Abstract T105

Longitudinal circulating tumor DNA sequencing may predict the response to PD1 blockade therapy in relapsed/refractory classical Hodgkin Lymphoma patients

Background: The introduction of immune checkpoint inhibitors (CPIs) has revolutionized the treatment of relapsed/refractory (R/R) classical Hodgkin lymphoma(cHL). However, nearly 65% fail to respond or progress after an initial response within 24 months. The evaluation of tumor-specific biomarkers of response currently requires invasive procedures, does not capture spatial tumor heterogeneity, and is not suitable for repeated evaluations. On the contrary, cell-free DNA sequencing represents a non-invasive tool for genotyping and response monitoring of several solid and hematological neoplasms. Here, we performed baseline and longitudinal liquid biopsies in 40 R/R cHL patients to identify biomarkers of response to CPIs

Methods: Peripheral blood samples were collected before treatment initiation, at each metabolic response assessment, and at the end of treatment. A targeted re-sequencing panel including the coding exons, splice sites, and Aberrant Somatic Hyper Mutation (ASHM) regions of 133 genes was designed for Cancer Personalized Profiling by deep Sequencing (CAPP-Seq). The target sequencing was performed in paired-end runs on the Nextseq 550 platform (Illumina), allowing > 2000x coverage.

Results: Baseline circulating tumor DNA (ctDNA) load positively correlated with the Total Metabolic Tumor Volume (Spearman coefficient= 0.67, p= .00003). Patients with higher levels of ctDNA showed lower overall response rates (65% vs. 100%, p= .038) and shorter Event-Free Survival [EFS] (HR 2.0, 95%CI [0.5-3.4], p= .009). TP53 mutation emerged as the only significant pre-treatment ctDNA mutation associated with a worse EFS (HR, 3.04; P = .03). Interestingly, after four cycles of treatment, a 1-Log reduction of the ctDNA load was associated with longer EFS (HR 0.33, 95%CI: 0.13 to 0.82, p= .02). Concomitantly, the persistence of an increased percentage of baseline variants was consistently detected in CPIs non-responsive vs responsive patients (median 0 vs 90%, P< .0001). We performed ROC analysis to assess the response classification performance of baseline ctDNA load, dynamic load reduction, and persistence of Non-Synonymous Variants. Of note, the latter feature yielded the best accuracy with an AUC of 0.95 (DeLong test p<.05).

Conclusions: Taken together, these findings highlight the predictive role of baseline and longitudinal ctDNA sequencing in the early identification of R/R cHL patients at high risk of failing CPIs.