Comprehensive diagnostics of acute myeloid leukemia by whole transcriptome RNA sequencing

Abstract

Acute myeloid leukemia (AML) is caused by genetic aberrations that also govern the prognosis of patients and guide risk-adapted and targeted therapy. Genetic aberrations in AML are structurally diverse and currently detected by different diagnostic assays. This study sought to establish whole transcriptome RNA sequencing as single, comprehensive, and flexible platform for AML diagnostics. We developed HAMLET (Human AML Expedited Transcriptomics) as bioinformatics pipeline for simultaneous detection of fusion genes, small variants, tandem duplications, and gene expression with all information assembled in an annotated, user-friendly output file. Whole transcriptome RNA sequencing was performed on 100 AML cases and HAMLET results were validated by reference assays and targeted resequencing. The data showed that HAMLET accurately detected all fusion genes and overexpression of EVI1 irrespective of 3q26 aberrations. In addition, small variants in 13 genes that are often mutated in AML were called with 99.2% sensitivity and 100% specificity, and tandem duplications in FLT3 and KMT2A were detected by a novel algorithm based on soft-clipped reads with 100% sensitivity and 97.1% specificity. In conclusion, HAMLET has the potential to provide accurate comprehensive diagnostic information relevant for AML classification, risk assessment and targeted therapy on a single technology platform.

Fig. 1. The bioinformatics pipeline HAMLET.

HAMLET was developed as a bioinformatics pipeline to call all relevant information for diagnosis and prognosis of AML from raw mRNAseq data. HAMLET integrates four modules using algorithms to detect fusion genes (STARfusion ∩ FusionCatcher), small variants (VARSCAN), large tandem duplications (ReSCU), and gene expression. HAMLET permits immediate AML classification into subtypes according to WHO 2016 or genomic classifications, provides additional prognostic information according to the European Leukemia Net, and identifies molecular targets for therapy.

Fig. 2. Detection of fusion genes by HAMLET.

a Detection of fusion genes with prognostic relevance for AML included in the WHO 2016 classification. Examples shown: CBFB-MYH11 of the inv(16)(p13q22) (case 2-004), RUNX1-RUNX1T1 of the t(8;21)(q22;q22) (case 3-007), and KMT2A-MLLT3 of the t(9;11)(p21;q23) (case 3-001). b Detection of fusion genes with prognostic relevance for AML not included in the WHO 2016 classification. Examples shown: KMT2A-MLLT6 of the t(11;17)(q23;q21) (case 3-010), NUP98-NSD1 of the cryptic t(5;11)(q35;p15.5) (case 2-031 with normal karyotype), and ETV6-LYN of a cryptic ins(12;8)(p13;q11q21) in a case (2-020) with add(8)(q24) and del(12)(p13). Gels adjacent to the circus plots depict validation by RT-PCR with custom primers (Table SVI). An amplicon of 279 bp demonstrated fusion of KMT2A exon 7 to MLLT6 intron 10 (case 3-010). Amplicons of 209 and 350 bp demonstrated fusion of NUP98 exon 11 or 12 to NSD1 exon 7 (cases 2-003, 2-031, 2-032). An amplicon of 181 bp demonstrated fusion of ETV6 exon 5 to LYN exon 8 (case 2-020). All fusion transcripts were validated by sequencing (Tables SIII and SV).

Fig. 3. Detection of double CEBPA mutants.

Expression was analyzed by RNAseq for 16 genes included in the 19-probe signature associated with bi-allelic CEBPA mutants [14]. Up- and down-regulated expression levels are shown in green and red, respectively. All 100 AML cases are indicated on the Y-axis. All three double CEBPA mutants shared the same gene signature (2-009, 2-039, and 2-045). Two single CEBPA mutant cases (2-025, 3-003) clustered with the double mutants. Sanger sequencing revealed that one case (3-003) had a 30 bp C-terminal insertion that was missed by HAMLET, while the other case was a true single N-terminal mutant without C-terminal mutation (2-025). Similar results were obtained for the proposed 55-probe signature (data not shown) [13].

