Abstract
Next-generation sequencing (NGS)-based mutation panels profile multiple genes simultaneously, allowing the reporting of numerous genes while saving labor and resources. However, one drawback of using NGS is that the turnaround time is often longer than conventional single gene tests. This delay can be problematic if molecular results are required to guide therapy in patients with clinically aggressive diseases, such as acute myeloid leukemia. To overcome this limitation, we developed a novel custom platform designated as Ultra-rapid Reporting of GENomic Targets (URGENTseq), an integrated solution that includes workflow optimization and an innovative custom bioinformatics pipeline to provide targeted NGS results on fresh peripheral blood and bone marrow samples within an actionable time period. URGENTseq was validated for clinical use by determining mutant allelic frequency and minimum coverage in silico to achieve 100% concordance for all positive and negative calls between the URGENTseq and conventional sequencing approach. URGENTseq enables the reporting of selected genes useful for immediate diagnosis (CALR, CSF3R, JAK2, KRAS, MPL, NPM1, NRAS, SF3B1) and treatment decisions (IDH1, IDH2) in hematologic malignancies within 48 hours of specimen collection. In addition, we summarize the molecular findings of the first 272 clinical test results performed using the URGENTseq platform.
Molecular biomarkers have become an integral part of patient care across the entire spectrum of hematologic malignancies. Acute myeloid leukemia (AML), myelodysplastic syndromes (MDSs), and myeloproliferative neoplasms (MPNs) represent neoplasms in which analysis of multiple biomarkers is essential for diagnosis, classification, risk stratification, and treatment decisions.1 The 2017 World Health Organization classification of myeloid neoplasms incorporates molecular biomarkers in the diagnostic criteria for MDS (SF3B1), MPN (CALR, CSF3R, JAK2, MPL), MDS/MPN (KRAS, NRAS), and AML (NPM1, CEBPA, RUNX1).2, 3 Additional biomarkers such as ASXL1, DNMT3A, EZH2, FLT3, IDH1, IDH2, and TP53 provide important prognostic and predictive information.4, 5 With the development and Food and Drug Administration approval of targeted inhibitors for FLT3, IDH1, and IDH2, predictive biomarkers are also currently required to select patients for targeted therapy.6, 7, 8, 9 Compared with prognostic biomarkers that are required for long-term patient care decisions, both diagnostic and predictive biomarkers are required in a more time-sensitive manner to guide clinical decision making.
Massively parallel next-generation sequencing (NGS) interrogates the mutational status of multiple genes simultaneously, using small amounts of DNA with a much higher throughput and lower cost per base compared with previous methods.10 Turnaround time (TAT) of NGS-based assays usually ranges from 3.5 to 7 days, but it can reach up to several weeks, depending on the type of test and instrument used.11, 12, 13 In clinical settings, diagnostic biomarkers are needed more urgently by pathologists to confirm or rule out a morphologic impression, and predictive biomarkers are required by oncologists to initiate or withhold targeted therapy or to enroll patients in specific clinical trials or treatment pathways. Therefore, a rapid TAT becomes critical for diagnostic and predictive markers compared with prognostic markers. Although NGS provides a much broader coverage for genotyping compared with the conventional single gene assays, the ability to provide faster results for selected genes for immediate clinical decision making is compromised, posing challenges in the selection of patients for treatment with targeted inhibitors.
Herein, we describe the process and challenges of developing and implementing an integrated clinical testing platform, Ultra-rapid Reporting of GENomic Targets (URGENTseq), which facilitates the reporting of targeted NGS results within 48 hours from sample collection, thereby enabling more rapid diagnosis and treatment selection.
Materials and Methods
Our College of American Pathologists–accredited and Clinical Laboratory Improvement Amendments–certified clinical molecular diagnostics laboratory supports active hematology services with large patient volumes. There is a constant need for rapid TAT to facilitate the use of targeted inhibitors or to place patients on appropriate clinical trials or treatment plans based on the molecular biomarker data. To improve the TAT of NGS results to support the clinical services, the following actions were taken: i) a customized NGS bioinformatics solution was developed for real-time data analysis of an ongoing sequencing run to generate URGENTseq results for specific mutation hotspots on a 28- or an 81-gene panel at least 24 hours before the results are available in the standard pipeline; ii) the URGENTseq and conventional NGS results were compared to calculate sensitivity, positive predictive value, concordance, and lower limit of detection (LOD); and iii) URGENTseq was validated to report in 48 hours NGS results that include hotspots of 10 selected genes to facilitate specific diagnosis and/or to initiate appropriate treatment promptly.
