对调控区域cfDNA WGS的片段组学分析生成基因水平的类表达特征用于乳腺癌亚型分析

Introduction: Analysis of cell-free DNA (cfDNA) from liquid biopsy provides a rapid, repeatable, and non-invasive means to study tumor biology. In breast cancer, invasive tissue biopsies and immunohistochemistry (IHC) remain the standard for determining molecular subtype and treatment selection. However, fragmentomic analysis of cfDNA whole genome sequencing (WGS) can yield biologically meaningful surrogates of tumor transcriptional states through its relationship with nucleosome positioning. Here, we describe a WGS-based fragmentomic pipeline that recapitulates gene-level, expression-like traits from cfDNA. These traits enable the development of fragmentomic classifiers reflective of tissue expression markers such as hormone receptor and HER2. Experimental Procedures: Matched FFPE tumor blocks and plasma samples from 25 consenting patients with breast cancer were analyzed. Plasma cfDNA underwent WGS to 30-40× depth, and matched FFPE RNA was profiled by RNA sequencing. cfDNA reads were processed through a fragmentomic pipeline quantifying fragment size and coverage properties across predefined regulatory regions assigned to nearby genes. Gene-level expression from RNA-seq was analyzed and established breast cancer molecular subtype classifiers were applied. Fragmentomic signals were correlated with RNA expression for the same genes. Samples were split into training (n = 17) and test (n = 8) sets, and features with Pearson correlation > 0.4 were used in an elastic net model to predict ESR1 expression from cfDNA fragmentomic data. Results: The elastic net model's predicted ESR1 score showed a Spearman correlation of 0.99 (p = 0.003) in the training set and 0.64 (p = 0.10) in the test set. Heatmaps of fragmentomic signal at model-selected regions mirrored corresponding RNA-seq expression patterns. Genes identified by the model included DLG5 and MTUS1 (both upregulated in samples with high ESR1 expression), and SMC4 (downregulated in samples with high ESR1 expression), consistent with known ER-associated biology. Summary and Conclusions: These findings demonstrate that cfDNA fragmentomic patterns derived from high-depth WGS can recapitulate gene-level, expression-like traits reflective of tumor biology in breast cancer. Integrating regulatory-region fragmentomics with expression-informed modeling enables non-invasive inference of molecular phenotypes such as estrogen receptor status directly from plasma. The deep WGS data, supported by RNA-seq, could also be leveraged to define target regions to enable gene-level expression fragmentomics in targeted hybrid capture panels. This approach highlights the potential of cfDNA fragmentomics as a surrogate for tissue-based transcriptomic and IHC profiling, supporting development of liquid biopsy-based classifiers for breast cancer subtyping and therapeutic stratification.