Al Bakir et al. (2023) — The Evolution of NSCLC Metastases in TRACERx

Bibliographic Reference

Al Bakir, M., Huebner, A., Martínez-Ruiz, C., Grigoriadis, K., Watkins, T. B. K., Pich, O., … McGranahan, N., & Swanton, C. (2023). The evolution of non-small cell lung cancer metastases in TRACERx. Nature, 616(7957), 534–542. https://doi.org/10.1038/s41586-023-05729-x

Core Argument

By longitudinally tracking 126 NSCLC patients who developed metastatic disease within the TRACERx prospective cohort, the timing, mode, and selective forces governing metastatic dissemination can be reconstructed from multi-region whole-exome sequencing of paired primary tumours and metastases. The majority of metastases (~75%) diverge late, after the last clonal sweep in the primary tumour. Metastasis-seeding subclones exhibit evidence of positive selection — they are larger and more dispersed within primary tumours than non-seeding counterparts. This evolutionary framework has direct implications for early detection, sampling strategy, and adjuvant therapeutic targeting.

Methods

TRACERx 421 cohort: 421 prospectively recruited patients with untreated early-stage (I–III) NSCLC, followed for median 4.66 years across 19 UK hospital sites (2014–2021). WES at median 398× depth of 476 primary tumour regions paired with 218 metastatic samples from 126 patients. Phylogenetic trees constructed using CONIPHER. Timing of metastatic divergence defined relative to the last clonal sweep in the primary tumour. Dissemination patterns classified as monoclonal vs polyclonal (subdivided into monophyletic/polyphyletic) at the case level. dNdScv used to detect selection in lung cancer genes within seeding vs non-seeding clusters. GISTIC2.0 for SCNA positive-selection scores in metastases vs non-metastatic primaries. MACHINA for orthogonal dissemination pattern validation. Spatially explicit in silico simulations (modified from Sun et al.) to model divergence timing vs tumour size. 144 non-metastatic TRACERx patients with ≥3 years follow-up used as controls.

Key Findings

• ~75% of metastases diverge late. In 74.6% (94/126) of cases, metastatic divergence occurred after the last clonal sweep in the primary tumour — all primary clonal mutations were present in the metastasis. In 25.4% (32/126), divergence was early: a complete clonal sweep occurred in the primary after the metastatic clone departed, leaving primary-ubiquitous mutations absent from the metastasis. Orthogonal methods (LOH, clonal WGD, proportion of shared ubiquitous mutations) supported late-divergence predominance.

• Early divergence enriched in smokers, occurs at small tumour size. Early divergence was independently associated with smoking status at resection (GLM, P = 0.016). Simulations showed that early divergence is more likely when the primary tumour diameter is <8 mm — below the typical actionable nodule threshold in CT screening protocols. At <1 mm simulated diameter, 78% of divergence was classified as early.

• Single-region sampling misclassifies 83% of late divergence as early. When only one randomly selected primary region was used, 75/90 late-divergence cases were incorrectly classified as early — a stark warning against under-sampling in metastasis evolution studies.

• Polyclonal dissemination in 32% of cases, associated with extrathoracic recurrence. 68.3% (86/126) monoclonal, 31.7% (40/126) polyclonal (21 monophyletic, 16 polyphyletic, 3 mixed). Polyclonal dissemination was enriched in primary LN/satellite lesions (P = 0.03) and associated with extrathoracic disease recurrence (P = 0.0056). The true rate of polyclonal dissemination is probably higher — monoclonal calls are confounded by undersampling of metastases.

• Primary LN metastases are a hallmark of metastatic potential, not a gateway. In only 3/19 cases with both primary LN and subsequent recurrence did the LN seed the recurrence. In 13/19 cases, dissemination occurred solely from the primary tumour. LN involvement reflects metastatic propensity rather than serving as a route to further disease.

• Metastasis-seeding clones are larger and under positive selection in the primary tumour. Seeding clusters had significantly higher maximum CCF (P = 6.4 × 10^−5) and greater spatial dispersion (P = 1.6 × 10^−8) than non-seeding clusters in the primary. In LUAD, seeding clusters showed significant positive selection (dN/dS = 1.97, 95% CI 1.14–3.38) for lung cancer genes. In LUSC, positive selection was observed only in seeding clones (dN/dS = 2.03, 95% CI 1.16–3.57); subclonal mutations in non-metastasizing LUSC primaries showed no significant selection (dN/dS = 0.89).