Conventional NGS Workflow
Mutation analysis of genes with diagnostic, therapeutic, and/or prognostic importance in hematologic malignancies is routinely performed by using an initial 28-gene (ABL1, ASXL1, BRAF, DNMT3A, EGFR, EZH2, FLT3, GATA1, GATA2, HRAS, IDH1, IDH2, IKZF2, JAK2, KIT, KMT2A, KRAS, MDM2, MPL, MYD88, NOTCH1, NPM1, NRAS, PTPN11, RUNX1, TET2, TP53, WT1) and an updated 81-gene (ANKRD26, ASXL1, ASXL2, BCOR, BCORL1, BRAF, BRINP3, CALR, CBL, CBLB, CBLC, CEBPA, CREBBP, CRLF2, CSF3R, CUX1, DDX41, DNMT3A, EED, ELANE, ETNK1, ETV6, EZH2, FBXW7, FLT3, GATA1, GATA2, GFI1, GNAS, HNRNPK, HRAS, IDH1, IDH2, IKZF1, IL2RG, IL7R, JAK1, JAK2, JAK3, KDM6A, KIT, KMT2A, KRAS, MAP2K1, MPL, NF1, NOTCH1, NPM1, NRAS, PAX5, PHF6, PIGA, PML, PRPF40B, PTEN, PTPN11, RAD21, RARA, RUNX1, SETBP1, SF1, SF3A1, SF3B1, SH2B3, SMC1A, SMC3, SRSF2, STAG1, STAG2, STAT3, STAT5A, STAT5B, SUZ12, TERC, TERT, TET2, TP53, U2AF1, U2AF2, WT1, ZRSR2) NGS-based panel. Sequencing libraries are prepared by using HaloPlex (Agilent Technologies, Santa Clara, CA) chemistry. The panels are run on a MiSeq V3 300 cycle kit (Illumina, San Diego, CA) with 151 paired-end sequence plus eight additional cycles that include the index reads, for a total of 310 cycles. Raw sequencing data are processed initially through MiSeq reporter (Illumina) for base calling. The resulting FASTQ files are then aligned to the GRCh37/hg19 reference genome by using SureCall 3.0 software (Agilent Technologies), and binary alignment map (BAM) and variant call format (VCF) files are analyzed for coverage and variant calling by using Oncoseek (version 1.3.1.277), an in-house, custom-developed annotation and reporting system.14 When the first 28-gene panel test was initially implemented, the conventional workflow had a TAT of at least 5 business days.
Development of Bioinformatics Workflow and Computing Infrastructure
A comprehensive review of existing end-to-end workflow from sample collection to final results was performed to identify opportunities for TAT improvement. Rate-limiting steps in both physical and informatics workflows were identified and optimized, leading to the development of a novel URGENTseq platform. Examples of physical workflow optimization included modifications in DNA extraction schedule, optical density measurement schedule, library preparation, loading, and triggering of data analysis steps. Specific protocols for the identification and handling of specimens that required a faster TAT were integrated into existing pathologists triage and wet bench workflows. By extending the working shifts of laboratory technologists by only 1 hour, the library could be loaded in a MiSeq sequencer (Illumina) the day after sample collection. This modification allowed optimizing the workflow and reducing the TAT to 3 business days (Supplemental Figure S1). To further improve the TAT of this assay, a customized pipeline was developed that allowed the interpretation of NGS data in real time while the sequencing run was still ongoing. The analysis was performed by using a workflow built in IBM Platform Project Manager and published on an IBM platform application center interface version 9.1.3 (Armonk, NY) supported by one of the institution's integrated high performance computing (HPC) clusters (IBM Corporation, Somers, NY). This HPC cluster allows for the use of 14 nodes with 24 CPU cores and 384 GB of RAM per node. Allocation of a dedicated spot in the HPC cluster was crucial to perform bioinformatics analyses required in the pipeline.
Samples and Sequencing Data for Proof of Concept
As proof of concept, a retrospective comparison of 50 consecutive conventional NGS runs was performed with 303 samples completed in the laboratory. The URGENTseq pipeline was deployed through the command line batch run mode on the HPC cluster to copy raw data from the first 159 cycles of all 50 consecutive runs, followed by demultiplexing, sequence alignment, and variant calling. The variants called by URGENTseq were computationally compared with the conventional NGS production results, as a VCF to VCF comparison. Seven genes in the 28-gene panel, harboring diagnostic or therapeutic implications: JAK2 (codon 617) and MPL (codon 515) for the diagnosis of MPNs, NPM1 (exon 11) for the diagnosis of AML with NPM1 mutation, KRAS and NRAS (codons 12, 13, 61, and 146 for both genes) for the diagnosis of MDS/MPNs, and IDH1 (codon 132) and IDH2 (codons 140, 172) for the use of targeted IDH inhibitors predominantly in AML patients were initially studied. Overall, concordance for the seven target genes and individual concordance for each of these genes between conventional NGS and URGENTseq pipelines were determined by comparing the calls and variant allelic frequency (VAF). Subsequently, CALR, CSF3R, and SF3B1 genes were validated on an updated 81-gene panel with adequate concordance to bring the list to 10 genes (data not shown). Concordance analysis was performed to determine sensitivity, positive predictive value, and the LOD between conventional NGS as the gold standard and URGENTseq. Appropriate confidence intervals were also calculated. Statistical significance was set at P < 0.05.
Prospective Parallel Review and Validation
After determining the pipeline variables, a parallel review of all 14 runs (109 samples) was performed by using the 28-gene panel within 1 week. The URGENTseq pipeline was deployed for sample demultiplexing and variant calling while the sequencing run was ongoing. Downstream analysis of the URGENTseq results included manual review by using the Integrative Genomics Viewer15, 16 by one of the authors (R.R.-C.) who was blinded to any previous clinical and mutational history of the patients. Emphasis was made to focus only on the hotspots of the seven target genes relevant for diagnosis and choice of therapy. The results obtained through the URGENTseq workflow were manually compared with the results obtained by using conventional NGS data analysis with the use of a 5% VAF cutoff for reporting.
