The Interlaboratory RObustness of Next-Generation Sequencing (IRON) Study: Deep-Sequencing Investigating TET2, CBL, and KRAS Mutations In 4464 Amplicons by An International Group Involving 8 Laboratories.

Abstract 1665

Massively parallel pyrosequencing in picoliter-sized wells is an innovative technique and allows highly-sensitive deep-sequencing to detect molecular aberrations. Thus far, limited data is available on the technical performance in a clinical diagnostic setting. Here, we investigated - as an international consortium - the robustness, precision, and reproducibility of 454 amplicon next-generation sequencing (NGS) across 8 laboratories from 6 countries. As a first candidate gene we selected TET2, a frequently mutated gene in myeloproliferative neoplasms. In total, 31 primer pairs including a 10-base molecular barcode sequence were designed and evaluated: All coding exons of TET2 were represented by 27 amplicons. In addition, 2 primer pairs were amplifying hotspot regions to characterize the RING finger domain and linker sequence for CBL and 2 amplicons covered KRAS exons 2 and 3. To execute our study, we used the small volume Titanium emulsion PCR setup (454 Life Sciences, Branford, CT). A cohort of 18 chronic myelomonocytic leukemia (CMML) patient samples were centrally collected by the Munich Leukemia Laboratory and characterized by conventional sequencing for mutations in TET2, CBL, and KRAS. In this selected cohort 33 distinct mutations in TET2, 7 mutations in CBL, and 3 mutations in KRAS, respectively, were detected by Sanger sequencing (plus 10 SNPs and one silent mutation). Each of the participating laboratories received anonymized aliquots of 1.6 μg of genomic DNA to be processed for the generation of PCR amplicons suitable for 454 deep-sequencing. In detail, a total of 31 × 18 (n=558) PCRs were locally performed at each laboratory, i.e. a total of 4464 PCR reactions across 8 centers. Subsequently, at each site each PCR product was individually purified and quantified and corresponding pools were generated by combining 31 amplicons in an equimolar ratio for each patient sample. After processing the samples using the 454 workflow, 3 patients each were loaded per lane on an 8-lane PicoTiterPlate on the GS FLX sequencer instrument. Overall, each of the 8 participating laboratories generated in median 432,606 reads across the 31 PCR amplicons (“Passed Filter Wells”). The median coverage per amplicon was 713-fold, ranging from 553-fold to 878-fold. Dropouts of single amplicons with no coverage obtained were observed in 4/8 laboratories in 61 of 4464 PCR products (1.4%). After alignment of the obtained sequences against the reference genome a total of 92 variants (44 distinct mutations and 10 SNPs) were observed across 22 amplicons. For this analysis, a given variant was scored if, in median, both forward and reverse reads were harboring the variant in at least 20% of reads, i.e. in line with the Sanger sequencing detection limit (GS FLX Amplicon Variant Analysis software v.2.3). In comparison to data available from Sanger sequencing, 454 amplicon deep-sequencing detected all mutations and SNPs that were previously known (few comparisons not possible due to single amplicon dropouts). In 90/92 variant comparisons all eight laboratories consistently detected the variant (two KRAS mutations being detected with a range from 18.0% - 22.6% of reads carrying the mutation). We did not observe a considerable bias in the measurements of the 92 variants between any two centers. Based on paired t-tests for equivalence, with equivalence limits for the standardized expected differences between two centers of -+ε (ε=0.5), the null hypothesis of dissimilar measurements was rejected for all pairs of centers (alpha=0.05). The estimated standard deviation of the measurements across centers was 3.1% (95% CI: [2.9%, 3.2%]), demonstrating the high precision of 454 sequencing to detect mutations. Additionally, we took advantage of the high sensitivity of deep-sequencing. As such, we observed 7 distinct novel mutations (n=2 TET2, n=3 CBL, n=2 KRAS) with frequencies below the Sanger sequencing cut-off value of 20% (median values ranging from 2.8% - 12.6%). These low-level mutations were consistently detected in all laboratories (one CBL mutation with <3% frequency detected in only 5/8 centers). In conclusion, we here demonstrate in a multicenter analysis that amplicon-based deep-sequencing is technically feasible, achieves a high concordance across multiple laboratories, and therefore allows a broad and in-depth molecular characterization of hematological malignancies with high diagnostic sensitivity.