Background. Normal karyotype Acute Myeloid Leukemia (nk-AML) accounts for 40% of all AML. Conventionally, nk-AML patients (pts) have been associated with the ELN favorable and intermediate risk category, characterized by extreme prognostic heterogeneity, representing an unmet therapeutic need especially as regards effective positioning of allogeneic transplantation (ASCT) as consolidation treatment. Aim. To improve risk stratification of nk-AML, on top of 2017 ELN criteria, by identification of structural variants (SV) with prognostic significance through the analysis of whole genome sequencing by long reads Nanopore platform. Patients and Methods. A total of 247 nk-AML pts receiving intensive therapy was included. The learning cohort included 95 pts from the prospective NILG 02/06 trial ( Bassan R Blood Adv 2019; 3:1103) comparing standard vs high-dose chemotherapy and 67 pts from the prospective GIMEMA AML1310 trial of risk-adapted, MRD-directed therapy ( Venditti A, Blood 2019;134:935). The validation cohort comprised 85 pts from Florence database. Whole genome long-read nanopore sequencing was performed on GridION platform to a median depth of 5x. After alignment to hgr38 genome, only those SV that had been concurrently identified by 4 independent callers were retained for analysis. Results. In the whole series (WS),197 pts (79.7%) obtained CR, 91 (46.2%) relapsed; 50 pts (20.2%) were primary refractory, 83 (33.6%) underwent ASCT. According to ELN criteria, 56.7%, 29.2% and 14.1% of the pts were in the favorable, intermediate and adverse risk category. Median survival (OS) was 16.9mo (9-24.7) in adverse risk category, 24.1mo (4.5-43.7) intermediate, and not reached in favorable risk category (P= 0.00007). However, adverse and intermediate category did not differ significantly (P=0.63), as expected considering the selection of nk-AML. Extensive filtering against 100 healthy donors and publicly available datasets of genomic variants to remove SV with >0.01% allele frequency resulted in 2,038 SV; after removal of private SV (ie, present in ≤2 pt), 122 SV were finally retained. Univariate regression analysis resulted in 10 SV associated with OS, of which 7 were validated with independent technology (Sanger and high-resolution melting, HRM). Multivariate analysis including all ELN-risk score variables showed that 5 of the 7 SV were independently associated with OS (hence, referred as high-risk variants, HRV). Three HRV affected the locus of coding genes on Chr4 q32.3 (inversion spanning 6,103bp), Chr5 p12 (333bp insertion) and Chr9 q34.3 (57bp deletion); Chr3 q26.1 (64bp insertion) in a Long Intergenic Non-Protein Coding RNA gene; Chr2 p13.2 (64bp deletion), in intergenic region. Presence of HRV impacted on OS with HR of 4.1 (95%CI,2.4-7; P=1.5e-7) in the learning cohort and HR 3.4 (95%CI, 1.5-7.6; P=0.003) in the validation cohort. Censoring at ASCT had no impact in either cohort. In total, 29 pts (11.7%) were HRV+, and their OS was 12.6mo (95%CI, 5.4-19.9) vs not reached (95%CI, 33.5-40.9; P=8.4183e-11) in HRV- pts (HR 3.9, 95%CI, 2.5-6; P=1.6261e-9) (Figure 1). DFS and EFS was 7mo (95%CI: 3.6-10.5) and 14.6mo (95%CI,2.1-27) in HRV+ pts vs not reached (95%CI, 31.7-40) and 30mo (95%CI: 29.9-30.1) (P<0.01 for both). Also, the rate of relapse was higher among HRV+, 58.6% vs 34.4% (P=0.01), with a trend to lower rate of CR attainment (72.4% vs 82.6%). There was no remarkable difference between HRV+ and HRV- pts as regarded frequency and type of ELN molecular variables. We finally evaluated whether integration of HRV within ELN risk score might improve the predictive performance of the latter. The resulting model enlisted 4 tiers, favorable (median OS not reached), intermediate (27.4mo, 95%CI: 3.6-51.1, HR:2.4; 95%CI,1.5-3.8), adverse (21.5mo, 95%CI: 5.4-37.6, HR:2.5; 95%CI,1.4-4.3) and HRV+ (12.6mo, 95%CI: 5.4-19.9, HR:5.9; 95%CI,3.6-9.8) categories (P=1.3395e-12) (Figure 2). Of note, the HRV+ category included 79.3% of pts otherwise categorized as ELN favorable or intermediate. Conclusions. We have identified, by long-read whole genome sequencing, a set of 5 structural variants that marks a subgroup of nk-AML pts with outmost dismal prognosis and may be integrated in ELN-risk score to help therapeutic decisions. Since HRV can be conveniently interrogated by routine approaches such as Sanger sequencing and HRM, further studies on larger cohorts might support these findings.