Enhanced sensitivity of flow cytometry for routine assessment of minimal residual disease

In a recent paper by Béné and Kaeda,1 technical approaches for minimal residual disease (MRD) assessment are extensively reviewed. PCR-based studies have proved to be 1-log more sensitive than flow cytometry (FC). For this reason, they are increasingly being preferred for MRD analysis, especially at the end of therapy or post hematopoietic stem cell transplantation.1,2 It would be valuable to develop MRD flow cytometry assays with this level of sensitivity that could be applied routinely. In the present work, we analyzed MRD samples with a level of infiltration below the limit of detection of routine FC, which is accepted as 10 and comes from the standard acquisition of 2–5×10 leukocytes.3–11 At least 10-fold more leukocytes must be acquired to increase sensitivity by 1-log; this large number of leukocytes can be acquired easily in digital cytometers by acquiring several individual tubes stained with the same combination of monoclonal antibodies, and putting them in a single file. Because the time of acquisition for each individual tube is not increased, no problems of cellular aggregation arise.

Using this approach, we acquired 6 million leukocytes from each sample using a FACSCanto flow cytometer (Becton-Dickinson). The stability of fluorescent parameters between tubes was verified (Online Supplementary Figure S1). The study was performed with the event rate used routinely in our laboratory, i.e. 2,000 events/sec. To explore the possibility of decreasing acquisition time, we analyzed the influence of increasing the event rate on the percentage of electronic aborts (due mainly coincidence events), and compared MRD measurements in samples acquired at different event rates (Online Supplementary Table S1). We found that the acquisition event rate could be increased reliably to at least 4,000 events/sec. Using this event rate, the time of acquisition of 6 million leukocytes is suitable for routine measurements.

We assessed that this approach enhanced sensitivity by 1-log by comparing MRD analysis by FC with real-time quantitative PCR for BCR-ABL1 transcripts in serial dilutions of a Philadelphia chromosome-positive B-II acute lymphoblastic leukemia sample (Table 1A). We then selected multiple myeloma (MM) and B-cell chronic lymphocytic leukemia (BCLL) MRD+ samples with about 0.01% infiltration and diluted them 10-fold with normal leukocytes. As shown in Table 1B, MRD was detected accurately in all diluted samples (10 infiltration). Observed values were as expected ±10% in all cases.

This increase in sensitivity did not compromise the specificity of the technique. Ten million leukocytes from the bone marrow of patients (n = 3) without hematologic neoplasias and from the peripheral blood of healthy donors (n = 3) were acquired and blindly tested for the presence of events with a myelomatous or B-CLL phenotype, respectively. The results were unambiguously negative (Online Supplementary Figure S2). We also acquired 10 million normal leukocytes and searched for events with phenotypic characteristics of other hematologic malignancies, such as follicular lymphoma and hairy cell leukemia, with negative results.

In addition to the total number of leukocytes analyzed, the sensitivity of FC depends on the number of neoplastic events that must be detected in order to obtain precise measurements.

Table 1.Accuracy of minimal residual disease assessment in samples with infiltration <10⁻⁴.

Based on the theory that rare events follow a Poisson distribution, it is accepted that 50–100 events are needed to reliably measure the frequency of a population.3,7,8,11,12 However, to our knowledge, no studies have really measured the influence of the number of detected tumoral cells on MRD measurement precision. Accordingly, we determined the coefficient of variation (CV) of the percentage of MRD in samples from patients with MM, B-CLL and T-cell lymphoproliferative disorders (T-CLPD) obtained when detecting increasing numbers of events in the malignant cluster. As shown in Table 2, the CVs of the B-CLL and T-CLPD samples were very close to those predicted by the Poisson distribution. Fifty to 60 events were required to obtain a CV less than 15%. Strikingly, in MM samples, the CVs were around 10% regardless of the size of the cluster, even when as few as 20 malignant events were detected (possibly because it is easier to identify malignant cells from MM than from other hematologic malignancies, since myelomatous plasma cells usually occupy a space in which background events are scarce).

Table 2.Precision of minimal residual disease assessment according to the number of events in the malignant cluster.

In summary, acquiring 6 million leukocytes is feasible with a digital cytometer on a routine basis. Because detection of 50–60 malignant cells is required to get a CV less than 15%, a sensitivity of 1×10 is achieved. Being able to routinely apply MRD FC assays with high sensitivity would be very valuable, especially in cases where molecular techniques cannot be used.