Abstract
Introduction
As flow cytometric MRD is increasingly used to tailor treatment according to the needs of individual CLL patients (pts), full standardization and increased sensitivity becomes mandatory. Having developed an automated 2-tube 8-color flow cytometric approach with a sensitivity of 10-4(Engelmann et al., AJH, 2025), the EuroFlow consortium now designed a 12-color panel and algorithm-driven operator-independent analysis strategy reaching sensitivities of at least 10-5 (CLL next generation flow, CLL NGF).
Methods
Based upon our 8-color panel, we tested multiple 12-color approaches and obtained an optimized single tube 13-antibody combination: CD27/CD180/CD79b/CD19/ROR1/IgK/IgL/CD5/CD81+CD38/ CD43/CD3/CD45. Globally, 415 samples stained with that combination were acquired at 9 centers using standardized EuroFlow methods. Pretherapeutic samples from 120 CLL pts and 23 normal peripheral blood (NPB) samples were used to construct a data base representing the 12-dimensional flow cytometric space in 2 dimensions (2-dimensional representation of robust Mahalanobis' distance, 2D-RC) while preserving the immunophenotypic differences between CLL and normal cells. NGF data analysis of samples with unknown MRD content was performed without operator intervention. We first generated clusters of highly similar cellular events in the multidimensional immunophenotypic space as previously described (Pedreira et al., JIM 2019). Next, clusters were designated as CLL if they fell within 1.5 standard deviations of the CLL immunophenotype of the 2D-RC data base (i.e. presented with a CLL immunophenotype) plus showed an immunoglobulin (Ig) light chain restriction (Ig kappa/Ig lambda < 0.3 or > 6) or the underexpression of both Ig light chains (< 1000 units Ig kappa, < 750 units Ig lambda). To evaluate NGF performance, we diluted CLL cells from 26 patients in NPB targeting at MRD levels of 10-3, 10-4, 10-5, 2x10-6 and 0 (total of 123 artificial MRD samples). In addition, 139 real MRD samples from patients treated with state-of-the-art therapies and 10 NPB samples were used to evaluate the approach. All real and artificial MRD samples were blinded and independently assessed by 4 internationally renowned experts in the field. Individual experts´ results were subsequently combined into an expert consensus (EC: 3/4 experts positive rates required for an overall positive MRD result, MRD level=mean of positive results). Samples with sufficient DNA available (n=151) were additionally assessed for MRD by IGH NGS as described by Hengeveld et al. (Blood, 2023). Results
We acquired a median of 5,752,261 (IQR: 4,554,181 to 6,711,469) nucleated cells per test sample. Using 31 NPB samples, limits of blank were determined as 2x10-6 (NGF) and 10-6 (EC).
We first evaluated different MRD approaches vs. expected values from dilution series. NGF achieved an overall accuracy of 93% in 123 artificial MRD samples (45 samples true negative, TN; 70 true positive, TP; 8 false negative, FN, 7%). The expected MRD level in FN samples was low (mean 6x10-6). Overall, NGF showed an excellent correlation to expected (R=0.97, p < 0.0001). Accurate scoring with respect to 10-4 was achieved in 94%, with respect to 10-5 in 94%, and with respect to 2x10-6 in 92% of all samples. EC was moderately less accurate (overall 89%, 3FP, 10 FN, 42 TP, 68 TP). The combined expert result was correct for the 10-4 threshold in 94%, for 10-5 in 91%, and for 2x10-6 in 88% of all samples. NGS results were obtained in 71 samples and yielded an overall accuracy of 88% (FP 3, FN 5, TP 63). Quantitatively, the NGS results were correct at 10-4 in 82%, at 10-5 in 80%, and at 2x 10-6 in 87% of all samples.
We next tested our approach in 139 real MRD samples (114 PB, 25 BM). Compared to the consensus of 4 experts, NGF was accurate in 95% of all samples (FP 4, FN 3, TN 21, TP 111) with good correlation (R=0.96, p < 0.0001). We found NGF to be accurate in 97%, 95%, and 95% of all cases at thresholds of 10-4, 10-5, and 2x10-6, respectively. NGS (80 samples analyzed) showed a similar agreement to the EC by flow: an overall accuracy of 92% and an agreement between the methods of 94% at 10-4, 90% at 10-5 and 91% at 2x10-6 thresholds.
ConclusionCLL NGF is the first fully operator-independent flow cytometric method for MRD detection in CLL with a sensitivity of at least 10-5. It improves previous flow cytometry approaches and provided superior accuracy compared to parallel NGS MRD assessments.
