Flow cytometry has become an essential tool for identification and characterization of hematological cancers and now, due to technological improvements, allows the identification and rapid enumeration of small tumor populations that may be present after induction therapy (minimal residual disease, MRD). The quantitation of MRD has been shown to correlate with relapse and survival rates in numerous diseases and in certain cases, and evidence of MRD is used to alter treatment protocols. Recent improvements in hardware allow for high data rate collection. Improved fluorochromes take advantage of violet laser excitation and maximize signal-to-noise ratio allowing the population of interest to be isolated in multiparameter space. This isolation, together with a low background rate, permits for detection of residual tumor populations in a background of normal cells. When counting such rare events, the distribution is governed by Poisson statistics, with precision increasing with higher numbers of cells collected. In several hematological malignancies, identification of populations at frequencies of 0.01% and lower has been attained. The choice of antibodies used in MRD detection facilitates the definition of a fingerprint to identify abnormal populations throughout treatment. Tumor populations can change phenotype, and an approach that relies on ‘different from normal’ has proven useful, particularly in the acute leukemias. Flow cytometry can and is used for detection of MRD in many hematological diseases; however, standardized approaches for specific diseases must be developed to ensure precise identification and enumeration that may alter the course of patient treatment.
Detection of rare cell populations in the blood by flow cytometry (FCM) was reported almost 30 years ago, with the enumeration of fetal red blood cells in the maternal circulation at a frequency of 1/10 000–1/100 000 by Cupp . Since that time, FCM has been used to detect and quantitate numerous rare cell populations in blood or bone marrow, from circulating endothelial cells to cancer stem cells [2-7]. FCM has become an essential tool in the diagnosis and monitoring of hematological cancers. While previously FCM was mainly used for lineage assignment in the acute leukemias, more recently, the improved ability of this technology over morphological assessment of blood and bone marrow to detect residual tumor populations, referred to as minimal residual disease (MRD), has led FCM, along with molecular analysis, to be incorporated into clinical trials and treatment decisions of pediatric acute precursor B cell and more recently acute myeloid leukemia [8-17]. The usefulness of FCM in detection small tumor population to the level of 0.01% has also been demonstrated in multiple myeloma , chronic lymphocytic leukemia [17, 18], and other lymphoproliferative disorders [9, 13, 19]. More recently, there has been growing interest in using MRD as a biomarker with the potential to allow rapid testing and evaluation of new drugs in inducing remission states in several diseases.
While detection of MRD has been shown to correlate with time to relapse in the acute leukemias and multiple myeloma [8, 9, 12-14, 16, 20-24], there are several technical and interpretative hurdles that need to be addressed before a laboratory undertakes such an analysis.
For a laboratory about to embark on developing MRD for a specific disease state, several fundamental questions need to be addressed even before the technical aspects can be considered.
Is there a published standardized method available? Developing an MRD protocol in-house is often less satisfactory than adopting a previously validated model. Such methodologies have been developed for CLL [17, 18, 25], myeloma [10, 23], and ALL [9, 14, 26]. For AML, there are few well-validated protocols, and this is complicated by the fact that this disease is often composed of multiple clones at presentation and can be affected by antigen modulation during treatment [13-16, 27]. Additionally, even within the acute lymphoblastic leukemias, down-regulation of several B [14, 28-30]- or T [19, 31-33]-cell markers of immaturity has been well documented.
This manuscript will focus mainly on the technical aspects of MRD as related to hematological disease states; however, many of the concepts are applicable to any rare event analysis by flow cytometry.
Sample Preparation Issues
One variable that needs to be taken into consideration is whether analysis should be performed on an enriched or nonenriched sample. Sample enrichment can be either positive or negative. The simplest way of enriching the sample is to deplete nonmononuclear cells over a Ficoll gradient. This method has been used by Campana in the analysis of MRD in pediatric ALL . Positive selection involves labeling the population of interest, usually with magnetic particles linked to an antibody, for example CD34 for hematopoietic stem and progenitor cells. A magnet is then used to hold the cells of interest within the tube, while unwanted cells can be poured off. Negative selection involves removing many of the unwanted cells from the sample of interest. In the RosetteSep™ (STEMCELL Technologies, Vancouver, BC, Canada) bone marrow progenitor cell pre-enrichment cocktail, cells are labeled with tetrameric antibody complexes recognizing CD3, CD11b, CD14, CD16, CD19, CD56, CD66b on white blood cells and glycophorin A on red blood cells (RBCs). When centrifuged over a buoyant density medium, the unwanted cells pellet along with the RBCs. The pre-enriched progenitor cells can then be removed at the interface between the plasma and the medium.
