Sensitive NPM1 Mutation Quantitation in Acute Myeloid Leukemia Using Ultradeep Next-Generation Sequencing in the Diagnostic Laboratory Piers Blombery, BSc (Biomed), MBBS (Hons), FRACP, FRCPA; Kate Jones, PhD; Ken Doig, BSc (Hons), MPhil (Melb); Georgina Ryland, PhD; Michelle McBean, BAppSc (Med Lab Sci); Ella Thompson, PhD; Costas K. Yannakou, MBBS(Hons), FRACP, FRCPA; David Westerman, MBBS, FRACP, FRCPA, FFSc
Context.—Detection of measurable residual disease after therapy is an important predictor of outcome in acute myeloid leukemia. Objective.—To investigate the feasibility of using nextgeneration sequencing (NGS) in the diagnostic laboratory to perform quantitative NPM1 mutation assessment using ultradeep (approximately 300 0003–500 0003) sequencing (NGS-qNPM1) as a method of assessing residual disease burden in patients with acute myeloid leukemia. Design.—A flexible NGS-based assay for the detection and quantitation of NPM1 mutations was developed by polymerase chain reaction amplification of target DNA sequences, sequencing on an Illumina (San Diego, California) MiSeq, and analyzing data with an in-house– designed bioinformatic pipeline. NGS-qNPM1 was com-
pared with current NPM1 quantitation methods (real-time quantitative-polymerase chain reaction and multiparameter flow cytometry). Results.—The NGS-qNPM1 assay had a sensitivity of between 104 and 105 and showed high concordance and correlation with reference methodologies. Moreover, the NGS-qNPM1 assay was able to be integrated into the laboratory’s existing, targeted amplicon-based sequencing workflow. Conclusions.—An NGS-based, quantitative NPM1-mutation assessment can be used to monitor patients with acute myeloid leukemia, and it has some practical advantages over existing modalities. (Arch Pathol Lab Med. 2018;142:606–612; doi: 10.5858/ arpa.2017-0229-OA)
A
demonstrated to have a mutation in NPM1. Each of those modalities (FC and RT-qPCR) has technical strengths and limitations for sensitive MRD assessment in AML.6 Given the adoption of next-generation sequencing (NGS) in many diagnostic laboratories as a modality for mutation detection in myeloid malignancy, it represents a potentially attractive modality for use in MRD assessment. Moreover, the possibility of sensitive NGS-based NPM1 detection has been demonstrated.7 We aimed to develop and investigate the performance of a sensitive, amplicon-based quantitative NPM1 assay (NGS-qNPM1) for use in disease monitoring in AML. In this article, we describe the development, performance, implementation, and initial experience of an NGSqNPM1 assay into routine use in a diagnostic laboratory (Molecular Haematology Laboratory, Peter MacCallum Cancer Centre, Victoria, Australia) that performs a range of amplicon- and hybridization-capture–based NGS assays for the assessment of hematologic malignancy.
ssessment of the response of acute myeloid leukemia (AML) to intensive therapy is central to therapeutic decision making. It is well-established that detection of measurable residual disease (MRD) after remission-induction chemotherapy is a predictor of outcome in AML.1–5 Outside of the context of acute promyelocytic leukemia and core-binding factor leukemias (ie, AML with RUNX1RUNXT1 and CBFB-MYH11 fusions), in which translocation-specific, real-time quantitative-polymerase chain reaction (RT-qPCR) assays exist, the sensitive detection of MRD in AML is most commonly performed using either (1) flow cytometry (FC) for detection of immunophenotypic aberrancies present in leukemic blasts, or (2) detection of the expression of NPM1 mutations by RT-qPCR in patients Accepted for publication August 16, 2017. Published as an Early Online Release February 9, 2018. From the Department of Pathology, Peter MacCallum Cancer Centre, Melbourne, Australia (Drs Blombery, Jones, Ryland, Thompson, Yannakou, and Westerman; Mr Doig; and Ms McBean); and the Sir Peter MacCallum Department of Oncology, University of Melbourne, Melbourne, Australia (Drs Blombery, Thompson, Yannakou, and Westerman). Drs Blombery and Jones contributed equally to this work. The authors have no relevant financial interest in the products or companies described in this article. Reprints: Piers Blombery, BSc (Biomed), MBBS (Hons), FRACP, FRCPA, Peter MacCallum Cancer Centre, 305 Grattan St, Melbourne VIC 3000, Australia (email:
[email protected]). 606 Arch Pathol Lab Med—Vol 142, May 2018
