Analytical validation and performance characteristics of a 48-gene next-generation sequencing panel for detecting potentially actionable genomic alterations in myeloid neoplasms

Identification of genomic mutations by molecular testing plays an important role in diagnosis, prognosis, and treatment of myeloid neoplasms. Next-generation sequencing (NGS) is an efficient method for simultaneous detection of clinically significant genomic mutations with high sensitivity. Various NGS based in-house developed and commercial myeloid neoplasm panels have been integrated into routine clinical practice. However, some genes frequently mutated in myeloid malignancies are particularly difficult to sequence with NGS panels (e.g., CEBPA, CARL, and FLT3). We report development and validation of a 48-gene NGS panel that includes genes that are technically challenging for molecular profiling of myeloid neoplasms including acute myeloid leukemia (AML), myelodysplastic syndrome (MDS), and myeloproliferative neoplasms (MPN). Target regions were captured by hybridization with complementary biotinylated DNA baits, and NGS was performed on an Illumina NextSeq500 instrument. A bioinformatics pipeline that was developed in-house was used to detect single nucleotide variations (SNVs), insertions/deletions (indels), and FLT3 internal tandem duplications (FLT3-ITD). An analytical validation study was performed on 184 unique specimens for variants with allele frequencies ≥5%. Variants identified by the 48-gene panel were compared to those identified by a 35-gene hematologic neoplasms panel using an additional 137 unique specimens. The developed assay was applied to a large cohort (n = 2,053) of patients with suspected myeloid neoplasms. Analytical validation yielded 99.6% sensitivity (95% CI: 98.9–99.9%) and 100% specificity (95% CI: 100%). Concordance of variants detected by the 2 tested panels was 100%. Among patients with suspected myeloid neoplasms (n = 2,053), 54.5% patients harbored at least one clinically significant mutation: 77% in AML patients, 48% in MDS, and 45% in MPN. Together, these findings demonstrate that the assay can identify mutations associated with diagnosis, prognosis, and treatment options of myeloid neoplasms even in technically challenging genes.

