Abbosh et al. (2017) — Phylogenetic ctDNA Analysis Depicts Early-Stage Lung Cancer Evolution

Bibliographic Reference

Abbosh, C., Birkbak, N. J., Wilson, G. A., Jamal-Hanjani, M., Constantin, T., Salari, R., Le Quesne, J., Moore, D. A., Veeriah, S., Rosenthal, R., Marafioti, T., Kirkizlar, E., Watkins, T. B. K., McGranahan, N., Ward, S., Martinson, L., Riley, J., Fraioli, F., Al Bakir, M., … Swanton, C. (2017). Phylogenetic ctDNA analysis depicts early-stage lung cancer evolution. Nature, 545, 446–451. https://doi.org/10.1038/nature22364

Core Argument

Circulating tumour DNA (ctDNA) can be combined with multi-region exome sequencing (M-seq) phylogenetic trees to non-invasively track the subclonal dynamics of early-stage non-small-cell lung cancer (NSCLC), enabling detection of postoperative residual disease, identification of adjuvant chemotherapy resistance, and phylogenetic characterization of metastatic relapse. Using bespoke multiplex-PCR NGS panels targeting patient-specific clonal and subclonal SNVs selected from tumour phylogenetic trees, the authors profile the first 100 patients of the prospective TRACERx study, establish independent predictors of ctDNA release, quantify the tumour-volume detection limit, and show that ctDNA profiling can resolve which primary tumour subclone seeds a given metastasis.

Methods

Study design and cohort. Prospective analysis of the first 100 patients from the TRACERx (Tracking Non-Small-Cell Lung Cancer Evolution Through Therapy) study. Multi-region exome sequencing of primary tumours was used to construct phylogenetic trees. Bespoke multiplex-PCR NGS assay panels were designed for each patient, targeting clonal and subclonal SNVs selected to represent phylogenetic branches.

Preoperative phase (n=96). Plasma collected before surgery was profiled to identify clinicopathological determinants of ctDNA detection. A median of 18 SNVs (range 10–22) were targeted per patient: median 11 clonal (range 2–20) and median 6 subclonal (range 0–16). ctDNA positivity threshold: at least 2 SNVs detected. The multiplex-PCR NGS platform achieved >99% sensitivity for SNVs at VAFs above 0.1% and 99.6% specificity for a single SNV.

Longitudinal phase (n=24). Pre- and postoperative plasma was profiled blinded to relapse status, including 10 relapse-free patients (median follow-up 775 days) and 14 confirmed NSCLC-relapse cases. Assay panels were redesigned for LUADs to include additional clonal SNVs (median 28 total SNVs). Patients were followed every 3 months for the first 2 years, then every 6 months.

Volume analysis. CT volumetric analysis was performed on 38/46 ctDNA-positive patients. Tumour volume was correlated with mean clonal plasma VAF using linear regression on log-transformed values. Effective subclone size was defined as mean CCF of the subclone multiplied by tumour volume and mean tumour purity.

Metastatic validation. Relapse tissue biopsies (n=4 patients) and post-mortem tissue (n=1 patient, CRUK0063, via the PEACE study) were exome-sequenced and integrated with primary tumour phylogenies to validate ctDNA-based phylogenetic characterization.

PEACE post-mortem extended analysis. For CRUK0063, a 103-SNV bespoke panel was designed to track 9 metastatic subclonal clusters across longitudinal time points, with updated calling thresholds to control false-positive rates (predicted false-positive risk 10.7% for a single SNV, 0.5% for two SNVs per time point).

Key Findings

• ctDNA detection in early-stage NSCLC is histology-dependent and predicted by independent clinical factors. 48% (46/96) of early-stage NSCLCs had ≥2 SNVs detected preoperatively. 97% (30/31) of LUSCs were ctDNA-positive compared with 19% (11/58) of LUADs. In multivariable analysis, non-adenocarcinoma histology (OR 40.76, 95% CI 4.55–365.14, P = 0.001), lymphovascular invasion (OR 5.84, 95% CI 1.07–32.03, P = 0.042), and high Ki67 proliferation index (OR 1.40 per 10% increase, 95% CI 1.05–1.84, P = 0.022) were independent predictors of ctDNA detection. Necrosis was significant in univariable but not multivariable analysis.