MATERIALS AND METHODS Patient Samples Samples were selected from stored, routine, diagnostic cases from patients with NPM1 mutations identified from institutional databases. Bone marrow aspirate samples were treated with red blood cell lysis buffer (10 mM NaHCO3, 150 mM NH4Cl, 0.1 mM Na2EDTA, pH7.4) at room temperature for 10 minutes. Cells were then pelleted at 440g for 5 minutes, and the supernatant was discarded. The cell pellet was resuspended in fresh red blood cell NGS-qNPM1—Blombery et al
Table 1. Amplicon Name NPM1_Ex11 NPM1_Ex11_A NPM1_Ex11_B
Primer Sequences for Amplicons Used for Quantitative NPM1 Sequencing Forward Primer
Reverse Primer
Genome Coordinates
TGTCTATGAAGTGTTGTGGTTCC GGGAAAGTTCTCACTCTGCATT chr5:170837478þ170837673 CTAGAGTTAACTCTCTGGTGGTAGAATGA GACAGCCAGATATCAACTGTTACAGAA chr5:170837406þ170837643 TTTTCCAGGCTATTCAAGATC TTTGGACAACACATTCTTGGCAA chr5:170837523þ170837728
lysis buffer and incubated at room temperature for 5 minutes. The sample was then split into 2 separate tubes, and the cells pelleted at 440g for 5 minutes, and the supernatant was discarded. One of the resulting cell pellets was then frozen and stored at 808C for subsequent DNA extraction. The other cell pellet was lysed in TRIzol reagent (Thermo Fisher Scientific, Waltham, Massachusetts) at approximately 1 3 107 cells/mL and stored at 808C for subsequent RNA extraction. Reference genome DNA NA12878 was obtained from Coriell Institute for Medical Research (Camden, New Jersey).
DNA Extraction DNA from bone marrow aspirate samples was extracted with the EZ1 DNA Tissue kit on the EZ1 Advanced XL system (Qiagen, Hilden, Germany), per the manufacturer’s instructions. DNA quantification was performed using the Qubit double-stranded DNA (dsDNA) BR Assay kit using the Qubit 2.0 Fluorometer (Thermo Fisher Scientific). Qubit readings were used as a guide for DNA input into polymerase chain reactions (PCRs).
PCR and NGS Amplification of exon 11 (accession NM_002520.6) of NPM1 was performed using a single amplicon containing the Fluidigm (South San Francisco, California) universal forward and reverse sequencing tags (CS1 and CS2). Three different amplicons were designed (see Table 1) to overcome the issue of misattributed reads (see Results section). The PCR amplification was performed in duplicate on approximately 500 ng of genomic DNA using FastStart High Fidelity PCR System (Roche Diagnostics, Basel, Switzerland). The PCR conditions consisted of an initial denaturation step of 958C for 7 minutes, followed by 35 cycles of 958C for 30 seconds, 638C for 30 seconds, and 728C for 30 seconds, and a final elongation step at 728C for 7 minutes. The harvested PCR products were then used as template in a second PCR reaction with sample-specific barcode primers (Fluidigm), per the manufacturer’s instructions. Uniquely indexed samples were pooled and the resulting library was purified using the Agencourt AMPure XP system (Beckman Coulter, Brea, California). The resultant library was quantified on a TapeStation 2200 (Agilent Technologies, Santa Clara, California). Libraries were denatured and diluted, per manufacturer’s instructions, and 150– base pair (bp), paired-end sequencing was performed on an Illumina (San Diego, California) MiSeq sequencer using MiSeq version 2 chemistry. When used in diagnostic practice, the NGSqNPM1 samples were either sequenced alongside a routinediagnostic, 26-gene, targeted amplicon panel or libraries were spiked with a preexisting library to add complexity. To avoid carryover, the sequencers undergo a postrun wash after use and a maintenance wash every 7 days, per manufacturer’s instructions. In addition, runs containing NGS-qNPM1 samples were rotated sequentially among the 3 available Illumina MiSeq sequencers. Moreover, the 8 barcode plates were rotated so the same barcode plate was never used consecutively on the same instrument.
Bioinformatic Alignment and Variant Calling Raw reads were de-indexed and aligned using an in-house bioinformatic analysis pipeline. CASAVA (version 1.8.2, Illumina) was used to convert .bcl (base call) files generated from the MiSeq instrument into FASTQ files containing short-read data and to perform sample de-multiplexing. Reads were assembled, aligned, and variant called using a novel in-house–designed combination nonglobal alignment and variant caller (known as CANARY) with subsequent interpretation of the output using an in-house–designed Arch Pathol Lab Med—Vol 142, May 2018
clinical informatics software. All samples were tested in duplicate, and an average variant allele frequency (VAF) was calculated. A positive result was defined as the presence of 1 or more reads containing a pathogenic NPM1 mutation in either duplicate.