. Coupled with clinical, morphologic, and immunophenotypic abnormalities, identification of genetic alterations by molecular testing has an important role in the classification, risk stratification, and management of myeloid neoplasms [2]. For example, mutations in NPM1, CEBPA, and RUNX1 can identify specific subclass of AML, and mutations in FLT3, IDH1, IDH2, and JAK2 can guide targeted therapies for AML [3]. For patients with MDS, identification of mutations in genes including TET2, SF3B1, ASXL1 and TP53 is particularly useful to help establish clonal hematopoiesis to make a definitive diagnosis of MDS [4]. In addition, for MPN, certain mutations such as JAK2 V617F or exon 12 mutation satisfy diagnostic criteria to help establish a diagnosis of MPN [5]; mutations in ASXL1, SRSF2, EZH2, IDH1, and IDH2 categorize patients to high molecular risk in Primary Myelofibrosis (PMF) [6]; and mutations in IDH2, U2AF1, EZH2, TP53, SH2B3, and SF3B1 indicate adverse prognostic value in Essential Thrombocythemia (ET) and Polycythemia Vera (PV) [7,8].
Historically, single gene testing using Sanger sequencing or real-time PCR have been used to identify genetic alterations in myeloid neoplasms [9]. Although these methods are readily adaptive and widely used in clinical laboratories, their major limitation is that very few selected mutations can be detected at a time. In recent years, whole genome sequencing (WGS) and whole exome sequencing (WES) studies have been used to identify oncogenic mutations in hundreds of genes, revealing a wide genetic heterogeneity in myeloid neoplasms [9]. Indeed, more molecular genetic markers have been added in the most recent revision of the WHO classification [1]. Since the number of mutations that can guide diagnosis, prognosis, and treatment options are increasing, using single-gene testing for myeloid malignancies is becoming impractical. Although WGS and WES are standard in research settings, targeted next generation sequencing (NGS) panel assays that are composed of genes associated with a disease and recurrently mutated are more commonly used in clinical settings. Compared to WGS or WES, targeted NGS assays are highly sensitive for detecting low-frequency variants and can identify a number of mutations that are critical in diagnosis and risk stratification in a relatively short time [10].
In recent years, various NGS based myeloid neoplasm panels developed in-house and commercially have been integrated into routine clinical practice [10][11][12][13]. Panels developed inhouse can differ substantially between laboratories in many aspects including gene content, the analysis of genes in a panel, sequencing library preparation chemistry, sequencing platform, and variant types detected [14][15][16][17][18][19][20]. Among many others, amplicon-based commercial panels, e.g. Illumina TruSight Myeloid panel, Thermo Fisher Oncomine Myeloid research panel, and Archer VariantPlex Core Myeloid panel are most commonly used in clinical laboratories [12,21,22]. However, some genes frequently mutated in myeloid malignancies are particularly difficult to sequence with NGS, especially with amplicon-based or RNA bait capture-based panels. For example, technical difficulties hinder the ability to capture targets with high GC content, such as CEBPA, which is associated with poor prognosis of AML, or repetitive genomic regions, such as FLT3-ITD, which is associated with poor prognosis of AML in the absence of NPM1 mutation [10,12,[21][22][23][24][25][26][27]. Aguilera-Diaz and colleagues evaluate the performance of four different targeted NGS gene panels, three commercial panels and one in-house developed panel, and noted that CEBPA, CALR and FLT3 genes remain challenging the use of NGS for diagnosis of myeloid neoplasms in compliance with current guidelines [21]. Similarly, Lzevbaye and colleagues evaluated an amplicon based commercial panel and noted that the panel is unable to detect specific variants in ASXL1 and CEBPA in long homopolymer regions [22]. The application of NGS in clinical settings has additional challenges. For example, many artifacts are known to arise during NGS library preparation, sequencing, and data analysis (e.g., read mapping, variant calling) and these may cause challenges in discriminating true genetic alterations from artifacts caused by PCR, sequencing, and post-sequencing steps. The lack of uniform practice standards for quality assessment of NGS data also challenges implementation of NGS in clinics [11,28]. As such, careful design of panel contents and workflow as well as through analytical validation of each clinical NGS assay for its defined performance requirements for the intended use is critical to ensure the clinically impactful high-quality sequencing results in clinical settings.
In this study, we report development and validation of a 48-gene NGS panel that includes technically challenging genes for the detection of alterations that have a putative role in diagnosis, prognosis, or therapy of myeloid neoplasms. We also report our experience with the first 2,053 clinical specimens of suspected myeloid neoplasms submitted for the 48-gene panel and estimate the frequency of actionable alterations in a clinical laboratory.

Validation specimens
A total of 184 unique specimens (S2 Table) were included: 96 whole blood, 20 bone marrow aspirate, 20 cell pellet, 17 FFPE, and 31 extracted DNA. The 184 specimens included 32 commercial controls, 25 from healthy individuals, and 127 clinical specimens previously tested by alternative analytical methods for indications of myeloid neoplasms (n = 105), unknown history (n = 18), and adjacent normal tissue from FFPE section (n = 4). All patient specimens were de-identified before use. Alternative analytical methods included Sanger sequencing, fragment analysis for FLT3-ITD, and 4 different CLIA-validated NGS assays developed in our laboratory (see Alternative analytical methods, below). The 32 commercial controls included 6 multiplex mutation controls (5% Tier, Horizon discovery, Lafayette, CO), 19 controls with known mutation(s) within the 48 genes in the panel (Coriell Institute, Camden, NJ), and 7 well-characterized Genome In a Bottle (GIAB) reference specimens (Coriell Institute, Camden, NJ). For the 7 GIAB specimens, high-quality public sequence data were downloaded from http://ftp-trace.ncbi.nlm.nih.gov/giab/ftp/release/ for comparison.

