Portal de Periódicos da CAPES

Validation and Implementation of a Custom Next-Generation Sequencing Clinical Assay for Hematologic Malignancies

2016; Elsevier BV; Volume: 18; Issue: 4 Linguagem: Inglês

10.1016/j.jmoldx.2016.02.003

1943-7811

Michael Kluk, R. Coleman Lindsley, Jon C. Aster, Neal I. Lindeman, David Szeto, Dimity Hall, Frank C. Kuo,

Myeloproliferative Neoplasms: Diagnosis and Treatment

Targeted next-generation sequencing panels to identify genetic alterations in cancers are increasingly becoming an integral part of clinical practice. We report here the design, validation, and implementation of a comprehensive 95-gene next-generation sequencing panel targeted for hematologic malignancies that we named rapid heme panel. Rapid heme panel is amplicon based and covers hotspot regions of oncogenes and most of the coding regions of tumor suppressor genes. It is composed of 1330 amplicons and covers 175 kb of genomic sequence in total. Rapid heme panel's average coverage is 1500× with <5% of the amplicons with 95% concordance among a wide array of variants across a range of allele frequencies. Read count analyses that used rapid heme panel showed high concordance with karyotypic results when tumor content was >30%. The average turnaround time was 7 days over a 6-month span with an average volume of ≥40 specimens per week and a low sample fail rate (<1%), demonstrating its suitability for clinical application. Targeted next-generation sequencing panels to identify genetic alterations in cancers are increasingly becoming an integral part of clinical practice. We report here the design, validation, and implementation of a comprehensive 95-gene next-generation sequencing panel targeted for hematologic malignancies that we named rapid heme panel. Rapid heme panel is amplicon based and covers hotspot regions of oncogenes and most of the coding regions of tumor suppressor genes. It is composed of 1330 amplicons and covers 175 kb of genomic sequence in total. Rapid heme panel's average coverage is 1500× with <5% of the amplicons with 95% concordance among a wide array of variants across a range of allele frequencies. Read count analyses that used rapid heme panel showed high concordance with karyotypic results when tumor content was >30%. The average turnaround time was 7 days over a 6-month span with an average volume of ≥40 specimens per week and a low sample fail rate ( 120 genes. The criteria used to determine whether a gene is included in the final panel were i) importance for classification according to World Health Organization; ii) risk stratification with National Comprehensive Cancer Network category 2B or above of evidence; iii) frequency of mutations; iv) germline predisposition of myeloid neoplasms; and v) primer compatibility in a multiplex panel. The amplicons (N = 1330) span 757 coding exons and 175 kb of genomic sequence in aggregate. The final design had a success rate of 98.5% and a total gap of 80%. Tumor samples were sequenced alone without a matched normal sample. DNA isolated from other fresh/frozen and/or alcohol-fixed sources such as coverslips of bone marrow aspirate, fine needle aspirate (both stained and unstained), and cerebrospinal fluid have also yielded satisfactory results. DNA from formalin-fixed, paraffin-embedded was not tested in this panel. The analysis pipeline is outlined in Supplemental Figure S1. Paired-end reads (150 bp) were de-multiplexed into individual BAM files and aligned to Hg19 by the onboard Illumina Reporter software version 2.5.1.3 with the use of the TruSeq Amplicon Workflow with one modification of the default variables (maximum indel size, 300). The genome vcf file was scanned for any nucleotide position with a variant read count >9 regardless of variant allele frequency (VAF) or a position with variant read count of 5 to 9 and a VAF >33%. The candidate list was then filtered for the presence of missense, nonsense, frameshift, splice-site, insertion, and deletion mutations. Recurrent assay-specific false positives (those found in the run control) were filtered out. The resulting candidate list was loaded into a custom database and was reviewed with a web-based interface with links that allow autoloading of BAM files and viewing of