Comparison Methodologies Mutant NPM1 RT-qPCR.—Bone marrow aspirate samples were aliquoted and stored in TRIzol reagent for subsequent RNA isolation and RT-qPCR analysis (Department of Medical & Molecular Genetics, King’s College London, London, England). RNA extraction and quantitation of NPM1 mutations by RT-qPCR was performed as previously described.2,8 Briefly, NPM1-mutated transcripts were detected using a common primer and probe with a mutation-specific primer for type A or type B mutations.8 Assays were run in triplicate on the ABI 7900 platform (Thermo Fisher Scientific), and mutated transcript levels were compared with expression of the ABL1 reference gene using plasmid standards (Qiagen). The positivity of RT-qPCR results for mutated NPM1 was defined according to amplification in at least 2 of 3 replicates with cycle-threshold values of 40 or less (using a threshold setting of 0.1). The RT-qPCR results are expressed as a normalized copy number—NPM1 mutant copies per 100 ABL1 copies. Flow Cytometry.—Detection of AML MRD populations in bone marrow aspirate samples was performed by staining white blood cells for 15 minutes in the dark with a 4-tube, 8-color panel. Red cells were lysed with FACSLyse (BD Biosciences, San Jose, California) for 10 minutes in the dark at room temperature. Cells were then washed, and the sample was acquired on an 8-color FACSCanto II cytometer (BD Biosciences). More than 400 000 white blood cell count events acquired per tube was the standard aim. Abnormal clusters of blasts were determined after comparison to healthy pooled marrow, as well as to previously known abnormal diagnostic immunophenotypes, where available. Data were analyzed with Kaluza Analysis Software (Beckman Coulter). Droplet Digital PCR.—DNA samples were also quantified by droplet digital PCR, using the QX200 Droplet Digital PCR System (Bio-Rad Laboratories, Hercules, California). A 20-lL PCR reaction, containing 40 ng DNA (as quantified by Qubit fluorometer, Thermo Fisher Scientific), 10 lL droplet digital PCR Supermix for probes (no deoxyuridine triphosphate), 5 lL nuclease-free water, and 900 nM primers, and 250 nM hexachloro-fluorescein–labeled reference probe targeting KIT was partitioned into droplets using the Automated Droplet Generator (Bio-Rad). Amplification was performed on a C1000 Touch Thermal Cycler (Bio-Rad, 958C for 10 minutes, 40 cycles of 948C for 30 seconds, 558C for 1 minute, 988C for 10 minutes, and a final hold at 48C) followed by detection on a Bio-Rad QX200 Droplet Reader using QuantaSoft (version 1.7.4) software (Bio-Rad). Measured genome copies in 40 ng were extrapolated to the equivalent number of copies in 500 ng DNA.
Ethics This study was performed with approval by the ethics committee of the Peter MacCallum Cancer Centre (03/09, ethical oversight of pathology activity) and was conducted in accordance with the Helsinki Declaration of 1975, as revised in 2008.
RESULTS Lower Limit of Detection and Specificity The number of individual DNA molecules present in 500 ng of DNA was determined by droplet digital PCR NGS-qNPM1—Blombery et al 607
Table 2. Illustrative Single-Dilution Series of NPM1 Type-A Mutation Detection by Next-Generation Sequencing–Quantitative NPM1 Mutation Assessment Sample
Expected VAF, %
NPM1-10 NPM1-1 NPM1-0.1 NPM1-0.01 NPM1-0.001 NPM1-0.0001 NPM1-0.00001
10 1 0.1 0.01 0.001 0.0001 0.00001
Total Reads, No. 302 331 284 319 282 294 298
793 995 838 060 236 902 363
Mutated Reads, No.
Observed VAF, %
30 052 3955 308 44 8 1 0
9.9249 1.1913 0.1081 0.0138 0.0028 0.0003 0
Abbreviation: VAF, variant allele frequency.
