Cracking the POD24 code in follicular lymphoma Free

In this issue of Blood, Claudel et al¹ investigate the utility of end-of-induction (EOI) positron emission tomography/computed tomography (PET/CT) combined with circulating tumor DNA (ctDNA) as a composite biomarker for the prediction of progression of disease within 24 months (POD24) in patients with follicular lymphoma (FL), the most common indolent non-Hodgkin lymphoma worldwide.

FL is a biologically and clinically heterogeneous disease. Although most patients respond well to first-line immunochemotherapy, those who experience early progression (the POD24 group) have significantly inferior outcomes, underscoring the clear need for improved therapeutic concepts in this high-risk population.² Yet, the prediction of POD24 remains a major challenge in clinical practice, because existing tools that are based on clinical or genetic features have demonstrated suboptimal predictive value (see table). For example, although clinical prognostic scores such as the FL International Prognostic Index (FLIPI) and integrated clinicogenetic risk models such as the m7-FLIPI yield relatively high negative predictive values for POD24 at baseline, their positive predictive values (PPVs) remain <50%, limiting their use for guiding risk-adapted up-front treatment decisions (see table).^3,4 

Claudel et al address this problem by evaluating a novel combined biomarker approach that leverages PET/CT response and ctDNA detection at the EOI time point to improve POD24 prediction. This correlative biomarker analysis was conducted in 141 patients with FL enrolled in the RELEVANCE trial, which compared rituximab plus lenalidomide (R2) with rituximab chemotherapy as induction treatment, followed by R2 or rituximab maintenance.⁷ PET/CT response was assessed using the Deauville criteria, and ctDNA positivity in EOI serum was defined by the detection of at least 1 triplet-phased variant in the Switch μ region of the immunoglobulin gene or 1 doublet-phased variant in any other genomic locus.⁸ The key finding of the study is that the composite risk model markedly improved PPV for POD24 to 86%, outperforming the predictive performance of either modality alone¹ (see table).

The observed improvement of PPV might open avenues for future trial designs that could enable personalized clinical management after first-line induction treatment. Such trials would evaluate whether preemptive initiation of second-line therapy in patients identified as high risk for POD24 confers a survival advantage compared with initiation at the time of clinical progression. Of particularly interest in this clinical setting is certainly the use of chimeric antigen receptor (CAR) T-cell therapies or bispecific monoclonal antibodies (BsA), both of which have demonstrated high remission rates and durable responses in relapsed/refractory FL, including patients with POD24.⁹ However, the feasibility and rationale of such trials remain rather uncertain, mainly for 2 reasons: first, the study by Claudel et al demonstrates that only ∼6% of patients (7/124 with ctDNA⁺/PET⁺ profiles) would meet the eligibility criteria, implying that advanced ctDNA testing, which is both technically complex and costly, would have to be performed in most patients (∼94%) who ultimately do not qualify for inclusion. Second, given the favorable efficacy profiles observed in CAR T-cell and BsA trials in patients with POD24, demonstrating superiority of earlier intervention over standard timing, may be challenging and would likely necessitate large multicenter cohorts to achieve sufficient statistical power.⁹ 

The findings and conclusions presented in this article should be interpreted with caution, given the inherent risk of overfitting commonly associated with correlative biomarker studies and the absence of independent validation. Furthermore, the study presents 2 key limitations related to the ctDNA analysis, of which 1 has greater implications for the assay’s performance. First, serum was used as the primary analyte for ctDNA profiling, which generally contains an excess of contaminating leukocyte-derived DNA resulting in lower ctDNA allelic fractions (AFs) and, therefore, inaccurate estimation of tumor burden.¹⁰ Although this may have limited the assay’s sensitivity to a certain extent, it likely did not fundamentally affect its general performance, because ctDNA detection was based on absolute molecule counts rather than AF thresholds. Nevertheless, plasma is strongly recommended and widely accepted as the preferred analyte for liquid biopsy analyses. Second, the median number of phased variants detected at FL diagnosis was 47, which is substantially lower than what has been reported in comparable studies of germinal center B-cell non-Hodgkin lymphomas.⁸ This can likely be attributed to the panel design and imposes limitations on accurate longitudinal tracking of phased variants; an issue reflected in the relatively low sensitivity of 46% for predicting POD24 with the combined method (7/13 patients missed) and 40% with ctDNA monitoring alone (9/15 patients; see table).

Collectively, Claudel et al present a strong and novel biomarker model that provides an encouraging rationale for personalized and risk-adapted treatment strategies in FL. Yet, several further steps and improvements are warranted. These include independent validation of the composite biomarker approach in external cohorts and an optimization of the phased variant technology to enhance sensitivity by expanding the coverage of genomic regions affected by (aberrant) somatic hypermutation (eg, IGH gene V-regions, IGK/IGL). Moreover, although PET/CT is widely available as a standard imaging modality in FL, harmonization efforts for ctDNA detection and interpretation will be essential to ensure clinical utility across centers.

Conflict-of-interest disclosure: F.S. reports research funding from Roche Sequencing Solutions, Gilead, and Takeda and honoraria from AstraZeneca and Servier.