Detection of Multiple Types of Cancer Driver Mutations Using Targeted RNA Sequencing in NSCLC

Abstract

Currently, DNA and RNA are used separately to capture different types of gene mutations. DNA is commonly used for the detection of SNVs, indels and CNVs; RNA is used for analysis of gene fusion and gene expression. To perform both DNA sequencing (DNA-seq) and RNA-seq, material is divided into two copies, and two different proce-dures are required for sequencing. Due to overconsumption of samples and experimental process complexity, it is necessary to create an experimental method capable of analyzing SNVs, indels, fusions and expression. We developed an RNA-based hybridization capture panel targeting actionable driver oncogenes in solid tumors and corresponding sample preparation and bioinformatics workflows. Analytical validation with an RNA standard reference containing 16 known fusion mutations and 6 SNV mutations demonstrated a detection specificity of 100.0% [95% CI 88.7%~100.0%] for SNVs and 100.0% [95% CI 95.4%~100.0%] for fusions. The targeted RNA panel achieved a 0.73-2.63 copies/ng RNA lower limit of detection (LOD) for SNVs and 0.21-6.48 copies/ng RNA for fusions. Gene expression analysis revealed a correlation greater than 0.9 across all 15 cancer-related genes between the RNA-seq re-sults and targeted RNA panel. Among 1253 NSCLC FFPE tumor samples, multiple mutation types were called from DNA- and RNA-seq data and compared between the two assays. The DNA panel detected 103 fusions and 21 METex14 skipping events; 124 fusions and 26 METex14 skipping events were detected by the target RNA panel; 21 fusions and 4 METex14 skipping events were only detected by the target RNA panel. Among the 173 NSCLC samples negative for targetable mutations by DNA-seq, 15 (15/173, 8.67%) showed targetable gene fusions that may change clinical decisions with RNA-seq. In total, 226 tier I and tier II missense variants for NSCLC were analyzed at ge-nomic (DNA-seq) and transcriptomic (RNA-seq) levels. The positive percent agreement (PPA) was 97.8%, and the positive predictive value (PPV) was 98.6%. Interestingly, var-iant allele frequencies were generally higher at the RNA level than at the DNA level, suggesting relatively dominant expression of mutant alleles. PPA was 97.6% and PPV 99.38% for EGFR 19del and 20ins variants. We also explored the relationship of RNA expression with gene copy number and protein expression. The RPKM of EGFR transcripts assessed by the RNA panel showed a linear relationship with copy number quantified by the DNA panel, with an R of 0.8 in 1253 samples. In contrast, MET gene expression is regulated in a more complex manner. In IHC analysis, all 3+ samples exhibited higher RPKM levels; IHC level of 2+ and below showed lower RNA expression. Parallel DNA- and RNA-seq and systematic analysis demonstrated the accuracy and robustness of the RNA sequencing panel in identifying multiple types of variants for cancer therapy.

Uncommon/rare oncogenic drivers in non‐small cell lung cancer: Consensus and contention

“…However, on the other hand, two samples with positive results by DNA‐NGS, failed to have RNA‐NGS results due to insufficient RNA quality, indicating the higher sample quality requirements of RNA‐NGS. Another study conducted head‐to‐head comparison of DNA‐NGS and RNA‐NGS in clinical formalin fixed paraffin embedded (FFPE) samples of 1253 NSCLC patients and found that 6% (10/168) of patients with negative results of DNA‐NGS oncogenic drivers test were rediscovered by RNA‐NGS detection (including 7 samples of fusion mutations, 3 samples of MET exon 14 skipping mutations), and these patients were offered targeted therapy [23].…”

“…Using an NGS panel composed of a specific number of genes, typically covering at least 50 genes, is a feasible strategy. If no oncogenic drivers are detected by a DNA‐based NGS panel, RNA‐based NGS should be considered [23]. In addition to considering the number of genes, it is also important to consider the tiers of somatic variants on the covered genes.…”

ISMI-VAE: A Deep Learning Model for Classifying Disease Cells Using Gene Expression and SNV Data

“…On the one hand, gene expression data can indirectly reflect RNA-based protein expression levels, but cannot reflect gene sequence information, which SNV happens to reflect [11]. On the other hand, many diseases not only affect gene expression data, but more importantly, it is influenced by SNVs or even caused by SNVs [12,13]. Therefore, combining gene expression data and SNV data can better reveal the mechanism of action of diseases.…”