performed on a series of consecutive bone marrow aspirate samples. These data were used to determine a theoretical maximal level of sensitivity for the NGS-qNPM1 assay. The mean number of wild-type alleles present in those representative samples from different patients was 182 661 (range, 165 132–199 793, n ¼ 9) giving a theoretical lower limit of detection of approximately 1 NPM1 mutant cell (assuming heterozygosity) in 105 cells. When used in routine diagnostic practice, the NGSqNPM1 samples were sequenced alongside a targeted, routine, diagnostic, 26-gene amplicon panel using a Fluidigm access array as the library preparation and sequenced on an Illumina MiSeq. Each run typically contained 18 to 20 targeted amplicon panel samples (run in duplicate) and 1 to 2 NGS-qNPM1 samples (see section on Integration Into Existing Workflow). Using this strategy, the mean read depth obtained for the NPM1 amplicon was 305 7243 (n ¼ 167). The lower limit of detection for the assay was empirically determined by diluting DNA from a patient sample containing a type-A NPM1 mutation (accession NM_002520:c.860_863dup) of known VAF with NPM1 wild-type DNA in a 10-fold dilution series to obtain a range of VAFs from 10% to 0.00001% (107). The NPM1 mutation was detectable in all samples down to the 0.001% (105) sample assayed on 5 separate runs. The NPM1 mutation was detectable in the 0.0001% (106) sample in only 2 of 5 runs (40%). The NPM1 mutation was never detectable in the 107 sample. An illustrative single-dilution series is shown in Table 2. Based on these data, the low positive control for the assay was chosen as 0.01% (104), which was assayed with each run. False-positive NPM1 mutant detection was assessed using a Coriell Cell Repository reference genome (accession NA12878). In pilot experiments, typical 4-bp NPM1 mutations were detected in the NA12878 reference genome at a level of approximately 5 to 10 reads when sequencing to a depth of 300 0003 to 500 0003. This phenomenon was eventually determined to be a manifestation of misattributed reads during sequencing (discussed below). When unique amplicons were used for each sample containing an NPM1 mutation, there were no background NPM1 4-bp insertions detectable at routine sequencing depth (300 0003– 500 0003). In addition, bone marrow aspirate samples from 5 patients without AML were sequenced at 300 0003 to 500 0003. No NPM1 mutant reads were detected in those patients.
and the observed VAF (R2 ¼ 0.9969) in the dilution series (Figure 1). Intrarun reproducibility was determined by 4 replicates of samples containing an NPM1 type-A mutation at 2 separate dilutions (1% and 0.01%). Replicates consisted of the same first-round PCR product sequenced with different barcode indexes on the same sequencing run. The replicate samples demonstrated highly reproducible VAFs with a coefficient of variation of 0.004 and 0.123 for the 1% and 0.01% dilutions, respectively. Interrun reproducibility was assessed by evaluating the results of the 0.01% positive control on 47 separate runs. The mean (SD) VAF was 0.00998% (0.0038%), giving a coefficient of variation of 0.38 (38.1%) at a VAF of 0.01%. Comparison With Orthogonal Methodologies A cohort of samples from patients with NPM1 mutations being monitored after AML treatment (n ¼ 40) was run concurrently on both the NGS-qNPM1 assay and an established RT-qPCR assay.2 Of the 40 samples, 39 (98%) showed concordant categorization for detection of the NPM1 mutation (Table 3). The discordant sample was detected by the NGS-qNPM1 assay but not by the RT-qPCR assay because degraded RNA quality adversely affected the RT-qPCR sensitivity, as evidenced by the low ABL1 copy number (Table 3). Given the different assay methodologies (eg, expression in RT-qPCR versus direct mutation detection in NGS-qNPM1) and the different specimens used (RNA for RT-qPCR and DNA for NGS-qNPM1), the quantitative values for each method cannot be directly compared, and the analysis was limited to categorization detection of the
Assay Performance The quantitative performance of the NGS-qNPM1 assay was assessed by performing serial 10-fold dilutions of DNA with a known NPM1 mutation in wild-type DNA. A high level of correlation was observed between the expected VAF 608 Arch Pathol Lab Med—Vol 142, May 2018
Figure 1. Serial dilution curve for patient sample DNA containing a type-A, 4-bp insertion in NPM1 assessed with next-generational sequencing–quantitative NPM1 mutation assessment. Abbreviation: VAF, variant allele frequency. NGS-qNPM1—Blombery et al
Table 3.
Case No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 a b c
Comparison of Next-Generation Sequencing–Quantitative NPM1 Mutation Assessment (NGS-qNPM1) With NPM1 Real-Time Quantitative-Polymerase Chain Reaction (RT-qPCR)
NGS-qNPM1 Total Reads,a No. 158 249 194 235 332 195 228 163 335 278 310 439 253 323 421 413 367 277 206 224 355 297 359 412 428 305 359 399 437 252 227 232 267 314 385 255 357 371 241 332
638.5 602 636.5 138 593 468 693 707 646 908 862 856 715 900 478.5 809 026 523.5 924 133 455 948.5 413.5 336 539 357 195 632.5 179 476.5 539.5 780 789 587 558 716.5 461.5 325.5 262.5 976
NGS-qNPM1 Mutant Reads,a No.