Clinical Implementation and TAT Comparison
After determination of the LOD as well as analytical sensitivity, guidelines and a standard operating procedure were developed to implement URGENTseq workflow in routine clinical practice. The t-test was used to compare the mean TAT between rush cases tested on URGENTseq and conventional NGS platforms. Statistical significance was set at P < 0.05.
Results
Proof of Concept and Successful Pipeline Development
NGS technologies that used sequencing by synthesis were based on replicating DNA templates through a number of sequencing cycles in a flow-cell incorporated into the NGS instrument. The base incorporated into DNA strands was detected at the end of each sequencing cycle.17 Typically, DNA templates were 100 to 200 bases long and were sequenced through paired-end bidirectional sequencing in both forward and reverse directions. Because only one nucleotide was incorporated in each cycle according to the sequence of the template, the read length determined the total number of cycles and the sequencing time. In the conventional paired-end sequencing scenario, DNA strands needed to be sequenced in both forward and reverse directions, a process that took 36 to 48 hours, depending on the read length, before downstream data analysis could begin. This constituted the most time-consuming step of conventional NGS platforms. An innovative informatics pipeline (URGENTseq) was designed for real-time NGS data analysis with ongoing sequencing run (see below). Supplemental Figure S1 illustrates all the steps included in the workflow and highlights the differences between the conventional/optimized pipeline and URGENTseq. In summary, the URGENTseq pipeline intercepted the base call files that corresponded only to the first 159 cycles located in the general parallel file system folder, without disrupting the run, and converted them to FASTQ → BAM → VCF files. The 159-cycle benchmark was selected because cycles 1 to 151 spanned the entire first read sequence, that is, the forward read in F1R2 and the reverse read in F2R1 orientation. Cycles 152 to 159 included the eight-base sample index sequence were necessary for demultiplexing. Finally, FASTQ files were aligned to the hg19 reference genome on the HPC cluster through the IBM platform application center interface by using the Burrows-Wheeler transformation. Resulting BAM files were then indexed by using SAMtools18 and were run through Fastgc to obtain quality control metrics. BEDtools and GATK were used for the coverage and drop-out region analysis, and the variant caller LoFreq was used to provide VCF files for interpretation.
Retrospective Comparison of 50 Runs Shows High Concordance between URGENTseq and Conventional NGS
URGENTseq was validated against the existing pipeline through an extensive retrospective correlation of variant calls for 50 consecutive retrospective runs that encompassed a total of 303 patient samples. By comparing VCF to VCF the number of calls with VAF ≥5% obtained by URGENTseq versus the calls made by conventional NGS pipeline were determined. A total of 131 variants with VAF ≥5% were made by URGENTseq and conventional NGS (Table 1), resulting in a 100% concordance. As expected the concordance dropped slightly but still remained high at lower VAF cutoffs, 98% at 2% VAF and 96.3% at 1% VAF cutoff (data not shown). Figure 1 demonstrates the VAF correlations between URGENTseq (single read) and conventional NGS (paired-end) for all the calls in the initial seven genes. Of note, the NPM1 VAF correlation was the lowest (r2 = 0.97) because of variations in the alignment of single reads compared with paired reads (Figure 1). Concordance analysis at the sample and variant levels showed a sensitivity and positive predictive value of 100% (Table 1). Table 2 demonstrates the number of concordant (all) and discordant cases (none) at 5% VAF cutoff for the initial seven target genes.
Table 1.
Concordance Analysis Showing Number of Samples and Total Variants in Validation Studies That Compared Conventional NGS with URGENTseq Results
| URGENTseq | Conventional NGS |
|||||
|---|---|---|---|---|---|---|
| Retrospective (50 analysis runs, n = 303) |
Prospective (14 analysis runs, n = 109) |
|||||
| Positive | Negative | Total | Positive | Negative | Total | |
| Sample/variant | ||||||
| Positive | 107/131 | 0/0 | 107/131 | 39/55 | 0/0 | 39/55 |
| Negative | 0/0 | 196/0 | 196/0 | 0/0 | 70/0 | 70/0 |
| Total | 107/131 | 0/0 | 303/131 | 39/55 | 0/0 | 109/55 |
NGS, next-generation sequencing; URGENTseq, Ultra-rapid Reporting of GENomic Targets.
Figure 1.
Variant allelic frequency (VAF) correlations between Ultra-rapid Reporting of GENomic Targets (URGENTseq) and conventional next-generation sequencing (NGS). A: The VAF correlation between conventional NGS (paired-end) and URGENTseq (single read) among seven target genes. B–G and I: VAF correlation for individual genes. H: An Integrative Genomics Viewer (IGV) screen shot of a NPM1 positive case showing different alignments of common insertion reflected in a lower r2 value (0.9777). P values for all seven genes achieved statistical significance (P < 0.001) at VAF of ≥5%. n = 13 IDH1; n = 24 IDH2; n = 40 JAK2; n = 13 KRAS; n = 4 MPL; n = 15 NPM1; n = 22 NRAS.
Table 2.
Comparison of Variant Calls Detected by Conventional NGS and URGENTseq in Validation Studies (VAF ≥5%)
| Gene | Retrospective (50 analysis runs) |
Prospective (14 analysis runs) |
||
|---|---|---|---|---|
| Concordant | Discordant | Concordant | Discordant | |
| IDH1 | 13 | 0 | 16 | 0 |
| IDH2 | 24 | 0 | 10 | 0 |
| JAK2 | 40 | 0 | 8 | 0 |
| KRAS | 13 | 0 | 5 | 0 |
| MPL | 4 | 0 | 7 | 0 |
| NPM1 | 15 | 0 | 5 | 0 |
| NRAS | 22 | 0 | 4 | 0 |
NGS, next-generation sequencing; URGENTseq, Ultra-rapid Reporting of GENomic Targets; VAF, variant allelic frequency.
