Our retrospective blinded study was designed to assess as many sources as possible of variability commonly found in the molecular diagnostic setting, including pre-analytical, analytical and clinical parameters associated with different specimen collection and testing sites. At sites 1 and 4 the sample set represented a broad array of myeloid neoplasms and acute leukemias positive or negative by cytogenetics. It included 75 specimens (63%) with normal karyotype or various chromosomal abnormalities not included in the MMA panel to challenge the assay specificity. At sites 2 and 3 the sample set was enriched for relevant translocation-positive specimens, including pediatric cases, to challenge the assay sensitivity with multiple independent specimens for every target included in the MMA panel. Other parameters, such as cytogenetic methods and interpretation, specimen collection method, RNA extraction method and MMA operator or instrument variability, were confounded with each site. Overall, 281 samples were evaluated with a relatively balanced design enabling assessment of both clinical sensitivity and specificity (145 positive and 135 negative).
The MMA accuracy relative to cytogenetics and independent validated molecular LDT was high, 98.6% and 98.7%, respectively, with 95% confidence intervals ranging from 95.4 to 99.6%. Investigation of the two apparent false-positive and two apparent false-negative cases showed that these discrepant results were in fact in agreement with the design and analytical specificity and sensitivity of the MMA, resulting in a virtual accuracy of 100% (95% confidence interval: 98.6–100%). All the samples in our retrospective study were well characterized by clinically accepted reference methods such as karyotyping, FISH and independent molecular tests when available. We therefore did not evaluate cases where translocations involving regions with similar banding patterns can result in cryptic sites detected by molecular methods but missed by karyotyping. However, in a single-institution prospective study on 330 diagnostic adult acute leukemia, King et al.11
recently reported a 12% (7/57) false-negative rate for karyotyping relative to a molecular LDT based on the same technology as the MMA. This high rate was within the range reported in independent studies for various clinical populations (0.5–15%)6, 13
and highlighted the complementary role of molecular and cytogenetic methods in the clinical management of leukemia.
Conventional G-banding analysis has the intrinsic ability to detect any structural or numerical aberration, including novel, uncharacterized abnormalities.3
Multiplex RT-PCR methods such as the MMA will not detect specific chromosomal features such as numerical aberrations or deletions and therefore cannot be a substitute for karyotypic analyses. However, when poor quality cultures, lack of analyzable metaphases or complex, masked and cryptic variants affect cytogenetic results, molecular methods can provide entity-defining, therapy-determining and prognosis-relevant information.1, 3
For example, detection of BCR–ABL1
e13a2 or e14a2 in CML is generally associated with a favorable outcome with tyrosine kinase inhibitor treatment. Identification of the rare e1a2 variant would considerably affect the patient's risk profile and might mandate early allogeneic stem cell transplantation.27
In AML, where the favorable prognosis associated with t(8;21), inv(16), or t(15;17) is well documented, detection of RUNX1–RUNX1T1
variants, as well as NPM1
mutations in certain cases, would result in a significant change from an intermediate risk group for normal karyotype to a favorable risk group.1, 16, 28
Similarly, risk-based classification of ALL would be impacted by the detection of specific fusion transcripts with BCR–ABL1
being associated with unfavorable prognosis and TCF3–PBX1
with an intermediate to very good prognosis.1, 2
Not only are many of these genetic alterations characterized by frequent cytogenetically cryptic sites but in certain cases, such as t(12;21) ETV6–RUNX1
mutations (four-nucleotide long insertions), they cannot be detected by conventional banding techniques because of their submicroscopic nature.
Most of the archived RNA samples evaluated in this study were from cases at presentation (n
=243). Among the 18 cases at follow-up, 9 with no residual disease, normal cytogenetic, or with chromosomal abnormalities not included in the MMA panel and 7 in cytogenetic relapse for t(9;22), t(4;11) or t(8;21) were correctly classified with the MMA (Supplementary Table 1
). The MMA also detected BCR–ABL1
in two t(9;22) positive CML or ALL cases in cytogenetic remission following treatment. Sample no. 69 was from an ALL patient who had received bone marrow transplantation 5 weeks before the low-positive MMA signal. Although he was clinically and cytogenetically free of disease at that time, he relapsed and died of leukemia 6 months later. Independent from this study, we are aware of another patient with AML with t(8;21) who had successfully completed induction therapy, was cytogenetically and flow cytometrically normal, and had a low positive MMA value for RUNX1–RUNX1T1
4 months before relapsing. These observations are consistent with a demonstrated MMA analytical sensitivity equivalent to 1% blast or less () enabling detection of molecular relapse earlier than cytogenetics.