• Tumour volume predicts plasma VAF with a quantitative relationship applicable to detection limits. Tumour volume correlated significantly with mean clonal plasma VAF (Spearman’s rho = 0.61, P < 0.001, n = 38). Linear modelling predicted that a primary tumour burden of 10 cm³ would result in a mean clonal plasma VAF of 0.1% (95% CI 0.05–0.17%). A 4-mm spherical nodule (0.034 cm³, the limit of low-dose CT screening) would equate to a VAF of approximately 1.4 × 10⁻⁴% (95% CI 6.4 × 10⁻⁶ – 0.0031%), at the extreme of current ctDNA platform detection limits.

• Subclonal ctDNA detection is determined by subclone volume. Shared subclones (identified in >1 tumour region by M-seq) were detected in ctDNA more frequently than private subclones (35/57, 61% vs. 26/80, 33%). Effective subclone size (mean CCF × tumour volume × purity) was significantly higher for detected vs. undetected subclonal SNVs (Wilcoxon rank-sum test, P < 0.001, n = 272). Subclone volume correlated with subclonal plasma VAF (Spearman’s rho = 0.53, P < 0.001, n = 109).

• Postoperative ctDNA profiling detects NSCLC relapse with a median lead time of 70 days and identifies adjuvant chemotherapy resistance. 13/14 (93%) confirmed relapse patients had ≥2 SNVs detected before or at clinical relapse. The median interval between ctDNA detection and CT-confirmed relapse was 70 days (range 10–346 days); 4 cases had lead times >6 months. In two cases, ctDNA detection preceded CT imaging inconclusive for relapse by 157 and 163 days. Three patients with detectable ctDNA within 30 days of surgery showed increasing SNV counts despite adjuvant chemotherapy, with disease recurring within one year — indicating adjuvant chemotherapy resistance. One patient with detectable ctDNA postoperatively showed clearance after adjuvant chemoradiotherapy and remained relapse-free at 688 days.

• Phylogenetic ctDNA profiling can resolve the subclone of origin for metastatic relapse and track tiered subclonal burden. ctDNA detection from the CNS was demonstrated in one patient with intracerebral metastasis but no extracranial disease. In patients with detected postoperative subclonal SNVs, the pattern of plasma VAFs mapped back to M-seq phylogenetic trees identified whether relapse was dominated by a single subclone (4 patients) or involved subclones from multiple phylogenetic branches (3 patients). In the PEACE post-mortem case (CRUK0063), a 103-SNV panel revealed tiered subclonal burden concordant with the phylogenetic distance from the clonal cluster — and detected private para-aortic subclones 90 days before death, consistent with para-aortic progression found at autopsy.

Concepts Introduced or Used

• Phylogenetic ctDNA profiling — Targeting patient-specific clonal and subclonal SNVs selected from M-seq-derived phylogenetic trees, enabling non-invasive tracking of subclonal dynamics. A method for linking liquid biopsy results to tumour evolutionary structure.
• Effective subclone size — Mean cancer cell fraction (CCF) of a subclone multiplied by tumour volume and mean tumour purity. Used to predict subclonal ctDNA detectability.
• ctDNA detection lead time — The interval between first ctDNA detection and clinical/radiographic confirmation of relapse. Median 70 days in this study; ranged up to 346 days.
• Subclonal architecture — The composition and phylogenetic structure of subclones within a tumour. This study demonstrates ctDNA can non-invasively resolve subclonal architecture at relapse.
• Clonal vs. subclonal SNVs in ctDNA — Clonal SNVs were detectable in all ctDNA-positive patients and exhibited higher mean plasma VAF than subclonal SNVs, supporting clonal alterations as a more sensitive ctDNA detection method.
• Phylogenetic tree — Evolutionary tree constructed from M-seq data, with branch lengths proportional to mutation count. Used as template for ctDNA assay design and for mapping detected SNVs back to tumour evolutionary history.
• TRACERx — Tracking Non-Small-Cell Lung Cancer Evolution Through Therapy, the prospective cohort study that provided the patient population and multi-region sequencing data.