The average coverage per amplicon for all genes was 7306 reads (range, 19 to 114,340 reads), whereas the average coverage per candidate amplicon was 5390 (range, 1779 to 12,810). All the hotspots were adequately covered at >250× depth in a maximum batch size of 10 samples per run (Table 3). The average number of mutant reads was 1029 (range, 54 to 4874 mutant reads) with an average depth of coverage of 3764 (range, 429 to 14,824 depth of coverage).
Table 3.
Details of All Variants Detected in Seven Targeted Genes by URGENTseq in 50 Retrospective NGS Runs
| Variant | n | VAF, mean % (range) | Variant coverage, mean reads (range) | Coverage depth, mean reads (range) |
|---|---|---|---|---|
| NM_005896.2(IDH1):c.394C>G p.R132G | 3 | 19 (5–27) | 1817 (427–3983) | 7797 (3523–14,824) |
| NM_005896.2(IDH1):c.394C>T p.R132C | 5 | 18 (6–28) | 476 (93–1167) | 4078 (1668–6045) |
| NM_005896.2(IDH1):c.395G>A p.R132H | 5 | 26 (13–46) | 1297 (1174–3528) | 4523 (3404–5784) |
| NM_002168.2(IDH2):c.418C>T p.R140W | 1 | 42 (42–42) | 2909 (2909–2909) | 7005 (7005–7005) |
| NM_002168.2(IDH2):c.419G>A p.R140Q | 17 | 38 (13–49) | 2916 (701–4874) | 7810 (5258–11,159) |
| NM_002168.2(IDH2):c.515G>A p.R172K | 6 | 24 (5–46) | 1816 (442–2968) | 8250 (6352–10,170) |
| NM_004972.3(JAK2):c.1849G>T p.V617F | 40 | 43 (5–95) | 1315 (104–3905) | 3180 (1207–5252) |
| NM_004985.3(KRAS):c.34G>A p.G12S | 1 | 19 (19–19) | 163 (163–163) | 878 (878–878) |
| NM_004985.3(KRAS):c.34G>C p.G12R | 1 | 7 (7–7) | 173 (79–79) | 1884 (2471–2471) |
| NM_004985.3(KRAS):c.35G>A p.G12D | 4 | 29 (5–44) | 595 (64–955) | 1989 (1610–2498) |
| NM_004985.3(KRAS):c.35G>C p.G12A | 3 | 12 (6–16) | 213 (98–310) | 1770 (1465–2235) |
| NM_004985.3(KRAS):c.35G>T p.G12V | 1 | 28 (28–28) | 281 (281–281) | 998 (998–998) |
| NM_004985.3(KRAS):c.38_40dupGCG p.G13dup | 1 | 34 (34–34) | 1178 (1178–1178) | 3494 (3494–3494) |
| NM_004985.3(KRAS):c.38G>A p.G13D | 2 | 23 (15–29) | 360 (291–751) | 1989 (1976–2565) |
| NM_005373.2(MPL):c.1544G>T p.W515L | 4 | 49 (8–87) | 1642 (273–3359) | 3543 (2133–4471) |
| NM_002520.6(NPM1):c.860_863dupTCTG p.W288fs | 14 | 25 (7–62) | 531 (160–1405) | 2245 (429–7534) |
| NM_002520.6(NPM1):c.863_864insCATG p.W288fs | 1 | 5 (5–5) | 134 (134–134) | 2881 (2881–2881) |
| NM_002524.4(NRAS):c.182A>C p.Q61P | 1 | 10 (10–10) | 436 (436–436) | 4263 (4263–4263) |
| NM_002524.4(NRAS):c.183A>T p.Q61H | 2 | 12 (5–28) | 541 (73–1361) | 3942 (2817–4940) |
| NM_002524.4(NRAS):c.34G>A p.G12S | 1 | 17 (17–17) | 242 (391–391) | 2702 (3136–3136) |
| NM_002524.4(NRAS):c.34G>C p.G12R | 1 | 51 (51–51) | 1447 (1447–1447) | 2847 (2847–2847) |
| NM_002524.4(NRAS):c.35G>A p.G12D | 8 | 27 (7–49) | 582 (54–1205) | 2730 (1604–4126) |
| NM_002524.4(NRAS):c.35G>C p.G12A | 2 | 20 (5–34) | 435 (151–719) | 2437 (2103–2771) |
| NM_002524.4(NRAS):c.37G>C p.G13R | 1 | 35 (35–35) | 1365 (1365–1365) | 3875 (3875–3875) |
| NM_002524.4(NRAS):c.37G>T p.G13C | 1 | 21 (21–21) | 669 (669–669) | 3128 (3128–3128) |
| NM_002524.4(NRAS):c.38G>A p.G13D | 3 | 23 (10–45) | 550 (86–1532) | 2699 (2268–3400) |
| NM_002524.4(NRAS):c.38G>T p.G13V | 2 | 10 (6–16) | 147 (66–251) | 2013 (1538–2268) |
Source of accession numbers: National Center for Biotechnology Information reference sequence collection (https://www.ncbi.nlm.nih.gov/refseq).