each variant in Integrated Genome Viewer (Broad Institute, Cambridge, MA). A variant was classified as pathogenic if it was on a curated list of known pathogenic variants for an oncogene, or if it was a frameshift or nonsense mutation in known functional domains of tumor suppressor genes. The final diagnostic report (Supplemental Appendix S1) included Human Genome Variation Society gene name; VAF; Human Genome Variation Society coding nucleotide, and amino acid level annotation for any variants; read depth; and an interpretive comment from a custom in-house knowledge database that relies on information culled from the published literature. The genomic vcf file was also scanned against a must-call list of known clinically relevant codons (Supplemental Table S3), and any non-reference reads on these codons were separately recorded in the database for review. Findings in the must-call review, case-specific amplicon failures (defined as 200× read depth; and <5% of amplicons with 1 year ago (OncoPanel version 1). RHP was performed on genomic DNA from 24 peripheral blood and bone marrow aspirate specimens that had been previously analyzed by OncoPanel. Among the 24 samples, there were 130 SNVs and indels identified by the OncoPanel assay with VAF, ranging from 4% to 98%, distributed across 46 different genes that are shared between the two panels. The concordance rate for detection of these SNVs and indels in RHP was 99% (129 of 130). The only discrepancy was a variant (STAG2 c.436C>T, 19% VAF) detected by OncoPanel that was poorly covered in the RHP assay (only one read present for that sample). Furthermore, allele frequencies (AFs) between the two assays were highly concordant (correlation coefficient, R2 = 0.89) (Figure 2). Additional validation was performed for specific disease-associated variants in which known positive samples or cell lines were available, including 30 samples with a variety of SNVs and indels (Supplemental Table S4), with VAF of 3% to 80% that were previously identified by alternate methods. All of these variants were detected by RHP. We found that the default variable in the Illumina pipeline (maximum indel size = 25 bp) would not identify indels >25 bp in size. This could be overcome by changing the variable to 300 bp without a significant increase in execution time as stated in the Illumina manual. We found that most reads containing FLT3-ITD were not mapped, whereas their read-mates were. Our customized algorithm thus collects these one-side-mapped-reads whose mates were mapped to FLT3 exon 13 to 15 and re-aligned them to identify the ITD. All nine samples with FLT3-ITDs that ranged in size from 18 to 69 bp identified by OncoPanel or reference laboratory were also successfully detected by RHP with this approach (data not shown). The analytical sensitivity of RHP was evaluated by mixing a positive control sample with normal control DNA. Specifically, genomic DNA from five different cell lines with 22 different SNVs or indels in 19 different genes (Supplemental Table S5) at VAF, ranging from 3% to 60% (before mixing), was used to prepare a 1:1:1:1:1 mixture. This mixed-positive sample was then serially diluted into normal control DNA at dilutions of 1:2, 1:6, and 1:12. The undiluted mixed-positive sample was run in triplicate, and the 1:2 diluted samples were run in duplicate. Known variants were reproducibly detected down to a 5% AF threshold (Figure 3). Below this level, the reproducibility of detection was variant dependent. In addition, the mixed-positive sample was used to evaluate the effect of varying input DNA amounts from 50 to 500 ng. Input of as little as 50 ng of DNA was sufficient for detection of all tested variants across all AFs down to the 5% threshold (Figure 4). However, more false-positive variants [defined as variants with frequencies above the pipeline cutoff (>9 reads or, 5 to 9 reads and >33% AF, plus Illumina Q score > 30) that were not reproducible and that were not detected by other reference assays] were seen with small amounts of input DNA. For example, at an input of 250 ng of DNA, only one variant was determined to be a false positive, whereas 100 ng of input DNA produced four false-positive variants and 50 ng of input DNA produced 11 false