Integrating Genomic & Transcriptomic Features for Noninvasive Detection, Characterization, and Monitoring of T-Cell Lymphomas

Background: Mature T-cell lymphoma (TCL) outcomes remain inferior to aggressive B-cell counterparts, and thus represent unmet need. Given high relapse rates after remissions induced by current regimens, more sensitive MRD methods are needed to personalize treatment. We tackled these questions by integrating distinct host and viral molecular features for noninvasive detection, risk profiling, and monitoring of diverse mature TCLs.

Methods: We studied 132 patients diagnosed with diverse TCLs (PTCL-NOS n=21, ALCL n=13, TFH lymphoma n=21, ATLL n=54, ENKTL n=6, CTCL n=15, others n=2). 37% of pts had newly diagnosed TCL primarily treated with CHOP-based regimens, while 62% had relapsed/refractory disease treated with salvage chemotherapies (26%), nivolumab (5%), and/or allogeneic HSCT (31%).

We molecularly profiled 530 specimens, including serial blood plasma samples drawn before, during, and at the end-of-therapy (EOT) (n=274), baseline tumor tissues (n=128), and matched normal cells (n=128). Plasma cfDNA for all cases were profiled to detect somatic mutations in 259 recurrently mutated TCL genes by CAPP-Seq, genome-wide CNVs by CANARy (Chabon 2020 Nature), clonotypic VDJ at all TCR/BCR loci by SABER, T/B-cell abundance by QUARTZ (Sugio ASH 2024 abstract #193441), and inferred expression of 381 TCL/immune-related genes by EPIC-Seq (Esfahani Nature biotechnology 2022). Viral load and genotyping for EBV and HTLV-1 were assessed by VirCAPP-Seq (Garofalo ASH2022). Plasma cfRNA expression was profiled using RARE-Seq, by targeting 6.8k coding genes (Nesselbush 2024 Cancer Research).

Results: We found that mutation-based disease burden measurements in cfDNA by CAPP-Seq were significantly correlated with clonotypic cell-free TCR levels (cfTCR; R=0.95, p<0.001) across all patients and plasma HTLV-1 load in ATLL patients (R=0.84, p<0.001).

When considering outcomes after CHOP-based regimens, detection of ctDNA-MRD at EOT predicted significantly higher relapse risk (p= 0.003), inferior PFS (p=0.002) and OS (p=0.012). Among 30 patients with CR by PET, there were 4 patients with negative and all of them had PFS with 1.1 year of median observation time. All nonCR patients were MRD positive at EOT.

Moreover, higher baseline ctDNA levels predicted inferior outcomes in non-ALCL/ATLL PTCLs (relapse=0.003, PFS=0.003) and ATLL (relapse p=0.022, PFS p=0.037). Of interest, we confirmed that 5’ deletions of HTLV-1 genome were associated with significantly higher relapse rate after allo-HSCT (p=0.006).

Separately, when comparing plasma cfDNA vs cfRNA, mutation-based ctDNA burden was significantly correlated with cfRNA tumor burden in PTCL-NOS (R=0.58, n=18), ALCL (R=0.77, n=11), AITL (R=0.50, n=24), and ATLL (R=0.95, n=13) using corresponding gene expression signatures for each entity. However, for cfTCR defined from tumor biopsies, we found significantly higher cfTCR in cfDNA than cfRNA (p<0.001), where 43.2% had MRD exclusively detected in cfDNA, while 4.5% had MRD exclusively detected in cfRNA.

By comparing paired baseline cfDNA & cfRNA samples obtained at time of tumor biopsies, cfDNA had significantly higher overlap in TCR/BCR repertoires with the tumor tissue than cfRNA (TCR p<0.01, BCR p=0.02) or PBMC DNA (TCR p<0.001, BCR p=0.01). Furthermore, expression levels of TCL TME genes (30 marker genes for B cells, and M1&M2 macrophage, and eosinophil in LM22 [Newman, Nat. Methods 2015]) inferred from cfDNA by EPIC-Seq were significantly correlated with RNA expression of tumor tissues in 15 cases (mean R=0.56, PTCL-NOS n=5, TFH n=4, ENKTL n=3, ALCL n=3). Collectively, these results suggest that cfDNA may reflects not only TCL tumor cells, but also TME.

When assessing baseline plasma samples by EPIC-Seq, we found significant clustering of cases based on inferred expression of TME constituents. In patients treated with chemotherapies (n=53), a 'Cold TME' phenotype (low expression of B cells, and myeloid cells, but not T/NK cells) was significantly associated with poor outcome.

Conclusions: Integrated noninvasive profiling of diverse TCLs is feasible and can predict outcomes by capturing clinically relevant tumor burden before therapy, as well as post-treatment MRD. Additionally, cfRNA and viral genome analysis complement cfDNA monitoring results. Importantly, our results suggest that when compared with cfRNA, cfDNA better reflects both tumor burden and tumor immune microenvironments.