Clonal evolution and risk assessment in myelodysplastic syndromes (MDS): A prospective Study of dynamic IPSS-m validation and evolutionary trajectory modeling by the italian MDS foundation (FISIM)

Abstract

Abstract Background Myelodysplastic syndromes (MDS) are associated with an increased risk of progression to acute myeloid leukemia (AML). Disease evolution involves the sequential acquisition of somatic mutations and clonal selection over time. The Molecular International Prognostic Scoring System (IPSS-M) represents the state-of-the-art for risk stratification at diagnosis but its applicability in a longitudinal context has not been validated. The FISIM-NGS-MDS study was designed to prospectively collect longitudinal clinical and molecular data (from peripheral blood, PB) to investigate clonal evolution and identify patterns predictive of progression (NCT04212390). Methods: Adult patients with a diagnosis of MDS according to the 2016 WHO Classification were prospectively enrolled at diagnosis at 28 Italian hospitals. PB samples were collected at diagnosis, annually during follow-up, before/after treatment, and at disease progression or AML evolution. Targeted NGS was performed at Humanitas Research Hospital. To validate mutation detection accuracy, a subset of PB samples was analyzed in parallel with paired bone marrow (BM) samples. Dynamic validation of the IPSS-M was performed using time-dependent Cox regression models, with model performance assessed by concordance index (c-index). Clonal evolutionary trajectories were reconstructed using cancer cell fractions (CCF), derived from copy number–adjusted variant allele frequencies (VAF), focusing on mutations occurring in at least 1% of the study population. Patient-level directed acyclic graphs were generated through pairwise CCF comparisons and aggregated into cohort-level temporal graphs. A minimum-agony ranking algorithm was applied to infer the most likely order of mutation acquisition. Baseline findings were validated using the original IPSS-M development cohort. Longitudinal validation, comparing inferred versus observed evolutionary directionality, was performed using serial sequencing data from the FISIM cohort. Finally, a time-dependent Cox model with 100-iteration bootstrap validation was fitted to identify CCF dynamics consistently associated with AML evolution. Results The study included 1,002 patients with a median age at diagnosis of 74 years; 315 patients (31%) were classified as IPSS-M Moderate High or higher at baseline. Analysis of paired PB/BM samples from 115 patients revealed 97.1% concordant variants. The few discordant variants had VAF <3%. The rate of discordant events remained low (3.8%) for variants with VAF <5%. VAF for BM was slightly higher than paired PB (median difference 2%, p<0.01). IPSS-M risk classification changed over time in 217 patients (28.6%). Compared with baseline assessment, dynamic IPSS-M showed improved predictive performance across all clinically relevant outcomes, with c-index for overall survival of 0.80 vs 0.74 for baseline IPSS-M, and for leukemia-free survival 0.81 vs 0.77, respectively. Evolutionary modelling using CCF at diagnosis identified 46 recurrent mutation trajectories (present in >10 patients), defined as pairs of co-occurring mutations with a consistent temporal relationship—i.e., the presence of an earlier mutation increased the likelihood of acquiring a subsequent one. Longitudinal validation confirmed consistent directionality for 29 out of 46 trajectories. External baseline validation in the original IPSS-M cohort (n=2,957) was concordant, with only 5 trajectories showing divergent directionality. Time-dependent Cox regression, adjusted for IPSS-M and IPSS-R, showed that CCF for TP53, RUNX1, TET2, PHF6, U2AF1, STAG2, and PTPN11 independently predicted AML evolution. Each gene was retained in over 30% of bootstrap iterations. All associations had a positive hazard direction, suggesting that progressive clonal expansion and/or acquisition of new mutations within these evolutionary trajectories correlates with worse clinical outcomes. Conclusions. Dynamic IPSS-M validation showed superior prognostic performance vs conventional assessment, supporting its use for re-evaluating patient risk over time. CCF-based evolutionary modeling reconstructed mutation sequences and identified genes whose clonal expansion independently predicts AML evolution. Longitudinal clonal monitoring with PB samples was reliable, enabling early, non-invasive identification of high-risk trajectories and improved patient management. Overall, these findings support the concept that novel MDS prognostic tools should be based on longitudinal data.