Alternative analytical methods
Sanger sequencing on CEBPA or JAK2 were performed using gene-specific PCR primers for target amplification and incorporation of BigDye Terminator (Applied biosystems, Foster City, CA, USA). T7 promoter and terminator sequences fused to gene specific primers were used for sequencing on an ABI 3730 fragment analyzer (Applied biosystems, Foster City, CA, USA).
FLT3-ITD fragment analysis was performed by Laboratory for Personalized Molecular Medicine (LabPMM, San Diego, CA) or in-house using a modified protocol based on one reported previously [31]. In brief, FLT3-ITD was PCR-amplified with a fluorescence-labeled forward primer, and a non-labeled reverse primer. The PCR products were analyzed using an ABI3730 genetic analyzer (Applied Biosystems, Foster City, CA, USA), and the amplicons with a size greater than that of wild type (324-326 bp) were interpreted as positive for the FLT3-ITD.
Four separate CLIA-validated NGS panel assays were used for the accuracy study to

DNA isolation, library preparation, and NGS
Genomic DNA was isolated from whole blood, bone marrow, or cell pellet using Qiagen EZ1 kit or DSP DNA mini kit (Qiagen, Mississauga, ON). FFPE DNA was extracted using QIAAMP DNA FFPE tissue kit (Qiagen, Mississauga, ON), or Arcturus PicoPure kit (Thermo Fisher Scientific, Markham, ON). Isolated genomic DNA was mechanically sheared to an average size of 250 bases using a Covaris instrument LE220 (Covaris Inc., Woburn, MA). The fragmented DNA was enzymatically repaired and end-modified with adenosine (NEBNext1 Ultra™ II End Repair/dA- Tailing

Bioinformatics processing
De-multiplexing and conversion of NextSeq500 BCL files were done by using Illumina's bcl2fastq software utility. The raw sequence reads in FASTQ files were then aligned to the Genome Reference Consortium human genome build 37(GRCh37) by using BWA alignment package. Reads were sorted and indexed using SAMtools with subsequent reads duplication removal by Picard Tools. Local realignment and base quality score recalibration were performed using the Genome Analysis Toolkit (GATK). Mapped reads were further filtered by mapping score � 30 (� 99.9% accuracy) and base quality score � 20 (� 99% accuracy) before downstream analysis. This assay covered 98,809 bp from a total of 999 target regions, approxi-mately100 bp for each target, across the 48 genes in the panel. Average and minimum depth of coverage for every region of interest (ROI) and depth of each targeted positions were computed using SAMtools Pysam. SNVs and short indels were called by MuTect2 and LoFreq. CALR indels and FLT3-ITD were called by PINDEL. Criteria used for specimen and variant quality control are provided in S3 Table. Variant call comparison

Clinical specimens
A total of 2,053 consecutive patient specimens submitted for the 48-gene NGS panel were included in this study. Patient results were de-identified before analysis. Based on the clinical information submitted, indications for testing included AML (23.9%, n = 490), MDS (49.6%, n = 1,018), or MPN (26.5%, n = 545). Patient characteristics are presented in Table 1. This

Variant classification
For the 2,053 clinical cases, clinical interpretation of variants was performed in conjunction with a third-party annotation group (N-of-One, Concord, MA). Detected variants were classified in the following categories based on the 2017 guideline recommendations by AMP/ ASCO/CAP [33]: Tier I, strong clinical significance; Tier II, potential clinical significance; Tier III, uncertain clinical significance; and Tier IV, benign or likely benign. A pathologist or a licensed clinical laboratory director reviewed the results and the clinical annotation, edited information as deemed necessary for specific cases and indications, and signed out the results for reporting. For this study, variants were considered clinically significant if they provided prognostic or diagnostic information for the disease or were clinically actionable. Variants were considered clinically actionable when a targeted therapy or an experimental drug was available for the disease or other disease.

Assay performance
The analytical validation study was performed on 184 unique specimens for a total of 427 trials in 20 consecutive sequencing runs. On average, each specimen generated 11.8 million reads; 100% (SD = 8.8%) of reads mapped to the reference sequence (hg19); 65% (SD = 8.8%) of reads were on-target; and average coverage depth across target regions was 1,767 (SD = 875) (S4 Table). Average and median coverage of each target region across all validation specimens are plotted in Fig 1A. Of the 184 specimens, 181 (98.3%) specimens exceeded the coverage requirement in 422 trials (98.8%) and 3 specimens (1 whole blood, 1 FFPE, and 1 extracted DNA) failed in 5 trials. This result demonstrated that the developed workflow is robust for routine clinical testing and compatible with different specimen types. We next assessed coverage at each target region as a function of target region % GC using a control specimen NA12878, which was tested 17 times. Throughout the whole % GC spectrum (range: 12%-87%), we observed narrow coverage distribution, even in the extreme end of % GC (Fig 1B). In addition, we reviewed specimen pass rates of each target for Q20 coverage of 500X, 250X, or 100X (Fig 1C). A limited number of target regions did not achieve 500X coverage in >10% of validation specimens. Those regions included STAG2 exon 11, 16, and 21, all of which were situated near polyA tracts and represented 0.36% (354/98,809bp) of our target capture region. All 3 target regions achieved 100X coverage in 100% of validation sets and 250X coverage in >96% of validation sets.