variants. The AFs of false-positive variants were generally 10 reads, and the most frequent false-positive mutations were transitions (C>T or T>C). Given these findings, the input DNA amount used in this assay was set at 250 ng, consistent with the manufacturer's recommendation for this platform.Figure 4Evaluation of optimal DNA input amount. DNA from the mixed positive cell line sample was tested, undiluted, using different amounts of input DNA for library preparation. Two hundred fifty nanograms was run in duplicate, and the other amounts were run once. On the x axis, the variants are sorted by increasing allele frequency and numbered 1 to 26.View Large Image Figure ViewerDownload Hi-res image Download (PPT) For evaluation of inter-run reproducibility, eight DNA samples that contained 59 SNVs and indels with AFs that ranged from 3% to 100% were tested in three separate runs; among these variants, 55 of 59 (93%) were detected in all three runs (Figure 5). To assess intra-run reproducibility a mixed-positive control DNA sample that contained 20 different SNVs and indels with AFs that ranged from 2% to 56% were tested in triplicate on two separate runs. Among the 40 total variants (20 variants × 2 runs), 88% (35 of 40) were detected in all three replicates; four of the five variants that were not detected in every replicate had AFs <5% (range, 2% to 4%), consistent with limit of analytical sensitivity of this assay previously determined as described in the previous section. Twelve DNA samples that contained 41 different SNVs and indels with AFs that ranged from 90% were tested on two separate MiSeq instruments in two different locations at our institution. The concordance rate for detection of the variants was 98% (40 of 41). In addition, the AFs of the different variants were highly reproducible between both instruments across the full range of AFs (correlation coefficient, R2 = 0.97) (Figure 6). Myeloid neoplasms are diseases of the hematopoietic stem/progenitor cells.22Bonnet D. Dick J.E. Human acute myeloid leukemia is organized as a hierarchy that originates from a primitive hematopoietic cell.Nat Med. 1997; 3: 730-737Crossref PubMed Scopus (5477) Google Scholar Mature neutrophils, monocytes, eosinophils, but typically not lymphocytes, in the peripheral blood are expected to harbor the same genomic alterations that are found in myeloid progenitors in the bone marrow. We therefore tested 18 paired DNA samples isolated from peripheral blood and bone marrow of the same individuals on the same day. Fifty-nine variants were identified in both paired samples (100% concordance) with allele frequencies that ranged from 2.5% to 90% (correlation coefficient, R2 = 0.77) (Figure 7). As expected, because lymphocytes make up a higher fraction of nucleated cells in the peripheral blood than in the marrow, there is a trend toward a slightly higher VAF in bone marrow. Several CNVs are common in myeloid neoplasms. These include 5q-, 7q-, 20q-, and trisomy 8, among others. We therefore evaluated the use of relative read counts to determine CNV by RHP (see Materials and Methods). Fifteen samples with normal karyotypes and 34 samples with karyotypic abnormalities were analyzed by RHP, and log2 ratio plots were visually inspected for deviation from normal (log2 ratio of ± −0.5 or greater) and scored for the presence of increased or decreased read counts (Figure 8). The results were then compared with the previously determined karyotypic findings. Forty of the 41 karyotypic abnormalities were identified by RHP read count analysis (Table 1). A few single gene deletions not included in the table (in TET2 and ETV6) were identified only by RHP, probably because of the presence of small deletions that are below the resolution of karyotypic analysis. By diluting samples with known CNV alterations into normal control DNA, we determined that identification of copy number gains and losses by read count analysis requires approximately 30% tumor content (data not shown).Table 1Comparison between Read Count Analysis of RHP and Karyotypic