Deciphering clonal progression in MDS via longitudinal sequencing

Abstract Introduction Next-generation sequencing (NGS) has redefined the molecular taxonomy of myelodysplastic syndromes (MDS), enabling classification into biologically distinct subgroups with prognostic significance. However, most efforts have focused on baseline genetics, with limited understanding of how these molecular subgroups evolve at relapse. Longitudinal sequencing offers a unique opportunity to track clonal dynamics and assess the stability of molecular classifications over time. We analyzed paired diagnostic and relapse samples to characterize patterns of clonal evolution and identify relapse-enriched genotypes. Methods We retrospectively analyzed 122 MDS patients (Karmanos Cancer Center 2016–2025) with paired diagnostic and relapse NGS (54-gene TruSight Myeloid Panel). Patients were classified into 18 molecular subgroups per Papaemmanuil et al. Chi-square tests were used for categorical comparisons, and Kaplan-Meier analysis with log-rank tests was used for survival analysis. Results: Among 122 MDS patients with paired samples at diagnosis and relapse, the median age was 66 years at both timepoints (p=0.4), and 65% were male (p=0.03). Race distribution was 65% White and 8% Black, with a significant difference between these groups (p=0.00001). Normal karyotype was observed in 38% at diagnosis and 33% at relapse (p=0.48), abnormal in 38% vs 45% (p=0.29), and complex in 24% vs 22% (p=0.75). Median blasts were 2% at both timepoints (p=0.54). No significant differences were seen in WBC (3.5 vs 2.9 × 10³/mm³, p=0.25), hemoglobin (9 vs 8.6 g/dL, p=0.28), or platelets (79 vs 56 × 10³/mm³, p=0.27). Transplant was performed in 56% of patients (p=0.37). The distribution of patients across the 18 molecular clusters shifted significantly between diagnosis and relapse. The AML-like cluster included 7 (5.7%) patients at diagnosis versus 9 (7.4%) at relapse (p=0.55), while the BCOR/L1 cluster increased from 1 (0.8%) to 6 (4.9%) (p=0.13). The CCUS-like group grew from 9 (7.4%) to 13 (10.7%) (p=0.52), and the DDX41 cluster decreased from 1 (0.8%) to 0 (0%) (p=1). EZH27) cluster was absent at diagnosis but appeared in 1 (0.8%) patient at relapse (p=1). Lastly, the -7/SETBP1 cluster had 5 (4.1%) patients at diagnosis versus 4 (3.3%) at relapse (p=1). Molecular cluster stability varied markedly. The most stable clusters were IDH STAG2, EZH2&ASXL1 and bi-TET2, with 100% of patients retaining their molecular classification at relapse. In contrast, the most unstable clusters were No-event, with 0% retention (p=0.00003), and mNOS, which showed significant reclassification with only 36.6% remaining in the same group (p=0.0005). OS differed significantly across the 18 molecular clusters. The DDX41 group had the longest median OS at 167.1 months, followed by SF3B1 (71 months), SRSF2 (60 months), and EZH2/ASXL1 (50 months). In contrast, the poorest survival was seen in the -7/SETBP1 (5 months) and TP53-complex (4 months) clusters. The mNOS group also had poor outcomes (median OS: 8 months), whereas the No-event cluster displayed greater heterogeneity (median OS: 49 months). Notably, the difference in OS between DDX41 and TP53-complex clusters was statistically significant (p = 0.02), underscoring the prognostic relevance of molecular classification. Median overall survival did not differ significantly between stable (IDH-STAG2, TP53-complex) and unstable (No-event, mNOS) clusters (p = 0.33). Conclusions: Molecular trajectories in MDS at different time points in clinical course impacts survival outcomes. This highlights the need for longitudinal sequencing to add further insight into prognostication as well as monitor therapeutic response.