Improved Clinical Assessment of Myeloid Malignancies By Transcriptome Profiling

Clinical genetic testing for myeloid malignancies including acute myeloid leukemia (AML) and myelodysplatic syndromes (MDS) currently relies on cytogenetic assays and targeted sequencing of clinically relevant genes. As more genes with diagnostic and prognostic relevance to these diseases are discovered, traditional testing methods will not be able to scale to effectively meet clinical needs. Next-generation sequencing platforms offer a potential way around this bottle-neck, but the analytic and clinical validity of these methods must be firmly established. To determine whether RNA-Seq, whole-genome sequencing (WGS), or whole-exome sequencing (WES) could meet or exceed the current clinical standard of care, we used these platforms to study retrospective and prospective cohorts of patients with myeloid malignancies.

From a retrospective cohort of 176 patients, we generated RNA-Seq, WES, WGS libraries (176, 92, and 18, respectively). The patient cohort spanned a broad range of myeloid malignancies, including AML, MDS, and therapy-related AML and MDS. We observed that, for similar sequencing efforts, RNA-Seq libraries provided the highest coverage depth over clinically relevant genes, while WGS and WES platforms suffered from recurrent coverage dropouts, particularly in the GC-rich CEBPA gene (median coverage depths of 406x, 115x, and 33x for RNA-Seq, WES, and WGS libraries). We used de novo assembly to assess gene fusions from the RNA-Seq libraries, and demonstrated that all gene fusions previously detected by karyotyping were detectable (41 distinct patients, including 13 inv(16), 9 t(15;17), 8 t(8;21), 3 t(9;11), and 2 t(9;11)). Additionally, we identified five patients with novel rearrangements in the MLL gene family, with potential therapy-altering relevance. We observed complete sensitivity (30 cases) for FLT3 -ITD detection, and detected nine novel low-frequency events undetectable by the previous assay, which were validated using a targeted sequencing panel. We generated SNV and short indel calls for all platforms, and observed that most variants were concordant between platforms (2,079 matched variants aross 92 patients for the RNA-Seq/WES comparison). Missed variant calls (102) in WES libraries were generally due to poor sequencing coverage, while missing calls (44) in RNA-Seq libraries were generally due to allele-specific expression. We generated HLA genotypes for all patients using the RNA-Seq patients, and observed near-total concordance with previous clinical results, increasing the clinical utility of RNA-Seq based testing (of 621 called alleles, only two mismatches were observed at two-digit resolution).

To determine whether gene expression information could be used to identify distinct risk groups within normal-karyotype AMLs, we performed unsupervised clustering analysis and network-based clustering analysis, identifying distinct clusters using both approaches. We demonstrate distinct co-occurrence of two expression clusters with NPM1 - and FLT3 alterations, and distinct survival outcomes for these novel clusters. Additionally, we observed distinct pathway enrichments associated with these clusters, linking molecular subtypes with gene expression signatures.

Altogether, we observed that RNA-Seq offered substantial clinical benefits, due to the ability to recover expressed gene fusions and whole-transcriptome expression information, SNV and short indel variants, and HLA genotype information. Additionally, we observed gene expression heterogeneity which can be used to better stratify intermediate-risk patients. WGS and WES platforms showed recurrent weaknesses in coverage for clinically-relevant genes, and provided less actionable information overall. This work provides crucial analytic validation for the incorporation of RNA-Seq based testing for patients with myeloid malignancies, which may lead to substantial clinical benefits for those patients.