With the latest generation of clinical flow cytometers, it is now possible to analyze cells at rates of above 20 000/s, allowing for unselected samples to be analyzed with greater than 1 × 106 events collected per minute. An advantage of this method is that any selection process will undoubtedly lead to some cell loss of the target population. In leukemia MRD analysis by flow cytometry, only red blood cells are lysed in blood or bone marrow before analysis, as they outnumber WBCs by approximately 1000–1.
Normally samples are processed for flow cytometry with a cell count of approximately 5–10 × 106/mL, with 100 μL of blood containing 0.5–1 × 106 WBCs being stained for analysis. This is often insufficient for MRD, and cell concentrations up to 40–50 × 106/mL can safely be used, allowing for collection of several million events per tube. Cells should be concentrated in minimal volume while staining, as the total protein concentration is important for effective cell labeling with antibody. 5 million cells in 100 μL buffer will stain more effectively than those in 1 mL.
Antibody and Fluorochrome Selection
Detection of rare cell populations is enhanced by maximizing the signal-to-noise ratio of the cells of interest from background. Antibodies should be targeted toward specific cell types and, where possible, maintain the same level of expression in pre- and post-treatment samples. This is not always possible as the key markers CD10 and CD34 in detecting MRD in progenitor B-ALL are known to be down-regulated in patients treated with corticosteroids . In T-cell ALL, it has been shown that markers of immaturity, for example CD1a, TdT, and CD99, are lost during treatment (see Figure 1); however, the more mature T-cell markers remain fairly constant and can be used to monitor persistence of disease [19, 33].
The list of available fluorochromes seems to grow almost daily, and there are now several with large stokes shifts and high quantum efficiency for each of the common lasers used in the clinical flow laboratory, namely 488-nm, 633-nm, and more recently the violet 405-nm lasers. For the 488-nm laser, PE or tandem conjugates of PE (e.g., PE-Cy5 or PE-Cy7) have the greatest signal-to-noise ratio. For the 633-nm laser, APC and its conjugates (e.g., APC-Alexafluor700 or APC-H7) are best. For the violet diode laser, Krome Orange and V450 among others are both excellent fluorochromes with good stability.
Current clinical flow cytometers have the capacity to detect 8–10 fluorochromes simultaneously allowing for complex combinations of antibodies to be used to detect aberrant expression. While this is potentially an excellent advantage, consideration must be given to the impact of fluorescence compensation on signal detection. Additionally, many of the tandems dyes have reduced stability compared with nontandems, and their breakdown product will show up as a positive signal in the PE or APC channel depending on the tandem, potentially leading to a false-positive signal in that channel. In general, combining key markers with the brightest fluorochromes or those with low copy numbers of antigens on the cell surface is advisable. To simplify compensation issues, it is as important to ensure that expression of any potential antigen/antibody is not too bright as it is that it is not overly dim.
In the standard Cooperative Oncology Group (COG) analysis for precursor B-ALL , CD19, which is present in over 95% of patients, is the anchor marker used in all analyses (alternative protocols must be utilized when faced with CD19 negative patients). Presence or absence of other markers, for example CD10, CD20, CD58, and CD81, and their level of expression compared with normal cell counterparts are all used to detect the abnormal population. Intimate knowledge of normal cell development is essential in distinguishing normal regenerating hematogones from residual leukemic cells.
Rawstron used cell doublets in chronic lymphocytic leukemia to increase the sensitivity of detection of residual CLL cells . He noted that occasionally doublets of normal CD19+ and CD3+ cells would form, and these doublets would express similar markers to true CLL cells. Contamination was corrected for the numbers of CD19-gated events binding CD3, or using CD3 as a threshold to determine the limit of detection (that is, only classifying a sample as having detectable residual disease if the number of CLL events was greater than the number of CD19+ CD3+ events). In the 2007 study, it was also shown that the use of a standard operating procedure resulted in a 19% improvement in accuracy and 44% improvement in specificity (17).
Patients with acute myeloid leukemia have proven harder to monitor for MRD, due to the difficulty not only in detecting differences from normal in the leukemic blast population, but in the presence of leukemic stem cells (LSCs). This population that resides within the CD34+ CD38- subset of blasts has been shown to overexpress surface markers CD44 and CD123 among others, compared with normal stem cells. Of great interest is the antibody CLL-1, which has been shown to be present on the LSC and not on normal myeloid stem cells. The fact that it is present both at diagnosis and postinduction therapy should prove useful in detecting this small population in a background of normal and regenerating bone marrow cells .
