henhao
Sign in
Cancer BiologybioRxiv
Preprint

Minimally invasive monitoring of clonal evolution through integrated single cell and ctDNA analysis

Farhia KabeeriD, Matteo Lepur, Branden J. LynchiD, Emilia HurtadoiD, Elena ZaikovaiD, Janine SenziD, Vinci AuiD, Caroline BariliD, Ding Ma, Sheila Nicholson, Gavin HaiD, JESSICA N. MCALPINEiD, Samuel A. J. R. AparicioiD, David D. HuntsmaniD, Alexandre Bouchard‐CôtéiD, Yvette DrewiD, Andrew J. L. RothiD

Abstract

Circulating cell-free DNA (cfDNA) offers a minimally invasive lens into tumor evolution. However, the accurate quantification of clonal composition from cfDNA remains challenging. Existing methods for clone tracking using liquid biopsies are constrained by issues such as their reliance on bulk tissue references, incomplete representations of clonal architecture or limited breadth of mutation panels. To address these limitations, we developed cfClone, a Bayesian framework that integrates single-cell whole-genome sequencing (scWGS) derived clonal structures with cfDNA whole-genome sequencing (cfDNA-WGS) data to enable high-resolution tissue-informed clonal tracking. By jointly modeling local copy-number alterations and haplotype-specific signals via Bayesian model selection and Markov chain Monte Carlo (MCMC) sampling, the algorithm yields uncertainty-aware estimates of clonal prevalence and tumour fraction (TF). Applied to longitudinal clinical cohorts, cfClone can be used to reconstruct evolutionary trajectories and uncovers clonal selection driving therapeutic resistance, including the de novo detection of emergent clonal populations. We demonstrate that cfClone achieves accurate TF estimates and circulating tumor DNA (ctDNA) detection in malignancies with varying degrees of copy-number variant (CNV) burden using semi-synthetic data. We then compare to the state of the art scWGS informed panel based approach, and demonstrate cfClone provides comparable accuracy while allowing for the tracking of more clones and detection of novel clones. Finally, we show how cfClone can be used to quantitatively track clonal dynamics in response to treatment in high grade serous ovarian cancer. Github link: https://github.com/Roth-Lab/cfclone

PDFDOICode
Full-text digest ↓

02 /Specification & Classification

Click any tag to browse matching papers
Citations
0
OpenAlex Topics

In-depth summary

✦ Multi-agent verified full-text synthesis

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 4 (19 Aug 2026) · CC BY 4.0

cfClone, a Bayesian model, deconvolves cfDNA whole-genome sequencing using scWGS-derived haplotype-specific clone copy-number profiles to estimate tumour fraction, clonal prevalence, and flag missed clones.

Tumour-naive cfDNA tools such as ichorCNA estimate tumour fraction but not clonal structure, bulk-tissue references limit clone genotyping resolution, and CloneSeq-SV needs duplex sequencing of bespoke SV/SNV panels that cannot reveal clones unseen by scWGS. cfClone instead combines genome-wide copy-number and allelic signals with scWGS clone profiles to track all scWGS-resolved clones and flag missing ones.

Key results

  1. SuggestedWith scWGS priors, cfClone matched tumour-naive ichorCNA on tumour fraction (TF) in low-aneuploidy FL and beat it in DLBCL and HGSOC semi-synthetic data, and detected ctDNA at ultra-low coverage or very low TF in DLBCL and HGSOC.

    At 30X, TF mean absolute error (MAE) for cfClone vs ichorCNA: 0.243% vs 0.694% (FL), 0.195% vs 0.944% (DLBCL), 0.100% vs 1.57% (HGSOC). ctDNA detected in 60% FL and 100% DLBCL/HGSOC samples at 0.1x and 2% TF; 70% FL and 100% DLBCL/HGSOC at 100X and 0.1% TF. Patient cfDNA in-silico mixtures at 30X: MAE 0.0658% and 0.966%. · Semi-synthetic mixtures of cell populations; Semi-synthetic mixtures of patient derived cfDNA…

  2. SuggestedIn in-silico mixtures of patient-matched FL and DLBCL clones, cfClone recovered clonal prevalence; error fell sharply with coverage for the clone with a distinct haplotype-specific CNV (HSCNV) profile but stayed higher for two DLBCL clones with similar profiles.

    1 FL clone mixed with 2 DLBCL clones; TF MAE 0.300% at 100X. Prevalence MAE at 0.1X (TF 0.1% to 75%): FL-A 6.06%, DLBCL-B 9.00%, DLBCL-C 7.57%; at 100X: 0.496%, 4.25%, 4.15% respectively. · Semi-synthetic mixtures of cell populations

  3. ShownOn MSK-SPECTRUM cfDNA, cfClone tumour fraction correlated strongly with CloneSeq-SV panel estimates, and its clonal prevalence agreed with CloneSeq-SV about as well as CloneSeq-SV's SV-based and SNV-based prevalences agree with each other.

    TF vs duplex-sequenced clonal TP53 VAF: Spearman's ρ = 0.85 (6 patients, 12 samples); vs clonal SV panels ρ = 0.93, SNV panels ρ = 0.96. Clonal prevalence (3 patients; 6 samples for SV): ρ = 0.76 vs SV and 0.89 vs SNV estimates, comparable to CloneSeq-SV internal SV vs SNV ρ = 0.83; all p < 0.01. · Real data validation; Figure legends

  4. SuggestedModelling all four scWGS clones instead of the two panel-resolvable ones shifted inferred clonal composition in SPECTRUM-OV-107, and genome-wide residuals flagged copy-number events from clones missing from scWGS in SPECTRUM-OV-044.