Precision
To test inter-assay precision, 22 specimens that harbored at least one pathogenic mutation by alternative methods and a negative control (NA12878) were analyzed 3 times in 3 independently prepared sequencing runs. A total of 141 variants (98 SNVs, 39 indels, and 4 FLT3-ITD) were detected and were 100% concordant among 3 repeated runs (S5A Table). As expected, no reportable variant was detected from the negative control (NA12878) after applying variant filtering rules (S3 Table). Intra-assay precision was tested on 9 specimens that were tested in triplicate within a run. A total of 40 variants (30 SNVs, 8 indels, and 2 FLT3-ITD) were detected and all were concordant among the triplicates (S5B Table). In addition, detected variant frequency among replicates was reproducible within coefficient of variation <0.2 in 98.3% and <0.25 in 100% of variants (n = 181) tested in the precision studies (Fig 2A and 2B). In summary, both intra-and inter-assay variant call concordance was 100%.

Accuracy
A total of 140 unique specimens harboring variant(s) identified by alternative analytical methods, such as Sanger sequencing, fragment analysis, and other CLIA validated NGS assays, were

PLOS ONE
Analytical validation of a 48-gene next-generation sequencing panel for myeloid neoplasms analyzed for a total of 237 trials (S6 Table). The identified 165 unique variants by alternative methods consisted of 97 SNVs, 49 indels (1-33 bp insertions, 1-52 bp deletions), and 19 FLT3-ITDs (18-117 bp). From the 237 trials, collectively, 405 variants of various frequencies were analyzed including 229 SNVs, 117 indels, and 59 FLT3-ITDs in 29 genes (Fig 3A and 3B). All expected SNV and indel variants were accurately detected. If variant frequency was provided by an alternative method (n = 106), frequency of detected variant was highly concordant with expected (R 2 = 0.966, Fig 3C). In addition, all expected FLT3-ITD variants from 59 trials of 27 unique specimens were detected by this assay ( Table 2). In 18.5% (5/27) of FLT3-ITD positive specimens, additional ITD sizes that were not reported by a fragment analysis method were detected; this result may indicate higher sensitivity of NGS assay for FLT3-ITD detection.
We extended the accuracy study to 7 well-characterized GIAB reference specimens by comparing variant calls made by our pipeline with publicly available data. The variant concordance study was limited to those regions where high-quality public sequence data were available, approximately 62% to 86% of our target regions for each GIAB reference specimen ( Table 3). Population SNVs that would normally be excluded from reporting were also included for this study. Across the 7 reference specimens, a total of 464,162 bp were analyzed. All 142 expected variants were correctly detected ( Table 3), and no false positives were called, yielding 100% sensitivity (95% CI: 97.3-100%) and 100% specificity (95% CI: 100%). Review of the detected variant frequency at the expected homozygous and heterozygous positions revealed the mean observed variant frequency to be 99.9% (N = 70, SD = 0.1) and 49.6% (N = 72, SD = 2.6), respectively, demonstrating good agreement with the expected values.
In addition, 3 FLT3-ITD positive specimens were serially diluted up to 16-fold with NA12878. We considered FLT3-ITD positive when at least one ITD size is called. All dilution series yielded correct ITD calls (S7D Table). As the orthogonal method does not provide absolute FLT3-ITD frequency, in order to estimate sensitivity, we simulated expected variant frequency based on the relative fraction of different ITD sizes in a specimen, WB52. WB52 contained 3 different ITD sizes (21b, 51b, and 90b) with relative ITD fraction of 1.4, 75.7, and 23.0%. The major ITD size (51b) was detected from all dilution series including 16-fold dilution with simulated VAF of 4.7% (16-fold dilution of 75.7%). This result demonstrated that FLT3-ITD detection sensitivity to be at least 5%.
A summary of analytical sensitivity study results is provided in Table 4. Inspection of the discordant variants from this study (12 at �3.1% VAF and 3 at 5% VAF) showed that those variants had been detected by a variant caller, but the calls were filtered out because of low frequency (<3%) or low variant count (<25) (S8 Table).