NGS-qNPM1 Variant Allele Frequency, %
RT-qPCR ABL1 Copies, No.
RT-qPCR Mutant NPM1 Copies, No.
Normalized Copy No.b
Concordant
0 0 0 0 0 0 0 0 19 0.5 12.5 37 7 5.5 7.5 265.5 35 15 1654.5 6 17.5 9.5 87.5 1 245 16.5 71 90 605.5 187 93.5 556 186 44 48 119.5 292 687.5 1782.5 195
0 0 0 0 0 0 0 0 0.0057 0.0002 0.0040 0.0084 0.0028 0.0017 0.0018 0.0642 0.0095 0.0054 0.7996 0.0027 0.0049 0.0032 0.0243 0.0002 0.0572 0.0054 0.0198 0.0225 0.1385 0.0657 0.0411 0.2389 0.0695 0.0140 0.0125 0.0467 0.0817 0.1851 0.7388 0.0586
69 071 6979 58 959 38 272 10 895 11 664 16 269 1101 673 77 022 41 379 49 394 47 152 164 810 56 890 63 236 21 018 16 105 10 663 9931 53 185 58 849 38 628 47 958 6999 35 152 5910 35 544 73 890 74 919 58 313 5795 18 538 40 561 41 760 45 693 48810 60 660 23 480 100 349
0 0 0 0 0 0 0 0 0 1 1 4 6 26 11 19 7 6 5 5 33 43 41 75 25 172 33 220 517 717 743 74 250 1079 1189 2343 3670 4776 2757 40 400
0 0 0 0 0 0 0 0 0 0.0013 0.0024 0.0081 0.0127 0.0158 0.0193 0.0300 0.0333 0.0373 0.0469 0.0503 0.0620 0.0731 0.1061 0.1564 0.3572 0.4893 0.5584 0.6190 0.6997 0.9570 1.2742 1.2770 1.3486 2.6602 2.8472 5.1277 7.5190 7.8734 11.7419 40.2595
Yes Yes Yes Yes Yes Yes Yes Yes Noc Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Average across duplicates. Mutant NPM1 copies per 100 ABL1 copies. Reduced sensitivity because of low reference transcript numbers.
NPM1 mutation; however, samples with a range of leukemic burdens were selected (see Table 3). In addition, a different 48 samples were selected that had concurrent flow cytometry performed as part of the patients’ diagnostic assessments. Of 48 samples, 23 (48%) had concordant categorization as either positive or negative. In 22 of the 48 samples (46%), flow cytometry did not detect an abnormal blast population, whereas NPM1 mutations were detectable by NGS-qNPM1, reflecting the higher sensitivity of the NGS-qNPM1 than flow cytometry. Misattributed Reads Although NGS experiments were designed to minimize any sample crossover that might interfere with accurate detection of mutated reads at ultralow levels, the occurrence of a few mutated reads in negative control samples was observed in early validation experiments. Those unexpected mutated reads were present only when samples with ‘‘high’’ VAFs (eg, 30%–40%) were sequenced at ultradeep levels (eg, 300,0003–500,0003) on the flow cell. It was hypothesized Arch Pathol Lab Med—Vol 142, May 2018
that the mutated reads observed in the negative control actually represented mutated reads that were present in other samples in the run being ‘‘misattributed’’ to the negative control sample. That hypothesis was confirmed by using amplicons with different primers for the negative control sample and samples with a known significant burden of mutated reads and directly observing the misattributed reads in the sequence alignment data. Those experiments demonstrated that an average of 0.047% (range, 0.013%–0.199%) of reads from deeply sequenced amplicons were misattributed to other samples. To avoid that source of bioinformatic contamination, 3 unique amplicons (ie, with different start and stop positions) were designed covering the region of NPM1 mutation and used for different samples carrying the same mutation (Table 1). No further low-level interference was subsequently observed. Integration Into Existing Workflow The molecular hematology laboratory at the Peter MacCallum Cancer Centre has an existing NGS workflow NGS-qNPM1—Blombery et al 609
Figure 2. Combined sequencing workflow for next-generation sequencing (NGS) measurable residual disease (MRD) and multigene amplicon panels. Specific MRD steps are shown in yellow, amplicon panel laboratory (lab) steps are shown in blue, and amplicon panel analysis steps are shown in green. Abbreviations: BAM, .bam file; GATK, Genome Analysis Toolkit; HGVS, Human Genome Variation Society; PCR, polymerase chain reaction; VAF, variant allele frequency; VCF, .vcf file; VEP, variant effect predictor. PathOS is in-house–developed software at the Peter MacCallum Cancer Centre, Melbourne, Australia.