NGS, next-generation sequencing; URGENTseq, Ultra-rapid Reporting of GENomic Targets; VAF, variant allelic frequency.
Parallel Testing for Real-Time Assessment Shows Adequate Performance of the URGENTseq Pipeline Useful for Validation and Development of Clinical Guidelines
From the 100% concordance obtained with a 5% VAF in the 50-run comparison, the LOD was established at 5% VAF with a minimum coverage of 250× per variant. By using these criteria a parallel review was performed on a total of 109 patient samples sequenced on 14 consecutive analysis runs within 1 week. This ensured that the real-time data extraction did not affect the ongoing sequencing run. One individual (R.R.-C.), blinded to clinical and previous mutational data, analyzed the calls by using OncoSeek and Integrative Genomics Viewer for confirmation, populated a database with URGENTseq results, and compared them with the final results issued in the patient's report by a molecular pathologist. The overall accuracy of URGENTseq compared with the conventional NGS pipeline was 100% (Tables 1 and 2). The average number of mutant reads in positive cases was 868 (range, 123 to 2916 mutant reads) with an average depth of coverage of 3544 (range, 878 to 8250 depth of coverage). During this week the MiSeq instrument took approximately 12 hours to generate the 159 base call files , in comparison with the >24 hours used for one 310-regular cycle run in addition to approximately 40 hours of analysis. From the high concordance for 412 unique samples between URGENTseq and conventional NGS, URGENTseq was validated to report real-time NGS results on seven target genes and subsequently on three additional genes in 2 days. As part of the guidelines, calls detected in genes other than the targeted genes were reported only on the final report. Similarly, calls with VAF of <5% in any of the selected genes could be deferred for final review when the conventional NGS data were available. Thus, URGENTseq results could be reported as positive or negative for mutation, or deferred for conventional analysis (see below). It is important to mention that during the evaluation of conventional NGS results, a correlation with URGENTseq NGS reports was always performed to avoid and to investigate any discrepancies. This was performed by comparing the diagnostic reports and, if needed, by additional review of the vcf files manually with the use of an in-house–developed reporting software and the review of BAM files in Integrative Genomics Viewer. Any discrepancies between URGENTseq report and final NGS results were required to be investigated and resolved before issuing a conventional NGS report, although no such discrepancies have been noted so far by using 5% VAF cutoff for reporting. In addition, the pathologist had an option to defer the URGENTseq results for variants with <5% VAF for subsequent conventional NGS reports, which was clearly documented in the URGENTseq report. The pathologist completed a QA Review step in the laboratory information system for all the reports documenting the review and concordance before issuing the conventional NGS report. The cumulative data from the QA review step were reviewed for quality assurance. This carefully designed and monitored workflow provided quality assurance and patient safety measures for clinical reporting.
Successful URGENTseq Workflow Implementation and Current Clinical Experience Demonstrate Decreased TAT
On validation of the URGENTseq pipeline, an end-to-end clinical workflow was developed for cases that needed expedited reporting. Specific eligibility criteria and rush request notification algorithms were developed (Figure 2). Currently, all AML patients with >20% blasts were automatically flagged for URGENTseq reporting of IDH1 and IDH2 mutations through an in-house–designed software called Processing ALgorithms-PAL version 114 or by the molecular pathologists triaging the test orders. In addition, any oncologist or pathologist could request URGENTseq results for diagnostic markers (CALR, CSF3R, JAK2, KRAS, MPL, NPM1, NRAS, SF3B1) via e-mail. URGENTseq reports for 10 genes were issued at least 24 hours ahead of the conventional NGS reports for all 81 genes on the panel.
Figure 2.
Clinical workflow. The workflow diagram shows how a request was usually initiated, as well as the important steps that needed to be taken to successfully identify and report next-generation sequencing (NGS) cases by using the Ultra-rapid Reporting of GENomic Targets (URGENTseq) pipeline with a faster turnaround time (TAT). Timely identification and laboratory notification based on available clinical history and/or pathology findings was critical in ensuring inclusion of the sample in the earliest possible analysis runs as well as initiation of parallel URGENTseq pipeline. Under optimal conditions the results were available within 48 hours; however, a delay in any of the steps listed may affect the TAT.
Since go-live, 272 cases have been tested on the URGENTseq pipeline, including 46 on the 28-gene panel and 226 on the 81-gene panel (Figure 3A). URGENTseq real-time NGS reports have been issued successfully for all cases with 100% concordance with the subsequent conventional NGS results (Figure 3C). The cases assessed included acute leukemia (N = 141; 52%), MDS (n = 59; 52%), MPN (n = 38; 14%), MDS/MPN (n = 1; 0.3%), and other diseases (n = 33; 12%). An actionable mutation was noted in 118 cases (43%). The observed mutation frequency was as follows: NRAS (n = 31; 11%), NPM1 (n = 24; 9%), JAK2 (n = 23; 8%), IDH2 (n = 21; 8%), IDH1 (n = 19; 7%), SF3B1 (n = 15; 6%), KRAS (n = 8; 3%), CALR (n = 4; 1%), CSF3R (n = 1), and MPL (n = 1). There have been 14 deferred cases, predominantly for low level mutations (<5%) in the NRAS gene (Figure 3A). The median 2.1-business day TAT of URGENTseq reports (range, 1.65 to 5.85 days) was significantly different from the 3.8 and 4.2 business day median TAT (ranges, 2.7 to 9.7 and 3.1 to 11.1 days, respectively) of the matched and unmatched conventional NGS reports that used the optimized workflow, respectively; t-test P value < 0.0001 (Figure 3B). Significantly, the faster TAT in AML patients (n = 31) allowed more rapid treatment decisions for administration of IDH1 and IDH2 inhibitors (median, 2.1 days; range: 1.7 to 3.1 days). A total of 18 patients with IDH1 or IDH2 could be treated with targeted inhibitors because of timely availability of mutation results. Importantly, treatment decisions for all AML cases with wild-type IDH1/IDH2 could also be made in similar expedited time frame because of early availability of IDH1/IDH2 mutation status.
