Clonal Architecture of MDS Somatic Mutations Dynamically Changes during Azacitidine Therapy and Has Very Limited Potential to Predict Patient Outcome

Introduction: Myelodysplastic syndromes (MDS) are clonal disorders of myeloid hematopoietic stem cells. Recent studies has shown that nearly 90% of patients with MDS carry somatic mutations in bone marrow (BM). These findings triggered a number of studies to identify potential uses of these mutations for diagnostics and prognostics purposes. We focused on a group of 38 patients with advanced stages of the disease that were selected for Azacitidine (AZA) therapy. We then utilized a set of 98 BM samples from the patient cohort that were collected in different stages before, during, and after the period of 4-12 cycles of the therapy. Each patient provided 3 samples on average. This study excludes patients that died early on AZA. Median OS on AZA therapy was 31 Mo. Most prevalent MDS subtypes were RAEB-2 (55%), RAEB-1 (24%), and MDS/AML (13%). 20% of patients had complex karyotype or poor cytogenetics (MedOS=22Mo) and the rest had mostly normal karyotype or intermediate cytogenetics (MedOS=40Mo) prior to AZA. Progression to AML was observed in 55% of patients (PFS= 16 Mo). After 4 cycles, PR was achieved in 59% of patients, CR in 12%, while SD was maintained in 21%, and 9% of patients progressed (PD) to AML.

Methods: We detected relevant mutations in MDS samples using the following approach. We collected genomic DNA from separated BM samples: either a CD3-negative population containing the myeloid compartment, or CD3-positive T cells representing an internal control. We prepared amplicon libraries from these samples using the Illumina TruSight Myeloid Panel that covers 54 key genes involved in myeloid malignancies (notably MDS and AML). We sequenced these libraries using the Illumina NGS platform. To achieve greater sensitivity in detecting SNVs and InDels we utilized two different variant calling pipelines (using samtools mpileup or FreeBayes). Since the (PCR) validation efficacy of each mutation from the single NGS run was below 60%, we improved specificity by using two independently prepared sequencing libraries. The intersection of the variant detections from both libraries was considered accurate and only these data were reported as variants.

Results: When we excluded all germinal variants, 43 somatic variants in ~18 genes were identified per patient on average. The majority (31/43) of these variants had an intermediate impact (on amino acid sequence), while 12/43 had high impact on the protein structure. Importantly, the majority of them had ~1% VAF (variant allele frequency) representing putative clones with low proliferative potential. In contrast, only 8 genes (~14 variants) were mutated with VAF>2%. The following genes were mutated most frequently: TET2, STAG2, ASXL1 in approximately 60-80% of patients. Data from repeatedly analyzed patient samples on AZA therapy led to an unexpected observation that the variants with WAF>2% often exhibit dynamically changing mutation pattern while the variants of non-proliferating clones (with VAF ~1%) remain very stable. We observed prominent development of some variants (ASXL1, STAG2, CUX1, BCOR) as well as an increase in VAF of others (TP53, RUNX1, CUX1) on AZA therapy. Most of these genes when mutated were reported previously as altering prognosis of MDS (Bejar R et al, 2014). Surprisingly, in some samples we found a mutation in the RUNX1 gene before AZA therapy that was not present after the treatment however, after the treatment another not previously observed mutation of RUNX1 emerged. Furthermore, the presence of any of the mutations before AZA including SF3B1 or TP53 did not have any prognostically significant association with OS or PFS. This contention is supported by the fact that many mutations actually disappeared on AZA.

Conclusions: Using an internal sample control combined with a duplicate NGS library preparation we achieved a very high accuracy of detecting somatic variants in MDS-BM sub-separated samples. We observed that variants above 2% VAF change dynamically over the course of AZA therapy while the variants with ~1% VAF remain stable. Our data suggest that development of somatic mutations in AZA-treated MDS patients is a dynamic process, which involves previously identified high risk genes including TP53, RUNX1, CUX1, ASXL1 and BCOR.

Grant support: GAČR 16-05649S & P305/12/1033, AZV: 16-27790A, CZ.1.05/1.1.00/02.0109, UNCE 204021, LH15170, PRVOUK P24, LQ1604 and RVO-VFN64165.