Kan Etoh

Probe information used for deep sequencing

Multimodal integration model for predicting breast cancer recurrence. A, Strategy for integrating multimodal cfDNA features using a LASSO logistic regression model. B, Comparison of ROC curves from multimodal analysis in the training set (N = 138; pBC = 99, mBC = 39) and the testing set consisting of patient-matched samples (N = 12; pBC = 6, mBC = 6). The 95% confidential intervals for the AUC are shown. C, Comparison of recurrence score in six patient-matched primary (pBC) and metastatic (mBC) samples.

Figure S7. Quantification of cfDNA coverage across 26 genomic loci and nucleosome scores in the HER2-positive subset.

Patients’ characteristics and clinical information.

Sample characteristics and clinical information.

Cell-free DNA (cfDNA) offers a minimally invasive approach to capture genomic and epigenetic dynamics during cancer progression. We performed targeted sequencing of 26 gene loci transcriptionally regulated during the acquisition of therapy resistance in breast cancer and analyzed blood-derived cfDNA from 150 breast cancer samples (105 primary and 45 recurrent). Recurrent samples exhibited increased genomic variant counts in both coding and noncoding regions, accompanied by shorter cfDNA fragment lengths. Furthermore, cfDNA fragmentation profiles were variable in recurrent samples, with frequently amplified loci such as ERBB2 and concurrent reductions at loci, including RERE and SYNPO2. Notably, nucleosome occupancy–derived scores from RERE and SYNPO2 distinguished recurrent from primary cancer with high accuracy (area under the curve = 0.826). Using a machine-learning approach, integration of these cfDNA features accurately predicted breast cancer relapse. Collectively, these findings demonstrate that cfDNA-based profiling focused on transcriptional alterations provides a sensitive strategy for detecting breast cancer recurrence.Significance:cfDNA-based (epi)genomic profiling captures transcriptionally regulated chromatin and nucleosome remodeling during the acquisition of therapy resistance and relapse, enabling minimally invasive, mechanistically informed detection of breast cancer recurrence. Targeting transcriptionally relevant genomic loci provide clinically actionable biomarkers to monitor therapy resistance and guide precision treatment decisions in real time.

Distribution of fragment lengths and cfDNA coverage across 26 genomic loci. A, Fragment length frequency distribution of genomic (top) and mitochondrial (bottom) fraction from primary (gray) and recurrent (pink) samples. B, Comparison of fragment ratios of targeted genes in relation to CNA, calculated as the ratio of fragments shorter than 140 bp to peak fragment counts, in primary (N = 105; gray) and recurrent (N = 45; pink) samples. C, Quantification of cfDNA coverage across the 26 genomic loci. D, Comparison of cfDNA coverage of targeted genes between primary and recurrent/metastatic samples. Asterisks denote statistical significance with Welch’s t test. *, P < 0.05; **, P < 0.005; ***, P < 0.0005; ****, P < 0.0001 for the indicated comparison. E, Comparison of the SD of CNAs between primary (N = 105; gray) and recurrent/metastatic (N = 45; pink) samples. F, Relationship between CNAs and ctDNA aneuploidy fraction assessed by Spearman’s correlation in primary (N = 105; pink) and recurrent (N = 45; blue) samples.

Figure S6. Oncoprint analysis of the 26 targeted loci in clinical breast cancer samples from the TCGA study.

WPS classification by log2 fold change between recurrent and primary samples

Somatic variants analyzed using the maftools