Abstract Background: Stem cell transplantation (SCT) is the only curative option for higher-risk myelodysplastic syndromes (MDS). Nevertheless, relapse after SCT is common, leading to poor outcomes. Increasingly sensitive measurable residual disease (MRD) detection methods (multiparameter flow cytometry (MFC) and NGS) have established MRD as a strong predictor of relapse. However, these approaches offer limited insights beyond quantitative MRD assessments. Consequently, critical gaps remain in understanding MRD molecular identities and the mechanisms underlying its progression. Methods: We developed a novel single-cell (sc) multi-omic workflow, CARAMEL-seq, which simultaneously integrates scRNA-seq, scATAC-seq, surface-protein profiling, and lineage-informing mutation analysis with sc resolution. From 52 longitudinal bone marrow (BM) samples representing pre-SCT, relapse, and complete remission timepoints from 10 patients (pts) who relapsed and 3 who remained relapse-free, we sorted CD34⁺ and CD34⁻ BM mononuclear cells to profile both malignant and immune compartments. To enable tracking of coordinated genotype-phenotype dynamics of MRD cells using CARAMEL-seq, we developed a clonal tracing method by integrating posterior probabilities of copy number variants from Numbat and mitochondrial variant probabilities inferred using an Expectation-Maximization algorithm. Results: CARAMEL-seq enabled MRD detection at sensitivities near 10⁻⁷ in samples previously deemed MRD-negative by MFC and NGS and captured MRD cells in 9 of 10 pts who relapsed, but in none of the 3 non-relapsing pts (p=0.01). MRD cells were enriched 24-fold in the CD34⁺ compartment; pseudotime and cell-mapping analyses revealed their specific enrichment in hematopoietic stem and multipotent progenitor (HSC/MPP)-like phenotypes. The comparison between MRD- and donor-derived normal HSPCs revealed consistent, across-patient upregulation (e.g. CD45RA, GPR56, CD49f) and downregulation (e.g. MHC class II) of surface proteins not previously identified as markers of MRD (p0.01). We also detected increased expression of TGF-β, hypoxia, and IFNγ pathways in MRD cells. These findings confirm previous reports and reveal potential novel markers for improving MRD identification in MDS. Longitudinally, we found significant phenotypic and clonal evolution at post-SCT relapse with highest degree of cell state shift in the HSC/MPP compartment and decreased along differentiation trajectories. These phenotypic changes are associated with subclonal shifts and genetic evolution (whole genome seq), suggesting that SCT can selectively eliminate dominant clones while allowing the emergence of resistant subclones. IFNγ and TNF-α signaling pathways were the most consistently upregulated pathways in MDS at relapse compared to preSCT across ATAC, RNA, and protein layers. Moreover, dominantly expanding subclones expressed higher cytokine signaling than contracting ones across pts, suggesting their role in MRD progression. Post-SCT CD8+ T-cells were identified as the IFNγ source. To determine the functional consequences of differential cytokine signaling in preSCT and relapse MDS cells, we treated them in vitro with IFNγ. We observed identical, transcriptome- and protein-wide responses except for HLA class II molecules and their primary regulator (CIITA), whose classical upregulation by IFNγ remained intact in preSCT but not at relapse, suggesting that IFNγ exerts selective pressure post-SCT to promote the outgrowth of MDS clones that resist IFNγ induction of HLA class II expression, potentially contributing to T-cell evasion and relapse. Conclusions: We developed a highly sensitive sc assay for MRD detection and characterization. We demonstrated MRD progression as a phenotypically and genetically evolutionary, rather than static, expansionary process that comprised robust competition among MDS subclones. The “winning” subclones, across all genetically heterogeneous patients, converged on a non-genetic and leukemic adaptation to infiltrating T cells that rewired IFN responses to unlock a key immunoevasive pathway. These data suggest a dynamic interplay between cancer cell evolutionary trajectories and the selective, ecological pressures of the BM microenvironment in driving MRD progression. Furthermore, our study demonstrates the power of longitudinal, sc multi-omic analysis for identifying, tracking, and characterizing MRD cells, opening new avenues to target MRD persistence and progression.
Lee et al. (Mon,) studied this question.