for multigene amplicon-sequencing panels. To streamline and simplify the NGS-qNPM1 testing process, we aimed to integrate the 2 workflows. A flow chart outlining the combined workflow is shown in Figure 2. The NGS-qNPM1 samples, along with appropriate controls, were PCR amplified in a separate reaction and added into the NGS sequencing panel workflow at the point of the addition of sample barcodes. From that point onward, all samples followed the same workflow and were processed together as a single sequencing library. To flag NGS-qNPM1 samples to 610 Arch Pathol Lab Med—Vol 142, May 2018
be automatically processed by the NGS-qNPM1–specific bioinformatics pipeline, those samples were assigned an NGS-qNPM1 manifest file at the sequencing stage. Clinical Utility After validation of the NGS-qNPM1 assay, it has been implemented into routine use as a diagnostic test. Approximately 100 clinical patient samples have been run on this assay to date in a diagnostic context, with results observed NGS-qNPM1—Blombery et al
across the spectrum of VAFs, proving clinical utility in 2 major contexts: 1. Confirmation of remission status in the context of morphological/immunophenotypically ambiguous disease. An example of this clinical context was a 65year-old man with healthy-karyotype AML and monocytic differentiation. Targeted amplicon sequencing at baseline revealed an NPM1 (VAF, 37.7%; NM_002520.6, c.860_863dup; p.Trp288Cysfs*12) mutation as well as mutations in IDH1 (NM_005896.2, c.394C.T; p.Arg132Cys; VAF, 44.3%) and NRAS (NM_002524.4, c.181C.A; p.Gln61Lys; VAF, 49.9%). The patient underwent remission-induction therapy with HiDAC (high-dose Ara-C) and had morphological evidence of disease at the end of induction. He received reinduction therapy with fludarabine and cytarabine (FLAG), and on a bone marrow test performed on day 28 was noted to have 15% blasts by morphological assessment but with a spectrum of blast morphology, including undifferentiated blasts and primitive cells with morphological evidence of monocytic differentiation. Flow cytometry was difficult to interpret in this case given the immunophenotypic monocytic differentiation. NGSqNPM1 was performed at that point, which showed an NPM1 VAF of 0.0018%, suggesting a significantly lower leukemic burden than suspected based on morphology. The bone marrow was repeated 1 month later and showed an ongoing reduction in morphological blast percentage (7%), despite no intervening therapy. 2. Detection of early relapse. An example of this clinical context was a 68-year-old man with NPM1-mutated, healthy-karyotype AML who was treated to morphological remission with induction chemotherapy, followed by 2 cycles of consolidation chemotherapy. His NGS-qNPM1 results were negative at the end of treatment. He was followed with 3 monthly bone marrow biopsies with NGS-qNPM1 testing. Six months after finishing therapy, his NGS-qNPM1 was measured at 0.061%. Bone marrow biopsy at that stage showed an ongoing morphological and immunophenotypic remission. He underwent a repeat bone marrow biopsy 4 weeks later, and the NGS-qNPM1 was measured at 0.27%, and he subsequently underwent an allogeneic bone marrow transplant although still in a morphological remission. Of note, we have also observed one case of relapsed AML that had NPM1 wild type, whereas it had been NPM1 mutated at diagnosis. The patient had normal-karyotype AML with an NPM1 c.863_864insCATG and an IDH1 Arg132Cys mutation detected at diagnosis. The patient achieved morphological and immunophenotypic remission after induction and consolidation chemotherapy, and both the NPM1 and IDH1 mutations became undetectable. The patient’s NGS-qNPM1 was consistently negative in remission. Despite the negative NGS-qNPM1 testing, she developed progressive cytopenias and was found to have overt morphological relapse on bone marrow biopsy approximately 1 year after remission. Although her original NPM1 remained undetectable, the IDH1 Arg132Cys was detectable again in the bone marrow aspirate taken at relapse. Arch Pathol Lab Med—Vol 142, May 2018
DISCUSSION We have described the validation and clinical utility of an NGS assay for quantitative assessment of NPM1 in AML. With 500 ng of DNA input, this NGS-qNPM1 assay has an empirically determined lower limit of detection of 105 with performance characteristics comparable to other published quantitative molecular methods.9 Moreover, this assay has been integrated into a diagnostic laboratory with an NGSamplicon workflow and has shown clinical utility in the monitoring of patients with AML. One of the advantages of the NGS-qNPM1 assay, rather than RT-qPCR NPM1 testing, is that RT-qPCR assays are allele specific, and therefore, the specific mutation present in the patient at diagnosis needs to be known, and mutationspecific primers are required. That can limit testing