Background: MDS is a genetically complex, oligoclonal disease consisting of a founding clone and typically one or more subclones derived from the founding clone. Previously we have shown that in MDS patients treated with chemotherapy, a minor subclone present at diagnosis can expand during disease progression from MDS to secondary AML, highlighting the clinical implications of clonal heterogeneity. Whether a similar pattern of clonal evolution occurs in MDS patients that relapse following allogeneic hematopoietic stem cell transplant (alloHSCT) is not known.Methods: We identified 9 MDS patients who progressed after receiving either a myeloablative (n=3) or reduced-intensity (n=6) alloHSCT (median time to progression 309 days, range 98-881). We performed enhanced exome sequencing (EES) to define the clonal architecture of 23 tumor samples at the following clinical landmarks (with matched skin as a source of normal DNA): diagnosis, <2 months pre-alloHSCT (where available), and relapse post-alloHSCT. Somatic mutations were validated in the 23 discovery samples and genotyped in 35 additional serial banked samples at various time-points post-alloHSCT, including day 30 and 100, using capture probes targeting all putative single-nucleotide variants (SNV) and short insertions and deletions (INDELs) identified by EES. The variant allele fraction (VAF) of all validated somatic mutations was determined. Ultra-deep, error-corrected sequencing (i.e., barcoded sequencing) was performed on 49 tumor samples to provide increased sensitivity to detect low-level mutations post-alloHSCT. Copy number alterations were identified using exome and SNP array data.Results: Averaged sequencing coverage depth was 246x for tumors subjected to EES; 537x for validation sequencing, and 24,150x total and 5,180x unique for the ultra-deep sequencing. In all cases, we observed that mutations found in the diagnostic founding clone were always detected at relapse. However, using SNVs, INDELs, and copy number analysis, we show that the dominant clone at relapse was often derived from a population that was subclonal at presentation. We observed the following, non-mutually exclusive patterns of clonal evolution at relapse: i) A subclone expanded or emerged and became the dominant clone at relapse as compared to presentation (n=6). In 2 of these cases, the subclone contained mutations that were not detected at presentation even via ultra-deep sequencing. ii) A subclone was cleared with alloHSCT (defined as VAF<1% by EES, n=4), confirmed by ultra-deep sequencing when available (n=2). iii) The founding/dominant clone at diagnosis was also the dominant clone at relapse (n=3). However, in 2 of these 3 cases, changes in clonal architecture were observed with evidence of rising or contracting subclones. Although our sample size is relatively small, the intensity of the alloHSCT conditioning regimen did not impact the relapse pattern. No acquired abnormalities were detected in the MHC locus, and no mutations in a particular gene family or cellular pathway were consistently observed in rising or contracting subclones.Finally, we used ultra-deep sequencing to determine if mutations (i.e., tumor cells) could be detected at day 30 post-alloHSCT in 7 of the 8 patients with no evidence of disease, who had available data. Mutations were detected in 6 of 7 patients. The average detectable mutation VAF per patient was 0.37% (ranged from 0.04% to 0.93%)(i.e., 1 mutant cell in 135).Conclusion: Complex clonal dynamics and clonal evolution are observed at relapse post-alloHSCT for MDS. Although minor subclones rise and may become the dominant clone at relapse, mutations present in the dominant (i.e., founding) clone of the diagnostic MDS sample pre-alloHSCT are always detected at relapse. This is similar to the pattern of clonal evolution previously observed for MDS progression to secondary AML following chemotherapy. These observations have implications for targeted therapy and tumor burden monitoring. Ultra-deep sequencing can detect persistent or emerging mutations at early time-points post-alloHSCT that are associated with subsequent relapse. The predictive value of detecting persistent mutations early after post-alloHSCT merits testing in future studies. DisclosuresJacoby:Quintiles: Consultancy; Sunesis: Research Funding; Celgene: Speakers Bureau. DiPersio:Incyte Corporation: Research Funding.