Fast‐mode duplex qPCR for <i>BCR‐ABL1</i> molecular monitoring: Innovation, automation, and harmonization

Abstract

Reverse transcription quantitative polymerase chain reaction (RT-qPCR) is currently the most sensitive tool available for the routine monitoring of disease level in patients undergoing treatment for BCR-ABL1 associated malignancies. Considerable effort has been invested at both the local and international levels to standardise the methodology and reporting criteria used to assess this critical metric. In an effort to accommodate the demands of increasing sample throughput and greater standardization, we adapted the current best-practice guidelines to encompass automation platforms and improved multiplex RT-qPCR technology. Chronic myeloid leukemia (CML) and Philadelphia positive acute lymphoblastic leukemia (Ph+ALL) are the two hematological malignancies most commonly associated with the t(9;22)(q34;q11) translocation (Philadelphia Chromosome, Ph) and the resultant BCR-ABL1 fusion gene. This in turn encodes a dysregulated tyrosine kinase oncoprotein, affecting a multitude of downstream targets leading to aberrant mitogenic signalling, altered cell adhesion, suppression of cell death pathways and leukemogenesis [1]. The fusion-gene transcript, being unique to the leukemic clone offers both a diagnostic and therapeutic target, exploited via RT-qPCR and tyrosine kinase inhibitors (TKI), respectively. The transcript reduction kinetics in response to TKI therapy is a powerful prognostic tool, allowing the clinician to make informed choices about an individual's therapeutic course at defined time points [2, 3]. Currently, many centers in the United Kingdom and Europe use RT-qPCR protocols based on the Europe Against Cancer (EAC) collaboration, which produced standardized sample preparation, primer-probe sequences, and reaction conditions for use in the molecular monitoring of fusion-gene positive leukemias [4, 5]. The standardization process has since continued through national and international collaborative groups, resulting in the publication of a number of guidelines [6-8]. However, due to the dual pressures of increasing sample throughput and economic restriction, it has become necessary to employ new technological solutions whilst retaining compliance within the ongoing standardization frameworks. Consequently, we adapted the EAC BCR-ABL1 qPCR protocol to make use of current hardware and chemistry capabilities to reduce cost and increase throughput. We had previously introduced a duplex assay that was an incremental modification of our incumbent in-house qPCR methodology, but it suffered from an increasingly problematic high background and an unexplained sensitivity to interbatch variations in qPCR master mix [9]. In addition, because of the use of legacy standard controls, the assay was not EAC compatible, which was problematic when undertaking cooperative standardization exercises. We, therefore, decided to overhaul the qPCR assay and introduce automation at key points in the processing workflow and then perform validation to ensure performance. A detailed summary of the changes made can be found in Table I and the primer/probe sequences in Table II. To reduce the problem with assay noise levels, minor-groove binding (MGB) probes [Applied Biosciences (ABI), UK], using a non-fluorescent quencher, were chosen. The corresponding EAC probe sequences were then truncated to compensate for the inherent increase in Tm associated with the MGB moiety [10]. Additionally, a fast-mode qPCR master mix was used to take advantage of the fast-run capabilities of our ABI 7900HT real-time analyser, and in so doing reduced the run time from 1 hr 45 min to 55 min. The reaction volume was also decreased from 25 to 20 μl, ensuring optimal fast-mode reaction kinetics. The BCR-ABL1 and ABL1 probes were differentially labelled with 6-FAM and VIC, respectively, to allow duplexing of the qPCR reaction. To replace our EAC incompatible standard curve plasmid, we chose a four-target plasmid construct, developed and kindly provided by the Wessex Regional Genetics Laboratory (WRGL), which contains not only the BCR-ABL1 target sequence necessary for BCR-ABL1 and ABL1 measurement (as per EAC primer-probe set) but also the BCR (not associated with the fusion, as per IRIS primer-probe set [11]) and GUSB (EAC primer-probe) control gene sequence elements in a 1:1:1:1 ratio. This plasmid also benefited from being used in several national control rounds, so