in the ‘‘real-world’’ diagnostic setting when the mutation-type present at baseline may not be known. Another potential advantage of the NGS-qNPM1 assay is the use of DNA as the input material, which is easier to work with technically in the laboratory (because of its stability and transport requirements) compared with the RNA used in RT-qPCR assays. This was demonstrated in our orthogonal testing cohort in which the only discordant result was due to degraded RNA resulting in reduced sensitivity from the RTqPCR assay to a level that would have precluded its use in clinical reporting. It is notable that, although there is a general positive correlation between the values obtained for RT-qPCR and NGS-qNPM1 quantitation, there are outlier cases in both directions (ie, high RT-qPCR with a relatively low NGSqNPM1 level and vice versa). That may be related to RTqPCR being an expression-based assay with discrepant cases being the result of individual AML patients with either relatively high or low NPM1 expression within their leukemic blasts. The 2 different assay methodologies may, therefore, provide complementary insights into the leukemic biology of individual patients, with the NGS-qNPM1 assay providing a purely quantitative measure of leukemic burden, and the RT-qPCR assay providing an integration of NPM1 quantitation and expression. Although NGS-qNPM1 had similar sensitivity to the RTqPCR assay, the NGS-qNPM1 was clearly superior to flow cytometry for the detection of low levels of disease. The lower limit of detection of FC in AML MRD is determined by numerous factors, including the number of events acquired, the specific leukemia-associated phenotype, and the degree of background regenerating bone marrow observed. All of these factors result in a typical, sensitive lower limit of detection for AML flow cytometry in the diagnostic laboratory of approximately 104 to 103. In addition, significant interlaboratory and interobserver variability has been observed in the interpretation of AML MRD FC results. The NGS-qNPM1 assay, therefore, offers a more-objective and quantitative measure than FC in this setting. Of note, the literature regarding the predictive value of NPM1 MRD detection has largely been performed using RTqPCR.2,10,11 Those studies have generally identified different time points in treatment in which detectable NPM1 mutant expression is predictive of inferior outcomes.1,2 Importantly, the ability to predict patient outcome at defined time points is dependent on the nature (ie, type and intensity) of induction and consolidation treatment. Therefore, the applicability of a given NPM1 quantitative level at any NGS-qNPM1—Blombery et al 611
absolute time point in treatment cannot be assumed with certainty, outside the specific treatments given in those studied cohorts. Further confounding the generalizability of those results at absolute time points is the lack of standardization for NPM1 RT-qPCR assays among centers. Therefore, outside of specified treatments and prospective trials, the predominant utility of any quantitative NPM1 assay (including both the NGS-qNPM1 and RT-qPCR) is in informing treatment decisions based on relative changes in an individual patient over time (eg, detecting early relapse when early intervention may be possible) and in the resolution of cases with difficult-to-interpret morphology and flow cytometric findings (as illustrated in the 2 case examples presented). Simultaneous sequencing of multiple samples in NGS is possible because of DNA bar-coding, which allows subsequent bioinformatic de-multiplexing of sample-specific data. False assignment of reads to unique sample barcodes has been previously observed using the Illumina sequencing platform and is thought to be either a result of mixed clusters on the flow cell or jumping PCR.12 The accuracy of barcode sequence assignment to specific reads during data acquisition and bioinformatic handling of those data is crucial to the correct detection of low-level mutations. The phenomenon of misattributed reads within multiplexed data sets has been previously described,13 and low-level crosstalk among samples appears to be an unavoidable characteristic of the sequencing by synthesis technology.12 Although these misattributed reads are generally at a level that does not interfere with standard NGS variant detection, they are potentially a significant hurdle to detection of ultralow-frequency mutations. A level of misattributed reads of 0.047%, as observed in our initial experiments, can potentially mean significant ‘‘bioinformatic contamination’’ of other samples. For example, if a sample with an NPM1 mutation present at a VAF of 1% is sequenced to 500 0003, then approximately 200 to 300 reads would be expected to be misattributed to each sample on the run, and of those, 2 to 3 would be expected to