7538 Background: Patients with diffuse large B cell lymphoma (DLBCL) exhibit significant differences in clinical outcome based on cell-of-origin (COO). Patients are categorized as having germinal-center-like (GCB) or activated-B-cell-like (ABC) disease based on RNA microarray and histopathological analyses of tumor biopsies. We recently described an accurate sequencing-based method for determination of COO in DLBCL utilizing stereotyped differences in mutations (Scherer et al., 2016). Here, we further explore the mutational patterns in patients with differing molecular subtypes of DLBCL based on sequencing of circulating tumor DNA. Methods: We applied cancer personalized profiling by deep sequencing (CAPP-Seq) to pretreatment plasma samples and matched germline from a cohort of 115 patients with DLBCL. We then identified somatic alterations, which were used to determine COO molecular subtypes as previously described. Finally, we compared mutational patterns in patients with GCB and non-GCB DLBCL. Results: We detected a significantly greater number of total mutations (GCB: 1766 ± 160 mutations per Mb of targeted sequencing; non-GCB: 1364 ± 150 mutations per Mb of targeted sequencing; p < 0.05) and coding mutations (GCB: 145 ± 21 mutations per Mb of targeted sequencing; non-GCB: 28 ± 8.5 mutations per Mb of targeted sequencing; p < 0.001), particularly in immunoglobulin (Ig) regions (p < 0.05). In addition, GCB and non-GCB samples exhibited distinct mutational patterns within Ig regions. GCB samples were enriched for mutations in regions of switch mu (Sμ) (p < 0.01) and IGHV2-70 (p < 0.01), while non-GCB samples were enriched for mutations in regions of IGHG3 (p < 0.03), IGHV4-34 (p < 0.03), and IGLL5 (p < 0.05). GCB samples were also significantly enriched for coding mutations in SOCS1 (p < 0.01), a gene not included in our original COO classifier. Conclusions: Patients with GCB and non-GCB DLBCL exhibit distinct mutational patterns across both Ig and non-Ig loci of the genome. These differences in mutational patterns can be used to classify molecular subtypes noninvasively, potentially providing further utility to noninvasive genotyping and liquid biopsies.