Figure 3.
Current clinical experience showing turnaround time comparison between the Ultra-rapid Reporting of GENomic Targets (URGENTseq) and conventional workflows. A: The results of 272 clinical URGENTseq cases with tumor type, mutation pattern, and turnaround time (TAT). B: The boxplot highlights the significant decrease in TAT between the URGENTseq and conventional next-generation sequencing (NGS) pipelines. C: Examples of URGENTseq reports showing selected genes and the subsequent final reports showing all genes. The genes with mutation/s are highlighted by red circles for quick overview of wild-type (noncircled) and mutated (circled) genes. Specific details of each mutation by using standardized Human Genome Variation Society nomenclature and National Center for Biotechnology Information (https://www.ncbi.nlm.nih.gov/refseq) reference sequence (eg, NM_005896.2) are included in the subsequent section within the report. ∗∗∗∗P < 0.0001 versus final report (t-test). MDS, myelodysplastic syndrome; MPN, myeloproliferative neoplasm.
Discussion
The scope for clinical molecular testing in hematologic malignancies has rapidly evolved from providing correlative and prognostic information in selected conditions to delivering critical actionable information for a variety of neoplasms. NGS platforms allow comprehensive mutation profiling and have become a common practice in clinical molecular diagnostics laboratories. Clinical NGS implementation is generally achieved by a multidisciplinary effort that requires considerable amounts of human, computational, bioinformatics, and infrastructural resources.5, 6, 7, 8, 9 Because of recent development of targeted therapies, the results are needed more quickly than supported by current NGS panel approaches. The current delay in providing NGS results creates a challenge for the clinical laboratory and the oncology team in balancing the extent of desired genomic information with the TAT, especially for patients with clinically aggressive diseases such as acute leukemia. The need for faster TAT and timely availability of mutation data for selected genes has become even more critical with the availability of targeted inhibitors for IDH1 and IDH2. To address this issue, a novel real-time NGS platform, URGENTseq, was developed to provide genotypic data for selected high-yield gene targets quickly, followed by results from the conventional method later.
For the initial iteration of the URGENTseq workflow, diagnostic and predictive biomarkers with hotspot mutations were studied. Prognostic biomarker genes are equally important for overall clinical care, however, their reporting is not as time sensitive and the results need not be included in the URGENTseq report. This focused approach also helps to achieve an optimal balance for an additional clinical service requirement and the immediate clinical utility of the report generated. Higher coverage depth across known hotspots was intentionally added in the panel at the design stage by adding additional capture probes to achieve higher analytical sensitivity.
Many challenges were faced while successfully implementing URGENTseq. A key factor was obtaining access to the already available institutional computational resources to be able to run URGENTseq in parallel, without disrupting the ongoing run. One of the limitations of URGENTseq is the need for a HPC cluster, the cost of which can be prohibitive if not readily available and for which maintenance and separate technical support are crucial for the pipeline to work properly. However, by using an in silico proof-of-concept approach, the usual costs of a regular validation of any NGS pipeline were markedly decreased. One additional benefit of URGENTseq that was not foreseen before its implementation is its utility when a run fails. Although a re-run rate of similar NGS platforms has been reported at <5% with a fail rate of <1%,11 real-time NGS results can still be provided in an actionable timeframe, should this calamity occur.
The retrospective and parallel comparisons as well as the ongoing cases have reached 100% concordance when a 5% VAF cutoff and 250× coverage are used. Although the pipeline showed similarly high concordance for all genes of both 28- and 81-gene panels (data not shown), the overall concordance and clinical utility are optimized when targeting only the relevant hotspots in the IDH1, IDH2, MPL, KRAS, NRAS, NPM1, JAK2, CALR, CSFR3, and SF3B1 genes. Assessing the mutational status of these genes can facilitate prompt diagnosis and adequate targeted treatment selection with or without inclusion in one of the ongoing clinical trials at our institution. From the in silico analysis, URGENTseq performed better with single nucleotide variants than with insertions and deletions, as seen in cases with NPM1 mutations.
Commonly used NGS methods include single-read or bidirectional paired-end sequencing. Platforms that use the former approach have a tendency to perform faster because DNA is sequenced in one direction from only one end. However, these platforms are more prone to errors compared with bidirectional paired-end sequencing; the latter generates higher quality data by sequencing each molecule from both 5′ and 3′ ends.3 In addition, bidirectional reads offer the added advantage of improved detection of longer insertion, deletions, duplications, and translocations. Nevertheless, higher sequencing quality comes at the expense of longer sequencing time that usually accounts for approximately a 24-hour difference compared with unidirectional sequencing.4 URGENTseq introduces the concept of real-time data analysis and real-time NGS for commonly available NGS platforms, whereby reporting has traditionally been limited to sequencing end-point only. Such real-time NGS methods are likely to increase the clinical utility of NGS results by overcoming the TAT barrier associated with the conventional NGS approaches.