contain NPM1 mutated reads that could confound analysis. The presence of those small amounts of misattributed reads both reduces assay sensitivity and creates the potential for misinterpretation of results. The ability to confirm the correct sample of origin for a mutated read by an additional method to the standard sample barcode, either by an additional barcode (duplex sequencing) or by unique amplicon start and stop sites (such as in our method) is, therefore, required as part of any NGSqNPM1 protocol using the methodology described. For clinical diagnostic testing, we recommend the use of primer pairs with unique start and stop sites for each patient sample on the same sequencing run. Although the performance characteristics of our assay (eg, the coefficient of variation) are broadly similar with the initial description of NGS for sensitive NPM1 detection,7 our assay demonstrated a lower limit of detection (106) that is potentially the result of the greater DNA input used in our
612 Arch Pathol Lab Med—Vol 142, May 2018
assay (500 ng). However, although detection of NPM1 mutations is possible at 106 by our assay, the assay did not reproducibly detect mutations at that level. Future improvements to this assay include further optimizing the PCR conditions to enable greater DNA inputs and, therefore, lower limits of detection. In addition, the use of unique molecular indexes would be of value to precisely quantitate the degree of sequencing for PCR duplicates. In summary, we have developed, validated, and implemented an effective and flexible method for NPM1 quantitation, which can be used in the monitoring of patients with AML. This assay has attractive performance characteristics and is able to be used within the workflow of the NGS diagnostic laboratory. Moreover, this assay shows concordance with currently accepted standards for quantitative assessment at an MRD level (RT-qPCR and flow cytometry). With the accelerating uptake of NGS technology into diagnostic laboratories worldwide, we see this assay as a valuable addition to the suite of testing available in hematologic malignancy. We thank Adam Ivey, Msc, for performing the correlative NPM1 RT-qPCR assays and the Snowdome Foundation for their support of this work. References 1. Kronke J, Schlenk RF, Jensen KO, et al. Monitoring of minimal residual ¨ disease in NPM1-mutated acute myeloid leukemia: a study from the GermanAustrian acute myeloid leukemia study group. J Clin Oncol. 2011;29(19):2709– 2716. 2. Ivey A, Hills RK, Simpson MA, et al; UK National Cancer Research Institute AML Working Group. Assessment of minimal residual disease in standard-risk AML. N Engl J Med. 2016;374(5):422–433. 3. Terwijn M, van Putten WL, Kelder A, et al. High prognostic impact of flow cytometric minimal residual disease detection in acute myeloid leukemia: data from the HOVON/SAKK AML 42A study. J Clin Oncol. 2013;31(31):3889–3897. 4. Loken MR, Alonzo TA, Pardo L, et al. Residual disease detected by multidimensional flow cytometry signifies high relapse risk in patients with de novo acute myeloid leukemia: a report from Children’s Oncology Group. Blood. 2012;120(8):1581–1588. 5. Freeman SD, Virgo P, Couzens S, et al. Prognostic relevance of treatment response measured by flow cytometric residual disease detection in older patients with acute myeloid leukemia. J Clin Oncol. 2013;31(32):4123–4131. 6. Grimwade D, Freeman SD. Defining minimal residual disease in acute myeloid leukemia: which platforms are ready for ‘‘prime time’’? Hematology Am Soc Hematol Educ Program. 2014;2014(1):222–233. 7. Salipante SJ, Fromm JR, Shendure J, Wood BL, Wu D. Detection of minimal residual disease in NPM1-mutated acute myeloid leukemia by next-generation sequencing. Mod Pathol. 2014;27(11):1438–1446. 8. Gorello P, Cazzaniga G, Alberti F, et al. Quantitative assessment of minimal residual disease in acute myeloid leukemia carrying nucleophosmin (NPM1) gene mutations. Leukemia. 2006;20(6):1103–1108. 9. Jobbagy Z, van Atta R, Murphy KM, Eshleman JR, Gocke CD. Evaluation of the Cepheid GeneXpert BCR-ABL assay. J Mol Diagn. 2007;9(2):220–227. 10. Schnittger S, Kern W, Tschulik C, et al. Minimal residual disease levels assessed by NPM1 mutation-specific RQ-PCR provide important prognostic information in AML. Blood. 2009;114(11):2220–2231. 11. Kayser S, Benner A, Thiede C, et al. Pretransplant NPM1 MRD levels predict outcome after allogeneic hematopoietic stem cell transplantation in patients with acute myeloid leukemia. Blood Cancer J. 2016;6(7):e449. 12. Kircher M, Heyn P, Kelso J. Addressing challenges in the production and analysis of Illumina sequencing data. BMC Genomics. 2011;12:382. 13. Kircher M, Sawyer S, Meyer M. Double indexing overcomes inaccuracies in multiplex sequencing on the Illumina platform. Nucleic Acids Res. 2012;40(1): e3.
NGS-qNPM1—Blombery et al