Variant call comparison study
To increase per-sample data comparison efficiency, we compared variants of an additional 137 unique specimens identified by the validated 48-gene NGS panel for myeloid neoplasms to those detected by an independently developed and analytically validated 35-gene NGS panel for hematologic neoplasms (HemeSEQ, med fusion, Lewisville, TX). The 137 specimens were selected because they harbored at least one pathogenic mutation within overlapping genes (n = 26) between the 2 assays. From the 137 specimens (141 trials), a total of 1,094 variants (278 unique) of various frequencies were concordantly detected by the 2 assays: 1,007 (219 unique) SNVs and 87 (59 unique) indels (Fig 5A). In this study, benign variants (e.g., common population SNVs) were also included. In addition, agreement in VAF between the 2 assays was

Long-term test reproducibility
To evaluate long-term reproducibility, we examined variant calls for a multiplex mutation control (HD728) that was repeatedly tested, 119 times over 10 months, as a mutation positive control. The mutation control contained 7 alterations of �5% VAF (6 SNVs and 1 deletion, S7A Table). All alterations were successfully detected in all repeat tests, including 5 alterations occurring at 5% VAF. Detected VAF remained stable over the extended time period (Fig 6) with coefficient of variation less than 0.1x (range: 0.04-0.1x).

PLOS ONE
In 41.7% (856/2,053) of patients, at least 1 actionable mutation was identified (Fig 9): 27.5% (n = 565) of patients harbored mutations for which a targeted therapy is available either in the disease (n = 126) or another disease (n = 439); and 40.1% (n = 823) patients contained mutations for which an experimental drug is available. For this study, variants were considered clinically actionable when a targeted therapy or an experimental drug was available for the disease or other disease. In addition, 34.6% of patients (n = 711) had mutations with prognostic significance, and 36.4% (n = 748) of patients had mutations with diagnostic significance. In total, the assay identified clinically significant mutations in 51.7% (1,062/2,053) of patients. Mutations were considered clinically significant if they provided prognostic or diagnostic information for the disease or were clinically actionable.

Discussion
Molecular profiling can help diagnose, classify, and guide treatment of myeloid neoplasms [38,39]. In this study, we reported development and validation of a 48-gene NGS panel that includes genes that are technically challenging for molecular profiling of myeloid neoplasms, such as CEBPA, CALR, and FLT3. The assay demonstrated good inter-and intra-assay precision for SNVs, indels including CALR 52 bp deletion, and FLT3-ITDs (Fig 2 and S5A and S5B  Table). In addition, the assay detected 827 of 830 variants with VAF �5% reported by

