Phylogenetic ctDNA analysis depicts early stage lung cancer evolution

PMID: 28445469 · DOI: 10.1038/nature22364 · Journal: Nature (2017)

TL;DR

Abbosh et al. report the first ctDNA analysis from the TRACERx prospective lung cancer evolution study, profiling the first 100 surgically resected early-stage NSCLC patients with bespoke multiplex-PCR NGS panels (median 18 SNVs per patient) targeting clonal and subclonal variants identified by multi-region exome sequencing. Pre-operative ctDNA was detected in 48% (46/96) of tumors with a stark histology effect: 97% of LUSC versus 19% of LUAD. Post-operatively in a 24-patient longitudinal sub-cohort, ctDNA detected NSCLC relapse in 13/14 (93%) confirmed-relapse cases with a median lead-time of 70 days over imaging, identified adjuvant-chemotherapy resistance in real time, and resolved the subclonal architecture of metastatic seeding via phylogenetic tracking — including a targetable ERBB2 amplification subclone driving relapse in CRUK0004.

Cohort & data

• Cohort: First 100 patients of the prospective TRACERx NSCLC study (NCT01888601); 96 evaluable for pre-operative ctDNA. Histologies include 58 LUAD, 31 LUSC, and 7 other NSCLC subtypes.
• Longitudinal sub-cohort: 24 patients with pre- and post-surgical plasma profiling (14 confirmed relapse, 10 relapse-free with median follow-up 775 days).
• Tumor genomics: 327 spatially separated primary tumor regions plus 4 metastatic biopsies underwent whole-exome sequencing through the TRACERx multi-region exome (M-Seq) pipeline; copy number / purity / ploidy via ASCAT; subclonal clusters via a modified PyClone; phylogenetic trees constructed with CITUP (0.1.0) plus manual reconstruction for relapse/autopsy cases (PMID:28445469).
• ctDNA assay: Bespoke patient-specific multiplex-PCR NGS panels (Natera assay design pipeline; precursor to Signatera) — amplicon sequencing of clonal + subclonal SNVs, sequenced on Illumina HiSeq 2500 to ~40,000× target depth. Median 18 SNVs per panel pre-operatively; expanded to median 28 SNVs for LUAD panels in the longitudinal phase. Analytical sensitivity >99% at VAF ≥0.1%; specificity 99.6% per SNV.
• Imaging: Pre-operative PET/CT reviewed for 92/96 patients; primary tumor volume measured on CT using 3D Slicer. Ki67 immunohistochemistry on TMA cores.
• Post-mortem analysis: One patient (CRUK0063) co-recruited to the PEACE post-mortem study (NCT03004755); 6 metastatic regions sampled within 24 h of death.
• Validation: Cross-platform validation in 28 patients using the Oncomine Lung cfDNA Assay (ThermoFisher).
• Data deposition: EGA accessions EGAS00001002247 (primary) and EGAS00001002415 (metastatic).

