ULTRA-DEEP SEQUENCING WITH COMPUTED THRESHOLDS FOR SENSITIVE MUTATION ANALYSIS IN THE KINASE DOMAIN OF BCR-ABL: FOCUSED ON DEEP MUTATION DETECTION IN CHRONIC MYELOID LEUKEMIA IN CHRONIC PHASE

Background:

Sanger sequencing (SS) is widely used for mutation detection in the BCR-ABL kinase domain (KD) with a sensitivity of 15-20%. Due to its clonal nature, next generation sequencing (NGS) improves resolution through ultra-deep sequencing (UDS) of targeted genes. It is necessary to have tools separating errors produced in NGS pipeline from true mutational events.

Aims:

1) We develop a model for error correction and estimating thresholds for detection of mutations in the KD of BCR-ABL by UDS of an amplicon library that was prepared using BCR-ABL plates developed within the IRON-II study (Roche Applied Science). 2) We applied the thresholds in a retrospective UDS analysis of BCR-ABL mutations associated with resistance under TKI treatment in CP-CML patients to assess applicability to future clinical practice.

Methods:

BCR-ABL mutations were analyzed in 129 samples from 15 CP-CML patients by UDS on a 454 platform (Roche Applied Science). Each patient received 2 or 3 lines of TKIs. BCR-ABL oligonucleotide primer plates containing fusion primers were used to create 4 partially overlapping amplicons covering the KD coding region. For estimation of SNS (Single Nucleotide Substitution) error rates, we used 8 samples from healthy donors.  The SNS caller and software framework for estimation of errors was written in Java. We estimated error rates using the Lea Coulson distribution. The distribution parameter was estimated by the maximum likelihood method (Foster, PL in DNA Repair, 2006, Part B, Vol409). If the probability, that the same or higher count of SNS reads from single base occur in error distribution, is less than 1% then we call this SNS significant mutation. On the level of 1% for each nucleotide substitution we also set a threshold.

Results:

1) The SNS error frequency was much higher for nucleotide transitions vs. transversions (e.g., mean 2.2 SNS events for T/C vs. 0.1 for T/A per 3000 reads). Therefore, computations for each type of nucleotide substitution were performed separately and resulted in higher thresholds for transitions, resulting in e.g. T315I and Y253H with 77 (2.57%) and 96 (3.20%) mutations per 3000 reads, respectively, in contrast to transversions, resulting in e.g. V299L with 10 (0.3%) per 3000 reads. 2) Using the computed thresholds UDS revealed insignificant mutations at the time of diagnosis that subsequently developed under TKIs in 8/15 patients. We examined, if UDS can outperform SS in detecting mutations earlier considering significant mutations only. UDS detected 5 mutations 3 months earlier, 4 mutations 4 months earlier and 2 mutations 17 months earlier during imatinib treatment. UDS detected other minor mutant populations (G250R, L387F, L364I, Y253H) in 4 patients. In 3 patients, UDS, but not SS, revealed minor populations with F317L, Y253H and T315I before therapy switch that emerged under selected 2^nd TKI. Finally, we compared the level of baseline mutations detected by UDS and SS after TKI switch. Baseline mutations were still detectable by UDS in 2 patients who achieved MMR at the time of analysis.

Summary / Conclusion:

Since enzymes create errors during library preparation, the threshold computation is essential for relevant interpretation of BCR-ABL KD mutations detected with the highly sensitive UDS technology. Higher SNS error rate was found for nucleotide transitions. UDS outperformed SS in detection sensitivity and thus provided more information. For future reference, we suggest using UDS even for critical clinical applications provided that the precise methods for error management presented here are utilized.

Supported by IGA NT11555 and NT13899.

Keywords: BCR-ABL, Mutation analysis

Abstrakta v časopise Haematologica 2013, Suppl1

Online Program