PLOS ONE
Analytical validation of a 48-gene next-generation sequencing panel for myeloid neoplasms alternative analytical methods for a sensitivity of 99.6% (95% CI: 98.9-99.9%) (Fig 4 and S8  Table). The specificity of this assay appears to approach 100% as no false positive call was made in 7 GIAB accuracy studies across 0.4 million bp ( Table 3) and in a cross-platform study using 137 unique specimens across 4.8 million bp (Fig 5).
The 48-gene NGS assay offers the advantage of detecting variants in genes that are technically difficult to sequence, including CEBPA, CALR, and FLT3, with high sensitivity and specificity from a wide range of specimen types. In NGS, the limitation of a target enrichment method often leads to low coverage in genes with high GC content, and this may cause suboptimal assay accuracy. Supplementary single-gene assays are recommended to improve low coverage targets in some clinical assays [21,22,40]. In this study, we demonstrated a broad reportable range of the 48-gene panel even in extreme % GC spectrum (Fig 1B). For example, despite the very high GC content of CEBPA gene (up to 87% GC within a 100-bp window), hybridization enrichment coupled with careful bait design achieved an excellent result: median depth of 1,918X. During our development, direct comparison of capture technologies demonstrated superior performance of DNA bait over RNA bait for targets with high GC contents (S2 Fig). FLT3-ITD is inherently difficult to detect using NGS approaches [12,41]. In our validation studies, we also identified FLT3-ITDs of varying lengths (range: 18-117 bp) with 100% sensitivity and specificity ( Table 2). In some trials, additional ITD size(s) were detected, and we reasoned that our assay is more sensitive than the method compared, fragment analysis.
In clinical settings, specimen types submitted for an assay may vary. Thus, an ideal assay must be able to deal with a wide range of specimen types appropriate for the assay. Overall, our assay produced robust results (98.8% pass rate) for all specimen types (S2 Table), suggesting that DNA recovered from various specimen types can be successfully and accurately sequenced by this protocol. During our validation, we showed that VAF of SNVs and indels detected by our assay were highly concordant with ones observed by reference methods (Figs  3C and 5B), indicating our assay is accurate in variant quantification. Long-term test reproducibility of an analytical method is critical in clinical settings following initial validation; consistent variant allele frequencies over a 10-month period in a positive control specimen highlighted the long-term reproducibility of our assay (Fig 6).
However, this assay has several limitations. First, the assay does not include fusion detection. Accessing fusion by RNA analysis is an important component of comprehensive genomic approaches for improved clinically impactful data [42]. Fusion analysis is especially important for AML, as a wide variety of recurrent chromosome rearrangements define different disease subtypes for AML [1]. Conventional methods such as fluorescent in situ hybridization (FISH) or real-time PCR should be supplemented for fusion detection along with our NGS panel. Second, interlaboratory reproducibility studies have not been performed for our in-house developed protocol and bioinformatics pipeline. In-depth multi-center analytical validation would be required prior considering implementation of the assay in other clinical laboratories. Combining this assay with RNA sequencing as well as additional multi-center analytical validation studies would provide improved clinical utility of the assay.
In our study of 2,053 clinical patients, 55.6% had at least one pathogenic variant and 51.7% harbored clinically significant mutations with prognostic, diagnostic, or therapeutic relevance (Fig 9). The clinical utility of our assay is underscored by our ability to identify clinically significant variants in specific diseases. For example, from our cohort of 490 patients with indications of AML, TET2 had the highest mutation rate (17% of patients) followed by ASXL1, TP53, FLT3, NPM1, SRSF2, DNMT3A, IDH2, RUNX1 and NRAS (10~15%), frequencies similar to literature [37]. Among those frequently mutated genes, ASXL1, RUNX1, TP53, and FLT3-ITD mutations have been associated with poor prognosis, whereas NPM1 mutations in the absence of FLT3-ITD have been associated with favorable outcomes in AML [43]. In the 2016 revision of the WHO classification, AML with an NPM1 mutation is recognized as a subtype of AML; and AML with an RUNX1 mutation has been added as a provisional category of AML [1]. In addition, FDA-approved targeted therapies for AML are available for FLT3-ITD and IDH2, as well as IDH1 [44] which was mutated in 5% of patients in our study.