its compatibility and stability had been established across a broad range of laboratory settings. A critical step in the sample processing workflow is the disruption of the white cell pellet in a chaotropic lysis buffer. This was performed manually, using shearing from an 18-gauge needle, and was identified as a laborious bottleneck that would benefit from automation. Consequently, we trialed the TissueLyser II platform (Qiagen, UK), which combines an in-tube bead with mechanical agitation to induce rapid, high-throughput disruption. A QIAgility (Qiagen, UK) robotic liquid handling platform was also introduced to prepare the qPCR reaction plate, allowing walk-away automation of a complex and labor-intensive step, whilst minimizing the risk of handling errors. Due to the whole-scale changes, we undertook a multi-phase process of optimization and validation to ensure that all data generated would be contiguous with previous results. Each step in the procedure was individually optimized and tested and the final validation round assayed the workflow as a whole. In addition, post-qPCR analysis was performed using a spreadsheet macro, using algorithms based on the current UK guidelines [6] for sample quality, replicate delta, and positivity cut-offs to increase compliance through the removal of subjective interpretation. Figure 1 illustrates the results of the statistical analysis of four consecutive paired RT-qPCR runs for ABL1 and BCR-ABL1 (96 patient samples plus controls) that formed the final validation of the duplex assay (Dx) against the gold-standard simplex assay (Sx). No outlier or other point removal was used, except for BCR-ABL1 samples (and the associated %Ratio metric) that were qPCR-undetectable for either the duplex or simplex assay (n = 11); these were assayed separately as a discrete variable. In Fig. 1A, the median ABL1 levels for the Sx and Dx are shown to diverge by a small, but statistically significant amount (24,299 and 23,200 transcripts, respectively; P = 0.0109). However, the BCR-ABL1 values were not shown to be significantly different (Sx = 11.45, Dx = 10.99; P = 0.1544) (Fig. 1B) and neither were the %Ratio values (Sx = 0.05505, Dx = 0.04656; P = 0.0768) (Fig. 1C). The dotted line in Fig. 1B represents the limit of positivity (0.5 transcripts; intercept Ct+1) and data points below this threshold are counted as undetectable for analysis, in accordance with the UK guidelines. There were six samples that showed low positivity by the Dx that were undetectable by the Sx (6.2%) and 5 that were Sx low positive and Dx undetectable (5.2%). A Fisher's exact 2 × 2 contingency test showed these errors to be non-significant (P = 1.00). Also, there were two patient samples that gained major molecular response (MMR) going from Sx to Dx (2.2%) and none that lost it (P = 0.8751). All other RT-qPCR run metrics were within recommended parameters [6], with a mean slope, intercept, r2 and efficiency of −3.37%, 39.644%, 0.999%, 97.93%, respectively, for Sx ABL1; −3.35%, 39.493%, 0.995%, 98.88% for Dx ABL1; −3.44%, 39.592%, 0.999%, 95.34% for Sx BCR-ABL1; and −3.38%, 37.968%, 0.998%, 97.55% for Dx BCR-ABL1. Results of the statistical analysis of the final four rounds of tests that represent the validation of the MGB duplex assay (Dx) against the gold-standard simplex assay (Sx). A–C: The paired Wilcoxon analysis, where the figures below the data points representing the median for each data set; the dotted line in (B) represents the limit of positivity (0.5). D–F: Bland-Altman plots illustration inter-assay bias. G–I: Least-square regression analysis of the log transformed transcript data (solid line) with 95% prediction bands (dotted lines) and additional Deming regression (dashed line). Although the drop in median ABL1 for the Dx was shown to be statistically significant, the actual deviation between the two tests was small, representing a median-paired sample difference of 3.1% (Fig. 1A). It can be seen from Fig. 1B,C that where discordance did occur between paired samples it was amongst the low BCR-ABL1 samples (<10 molecules), again reflecting the loss of precision associated with measuring rare events by qPCR. It is this inherent difficulty in measuring very low BCR-ABL1 that is currently a challenge for those involved in CML management because of the drive to define complete molecular response (CMR), which currently has multiple interpretations, and is an essential criterion for patients to be eligible for TKI de-escalation