Although the output of the MMA is a qualitative positive or negative call, quantitative analyses further shed light on the performance and robustness of the assay. The distributions and median probe signals varied between sites but were consistent with the pre-analytical and protocol differences. The positive signal distributions were narrow at sites 1 and 4 where a constant amount of RNA was used (400
ng) and were broad at sites 2 and 3 where a constant volume of RNA was used independently of the concentration (the recommended protocol is 400–1000
ng input in 1–6
μl volume). Review of historical total RNA concentration data at site 2 revealed a range 30–150
ng/μl. With a 5
μl input per RT reaction, the corresponding input range was therefore only 150–750
ng per RT with about 40% of the samples expected to be below the minimum recommended input of 400
ng. At site 3, the range of concentration was 80–1000
ng/μl with a 3-μl constant input. Thus, only about 5% of the samples were below the minimum input and more than 70% of the samples were expected to be in the 1000–3000
ng input range. This high input likely contributed to the uncharacteristic cross-hybridization signals between the PML–RARA
probes observed only at site 3 in two out of 26 t(15;17) positive samples. It also resulted in very high true-positive signals at site 3 where 30% (12/40) of the positive samples generated signals between 7006 and 9502 MFI, a range of signal intensity observed only at this site.
Overall, the median positive signal was 4020 MFI and 95% of the true-positive signals were at least twofold above 350 MFI. In contrast, the negative-probe signals were well below 350 MFI with maxima at 197, 208 and 245 MFI at sites 1, 2 and 4, respectively. Only five out of 2935 negative signals (0.17%) were above 250 MFI, all at site 3 where the highest RNA input was used. Furthermore, analytical studies showed that 95% of the probe signals generated in the absence of target analyte are expected to be at least twofold below 350 MFI (LOB=147–177 MFI for 968 probe signals; ). Supplementary calculations of the area under the curve for a one-sided normal distribution indicates that 99.999999% of the probe signals would be below 343 MFI (mean plus 5.612 times the s.d.) or the equivalent of 1 potential false-positive signal for every 100 million reactions performed with true negative samples (data not shown). This very favorable signal-to-noise ratio enabled positive detection of fusion transcripts in 1–10
ng of RNA input, an analytical sensitivity of at least 1% when 400
ng or more of RNA is used, and a very simple and safe qualitative data interpretation relative to a single fixed cutoff at 350 MFI.
Successful evaluation of the MMA at four independent sites also demonstrated that the method is compatible with the clinical molecular laboratory. Although formal workflow or time-motion comparative analyses were not performed, the potential gain in operational efficiency is evident. Simultaneous analysis by multiplex RT-PCR is a practical and time-saving solution relative to cytogenetics or repeated series of individual RT-PCR. The entire procedure can be completed in 6
h, up to 93 samples can be tested for 12 different analytes in each run, and a same-day turn-around time is achievable for urgent cases.11
However, one limitation of the dual amplification/detection specificity technology is the inability to detect alternative splicing isoforms and rare or novel variants unless specific primers and/or probes are added to the design. This was illustrated at site 3 where two MLL–AFF1
fusion transcripts were missed by the MMA. One was e11e4, a relatively rare variant, >5%, in non-infant t(4;11) ALL7, 18
and the other was e12e5, a variant to our knowledge not previously described in the literature. The only similar case identified by searching sequence databases was an e12e4 variant detected in an infant pre-B ALL (GenBank accession number JN169752.1).
Another potential drawback of the MMA is the inclusion of targets associated with leukemia of both myeloid and lymphoblastic lineages in a single reaction, which can result in unnecessary testing when the lineage of the neoplastic cells has already been established by morphological, cytochemical and/or immunophenotypic analyses. Both limitations can be overcome as shown by the successful development and evaluation of additional disease-focused panels with increased content. The alternative splicing variants ETV6–RUNX1
e5e3 and TCF3–PBX1
e13e2i27 are relevant in 5–10% of the t(12;21) or t(1;19)-positive cases29, 30
and the multiple MLL–AFF1
variants have a different prevalence in specific ALL populations. For example, the e11e4 variant is infrequent in pediatric and adult ALL but represents >50% of the t(4;11) infant cases.7, 18
In AML, both the CBFB–MYH11
type E and PML–RARA
bcr2 variants represent about 5% of the inv(16) and t(15;17)-positive populations31, 32
and small insertions in NPM1
exon 12 at 5q35 are the most frequently observed genetic abnormalities in AML, found in about 50% of the cytogenetically normal de novo
Independent studies also validated that the 23 variants described here and a total of 34 variants resulting from 18 different genetic alteration sites, including multiple MLL
(11q23) and RARA
(17q21) rearrangement partners relevant in acute leukemia, are detected with high sensitivity, specificity and precision using the MMA technology.17
In summary, our data demonstrate that multiplex molecular tests are a reliable and sensitive technology with a broad clinical utility to complement the morphological, immunophenotypic and cytogenetic diagnosis of leukemia. In the future, validation of expanded and disease-focused panels could provide additional flexibility and clinical sensitivity to reliably identify genetic abnormalities such as submicroscopic alterations or cryptic translocations that are paramount for optimal patient management.