Before collecting data on the flow cytometer, it is important to ensure the instrument is clean and the background level of noise is below the threshold that would interfere with collection of rare events. For example, a background rate of one event in 10 000 would preclude detection of MRD at a rate of 0.01%. The simplest way to check this is to run the staining buffer for a period of time equivalent to the expected time of sample acquisition and note the number of events collected. A second check is to run a known normal sample stained with the combinations of antibodies used in the patient test. There should be few, if any, events that are detectable in the region of interest. This exercise should be performed periodically or any time when there is an unexpected result. It is also prudent to run a saline or sheath fluid sample before MRD collection, to ensure there is no carryover from a previous sample.
It is essential the instrument is correctly aligned and PMTs are set to optimally detect the signal above noise. Negative populations should be off the axis, and the use of bi-exponential plots may be useful in determining appropriate setup. Compensation should be checked using either beads or stained cells for each parameter being examined. Postacquisition compensation may also be used.
It is helpful to have an idea at what frequency the target events occur in the sample population. When dealing with populations below 1% and as low as 0.01%, Poisson statistics apply – counting randomly distributed objects (cells) in a certain volume. A simple calculation can be used to determine the precision of analysis, where the standard deviation (SD) is equal to the square root of the number of target events counted and the 95% confidence interval = 2 × SD For example, if 1 × 106 total events are collected and within that 100 target events are collected, the coefficient of variation can be calculated:
Therefore, for 100 total events counted, the statistical 95% confidence interval would be 80–120. It is evident that by increasing the number of target events collected, irrespective of the number of total events, the precision increases correspondingly. Table 1 illustrates how this data can be used to determine total number of events required for a given precision. An excellent example of this was shown elegantly by Donnenberg in several situations, one in leukodepleted platelet concentrates  and more recently  in which more than 1 × 106 CD45+ events were collected, and using time as a parameter, this group showed the C.V. of CD34+ events present at a concentration of 0.1% decreasing with increasing number of target events collected in a predictable fashion as would be expected for a Poisson distribution.
|Desired CV (%):||1||5||10||20|
|r = no of events of interest:||10 000||400||100||25|
|When occurring at a frequency of:|
|%||1 : n||Total no of events which must be collectedb|
|10||10||105||4 × 103||103||2.5 × 102|
|1||100||106||4 × 104||104||2.5 × 103|
|0.1||1000||107||4 × 105||105||2.5 × 104|
|0.01||10 000||108||4 × 106||106||2.5 × 105|
|0.001||100 000||109||4 × 107||107||2.5 × 106|
|0.0001||1 000 000||1010||4 × 108||108||2.5 × 107|
Once the total number of events is collected, the protocol developed to perform the analysis and isolate the cells of interest in multiparameter space is critical .There are key areas that need to be addressed to ensure robust data. Collecting time as a parameter allows for the exclusion of any bursts of data during acquisition. Collecting time as a parameter also allows for exclusion of the end of a sample collection if the sample should run dry introducing significant artifact into the data set. Exclusion of doublets can be performed by plotting forward scatter peak versus integrated or total fluorescence. A region to identify dead cells and debris can be drawn to exclude these events from the analysis. Alternatively, a viability dye can be included to exclude dead and dying cells. The combination of light scatter and a vital dye, which will detect all nucleated cells both live and dead, is useful when determining a denominator by which to compare the population of interest to. The use of Boolean logic is essential in rare event analysis to isolate the population of interest in multidimensional space.
Improved computer speed and software that can manage large data files have made analysis of this information manageable in a fraction of the time previously required. Further improvements in software could potentially allow for the identification of abnormal populations in an automated fashion, and indeed, the Euroflow Consortium has shown evidence of this .
Detection of MRD by flow cytometry holds great promise for the future of hematological disease prognosis and could potentially be used as a biomarker to predict outcome after therapy with new or novel drug therapies. Recent advances in hardware, software, and fluorochrome technology have made this testing available to most large clinical laboratories. Availability of novel antibodies present in the leukemic cell of interest or expressed at levels not seen in normal cell populations will aid in the acceptance of this technology and should allow for improved sensitivity. Standardized methodologies for specific diseases, coupled with quality assurance programs, need to be developed to ensure precise identification and enumeration of residual tumor populations that may alter the course of patient treatment. To this end, efforts in Europe and North America are underway to standardize such approaches.