ctDNA MRD after CAR-T Open Access

In this issue of Blood Advances, Krasnow et al¹ report on a novel method of circulating tumor DNA (ctDNA) detection, termed pooled minor allele enriched sequencing through recognition oligonucleotides (MAESTRO-Pool), to assess measurable residual disease (MRD) following axicabtagene ciloleucel (axi-cel) chimeric antigen receptor T-cell (CAR-T) therapy for relapsed/refractory large B-cell lymphoma (R/R LBCL) at a single institute. This important study highlights the rapid advance of ctDNA MRD in LBCL and identifies important caveats and further tracks for therapeutic integration of such assays into routine practice.

Despite the established efficacy and widespread adoption of CAR-T therapy for R/R LBCL, half or more patients relapse or progress at some point following therapy, including approximately a third of patients in complete metabolic remission (CMR) on the initial day-28 positron emission tomography–computed tomography (PET-CT) scan.² Overt progression after CAR-T therapy carries a dismal prognosis.^3,4 Predicting subclinical residual disease or relapse by liquid biopsies of ctDNA would enable prognostication, closer follow-up, and/or early institution of salvage therapies. However, developing a cost-effective, high-sensitivity assay that functions at low levels of MRD while minimizing false positives is complicated by variable levels of shed ctDNA and inter- and intrapatient genetic heterogeneity in LBCL.⁵ 

MAESTRO-Pool is a mutation-enrichment approach that uses individualized probe sets to track cancer-specific, single-nucleotide variants (SNVs) identified by concomitant whole-genome sequencing (WGS) of patients’ somatic and cancer genomes. This approach is purported to improve specificity, sensitivity, and possibly affordability.^6,7 Krasnow et al report that early MRD negativity by MAESTRO-Pool is associated with favorable long-term outcomes.¹ In the analyzed cohort of 28 patients, including 15 responders, defined as complete (CR) or partial response (PR) at 12 months, and 13 nonresponders, 3 of whom achieved an initial PR and then relapsed before 12 months, 73% of responders demonstrated undetectable ctDNA within 4 weeks of treatment. The trajectory of a decrease in MRD as early as 2 and 4 weeks, as opposed to absolute decrease in ctDNA, was most predictive of outcome. Using a receiver operating characteristic curve, the log-fold change at 2 weeks demonstrated an impressive area under the curve of 0.97, albeit in a small patient cohort. Fascinatingly, the responders with late conversions to CR after an initial PR tended to show an equivalent trajectory as early responders, whereas late relapsers displayed a trajectory reminiscent of never responders.

Currently, the 3 most common commercially available approaches to ctDNA for LBCL are (1) identification and tracking of clonal immunoglobulin heavy chain (IGHV) gene rearrangements⁸; (2) identification of baseline variants from whole-exome sequencing (WES) or WGS, followed by tracking of mutations by duplex sequencing⁹; or (3) identification and tracking of multiple mutations on the same strand of ctDNA (ie, phased variants [PVs]) from a hybrid-capture panel that target genomic loci that are stereotypically mutated via somatic hypermutation.¹⁰ All methods require balancing breadth and depth against accuracy and efficiency of sequencing. Breadth can be improved with larger panels that analyze more mutated genes, but this limits sequencing depth because of costs and sequencing chip capacities. Duplex sequencing from panels may improve accuracy in detecting low levels of mutated tumor ctDNA in settings of excess normal ctDNA, but efficiency may be compromised by the requirement for intact complementary strands. Using PVs negate the need for complementary strands, thereby improving accuracy and efficiency without compromising depth or breadth, which leads to the detection of mutated DNA at concentrations as low as parts per million.¹¹ However, PV identification requires relatively long (170 bp) intact pieces of DNA, which may be compromised by transport or shearing forces. Furthermore, commercial approaches that use predefined panels may not cover all patients.¹⁰ 

MAESTRO-Pool addresses some of these limitations by analyzing short DNA pieces and using individualized panels. The pooling of samples amortizes the initial WES/WGS costs and provides internal negative controls against which to benchmark the detection of mutated ctDNA. A dynamic, probabilistic in silico MRD caller that accounts for ctDNA recovery and number of mutated SNVs on a per patient basis has been reported to further enhance the accuracy.⁶ However, the generation of individualized probe sets may not be achievable for all patients. Krasnow et al report that there were too few cancer-specific variants to proceed with MAESTRO-Pool for 3 of the original 35 recruited patients. Furthermore, the creation of probe sets and use of the dynamic MRD caller may also be difficult to validate across laboratory quality authorities, thereby limiting multisite use.

Several other ctDNA MRD methods have been studied in the post–CAR-T setting, although these were not compared directly. In a pivotal study, Frank et al¹² tracked ctDNA after axi-cel therapy using IGHV ctDNA sequencing. At day 28 after CAR-T infusion, patients with detectable ctDNA had a median progression-free survival (PFS) of 3 months and an overall survival (OS) of 19 months, whereas the median PFS and OS were not reached among those with undetectable ctDNA. The predictive value was greater in patients with stable disease or with PR at the day-28 PET-CT; 1 of 10 patients with concurrently undetectable ctDNA relapsed. In contrast, 15 of 17 patients with concurrently detectable ctDNA relapsed. Subsequently, Stepan et al¹³ evaluated the utility of PVs for the early prediction of outcome following lisocabtagene maraleucel using samples from the pivotal phase 3 TRANCEND study. Durably responding patients achieved undetectable ctDNA as early as day 15 after infusion, whereas residual detectable ctDNA was highly associated with subsequent relapse, even among patients in CMR as determined by PET-CT.

The most valuable applications of ctDNA testing following CAR-T infusion may be as an early biomarker of long-term outcome in routine practice, as a surrogate end point for clinical trials, and to facilitate early therapeutic interventions to eradicate MRD, such as in the EpLCART study of bispecific antibody therapy for patients who remain ctDNA MRD–positive (ClinicalTrials.gov identifier: NCT06414148).¹⁴ In the upfront setting, there are currently open trials that implement early CAR-T therapy to address MRD by ctDNA following standard frontline therapy.¹⁵ 

The work by Krasnow et al is valuable in presenting the capabilities of ctDNA MRD following CAR-T therapy. Further studies will be needed to validate these exciting data and to externally benchmark it against other methodologies. We are still in the early days of clinical trials using such approaches, and ongoing improvements in ctDNA testing strategies, performance characteristics, and laboratory workflows are likely to further improve the feasibility of ctDNA MRD in real-world settings.

Conflict-of-interest disclosure: S.F. reports having patents and receiving licensing fees from Johnson & Johnson for patents for improving the efficacy of chimeric antigen receptor T-cell therapy; is the chair of the clinical advisory board for Arovella Therapeutics; and reports honoraria from Kite/Gilead. M.R.D. reports honoraria from Kite/Gilead and Novartis; research funding from Kite/Gilead, AbbVie, and Johnson & Johnson; and royalties from AbbVie.