Reliable Detection of Abl Tyrosine Kinase Domain Mutations to <1% Using NGS Data Quality Parsing and Corroboration of Overlapping Paired-End Sequences

CML patients should be screened for Abl kinase domain (KD) mutations before changing tyrosine kinase inhibitor (TKI) therapy to ensure that the second-line TKI will be effective against mutations that arose on the first drug. Sanger sequencing (SS) can confidently detect mutations present in >20% of BCR-ABL molecules but will fail to detect minor resistant sub-clones. If missed, these minor sub-clones may be further selected by the second-line TKI and cause treatment failure. The T315I mutation can be particularly problematic as it is resistant to most TKIs except ponatinib. It seems reasonable to detect all mutations if possible, and avoid second line drugs that are known to be ineffective in their presence.

We have developed a next generation sequencing strategy (Illumina MiSeq, 2 x 300 bp) that enables confident detection of all Abl KD mutations present at a level of at least 1% in the BCR-ABL cDNA, this level being 3-5x above the background calling-error rate. Two 500 bp PCR products are sufficient to cover the entire kinase domain (aa M237 to E505). With 300 bp paired-end sequencing of the products, a 65 bp region containing the 315 codon (codons 306-326) is sequenced on both strands. By excluding base changes that are not corroborated on both strands, mutations in this region can be detected with a 10-fold higher accuracy (in 0.1% of BCR-ABL molecules). Samples with low disease burden (1% BCR-ABL/ABL positivity) can also be amplified sufficiently with 50 cycles of PCR, minimising artefactual DNA polymerase-induced mutations. Indexing allows the simultaneous analysis of 80 PCR's in a single MiSeq run (Abl KD of 40 patients). Important aspects of the method are:

1. 50% PhiX DNA is added to the library to increase complexity, and the flow cell is seeded at low density (300,000 clusters per mm²) to reduce sequencing errors.

2. Overlapping paired reads are combined to produce a single FASTQ sequence (modified FLASH source code). Any bases in the overlapping region that do not agree with their counterpart on the other strand are labelled "N" and given a quality score (Q score) of 20.

3. Combined sequences are quality parsed (FASTX Tool Kit) to exclude sequences that do not have a Q score of at least 20 at all bases

4. Parsed high quality sequences are compared to the reference sequence.

We have sequenced the BCR-ABL KD of patients who were sub-optimal responders (BCR-ABL/ABL ratio of >1% at >= 11 months on therapy) in the NCRI SPIRIT 2 trial of first-line imatinib vs dasatinib. Of 60 sub-optimally responding imatinib patients, 6 (10%) had high level mutations (in >20% of BCR-ABL molecules): T315I, L387F, G250E, N331D, M244V x 2. The patients with L387F and G250E were switched to dasatinib and proceeded to respond well. The patient with T315I was also initially switched to dasatinib but failed to respond (BCR-ABL/ABL 29% after 1 year). This patient eventually received ponatinib with good response (BCR-ABL/ABL <1%).

One patient with M244V was switched to nilotinib. Initially this caused relative selection of a pre-existing M387F mutated clone, this mutation increasing from <1% to 55% of BCR-ABL molecules sequenced. However the patient eventually responded well to the drug, indicating that M387F causes only partial resistance to nilotinib. 11/60 (18.3%) imatinib-treated patients had low level mutations present in <20% of BCR-ABL molecules and multiple low-level mutations were seen in 2 patients. One patient had a BCR-ABL/ABL ratio of 40% and 4 non-compound mutations (Y253H - 18%, M244V- 6.5%, K285E - 5.6%, Y312C - 2.6%). All were undetectable by SS. This patient was discontinued from the study and received nilotinib. Nilotinib, which is known to be ineffective against Y253H, caused selection of the Y253H clone to 90% of BCR-ABL molecules and an increase in the BCR-ABL/ABL ratio to 61%. This patient was subsequently switched to ponatinib and responded well (BCR-ABL/ABL ratio < 1%) before undergoing allogeneic transplantation. Of 28 dasatinib-treated patients, 14 had low level mutations including one patient with a T315I of 3.4%. No high-level or compound mutations have so far been discovered in this group at this time point (>11 months).

Our study demonstrates the value of using 2 x 300 bp paired-end sequencing to detect high and low level mutations, even in patients with low-level disease burden, to guide the choice of an appropriate second-line TKI.