Fig. 4. Detection of tandem repeats in FLT3 and KMT2A by the ReSCU algorithm.

Graphs depict the ratio of soft-clipped (SC) reads-to-total coverage (left Y-axis; yellow peaks) and total read coverage (right Y-axis; gray areas) for each position. Reciprocal events at identical and close positions are indicated by solid and dotted lines, while red and purple lines represent dominant and subdominant events, respectively. a FLT3 exons 14–15 (1787–2024 in ENST00000241453 and 1705–1942 in coding sequence). AML cases are representative for a reciprocal event with high SC reads-to-total coverage (2-029), a case with dominant and subdominant reciprocal events (1-003), a reciprocal event with low SC reads-to-total coverage (3-009), and a case without ITD (1-001). b Comparison of FLT3-ITD results by HAMLET and diagnostic PCR. Diagnostic PCR was performed on genomic DNA using NED-5′-GTAAAACGACGCCCAGTCTGAAGCAATTTAGGTATGAAAGC-3′ and VIC-5′-GGAAACAGCTATGACCATGTACCTTTCAGCATTTTGACG-3′ as forward and reverse primers, respectively. Left panel: SC reads-to-total coverage by HAMLET (X-axis) versus allelic ratios by diagnostic PCR calculated from areas under the curve for mutant and wild-type FLT3 fragments after capillary electrophoresis (Y-axis). SC reads-to-total coverage is the sum of dominant and subdominant ITD and average between start and end of ITD. Right panel: ITD length as determined by HAMLET (X-axis) versus diagnostic PCR (Y-axis). Two subdominant ITD in case 3-027 (green dots) were detected by HAMLET with <0.5% SC reads-to-total coverage but not by routine diagnostics. Three subdominant ITD in cases 2-015, 2-032, and 2-046 (red dots) were detected by routine diagnostics but not by HAMLET. All dominant clones were called by both tests. c KMT2A exons 2–13 (456–4719 in ENST00000534358 and 433–4696 in coding sequence). AML cases are representative for duplications of exon 2–8 (2-002), exon 2 and exon 2–3 (2-027), exon 3-6 (2-013), exon 2-10 (2-054), and a case without PTD (2-005).

Fig. 5. Detection of EVI1 overexpression.

a Schematic representation of two mRNA transcripts from the MECOM locus on chromosome 3. One transcript contains exon 1–2 of MDS1 fused to exon 2–15 of EVI1 (MDS1-EVI1 transcript), whereas the other transcript contains exon 1–15 of EVI1 (EVI1 transcript). b Comparison of EVI1 expression by RNAseq and quantitative RT-PCR. X-axis: Expression of the first exon of EVI1 normalized for expression of the PBGD housekeeping gene by quantitative RT-PCR (log2 EVI1/PBGD). Y-axis: Sum of base coverage of the first exon of EVI1 per kb transcript and one million mapped reads (log2 BPKM) by RNAseq.

Fig. 6. Classification, risk assessment, and actionable targets as derived from HAMLET output and metaphase cytogenetics.

AML cases are primarily ordered according to the WHO 2016 classification and secondarily according to the proposed genetic AML classification [9]. Relevant variants with respect to the genetic classification and potentially actionable targets are indicated per case. Red font indicates parameters provided by HAMLET that cannot be reliably addressed by metaphase cytogenetics and targeted NGS sequencing. Prognosis according to the ELN risk [1] is indicated at the bottom. In addition, cases whose ELN risk was altered by HAMLET information are indicated. Red boxes mark the HAMLET parameter that modifies ELN risk. Except for aneuploidies and complex karyotypes, HAMLET provides all information required for risk assessment. Arrows point from a diagnostic sample to the relapse in the same patient. AML with MDS AML with myelodysplasia-associated changes, tAML therapy-related AML, Mut mutated, AML NOS AML not otherwise specified, ELN European leukemia net.