Entities Referenced

• Genes/Drivers: KRAS, EGFR, TP53, ERBB2 (HER2), ARID1A, OR5D18, OR10K1, FANCM, TERT, PRF1, LMO2, EP300, NF1, RET, PIK3CA, CDKN2A, MSH2
• Histologies: Lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC)
• Methods: multiplex-PCR NGS, multi-region exome sequencing (M-seq), PyClone, CITUP, ASCAT, PET-CT volumetric analysis, 3D Slicer
• Studies: TRACERx (NCT01888601), PEACE post-mortem study (NCT03004755)
• Platform/Technology: Natera custom multiplex-PCR NGS platform, Illumina HiSeq 2500, Oncomine Lung cfDNA assay (ThermoFisher), Ion S5 System
• Cell count conversion: 1 cm³ tumour volume ~ 9.4 × 10⁷ cells (Del Monte, 2009)

Limitations (as stated by authors)

1. Cost. Targeted ctDNA profiling is estimated at US$1,750 per patient for sequencing a single tumour region, synthesis of a patient-specific assay panel, and profiling of five plasma samples.
2. ctDNA detection limits constrain early screening. Based on the volume-VAF relationship, the sensitivity of clonal-SNV ctDNA-directed early NSCLC screening “may be constrained by tumour size” when using current technologies.
3. The bespoke approach is tumour-informed. It requires prior sequencing of tumour tissue to design the patient-specific panel and knowledge of the tumour phylogenetic tree, limiting its application to settings where tumour tissue is available.
4. The ctDNA detection threshold of 2 SNVs is a pragmatic compromise. Selected to minimize type I errors when testing up to 30 SNVs per time point, but may miss genuine low-burden ctDNA cases.
5. The post-mortem (PEACE) extension case (CRUK0063) had a higher false-positive risk. With 103 SNVs tested per time point, the predicted false-positive rate translated to a 10.7% risk of a single false-positive SNV per time point, and 0.5% risk of two false positives.
6. No statistical methods were used to predetermine sample size. (Noted in Methods.)
7. The experiments were not randomized. (Noted in Methods.)

Relevance to Clonal Evolution

This paper is the foundational ctDNA demonstration from the TRACERx study and directly connects liquid biopsy to clonal evolution in several critical ways:

Phylogenetic ctDNA bridges tissue sequencing and non-invasive monitoring. By designing bespoke panels targeting SNVs that represent specific branches of the tumour phylogenetic tree, the authors show that ctDNA can report not just the presence of residual disease but its phylogenetic composition — which clones are expanding, which are contracting, and which subclone seeds relapse. This is a direct extension of intratumor-heterogeneity measurement from tissue to plasma.

Quantitative framework for ctDNA-based ITH estimation. The relationship between tumour volume and clonal VAF (10 cm³ → 0.1% VAF), the effective subclone size metric for subclonal detection, and the finding that shared subclones are detected more frequently than private subclones, together establish a quantitative basis for inferring subclonal architecture from ctDNA alone. This is directly relevant to the empirical ITH test design using ctDNA-based ITH estimation, as the volume-VAF calibration and subclone detection limits define the operational boundaries of such an assay.

Subclonal reconstruction from plasma. The ability to track subclonal SNVs in ctDNA and map them back to phylogenetic clusters provides a non-invasive window into subclonal-reconstruction. The tiered burden pattern observed in CRUK0063 (VAF of subclonal clusters mirrors phylogenetic distance from the clonal trunk) suggests that ctDNA VAF ratios encode phylogenetic information that could be exploited for plasma-only tree reconstruction.

Adjuvant therapy resistance as clonal evolution in action. The observation that ctDNA levels rise during adjuvant chemotherapy in patients destined to relapse demonstrates that resistance-conferring subclones can be detected and monitored non-invasively during therapy — a direct observation of therapy-resistance as an evolutionary phenomenon. The clearance of ctDNA after chemoradiotherapy in the one patient who remained relapse-free shows the opposite: successful elimination of residual disease.

Metastasis as a subclonal bottleneck. By characterizing which phylogenetic branches seed relapse and validating against metastatic tissue sequencing, this study provides real-time non-invasive evidence that metastasis is a clonal-sweep-like bottleneck event. The finding that polyclonal relapse (multiple branches seeding metastases) can be distinguished from monoclonal relapse via ctDNA has implications for understanding the evolutionary dynamics of metastasis.

TRACERx ctDNA framework. As the first TRACERx ctDNA paper, this study established the methodology and cohort that would underpin subsequent TRACERx ctDNA analyses (e.g., bakir2023-tracerx-metastasis), including the integration of M-seq phylogenetics with bespoke ctDNA monitoring.