From our cohort of 1,018 patients with indications of MDS, TET2, SF3B1, and ASXL1 were the most frequently mutated genes (in >10% of patients), followed by SRSF2, TP53, DNMT3A, and RUNX1 (in >5% of patients), similar to Haferlach and colleague's study on the mutational profiles of 944 patients with MDS [36]. Among those genes, ASXL1 and TP53 mutations have been associated with poor prognosis [45], whereas SF3B1 mutation in MDS patients with ringed sideroblasts has been associated with favorable prognosis [46]. All of the frequently mutated genes have been associated with clonal hematopoiesis, and support the diagnosis of several different myeloid malignancies, including MDS, when found in combination with other diagnostic features [47,48]. While there are currently no therapies directly targeting mutated genes in MDS, hematologic malignancies harboring SF3B1 and SRSF2 mutations have been reported to be sensitive to splicing factor 3B subunit 1 (SF3b155) inhibitors, which are in clinical development [49,50].
Certain mutations in the MPN-associated genes satisfy subclassification of the disease. In our study of 545 patients with indications of MPN, JAK2 was the most frequently mutated gene (23% of patients) followed by TET2 (12%) and ASXL1 (10%), frequencies similar to literature [51,52]. CALR, CSF3R, and MPL were mutated in 4%, 3%, and 1% patients, respectively. Presence of a mutation in JAK2, CALR, and MPL is among the major criteria for the diagnosis of myelofibrosis (MF) or essential thrombocythemia (ET), while presence of JAK2 V617 or exon 12 mutations is among the major criteria for the diagnosis of polycythemia vera (PV) [1,53,54]. Activating CSF3R mutations have been found in the majority of chronic neutrophilic leukemias (CNLs) [55]. FDA-approved therapies for JAK2 mutations in PV and MF are available [56]. In addition, mutations in CALR exon 9, CSF3R, and MPL have been shown to be sensitive to Jak inhibitors [57][58][59].
For BCR-ABL1-negative MPN, common mutations of JAK2, CALR, and MPL genes are often examined as diagnostic targets using a cascade single-gene assay [54]. In our cohort of 545 patients with indications of MPN, 159 patients (29%) were positive for those 3 genes. Detection of an additional clinically significant mutation is more common in an NGS panel assay than single-gene tests owing to the multiplicity of genes tested. An additional 86 patients (16%) who were negative for those 3 genes were positive for pathogenic mutations in other genes in the panel. Of the 86 patients, 23 had mutations in genes that can aid diagnosis when found in combination with other diagnostic features (CBL, NRAS, and PTPN11) [1,8,48], 43 had mutations in genes with poor prognosis (ASXL1, EZH2, IDH2, NRAS, SETBP1, SRSF2, TP53, and U2AF1), and 29 had mutations in genes for which a therapy is available (BRAF, CSF3R, IDH2, KRAS, NF1, NRAS, PTPN11, and STAG2). In addition, of the 159 patients who were positive for those 3 genes, 84 patients had additional mutation(s) in 16 other genes. Among the 16 genes, 6 genes (IDH1, IDH2, KRAS, NF1, NRAS, and STAG2) are linked to available therapies in other diseases [38,60]. Combinations of Jak inhibitors with other targeted therapies may be relevant for those patients who harbor additional mutations.
In contrast to single-gene assays, NGS allows assessment of co-occurring mutations that might have heterogeneity of response to targeted therapy and survival. According to the recent National Comprehensive Cancer Network guidelines, AML patients with FLT3-ITD and NPM1 double mutation (AML FLT3-ITD+/NPM1+) are categorized as favorable and intermediate risk levels, depending on the allelic ratio of FLT3-ITD, for whom allogenic stem cell transplantation (allo-HSCT) is not obligated. Loghavi and colleagues reported that AML FLT3-ITD+/NPM1+ patients with a DNMT3A mutation had shorter event-free survival compared to those in other mutation groups [61]. Similarly, recent studies have suggested that AML patients with concomitant DNMT3A R882+/FLT3-ITD+/NPM1+ mutations had a very poor prognosis, and allo-HSCT could moderately improve their survival [62, 63]. Our study cohort included 21 AML FLT3-ITD+/NPM1+ patients, 4 of whom had a DNMT3A R882 mutation. In addition, Ardestani and colleagues reported that DNMT3A R882 mutations alone do not affect the clinical outcomes of AML patients, but when accompanied by FLT3-ITD mutations, overall survival was reduced, even after allo-HSCT [64]. In our study, 5 of 26 DNMT3A R882-positive AML patients had FLT3-ITD. These results support the clinical utility of our assay for detecting mutations that can alter prognostic and therapeutic significance when they occur in combination.

Conclusions
We have developed and validated a 48-gene NGS assay that can detect SNVs, indels, and FLT3-ITD with high sensitivity and specificity, even in technically challenging genes. The assay detects variants with clinical significance from a substantial proportion of patients tested. The developed assay may be used to guide more precise and targeted therapeutic strategies, possibly leading better treatment outcomes for patients with myeloid neoplasms.
Supporting information S1