False Positives in Multiplex PCR-Based Next-Generation Sequencing Have Unique Signatures

Next-generation sequencing shows great promise by allowing rapid mutational analysis of multiple genes in human cancers. Recently, we implemented the multiplex PCR-based Ion AmpliSeq Cancer Hotspot Panel (>200 amplicons in 50 genes) to evaluate EGFR, KRAS, and BRAF in lung and colorectal adenocarcinomas. In 10% of samples, automated analysis identified a novel G873R substitution mutation in EGFR. By examining reads individually, we found this mutation in >5% of reads in 50 of 291 samples and also found similar events in 18 additional amplicons. These apparent mutations are present only in short reads and within 10 bases of either end of the read. We therefore hypothesized that these were from panel primers promiscuously binding to nearly complementary sequences of nontargeted amplicons. Sequences around the mutations matched primer binding sites in the panel in 18 of 19 cases, thus likely corresponding to panel primers. Furthermore, because most primers did not show this effect, we demonstrated that next-generation sequencing may be used to better design multiplex PCR primers through iterative elimination of offending primers to minimize mispriming. Our results indicate the need for careful sequence analysis to avoid false-positive mutations that can arise in multiplex PCR panels. The AmpliSeq Cancer panel is a valuable tool for clinical diagnostics, provided awareness of potential artifacts.Detecting driver mutations in cancer genomes is of increasing importance for patient care, both for prognostic significance and for allowing better utilization of targeted therapies. Determining the mutational status of specific genes, such as KRAS, BRAF, and EGFR in lung adenocarcinoma and KRAS and BRAF in colorectal adenocarcinoma, has become the standard of care in clinical oncology to direct epidermal growth factor receptor (EGFR) inhibitor therapy.^1,2 To achieve this, targeted sequencing of these genes with the use of Sanger sequencing and pyrosequencing is widely available. However, Sanger sequencing is labor intensive and has a relatively poor analytic sensitivity (approximately 20% mutant alleles), requiring specimens with a significant percentage of tumor nuclei (>40%) to detect heterozygous mutations.^3,4 Pyrosequencing, although less labor intensive with a better limit of detection (analytic sensitivity approximately 5%), is typically limited to short regions of DNA, requiring the clustering of mutations (eg, KRAS codons 12 and 13). With each of these approaches, a single amplicon from a single patient is analyzed in a single well or capillary.Massively parallel, next-generation sequencing (NGS) platforms, such as the Ion Torrent Personal Genome Machine (PGM) and the Illumina MiSeq, provide limits of detection superior to pyrosequencing combined with even broader genomic coverage.^5–7 Although NGS platforms have the capability to perform cancer whole genome/exome sequencing, targeted sequencing of panels of amplicons with actionable and hotspot mutations is currently more practical in a clinical laboratory setting.^8–11 One panel, the Ion Torrent AmpliSeq Cancer Hotspot Panel, uses multiplex-PCR to cover >200 amplicons in 50 genes known to be involved in carcinogenesis. We recently transitioned to this platform to determine KRAS, BRAF, and EGFR mutation status in formalin-fixed, paraffin-embedded (FFPE) lung and colon cancer specimens.A major issue with FFPE specimens is significant variability in both the quality and quantity of DNA that can be isolated. Sources of this variability include the amount of tumor in the biopsy, time from biopsy/resection to fixation, and time in formalin before processing.^12,13 As a result, we often are left with relatively low DNA concentrations, which may require use of less DNA than the recommended 10- to 30-ng amount for the AmpliSeq panel.Here, we report that off-target amplification is common in multiplex-PCR–based NGS, yielding 19 mispriming events in 208 amplicons (9%) in our study. We define the signature features to identify mispriming events and show that false-positive mutations can be avoided by using multiple bioinformatic analysis tools in the pipeline. We also show that these events are more common with lower input DNA amounts. We demonstrate that the phenomenon is due to multiplex PCR and is not seen when primers are used in monoplex reactions. Finally, because the vast majority of primers do not show significant mispriming, we hypothesize that NGS may be the ultimate multiplex PCR primer design tool by allowing for sensitive detection of off-target amplification and consequent iterative primer design.