• Two categories of metastasis-associated alterations. (i) Truncal/maintained: MDM2 amplification and TP53 mutations — present clonally before divergence, associated with metastatic propensity. (ii) Metastasis-favoured: HIST1H3B amplification in LUAD — frequently subclonal in the primary, may confer selective advantage in the metastatic niche. Significantly maintained driver mutations included KRAS, TP53, and KEAP1 in LUAD, and TP53 in LUSC.

• Platinum chemotherapy leaves a mutational footprint. The platinum signature (SBS31/35) was detected in 9/11 treated recurrence samples. One case showed two brain metastases with differential platinum-signature exposure, timing their divergence to the chemotherapy window.

• Metastasis-unique drivers in 33% of cases. 33.3% (42/126) of cases had metastasis-unique driver mutations not detected in the primary tumour. However, 68.6% of driver mutations were shared. NRAS, RB1, and EGFR mutations were always shared; KRAS showed both shared and primary-unique activating mutations — relevant for allele-specific targeted therapy stratification.

Concepts Introduced or Used

metastasis, clonal-evolution, clonal-sweep, driver-mutation, subclonal-architecture, intratumor-heterogeneity, phylogenetic-tree, monoclonal-dissemination, polyclonal-dissemination, early-divergence, late-divergence, seeding-cluster, positive-selection, dN-dS-ratio, whole-genome-duplication, platinum-signature, mutational-signature, lymph-node-metastasis, ctDNA, CONIPHER, MACHINA, GISTIC, extrapulmonary-metastasis

Entities Referenced

• TRACERx (TRAcking non-small cell lung Cancer Evolution through therapy, NCT01888601)
• PEACE post-mortem study (NCT03004755)
• 421 NSCLC patients, 19 UK hospital sites, 2014–2021
• Genes: TP53, KRAS, KEAP1, EGFR, NRAS, RB1, STK11, PIK3CA, SMARCA4, NF1, CDKN2A, ARID1A, CREBBP, SETD2, FAT1, KMT2D, B2M, MDM2, CCND1, NFE2L2, SETDB1, HIST1H3B, MYC, RBM5, PPP2R1A, CARD11, MACC1, RAC1
• Methods: CONIPHER, dNdScv, GISTIC2.0, MACHINA, whole-exome sequencing (median 398×)
• Drugs: platinum chemotherapy (adjuvant)

Limitations (as stated by authors)

• Metastatic samples could not be obtained from 95/142 patients with recurrence — brain metastases, patient frailty, and quality-control failures limit coverage.
• Single time-point sampling of metastases limits inference about ongoing evolution within the metastatic site.
• Monoclonal dissemination is probably overestimated due to undersampling of metastases — polyclonal dissemination requires sequencing multiple metastatic deposits per patient.
• No standardized methods exist for assessing timing of divergence or modes of dissemination across studies, limiting cross-study comparisons (breast and colorectal cancer studies report predominantly early divergence, potentially confounded by primary tumour undersampling).
• In silico simulations use biologically informed but simplified models of tumour growth; empirical validation of divergence-timing-to-tumour-size is limited to cross-sectional inferences.
• The study is restricted to NSCLC; generalizability to other cancer types requires independent validation.

Relevance to Clonal Evolution

This is one of the most comprehensive empirical studies of metastasis as an evolutionary process. It demonstrates that metastatic seeding is not random — it is driven by selection acting on subclones within the primary tumour. The late-divergence predominance challenges the assumption (from breast and colorectal cancer studies) that metastases typically seed early, and highlights the critical importance of multi-region sampling for accurate evolutionary inference. The finding that LN metastases rarely seed further metastases but instead signal metastatic propensity reframes the clinical interpretation of nodal disease. The two-category model of metastasis-associated alterations (truncal propensity markers vs subclonal niche-adaptation events) provides an evolutionary framework for adjuvant therapeutic strategy: target the truncal alterations shared by all metastases, and monitor for the subclonal ones that emerge under treatment.