Identification of Genomic Structural Variants (SVs) with Adverse Prognostic Significance in Normal-Karyotype (nk) Acute Myeloid Leukemia (AML) Patients

Background: nkAML accounts for 40% of all AML. Although nkAML pts are included into favorable (fav) and intermediate (int) European Leukemia Net (ELN2022) risk category, the outcome is highly heterogeneous, and poses therapeutic challenges particularly as regards indication to stem cell transplant (SCT).

Aim: To improve ELN-based prognostication, we characterized genomic SVs in nkAML pts through Long-Read Whole Genome Sequencing (WG-LRS).

Methods: A total of 311 intensively treated nkAML pts were included; a discovery cohort (DC) was made up of 162 pts from the prospective NILG 02/06 and GIMEMA 1310 trial, a validation cohort (VC) included 149 cases from Florence center. WG-LRS was performed on blast-enriched samples by GridION platform to median depth of 5x. SVs identified in the DC were correlated to overall (OS) and event-free (EFS) survival by machine learning and Cox regression.

Results: In the DC, 120 SVs were retained after extensive filtering based on technical parameters and database information; 80.3% insertions, 15.6% deletions, 2.5% duplications, 1.6% inversions, average span 269 bp (range, 52-3.1x10³). Feature selection identified 38 SVs negatively associated with OS that were introduced into a multivariable model including NPM1, FLT3-ITD, CEBPA, ASXL1, TP53 and RUNX1 mutation status. Cox-based model identified 5 SVs with statistical significance for OS: chr2p13.1_del (HR 2.8; 95%CI:2.2-6.6; P=0.016), chr3p34.3_ins (HR 7.1; 2.2-23.2; P=0.001), chr5p12_ins (HR 6.2; 1.9-20.4; P=0.002), chr9q34.3_del (HR 7.6; 2.7-21.6 P≤0.001), chr18q23_ins (HR 3.3; 1.2-9.2; P=0.002). All other input covariates lost significance. Overall, 21 (13%) DC pts had >1 high-risk SVs (HRSV+); their median OS (9.7mo, 95%CI:0-19.4) was significantly shorter compared to HRSV- (not reached, NR; P<0.001), HR of 4.1 (2.5-6.9; P<0.001), median FU of 52.4mo (47.4-57.3). Findings were confirmed in the VC: 15.4% of pts were HRSV+, median OS 15.3mo (11.9-18.6) vs 44.4mo (30.2-58.6) (P=0.014). No significant difference in mutation profile (ELN criteria) was found between HRSV+ and HRSV- pts. HRSV+ pts had lower CR attainment rate (65.9% vs 82.4%, P=0.009) and higher relapse rate (68.9% vs 45%, P=0.015). Median EFS was 5.3mo (2.1-8.6) for HRSV+ and 19.3mo (12.6-25.9; P<0.001) for HRSV- pts. Censoring at SCT had no impact. HRSVs were prognostically informative also in the NPM1+ subset (162 pts): 14.8% were NPM1+HRSV+, median OS 8.2mo (5.5-10.9) compared to NR in NPM1+HRSV- pts (P<0.001). Using ELN2022 criteria, median OS was NR in the fav category (44.3%), 22.6 and 20.4mo, respectively (P=0.39), in the int (35.6%) and adverse (adv) (20.1%) category, highlighting the limited resolution of ELN scoring in nkAML pts. Using ELN fav group as reference, the HR of death was 1.98 (1.3-3) and 2.14 (1.32-3.46), respectively, for ELN int and adv category, compared to 4.4 (2.78-6.95) for HRSV+ category (P<0.001); of note, 73.2% of HRSV+ pts belonged to ELN fav or int category. We then introduced HRSV+ as independent adv variable in the ELN2022 risk score. The resulting model led to improved outcome prediction for the int and adv group, with median OS of 27.4mo and 17.6mo (P=0.034) and HR of 1.9 and 3 compared to fav category (OS, NR). HRSV+ enhanced ELN model showed better Akaike and Concordance Index compared to ELN only: 1576.73 and 0.62 vs 1587.98 and 0.58, respectively. Since HRSVs mapped to genomic regions coding for CCL28, ATP9B, PMPCA, LINC0267, and a topologically associated domain involving ZNF638 and DYSF, we comparatively assessed the mRNA levels of those genes in blasts of HRSV+ and HRSV- pts. In HRSV+, ZNF638 and ATP9B mRNA was upregulated (2.5 and 2.62-fold, respectively), while CCL28 and DYSF mRNA was downregulated (0.72 and 0.27-fold) (all P<0.05). Immunohistochemistry analysis of bone marrow blasts was consistent with mRNA changes. Conventional PCR-based methods for each HRSVs were developed to facilitate integration in diagnostic paths.

Conclusions. We identified a set of 5 SVs that marks a subgroup of nkAML pts with outmost dismal prognosis and helps refining ELN 2022 risk stratification, including NPM1+ pts. Preliminary findings in a separate cohort of additional 100 pts with non-adverse (ELN 2022) abnormal karyotype (presented at meeting) support that HRSVs+ maintains prognostic informativeness across a wide range of genomic-defined categories of AML. Supp. by AIRC-MYNERVA, GIMEMA FI2018-ID.13, Zottola legacy.