Inferred Gene Expression By Cell-Free DNA Profiling Allows Noninvasive Lymphoma Classification

Background: Current diagnostic criteria for mature B-cell tumors require a combination of histopathological, immunophenotypic, cytogenetic, and genetic evaluations on tissue biopsy specimens (Campo, Blood 2022; Alaggio, Leuk 2022). Despite this battery of techniques, substantial heterogeneity remains in specific subgroups that can be further resolved using additional molecular diagnostic methods, including gene expression profiling (GEP). For example, among aggressive B-cell tumors, a subset recognized as high-grade B-cell lymphomas can be identified as harboring MYC+BCL2 lesions by FISH (HGBCL-DH-BCL2). However, a significant remaining high-risk subset also have a characteristic tumor gene expression “double-hit signature” or “dark zone signature” (DZsig) (Ennishi, JCO 2018; Alduaij, Blood 2023).

Yet, inadequate tissue specimens can limit the application of these techniques for some patients. Separately, the invasive nature of tissue biopsies and their attendant risks can preclude adequate diagnostic evaluations in some patients, and logistical barriers for obtaining surgical tissue specimens can result in diagnostic delays. To address these unmet needs, we tested the performance of a novel noninvasive strategy to detect and classify mature B-cell tumors using plasma cfDNA, with special attention to HGBCL-DH-BCL2 and DZsig.

Methods: We used EPIC-Seq to infer expression levels for genes of interest, using cfDNA fragmentomic signals at their transcription start sites (Esfahani, Nat Biotech 2022). We first curated and catalogued genes from prior GEP studies (n=56) of >10k diverse lymphoid tumors profiled by various RNA techniques. We systematically prioritized genes recurrently identified in multiple studies of tumor-specific expression profiles as compared with other tumor types and normal tissues. We included canonical markers used for immunodiagnosis by IHC and FACS, as well as key lineage and differentiation markers of normal B and T cells, and other leukocyte subsets. Finally, we also included recurrently mutated genomic regions and fusion hotspots for quantifying ctDNA levels using variant allelic fractions (VAF). To validate performance of DZsig in FFPE for HGBCL-DH-BCL2, we analyzed 95 FFPE biopsies by FISH (20% HGBCL-DH-BCL2), IHC, and RNA-seq. We then analyzed 36 corresponding pretreatment cfDNA samples (28% HGBCL-DH-BCL2) and used EPIC-Seq to infer gene expression.

Results: Starting from our curated catalogue, we designed a targeted lymphoid EPIC-Seq panel (1676 genes; 2.6 MB) to resolve 11 major mature lymphoid malignancies and their clinically relevant molecular subtypes (cHL, DLBCL, HGBCL, FL, MCL, BL, MZL, PMBCL, SLL/CLL, PTCL, and MM). We next applied this EPIC-Seq panel to profile cfDNA samples from 207 lymphoma patients, including cHL (n=113), DLBCL (n=66), FL (n=15), and MCL (n=13) as well as healthy adults to assess specificity. We tested histology specific signatures derived from several existing RNA GEP datasets for cHL from sorted HRS cells (~85k by scRNA-Seq) and for DLBCL (n=959), FL (n=635), and MCL (n=100) from bulk RNA-Seq. Using this approach, we found high correlations between EPIC-Seq signature scores from cfDNA and corresponding circulating tumor VAFs in each histology [cHL (R_p=0.83), DLBCL (R_p=0.83), FL (R_p=0.87), MCL (R_p=0.81); Fig A]. In addition, when measuring these tumor derived signatures in cfDNA, each histology was significantly distinguishable from healthy controls (AUCs: cHL 0.94, DLBCL 0.91, FL 0.77, MCL 0.96).

We next evaluated the DZsig when measured in LBCL patients and comparing paired tumor and cfDNA samples. We first confirmed the performance of the DZsig for detection of HGBCL-DH-BCL2 in 95 FFPE tumors by RNA-Seq (p<0.0001, AUC 0.85). We then noninvasively measured the DZsig in plasma cfDNA by EPIC-Seq (cfDZsig) and found significant discrimination of HGBCL-DH-BCL2 and DLBCL (p=0.0088, AUC 0.78, Fig B).

Conclusions: These results confirm that inferred gene expression by cfDNA profiling allows noninvasive diagnosis of multiple lymphoma subtypes, including classification of challenging diagnostic entities such as HGBCL and DZsig+ tumors. Our ability to design and validate a pan-lymphoid gene panel for detection and classification of several lymphoma subtypes suggests the promise of this approach for other diverse tumors as well as the refinement of cfDZsig using a larger cohort.