    SPECTRUM-OV-107 at diagnosis: clone A 97.6% (two-clone) vs A 66.5% and B 31.1% (four-clone), Bayes factor (BF) = 0.01 (no preference); after recurrence four-clone model favoured (BF = 36), B 50.9% (95% HDI [39.0%, 62.8%]); CloneSeq-SV SV-based clone D dominance rested on one SV. SPECTRUM-OV-044: missed HSCNV on a chromosome 4 arm, smaller misses on 11 and 13. · Real data validation

  5. SuggestedAcross three BCCA HGSOC patients, serial cfDNA tracking revealed distinct treatment dynamics: rising clonal diversity, an inversion of clonal prevalence, and one treatment-refractory clone that also dominated several peritoneal metastases.

    Case 6608: TF 1.32% to 3.04% after chemotherapy, clone K 45.1% to 19.9%. Case 6607: TF 30.3% pre-surgery, 1.07% after progression on niraparib, with clone P overtaking Q. Case 6606: clone I 70.2% after progression; in bulk WGS clone I was 66.0% (omentum), 78.0% (sigmoid colon), 79.5% (round ligament). · BCCA cohort analysis

Methods, data and reproducibility

Haplotype-specific clone copy-number profiles from DLP+ scWGS (called with HapClone) act as priors in a Bayesian model of 500 kb binned read-depth ratio and B-allele frequency from cfDNA-WGS; MCMC gives posterior tumour fraction and clonal prevalence, Bayes factors test for ctDNA, and residuals flag unmodelled clones.

  • Semi-synthetic cfDNA-like samples simulated from scWGS of follicular lymphoma, patient-matched transformed DLBCL, and HGSOC across coverages and tumour fractions · human (patient tumours, GRCh38-aligned)
  • In-silico mixtures of patient plasma cfDNA (VOA12433P with VOA10703P or VOA8262P1) and bulk tissue (A78136 with VOA7648C or VOA7648A) · human · n = two cfDNA and two bulk-tissue mixture series
  • MSK-SPECTRUM HGSOC cases: cfDNA RDR/BAF, scWGS clone profiles and published CloneSeq-SV duplex SV/SNV estimates, processed with spectrum-ctdna · human · n = 6 patients, 12 samples (TF); 3 patients, 6 samples (prevalence) · Synapse syn66399325 (MSK-SPECTRUM pre-processed data)
  • BCCA retrospective HGSOC cohort (GTB, 2016 to 2023): serial plasma cfDNA-WGS, buffy coat, surgical tumour WGS and DLP+ scWGS from xenograft-derived cells · human · n = 3 patients (cases 6606, 6607, 6608) · EGA EGAS00001006343 (raw scWGS); cfDNA/bulk WGS to EGA upon acceptance
  • Bulk WGS of metastases (diaphragm, cul-de-sac, omentum, sigmoid colon, round ligament) from case 6606, deconvolved with cfClone · human
Methods & tools
cfClone v0.3.1, HapClone v0.1.0 (in-house HSCNV caller), ichorCNA (TF benchmark), CloneSeq-SV (panel-based comparator), DLP+ single-cell WGS, MCMC sampling with Bayes factor model selection, bwa-mem v0.7.19, GATK v4.6.2.0 (MarkDuplicatesSpark, BaseRecalibrator), Eagle v2.4.1 (phasing), KAPA HyperPrep Kit (cfDNA libraries), spectrum-ctdna pipeline, GRCh38 reference genome
Code
https://github.com/Roth-Lab/cfclone
Software & environment
cfClone v0.3.1; HapClone v0.1.0; bwa-mem v0.7.19; GATK v4.6.2.0; Eagle v2.4.1; GRCh38 reference; 500 kb bins with modal quantile regression GC/mappability correction. No container or dependency specification stated.
Reproducibility gaps
Raw cfDNA, buffy coat and bulk tumour WGS will be deposited at EGA only upon acceptance; the in-house scWGS pipeline and HapClone v0.1.0 CNV caller have no stated code location.
Limitations (5)
  • cfClone can flag genomic regions pointing to clones missed by scWGS, but cannot tell whether one or several clones explain them or what their genotypes are; the authors leave this to future work. · stated by the authors
  • Clones with similar HSCNV profiles, as with the two DLBCL clones, are weakly identifiable, so clone-level tracking is expected to suit aneuploid cancers, while low-CNV cancers suit only TF estimation and ctDNA detection. · stated by the authors
  • In mixtures of patient cfDNA, cfClone slightly underestimated tumour fraction when tumour content was very low. · stated by the authors
  • The BCCA application spans only 3 HGSOC patients without an orthogonal clone-level reference, and in case 6606 CA125 rose continuously while cfClone TF stayed stable, a discordance the text does not reconcile. · noted in this summary
  • Real-data clonal prevalence validation relies on 3 MSK-SPECTRUM patients with CloneSeq-SV as an imperfect reference, and reported estimates come from a single chosen MCMC restart ('2' for MSK-SPECTRUM, '0' for BCCA) rather than pooled chains. · noted in this summary

Who should read:Read this if you want to track scWGS-defined tumour clones and tumour fraction in serial cfDNA whole-genome sequencing without patient-specific panels, particularly in aneuploid cancers such as HGSOC.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISFu '25Williams '25Lakatos '21Zill '17Wu '25Oroperv '25
66% semantic matchBioinformatics
Xuecong Fu, Zhicheng Luo, Yueqian Deng+3 more·Bioinformatics·Mar 29, 2025· 2 citations