Lymphoma Genomic pro fi ling reveals spatial intra-tumor heterogeneity in follicular lymphoma

Follicular lymphoma (FL) is an incurable B-cell malignancy characterized by advanced stage disease and a heterogeneous clinical course, with high-risk groups including those that transform to an aggressive lymphoma, or progress early (within 2 years) following treatment. Recent sequencing studies have established the diverse genomic landscape and the temporal clonal dynamics of FL [1–7]; however, our understanding of the degree of spatial or intra-tumor heterogeneity (ITH) that exists within an individual patient is limited. In contrast, multi-site profiling in solid organ malignancies has demonstrated profound ITH impacting mechanisms of drug resistance and compromising precision-medicine-based strategies to care [8]. In FL, the rise in trials adopting targeted therapies such as EZH2, PI3K, and BTK inhibitors reflects this paradigm shift in cancer care and with the development of biomarker-driven studies highlights the need to accurately define genomic alterations with clinical relevance. As most FL patients manifest disseminated tumor involvement, we sought to uncover the extent and clinical importance of spatial heterogeneity in FL by using a combination of whole-exome and targeted deep sequencing (Supplementary methods). Our study cohort comprised nine patients (SP1–SP9) each with two spatially separated synchronous biopsies including two patients (SP3 and SP4) with spatial samples at two timepoints (FL and transformation), yielding a total of 22 tumor samples (Table S1). To improve the sensitivity for variant detection, fluorescence-activated cell sorting (FACS) was performed on cell suspensions where available (15 of 22 tumors) (Supplementary methods and Tables S2, S3). Exome sequencing of both the tumor and paired germline DNA was performed (median depth 131×) (Table S3) and we identified between 35 and 130 nonsynonymous somatic variants (SNVs) per sample corresponding to 659 coding genes comprising missense (81%), indels (10%), nonsense (7%), and splice site (2%) changes (Tables S4, S5). We verified 195/198 (98%) SNVs using an orthogonal platform (Haloplex HS), with a high concordance of variant allele frequencies (VAFs) (r= 0.91) (Table S6). The tumor purity was predicted across samples using the mclust algorithm (Supplementary methods and Figure S1), demonstrating a mean purity of 92% in FACSsorted samples and 66% in non-sorted samples. Although the spatially separated tumors shared identical BCL2-IGH breakpoints, we observed variable degrees of ITH, with on average 82% (range 50–99%) of variants shared between sites. To quantify this heterogeneity, we calculated the Jaccard Similarity Coefficient (JSC) [9] for each patient, which represents the ratio of shared to total These authors contributed equally: Shamzah Araf, Jun Wang.