Distinct Chromatin Accessibility Profiles of Lymphoma Subtypes Revealed By Targeted Cell Free DNA Profiling

• Published in: Blood Vol. 132; no. Supplement 1; p. 672
• Main Authors: Mehrmohamadi, Mahya, Esfahani, Mohammad S, Soo, Joanne, Scherer, Florian, Schroers-Martin, Joseph G, Chen, Binbin, Kurtz, David M., Hamilton, Emily, Liu, Chih Long, Diehn, Maximilian, Alizadeh, Ash A.
• Format: Journal Article
• Language: English
• Published: Elsevier Inc 29.11.2018
• ISSN: 0006-4971; 1528-0020
• DOI: 10.1182/blood-2018-99-119361

Background Diffuse large B-cell lymphomas (DLBCL) can be divided into subtypes that relate to their cell-of-origin: germinal center B-cell (GCB) and activated B-cell (ABC). These clinically distinct subtypes were originally discovered based on their unique transcriptional signatures, but they also harbor distinct genetic aberrations and epigenetic profiles. We previously showed the utility of genetic mutations in circulating tumor DNA (ctDNA) for DLBCL subtyping using cancer personalized profiling by deep sequencing (CAPP-Seq) [Scherer et al., Sci Transl Med 2016, Newman et al., Nat Med 2014]. Here, we assess epigenetic information encoded in cell-free DNA (cfDNA) for addressing this distinction. Methods We extended recent observations revealing nucleosome depletion at transcription start sites (TSS) of highly expressed genes within cfDNA [Snyder et al. 2016 Cell and Ulz et al. 2016 Nat Genetics]. Specifically, to overcome the inherent sensitivity limitations of cfDNA whole genome sequencing (WGS), we focused on 32 genes differentially expressed between DLBCL subtypes (as well as 70 control genes). We targeted 2 kb regions flanking each TSS for these genes by CAPP-Seq [Newman et al. 2014 Nature Med]. Plasma samples were collected from 41 individuals and cfDNA subjected to ultra-deep sequencing (~4000x) by CAPP-Seq. DLBCL cases were labeled as either ABC or GCB based on gene expression in tumor biopsies. Results We observed significant heterogeneity in the TSS profiles of individual genes. Despite this variation, many genes in our targeted panel exhibited discriminatory profiles. For example, CD20 (MS4A1), reliably distinguished DLBCL cases from control subjects in nucleosome depletion at the TSS as expected (p= 1.5e-05, Fig1a). Remarkably, when considering 32 genes differentially expressed between DLBCL subtypes, we observed significant differences in their TSS profiles in plasma from patients with GCB vs ABC DLBCL (p=0.0002, Fig 1b). For the majority of the subtype-specific genes, the pattern of nucleosome depletion at the TSS was consistent with expected expression (Fig 1b). Specifically, genes typically over-expressed in GCB DLBCL (e.g., CD10/MME, LMO2, SERPINA9), had significantly more TSS nucleosome depletion in cfDNA from GCB-like DLBCL patients. Conversely, we observed significantly more nucleosome depletion in cfDNA of ABC DLBCL patients at the TSS of genes known to be transcriptionally more active in ABC DLBCL (e.g., IRF4, PIM1, CCND2 and IL16). When aggregating the epigenetic profiles across these genes into a single cell-of-origin score, we observed significant stratification of patients with distinct DLBCL subtypes (Fig 1c, p=0.008). Across the cohort, this epigenetic cell-of-origin score from cfDNA showed significant correlation (Spearman rho=0.67, p=0.05) with a score derived from somatic mutations genotyped from ctDNA [Scherer et al. 2016 Sci Transl Med]. Conclusions Lymphoma subtypes exhibit distinct chromatin accessibility profiles that can be captured by targeted cell free DNA profiling, and that can serve as surrogates for expected gene expression differences. These epigenetic features can be used to noninvasively classify DLBCL molecular subtypes using blood plasma and should be readily extensible to other cancer classification problems. Extension of this approach to other epigenetic features could serve to further refine potential applications. Figure legend. Figure 1. (A) Sequencing coverage around the TSS of the B-cell antigen CD20 (MS4A1) is shown for three representative DLBCL patients (red) and three normal controls (gray). The boxplot shows summary of normalized depth across all samples (n=10 DLBCL, n=31 control). (B) The y-axis shows 32 genes with known ABC-favoring (blue) and GCB-favoring (orange) expression. The x-axis shows a delta score defined as log2( mean normalized depth across GCB samples - mean normalized depth across GCB samples). (C) Boxplot of the GCBness score obtained by subtracting the overall depth around the TSSs of a set of GCB-genes from the same metric for a set of ABC-genes. No relevant conflicts of interest to declare.