or cessation [12]. Additionally, the Dx assay was not shown to introduce any loss of sensitivity nor increase the risk of false positivity when compared to the Sx assay, as evidenced by the comparable reaction metrics and the non-significant Fisher's exact test performed on the low-level positive/negative results. However, the Dx test was associated with a small but noticeable increase in qPCR background signal (typically >40 Ct and seen in BCR-ABL1 negative/very low samples, and never in the NTC), the effect of which was successfully abrogated by strict (and automated) appliance of the UK/EAC guidelines for positivity (intercept Ct+1) [6]. Figure 1D–F shows Bland-Altman plots assaying for systematic bias. It can be seen that there is a high level of symmetry for all three metrics, with no significant bias. It can also be seen from Fig. 1E (and by derivation in Fig. 1F) that the bulk of the data scatter occurs at values below 10 molecules of BCR-ABL1, which is consistent with the drop in precision associated with measuring rare events. Figure 1G–I shows the regression analysis of the quantification data, with least-square (solid line) and Deming regression (dashed line) superimposed on the data points. It can be seen that the correlation between Sx and Dx is very high (r2 = 0.9257, 0.9562, 0.9800 for ABL1, BCR-ABL1, and %Ratio, respectively). The %Ratio data was also tested using a Passing-Bablok analysis, which assays for both constant and proportional bias, along with an integrated linearity test. This showed a low constant bias of −0.01 (95% CI = 0.04 to −0.05); a low proportional bias of 1.03 (95% CI = 1.00 to 1.05); and the Cusum linearity test showed no significant loss of linearity (P > 0.1). These analyses indicate that there was no significant bias introduced by the Dx assay and associated workflow changes. As a whole, the validation data showed that there would be no noticeable change in the result reporting metric introduced by adopting the MGB duplex assay and automation workflow. Indeed, since implementing the new workflow in September 2011, our continual quality control monitoring has not shown any adverse events or loss of data integrity that could be associated with the changeover. Although the automation points might not be of interest to laboratories with a limited sample throughput, the economical benefits of switching to a fast-mode duplex assay should be considered. Several companies produce fast-multiplex qPCR mixes and since these are typically used at lower reaction volumes than standard assays the cost saving is twofold. In addition, there are several probe strategies available, including standard probes with non-fluorescent quenchers, with or without locked nucleic acid (LNA) moieties to reduce probe length and increase binding efficiency. In-house optimization and validation is essential to ensure contiguous result reporting and sensitivity, especially as long-term molecular response is becoming an ever more important patient milestone. In summary, we have developed measures, including automation points and an improved MGB probe based duplex BCR-ABL1 molecular monitoring RT-qPCR assay to increase both our sample throughput capacity and our inter-laboratory standardization profile, whilst concurrently reducing hands-on time and financial cost. Routine sample processing was performed as previously described [6, 13]; briefly, 10 ml patient peripheral blood (PB) samples were collected in EDTA tubes and processed within a maximum of 72 hr of draw (typically within 24 hr). Red cell lysis buffer (in-house, ammonium chloride based) was used, in addition to centrifugation and washing in PBS (Gibco, UK) to isolate whole white cell pellet. Either manual [18-guage non-beveled needle (part: BN1838; Terumo, UK) and 5 ml syringe (Kendall, US)] or automatic (Qiagen TissueLyser II) disruption of the pellet in RLT buffer was employed. A total of 92 patient samples were included in the final analysis, along with high (∼ 10%) and low (∼ 0.1%) BCR-ABL1 controls (cell line controls, provided by WRGL), BCR-ABL1 negative controls (cDNA from HL60 and Jurkat cell lines) along with water (NTC), ethanol, and RLT-buffer negative controls from the sample processing. Total RNA was extracted from RLT lysate using the QIAcube platform with RNeasy mini kits (part: 217004; Qiagen) with an elution volume of 60 μl and first-strand cDNA synthesis performed on the full volume of RNA solution (100 μl final volume), without measuring the quantity