Emphasis has been made to load the sequencing libraries on MiSeq instrument at the same time every afternoon (approximately 5:00 PM) to guarantee that the 159-cycle benchmark is reached at approximately 9:00 AM the next morning. This timing allows comfortable interpretation of the results and issuance of the URGENTseq real-time NGS report by noon. Alerts currently need to be constantly monitored to be able to successfully triage rush cases to URGENTseq. This process will be automated to decrease potential oversight of any eligible cases. By implementing the URGENTseq pipeline, patients may receive a result on 10 target genes as early as 48 hours within the sample collection. It is apparent that minimizing any downtime between multiple steps, starting from the point of sample collection to the final reporting, potentially benefit the TAT. For example, the time from sample collection to sample receipt in the laboratory is arguably shorter for in-house samples. Similarly, the ability to accumulate adequate samples to initiate an analysis run varies between high-volume and low-volume laboratories. Even with the specified limitations of applying URGENTseq model to other clinical settings, the overall approach involving optimization of physical workflow and informatics pipelines is applicable universally and allows a maximum achievable reduction in TAT.
The ability to provide diagnostic and predictive biomarker data with a faster TAT allows faster treatment decisions and diagnostic determinations. Specifically, a fixed and predictable time interval from sample acquisition to targetable biomarker reporting allows better planning and coordination of care from clinical effectiveness point of view. The availability of mutation data allows timely diagnosis and treatment decisions, including alternative options for patients lacking a targetable mutation. Indicators of clinical impact such as time to treatment, reduction in hospital stay, and overall outcome in patients reported using URGENTseq will be compared at a future date. Timely availability of diagnostic biomarker information also facilitates the hematopathology diagnosis and reporting. URGENTseq reporting is being actively requested and used by pathologists for identifying clonal abnormalities in challenging cases of MDS, MPN, and MDS/MPN. The key advantage of URGENTseq is that it does not require separate sample processing and additional sequencing cost. All samples are prepared by using conventional NGS processing only. When a request to rush specific results is received, the bioinformatics pipeline component of the URGENTseq workflow is deployed on a high-performance computing cluster to extract the information from an ongoing sequencing run without disrupting the run and at no additional cost. In addition, it provides an independent secondary bioinformatics pipeline that compliments the primary analysis.
Conclusions
The URGENTseq real-time NGS platform allows faster diagnosis and treatment decisions for improved patient care and satisfaction. This approach does not require additional reagent setup or benchwork cost and acts as a useful quality check for the conventional NGS pipeline.
Acknowledgment
We thank the high performance computing (HPC) team at the MD Anderson Cancer Center for their support.
Footnotes
See related Commentary on page 13
K.P.P. and R.R.-C. contributed equally to this work.
Disclosures: K.P.P. received one-time consulting fee to participate in the Celgene Corporation's Acute Myeloid Leukemia advisory board meeting; C.D.D. is on advisory boards of Celgene, Agios, and AbbVie; the authors have a patent application on the URGENTseq workflow and informatics components in progress.
Supplemental material for this article can be found at https://doi.org/10.1016/j.jmoldx.2018.08.002.
Supplemental Data
Supplemental Figure S1.
Explanation of Ultra-rapid Reporting of GENomic Targets (URGENTseq) pipeline. A: The different steps performed in the molecular laboratory from the time of sample collection to when the clinical report is issued with corresponding estimated time (in hours). The timeline for each of the workflows (baseline, conventional optimized, URGENTseq) showed significant reduction in turnaround time (TAT) with the URGENTseq reporting because multiple steps were accomplished within the short timeframe. Day 0 referred to the day when the sample was collected. Each following day has a specific color. B: An amplicon of a positive case for JAK2 V617F hotspot mutation highlighted by the red rectangle. Comparison of the data analysis strategies for the conventional (left) and URGENTseq (right) pipelines showed significant time savings with URGENTseq pipeline. BAM, binary alignment map; BCL, base call file; OD, optical density; PDF, Portable Document Format; TIFF, Tagged Image File Format; VCF, variant call format.