FindingsSampledel(20q)del(7q)/7-8+i(17)/17q-4+i(X)10+21+9+19+12+del(11q)Karyotype1YY46,XY,del(20)(q11.2)[8]/45,XY,-7,del(20)[10]/46,XY[2]2YY45,XY,-7,del(20)(q11.2)[cp17]/46,XY[3]3Y47,XY,+8[20]4YY45,XY,-7[7]/46,XY,i(17)(q10{4]46,XY[cp9]5Y45,XY,-7[14]/45,idem,?r(6)(p22q27)[6]6Y47,X,-Y,del(6)(q15q21),+8,+14[17]/47,idem,t(1;9)(q23;q34)[3]7Y46,XY,del(20)(q11.2)[cp18]/45,idem,-7[1]/46,XY[1]8Y46,XY,i(17)(q10)[cp20]9Y47,XY,+8,t(15;17)(q24;q21)[18]/46,XY[2]10YY48,XY,+8,+10[cp20]11YY48,XX,+4,+8[17]/46,XX[3]12YY45,XY,-7[8]/46,XY,-7,+21[cp12]13Y46,XX,del(7)(q22q36)[17]//46,XY[3]14Y45,XY,i(7)(q10),del(12)(p12),-15[7]/46,XY[13]15NY47,XX,+9,del(20)(q11.2)[5]16Y47,XX,+21[20]17Y47,XY,t(1;17)(q42;q25),+19[20]18Y46,XX,del(7)(q21q31),t(15;17)(q24;q21)[cp18]/46,XX[2]19Y47,XY,+21[20]20Y47,XX,+8[20]21Y47,XX,+10[17]/46,XX[cp3]22Y46,XY,r(7)(p2?2?q11.2)[6]/46,XY[14]23Y45,XY,-7[19]/46,XY,t(17;22)(q21;q13)[1]24Y47,XY,+i(X)(q10)[20]25Y47,XX,+8[cp4]/46,XX[7]26Y47,XX,+12[2]/46,XX[7]27Y45,XY,-7[6]/46,XY[10]28Y45,XX,-7[19]/46,XX[1]29Y46,XY,del(11)(q13q23)[14]/46,XY[6]30Y45,X,-X,del(6)(q21),add(16)(p13),del(17)(p11.2)[cp8]31Y46,XX,del(11)(q13q23)[cp15]/46,XX[5]32Y45,XX,add(3)(q26.2),-7[20]33Y46,XY,del(20)(q11.2)[3]/46,XY[12]34Y47,XY,+i(X)(q10)[20]Count5137312231112Del, deletion; i, isochromosome; N, no; RHP, rapid heme panel; Y, yes. Open table in a new tab Del, deletion; i, isochromosome; N, no; RHP, rapid heme panel; Y, yes. In this report, we summarize the design, validation, and implementation of a 95-gene panel test on the basis of NGS that is targeted to hematologic malignancies. It is one of the most comprehensive, up-to-date panels of its kind and includes many recently discovered genes (eg, CALR,23Nangalia J. Massie C.E. Baxter E.J. Nice F.L. Gundem G. Wedge D.C. et al.Somatic CALR mutations in myeloproliferative neoplasms with nonmutated JAK2.N Engl J Med. 2013; 369: 2391-2405Crossref PubMed Scopus (1360) Google Scholar, 24Klampfl T. Gisslinger H. Harutyunyan A.S. Nivarthi H. Rumi E. Milosevic J.D. Them N.C.C. Berg T. Gisslinger B. Pietra D. Chen D. Vladimer G.I. Bagienski K. Milanesi C. Casetti I.C. Sant'Antonio E. Ferretti V. Elena C. Schischlik F. Cleary C. Six M. Schalling M. Schönegger A. Bock C. Malcovati L. Pascutto C. Superti-Furga G. Cazzola M. Kralovics R. Somatic mutations of calreticulin in myeloproliferative neoplasms.N Engl J Med. 2013; 369: 2379-2390Crossref PubMed Scopus (1460) Google Scholar GNB1,25Gambacorti-Passerini C. Donadoni C. Parmiani A. Pirola A. Redaelli S. Signore G. Piazza V. Malcovati L. Fontana D. Spinelli R. Magistroni V. Gaipa G. Peronaci M. Morotti A. Panuzzo C. Saglio G. Usala E. Kim D. Rea D. Zervakis K. Viniou N. Symeonidis A. Becker H. Boutlwood J. Campiotti L. Carrabba M. Elli E. Bignell G.R. Papaemmanuil E. Campbell P.J. Cazzola M. Piazza R. Recurrent ETNK1 mutations in atypical chronic myeloid leukemia.Blood. 2015; 125: 499-503Crossref Scopus (90) Google Scholar and STAG220Kon A. Shih L.Y. Minamino M. Sanada M. Shiraishi Y. Nagata Y. et al.Recurrent mutations in multiple components of the cohesin complex in myeloid neoplasms.Nat Genet. 2013; 45: 1232-1237Crossref PubMed Scopus (276) Google Scholar, 26Thol F. Bollin R. Gehlhaar M. Walter C. Dugas M. Suchanek K.J. Kirchner A. Huang L. Chaturvedi A. Wichmann M. Wiehlmann L. Shahswar R. Damm F. Göhring G. Schlegelberger B. Schlenk R. Döhner K. Döhner H. Krauter J. Ganser A. Heuser M. Mutations in the cohesin complex in acute myeloid leukemia: clinical and prognostic implications.Blood. 2014; 123: 914-920Crossref PubMed Scopus (141) Google Scholar) and most oncogenes and tumor suppressor genes frequently mutated in hematologic malignancies. We chose the Illumina TSCA platform for its capacity (≤1500 amplicons), relative ease of use, and uniform performance across amplicons ( 5% are reproducibly identified at read depths of approximately 1500. Because myeloid neoplasms are stem/progenitor cell disorders, mutations are present in early progenitors and differentiated myeloid cells. Except in the setting of extreme neutropenia/monocytopenia, peripheral blood can substitute for bone marrow in evaluating the presence of myeloid-specific alterations (Figure 7). For patients who present with cytopenia or cytosis, panel tests such as RHP on peripheral blood may provide valuable information for the physicians to decide whether to proceed with a bone marrow biopsy and aspirate. For patients with a diagnosis of myelodysplastic/