large prospective cohort. Establishes the technical methodology for clonal decomposition from multiregion sequencing data.

• chromosomal-instability: Introduces mirrored subclonal allelic imbalance as a direct readout of ongoing dynamic CIN. Demonstrates that CIN drives parallel evolution of driver CNAs. Establishes the genome doubling → CIN → diversification cascade. The distinction between dynamic and static CIN measures is a conceptual contribution with direct clinical implications.

• APOBEC-mutagenesis: Provides direct evidence that APOBEC activity generates subclonal driver mutations (19 tumors with subclonal drivers in APOBEC context). Links APOBEC to late evolutionary diversification. Extends Burns et al. (2013) mechanistic findings to the population/evolutionary level.

• clonal-evolution and positive-selection: Temporal dN/dS analysis demonstrates that selection is ongoing and that evolutionary constraints relax over time. Early driver mutations are subtype-specific; late drivers are pan-cancer — the evolutionary landscape broadens as tumors progress.

Temporal architecture of tumor evolution

The paper’s most important conceptual contribution is the temporal ordering of genomic events in NSCLC:

Smoking exposure → Early clonal drivers (EGFR/MET/BRAF/TP53)
                 → Genome doubling (76% of tumors, early and clonal)
                 → Ongoing CIN → Parallel evolution of driver CNAs
                 → APOBEC mutagenesis → Late subclonal driver mutations
                 → Chromatin/DDR gene alterations → Permissive diversification

This temporal architecture is the empirical foundation for the wiki’s Bozic-Nowak sweep timing condition (CC5) and the compression-entrenchment hypothesis: early tumors have clean sequential sweeps (τ_k >> sweep_time); late tumors enter the clonal interference regime as population size grows and multiple drivers compete.

Clinical translation

The finding that copy-number heterogeneity (HR 4.9) but not mutational heterogeneity (HR 0.86, P = 0.70) predicts survival is the paper’s most clinically actionable result. It suggests that CIN — not point mutation burden — is the primary driver of clinically significant ITH. This has direct implications for:

• Prognostic biomarker development (CNA heterogeneity as a risk stratifier)
• Therapeutic targeting (clonal > subclonal targets; 71% of tumors with subclonal targets also have clonal targets)
• The bottleneck paradox (deep responses to therapy may select for CIN-driven diversification)
• Liquid biopsy applications (CNA detection in ctDNA as a non-invasive prognostic tool)

Relation to wiki infrastructure

This paper is the empirical anchor for intratumor-heterogeneity, branching-evolution, subclonal-architecture, and chromosomal-instability. It provides the NSCLC-specific evidence that complements Gerlinger 2012 (renal), PCAWG 2020 (pan-cancer), and Al Bakir 2023 (TRACERx metastasis). The TRACERx protocol paper (Jamal-Hanjani et al., 2014, PLoS Biology) describes the study design; this paper delivers the first results.

Revision history

(None — initial creation 2026-07-16)