References
- 1.Luthra R., Patel K.P., Reddy N.G., Haghshenas V., Routbort M.J., Harmon M.A., Barkoh B.A., Kanagal-Shamanna R., Ravandi F., Cortes J.E., Kantarjian H.M., Medeiros L.J., Singh R.R. Next-generation sequencing-based multigene mutational screening for acute myeloid leukemia using MiSeq: applicability for diagnostics and disease monitoring. Haematologica. 2014;99:465–473. doi: 10.3324/haematol.2013.093765. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Swerdlow S.H., Campo E., Pileri S.A., Harris N.L., Stein H., Siebert R., Advani R., Ghielmini M., Salles G.A., Zelenetz A.D., Jaffe E.S. The 2016 revision of the World Health Organization classification of lymphoid neoplasms. Blood. 2016;127:2375–2390. doi: 10.1182/blood-2016-01-643569. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Swerdlow S.H., Campo E., Harris N.L., Jaffe E.S., Pileri S.A., Stein H., Thiele J., Arber D.A., Hasserjian R.P., Le Beau M.M., Orazi A., Siebert R. ed 4. IARC; Lyon, France: 2017. WHO Classification of Tumours of Haematopoietic and Lymphoid Tissues. [Google Scholar]
- 4.Papaemmanuil E., Gerstung M., Bullinger L., Gaidzik V.I., Paschka P., Roberts N.D., Potter N.E., Heuser M., Thol F., Bolli N., Gundem G., Van Loo P., Martincorena I., Ganly P., Mudie L., McLaren S., O'Meara S., Raine K., Jones D.R., Teague J.W., Butler A.P., Greaves M.F., Ganser A., Dohner K., Schlenk R.F., Dohner H., Campbell P.J. Genomic classification and prognosis in acute myeloid leukemia. N Engl J Med. 2016;374:2209–2221. doi: 10.1056/NEJMoa1516192. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Papaemmanuil E., Gerstung M., Malcovati L., Tauro S., Gundem G., Van Loo P., et al. Clinical and biological implications of driver mutations in myelodysplastic syndromes. Blood. 2013;122:3616–3627. doi: 10.1182/blood-2013-08-518886. quiz 3699. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Stein E.M. Enasidenib, a targeted inhibitor of mutant IDH2 proteins for treatment of relapsed or refractory acute myeloid leukemia. Future Oncol. 2018;14:23–40. doi: 10.2217/fon-2017-0392. [DOI] [PubMed] [Google Scholar]
- 7.Kim E.S. Enasidenib: first global approval. Drugs. 2017;77:1705–1711. doi: 10.1007/s40265-017-0813-2. [DOI] [PubMed] [Google Scholar]
- 8.Birendra K.C., DiNardo C.D. Evidence for clinical differentiation and differentiation syndrome in patients with acute myeloid leukemia and IDH1 mutations treated with the targeted mutant IDH1 inhibitor, AG-120. Clin Lymphoma Myeloma Leuk. 2016;16:460–465. doi: 10.1016/j.clml.2016.04.006. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Wei A.H., Tiong I.S. Midostaurin, enasidenib, CPX-351, gemtuzumab ozogamicin, and venetoclax bring new hope to AML. Blood. 2017;130:2469–2474. doi: 10.1182/blood-2017-08-784066. [DOI] [PubMed] [Google Scholar]
- 10.Mardis E.R. A decade's perspective on DNA sequencing technology. Nature. 2011;470:198–203. doi: 10.1038/nature09796. [DOI] [PubMed] [Google Scholar]
- 11.Hagemann I.S., Devarakonda S., Lockwood C.M., Spencer D.H., Guebert K., Bredemeyer A.J., Al-Kateb H., Nguyen T.T., Duncavage E.J., Cottrell C.E., Kulkarni S., Nagarajan R., Seibert K., Baggstrom M., Waqar S.N., Pfeifer J.D., Morgensztern D., Govindan R. Clinical next-generation sequencing in patients with non-small cell lung cancer. Cancer. 2015;121:631–639. doi: 10.1002/cncr.29089. [DOI] [PubMed] [Google Scholar]
- 12.Hamblin A., Wordsworth S., Fermont J.M., Page S., Kaur K., Camps C., Kaisaki P., Gupta A., Talbot D., Middleton M., Henderson S., Cutts A., Vavoulis D.V., Housby N., Tomlinson I., Taylor J.C., Schuh A. Clinical applicability and cost of a 46-gene panel for genomic analysis of solid tumours: retrospective validation and prospective audit in the UK National Health Service. PLoS Med. 2017;14:e1002230. doi: 10.1371/journal.pmed.1002230. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Kluk M.J., Lindsley R.C., Aster J.C., Lindeman N.I., Szeto D., Hall D., Kuo F.C. Validation and implementation of a custom next-generation sequencing clinical assay for hematologic malignancies. J Mol Diagn. 2016;18:507–515. doi: 10.1016/j.jmoldx.2016.02.003. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Routbort M., Handal B., Patel K.P., Singh R., Aldape K., Reddy N., Barkoh B., Riben M., Medeiros J., Luthra R. OncoSeek: a versatile annotation and reporting system for next generation sequencing-based clinical mutation analysis of cancer specimens. AMP 2012 Meeting Abstracts. J Mol Diagn. 2012;14:637–748. abstract TT79. [Google Scholar]
- 15.Robinson J.T., Thorvaldsdottir H., Winckler W., Guttman M., Lander E.S., Getz G., Mesirov J.P. Integrative genomics viewer. Nat Biotechnol. 2011;29:24–26. doi: 10.1038/nbt.1754. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Thorvaldsdottir H., Robinson J.T., Mesirov J.P. Integrative genomics viewer (IGV): high-performance genomics data visualization and exploration. Brief Bioinform. 2013;14:178–192. doi: 10.1093/bib/bbs017. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Mardis E.R. Next-generation DNA sequencing methods. Annu Rev Genomics Hum Genet. 2008;9:387–402. doi: 10.1146/annurev.genom.9.081307.164359. [DOI] [PubMed] [Google Scholar]
- 18.Li H., Handsaker B., Wysoker A., Fennell T., Ruan J., Homer N., Marth G., Abecasis G., Durbin R., 1000 Genome Project Data Processing Subgroup The sequence alignment/map format and SAMtools. Bioinformatics. 2009;25:2078–2079. doi: 10.1093/bioinformatics/btp352. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.