of RNA, using M-MLV reverse transcriptase (part: 28025021; Invitrogen, UK) and random hexamer primers (part: N8080127; Invitrogen) in accordance with the EAC and current UK guidelines [4, 6]. Standard primers and dual-labeled TaqMan probes were supplied lyophilized from ABI and diluted in 10 mM TE buffer to 80 μM and 50 μM stock solutions, respectively. MGB TaqMan probes were supplied in liquid form at 100 μM stock (ABI part: 4316032). For the simplex reaction, the BCR-ABL1 primers and probe were the EAC set and the ABL1 were an in-house design spanning exons 3 and 4 of ABL1; both dual labeled with 5′ FAM and 3′ TAMRA. The primers were used at a final concentration of 300 nM and the BCR-ABL1 and ABL1 probes at 100 nM and 200 nM, respectively. For the duplex reaction, the BCR-ABL1 and ABL1 probes were truncated and MGB modified versions of the EAC sets, used in conjunction with the corresponding primers. The BCR-ABL1 was 5′ labeled with FAM and the ABL1 was labeled with VIC. The BCR-ABL1 duplex primers and probe were at 300 nM and 100 nM, respectively; for ABL1, the primers were at 150 nM and the probe was at 200 nM. Primer and probe sequences and concentrations are illustrated in Table II. Sx were set up manually in 25 μl reactions using 3 μl cDNA and ABI Universal Master Mix (part: 4305719) in separate 96 well plates (ABI MicroAmp Fast, part: 4346907) for BCR-ABL1 and ABL1, using three technical replicates for the former and two for the latter. The standard curve was derived from a serial dilution (1 × 106–1 × 101/3 μl) of an in-house BCR-ABL1 plasmid (nc-p210), calibrated (via a compatible intermediate) against a commercially available kit (part: CGRS-01-4; Ipsogen, France). MGB duplex assays (Dx) were set up in a single 96 well plate with a 20 μl reaction volume (3 μl cDNA; three technical replicates) and ABI Fast Advanced Master Mix (part: 4444557) using the QIAgility liquid handling platform. The standard curve was derived from a serial dilution (1 × 106–1 × 101/3 μl) of a four-target BCR-ABL1 plasmid (p210-GUSB-BCR, WRGL, UK), calibrated against both the Ipsogen standard and the Mannheim plasmid currently in use for EUTOS-based standardization exercises [14]. The qPCR reaction cycling conditions were set at the ABI 7900HT standard-mode default (software version SDS2.4) for the simplex assay and for the MGB duplex we used a modified version of the fast-mode, with the dissociation time changed from the default 1 to 3 sec and the annealing/extension time increased from 20 to 45 sec. These modifications increased Dx reaction efficiency and are to compensate for the greater associated reaction complexity. For data analysis, the SDS auto-baseline was chosen and a fluorescence threshold of 0.1 was used for the simplex and 0.12 for the duplex. Data interpretation was via exported spreadsheet and quantification and quality control calling via an in-house Excel macro. A cut-off of 1 × 104 molecules of ABL1 per reaction volume was used for minimum sample quality and other QC metrics as recommended in the UK guidelines [6]. The %Ratio was derived by the following calculation: (QuantityBCR-ABL1/QuantityABL1) × 100; where Quantity = the number of transcripts in the reaction volume as derived by the SDS software from the plasmid standard curve. Statistical data were principally compiled using Microsoft Excel 2007 and ported into Graphpad Prism 4.03 for the relevant analysis. Each data set was tested for normality using a D'Agostino & Pearson omnibus test and paired non-parametric comparisons carried out using a Wilcoxon signed rank test. Systematic bias was assayed using Bland-Altman, least-square linear regression, and Deming regression. Constant and proportional bias was assayed in the final %Ratio set using a Passing-Bablok test and linearity using a Cusum test (Analyse-It for Excel). Differences in low positive and negative results and loss/gain of major molecular response (MMR; <=0.1% BCR-ABL1/ABL1 ratio) were assayed using 2 × 2 contingency Fisher's exact test (Graphpad QuickCalcs). Gareth Gerrard* , Katherine Mudge*, Pierre Foskett*, David Stevens*, Mary Alikian*, Helen E. White , Nicholas C.P. Cross , Jane Apperley , Letizia Foroni*, * Imperial Molecular Pathology, Imperial College Academic Health Science Centre, London, United Kingdom, Centre for Haematology, Faculty of Medicine, Imperial College London, London, United Kingdom, National Genetics Reference Laboratory Wessex, Salisbury District Hospital, Salisbury, United Kingdom.