Key findings

• Pre-operative ctDNA detection rate is dominated by histology. ≥2 SNVs detected in 46/96 (48%) NSCLCs pre-operatively: 30/31 (97%) LUSC, 5/7 (71%) other NSCLC, 11/58 (19%) LUAD. Stage I subgroup: 16/17 (94%) LUSC vs 5/39 (13%) LUAD (PMID:28445469).
• Independent predictors of ctDNA detection (multivariable logistic regression): non-adenocarcinoma histology, lymphovascular invasion, and high Ki67 proliferation index. PET FDG-avidity (TBR) predicted ctDNA detection by ROC AUC 0.84 (P<0.001, n=92).
• Driver mutation status does not predict ctDNA in LUAD. Within the LUAD subgroup, KRAS, EGFR, and TP53 driver events were not associated with ctDNA detection (Extended Data Fig. 3e).
• Tumor volume linearly predicts plasma clonal VAF. Tumor volume vs mean clonal VAF: Spearman’s ρ = 0.63, P<0.001, n=37; pathological tumor size ρ = 0.405, P=0.005, n=46. Linear model predicts a 10 cm³ primary tumor → mean clonal plasma VAF of 0.1% (95% CI 0.06–0.18%), corresponding to ~302 million malignant cells.
• Implication for screening sensitivity: A 4 mm low-dose CT-detectable nodule (~0.034 cm³) would correspond to a predicted plasma VAF of 1.8 × 10⁻⁴ % (95% CI 9.8 × 10⁻⁶ to 0.0033%), at or beyond current ctDNA platform detection limits.
• Subclonal SNV detection scales with subclone volume. Shared subclones detected in 35/57 (61%) vs private subclones in 26/80 (33%); subclone volume (CCF × tumor volume × purity) correlated with subclonal plasma VAF.
• Post-operative ctDNA detects relapse with a median 70-day lead-time. ≥2 SNVs detected in 13/14 (93%) confirmed-relapse patients prior to or at clinical relapse vs 1/10 (10%) relapse-free patients (CRUK0013). Lead-time range 10–346 days; 4/13 cases had >6-month lead-times. In two cases ctDNA preceded inconclusive CT imaging by 157 (CRUK0004) and 163 days (CRUK0045).
• ctDNA marks adjuvant chemotherapy resistance in real time. CRUK0080, CRUK0004, and CRUK0062 had detectable ctDNA within 30 days of surgery, with SNV counts increasing through adjuvant chemotherapy — all recurred within 1 year. Conversely, CRUK0013 had 20 SNVs detectable post-op, became undetectable after adjuvant chemoradiotherapy, and remained relapse-free at 688 days.
• Phylogenetic ctDNA tracks the subclone(s) seeding metastasis. In CRUK0063, post-operative detection of an OR5D18 SNV traced back to a subclone private to primary region 3; metastatic biopsy confirmed this subclone gave rise to the para-vertebral metastasis (and contained an ARID1A stop-gain driver only retrospectively detectable at lower threshold in primary region 3).
• Targetable relapse-driving subclones can be identified. CRUK0004 relapse was dominated by a subclone harboring an ERBB2 (HER2) amplification (>15 copies, triploid background) — a potentially actionable target in NSCLC.
• Polyclonal vs monoclonal relapse can be distinguished. Relapses involving subclones from a single phylogenetic branch (CRUK0004, CRUK0063, CRUK0065, CRUK0044) vs multiple branches (CRUK0080, CRUK0062, CRUK0041) were resolvable from plasma.
• Expanded 103-SNV panel detects relapse 189 days earlier. In the post-mortem case CRUK0063, the expanded panel detected ctDNA at day 151 post-surgery — 189 days before the 19-SNV panel.
• Plasma VAF of subclonal clusters mirrors phylogenetic distance from the trunk. In CRUK0063, mean VAFs of clusters 11, 8, 12, 9, and 5 reflected their proximity to the clonal cluster on the M-Seq tree, supporting tiered-burden inference from plasma alone.
• An ancestral subclone with KRAS amplification (>15 copies, triploid background) drove residual lymph-node disease in CRUK0013 that responded to adjuvant chemoradiotherapy.

Genes & alterations

• KRAS — driver SNVs targeted in LUAD assay-panels; not associated with pre-operative ctDNA detection within LUADs. KRAS amplification (>15 copies) found as an ancestral subclonal event driving lymph-node residual disease in CRUK0013, which responded to adjuvant chemoradiotherapy (PMID:28445469).
• EGFR — driver SNVs in LUAD; not associated with pre-operative ctDNA detection within the LUAD subset.
• TP53 — driver SNVs in LUAD; not associated with ctDNA detection within LUADs.
• ERBB2 — focal amplification (>15 copies, triploid background) on a relapse-dominant subclone identified in plasma for patient CRUK0004; flagged by the authors as a potentially targetable relapse driver in NSCLC.
• ARID1A — stop-gain driver SNV in the para-vertebral metastasis of CRUK0063, originally below M-Seq calling threshold in primary tumor region 3 from which the metastatic subclone derived.

