Genomic clustering informs risk-adapted conditioning regimens in AML with pre-transplant deep remission: Augmented stratification beyond ELN-2022 criteria

Abstract

Abstract Background: The composition of conditioning regimens are critical determinants of post-transplant outcomes. However, there remains unclear how to personalize conditioning based on disease-specific genomic features. This study aims to explore the potential of genomic clustering as a tool for augmenting stratification and personalizing conditioning regimens, moving beyond the limitations of the ELN-2022 classification. Method: Based on the National Longitudinal Cohort of Hematological Diseases (NCT04645199) in China, this retrospective study included 575 patients who achieved both pre-transplant complete remission and measurable residual disease negativity. Partitioning Around Medoids (PAM) unsupervised clustering was performed based on molecular and cytogenetic abnormalities in the ELN-2022 risk stratification and those occurring in >3% of newly diagnosed cases after comparing clustering performance with Hierarchical Clustering for internal validation. To enhance clinical applicability, the PAM-based subgroups and conditioning regimens were streamlined into two to three categories. The minimal P value among the two/three pairwise comparisons was used as the score to rank each combination, and the combination yielding the lowest minimum P value was selected as the optimal stratification. Results: The median age was 36 (range, 14 to 66) years of this cohort with a median follow-up of 2.12 years with maximum 9.60 years. Patients underwent one of four conditioning regimens: R1 [Busulfan (Bu) + Cyclophosphamide (Cy) + Deoxyadenosine analogs + Cytarabine, n = 316], R2 [Bu + Cy + Fludarabine (Flu) + Idarubicin, n = 178], R3 [Bu + Melphalan, n = 44], R4 (Other, n = 37). Deoxyadenosine analogs include Flu and Cladribine. Firstly, we compared prognostic impact of ELN-2022 risk stratification across different conditioning regimens. A total of 214 (37.2%), 184 (32.0%) and 177 (30.7%) in ELN-2022 favorable, intermediate and adverse-risk groups, respectively. Among intermediate risk patients, regimens R2 and R3 were associated with significantly better 3-year overall survival (OS) compared to R4 (85.8% vs. 0, P = 0.041). However, in both favorable and adverse risk groups, survival differences among individual conditioning regimens were not well distinguished. In addition, favorable risk group presented superior 3-year OS compared to those in adverse risk group (84.4% vs. 69.7%, P = 0.038) for patients treated with R2, while ELN-2022 classification failed to distinguish prognostic subgroups in patients receiving R1 and R3. Then, PAM clustering identified seven distinct genetic subgroups: Cluster1 (C1), ASXL1/GATA2/FAT1; C2, inv(16)(p13.1q22)/CBFB::MYH11; C3, +8; C4, KMT2A-rearranged, NRAS, KRAS; C5, t(8;21)(q22;q22.1)/RUNX1::RUNX1T1 and KIT mutations; C6, FLT3-ITD, NPM1, DNMT3A, and FLT3-TKD mutations; C7, normal karyotype, bZIP and non-bZIP in-frame CEBPA mutations. Different genetic clusters benefit from distinct conditioning regimens. In patients treated with R1, 3-year OS was significantly higher in C2/C3 (96.8% vs. 71.4%, P = 0.009) and C5/C7 (78.0% vs. 71.4%, P = 0.029) compared to C1/C4/C6. For patients receiving R2, C2/C3/C6 had significantly higher 3-year OS than C5/C7 (95.6% vs. 72.2%, P = 0.005), and also better than C1/C4 (95.6% vs. 76.9%, P = 0.026). For those treated with R3, C1/C4/C6/C7 showed better 3-year OS than C5 (90.9% vs. 64.1%, P = 0.060). The 7 clusters were stratified into 3 prognostic groups based on 3-year OS: favorable (C2/C3, 92.0%), intermediate (C5/C6, 78.8%), and adverse (C1/C4/C7, 73.3%) (P = 0.018). Within each group, R1, R2 and R3 were further compared. In the favorable-risk group (C2/C3), patients receiving R1 (96.8% vs. 100%, P = 0.039) and R2 (100% vs. 76.2%, P = 0.036) showed significantly improved 3-year OS compared to those receiving R3. In the intermediate-risk group (C5/C6), R1/R2 was associated with superior 1-year OS versus R3/R5 (89.6% vs. 75.3%, P = 0.044), though the 3-year OS difference was not statistically significant (79.9% vs. 75.3%, P = 0.192). Among patients in the adverse-risk group (C1/C4/C7), 3-year OS did not differ significantly between R1/R2/R5 and R3/R4 (73.1% vs. 59.2%, P = 0.464). However, R1/R2/R5 still achieved a 3-year OS exceeding 70%. Conclusion: Based on PAM clustering, patients in C2/C3 derived significant benefit from R1 and R2 regimens. Patients in C5/C7 appeared to benefit from R1, while those in C6 responded better to R2.