Clinical implications

• Post-operative ctDNA as a residual-disease biomarker in early-stage NSCLC, with median 70-day lead-time over CT-confirmed relapse, supports adjuvant trials triggered by molecular-residual-disease detection rather than population-level treatment of all resected stage I–III patients (PMID:28445469).
• Adjuvant chemotherapy resistance can be detected in plasma within 1–6 months of surgery — patients whose ctDNA fails to clear (or whose SNV burden rises) on adjuvant cisplatin/carboplatin plus vinorelbine recurred within 1 year, while ctDNA clearance after adjuvant therapy was associated with sustained remission (CRUK0013, 688 days relapse-free at publication).
• Phylogenetic ctDNA tracking can identify targetable subclonal drivers of relapse, exemplified by an ERBB2-amplified subclone in CRUK0004, opening the prospect of relapse-directed (rather than primary-tumor-directed) targeted therapy.
• Histology must inform ctDNA assay design and clinical interpretation. With LUAD detection rates of only 13–19% for stage I disease using clonal-SNV panels of ~18 markers, LUAD-targeted residual-disease platforms require either expanded panels (28+ SNVs as used in the longitudinal cohort) or orthogonal approaches.
• Cost remains a constraint: the authors estimate ~$1,750 per patient for primary-tumor exome + bespoke panel synthesis + 5 plasma time points.

Limitations & open questions

• LUAD sensitivity ceiling. Even the expanded LUAD panel (median 28 SNVs) may underdetect ctDNA-low adenocarcinomas; the volume-VAF model predicts that small (<1 cm³) LUADs lie below current ctDNA detection thresholds.
• Cohort size for the longitudinal phase is small (n=24): the 93% relapse-detection sensitivity and 70-day lead-time estimates carry wide confidence intervals and require validation in the planned larger TRACERx 421-patient cohort.
• Sampling bias for relapse tissue. Metastatic biopsy was only available for 4 patients plus the CRUK0063 PEACE autopsy; the inferred dominance of monoclonal vs polyclonal relapse may shift with broader sampling.
• Subclone resolution depends on M-Seq density. Subclones present below the M-Seq detection threshold in the primary tumor (e.g., the ARID1A stop-gain in CRUK0063) can only be retrospectively validated; the platform cannot prospectively design assays against truly de-novo metastatic mutations.
• No validation against orthogonal MRD platforms beyond a 28-patient cross-platform comparison with the Oncomine Lung cfDNA hotspot panel.
• Type I error at scale. The 103-SNV expanded panel for CRUK0063 carried a 10.7% per-time-point risk of a single false-positive SNV call, motivating the ≥2-SNV positivity threshold but illustrating that panel expansion is not free.
• Generalizability beyond surgically resected early-stage disease to unresectable, oligometastatic, or non-NSCLC settings remains untested in this report.

Citations from this paper used in the wiki

• “At least two SNVs were detected in ctDNA pre-operatively in 46 of 96 (48%) early-stage NSCLCs… 97% (30/31) of lung squamous cell carcinomas (LUSCs)… compared with 19% (11/58) of lung adenocarcinomas (LUADs).” (Results, p. 2)
• “Multivariable analysis revealed non-adenocarcinoma histology, the presence of lympho-vascular invasion and high Ki67 proliferation index as independent predictors of ctDNA detection.” (Results, p. 3)
• “Tumor volume correlated with mean clonal plasma VAF (Spearman’s Rho = 0.63, P<0.001, n=37)… a primary tumor burden of 10cm3 would result in a mean clonal plasma VAF of 0.1%.” (Results, p. 3)
• “At least 2 SNVs were detected in 13 of 14 (93%) patients with confirmed NSCLC relapse prior to, or at, clinical relapse… the median interval between ctDNA detection and NSCLC relapse confirmed on clinically indicated CT imaging (lead-time) was 70 days (range 10 to 346 days).” (Results, p. 4)
• “The subclone implicated by ctDNA as driving the relapse in the case of CRUK0004 contained an ERRB2 (HER2) amplification event (>15 copies, triploid background), that may be targetable in NSCLC.” (Results, p. 5)
• “This ancestral subclone contained a KRAS amplification (>15 copies, triploid background) and was identified as present in primary tumor and sampled lymph-nodes by M-Seq.” (Results, p. 6, CRUK0013)
• “A limitation to targeted ctDNA profiling is cost, estimated at $1750 per patient for sequencing a single tumor region, synthesis of a patient-specific assay-panel and profiling of five plasma samples.” (Discussion, p. 7)

This page was processed by crosslinker on 2026-05-15.