Session: 618. Acute Myeloid Leukemias: Biomarkers and Molecular Markers in Diagnosis and Prognosis: Poster II
Hematology Disease Topics & Pathways:
Research, Acute Myeloid Malignancies, AML, Adult, Translational Research, Diseases, Myeloid Malignancies, Technology and Procedures, Study Population, Human, Omics technologies
METHODS: The levels of 429 proteins (339 total, 90 post-translational modified) were measured in fresh samples of 805 newly diagnosed AML patients using Reverse Phase Protein Arrays. Protein expression was normalized to normal CD34+ cells. We selected 67 patients with AMoL diagnosis, which were treated with either AraC-based (N=48), or Hypomethylating agent-based (HMA, N=19) chemotherapy. Identification of prognostic proteins was carried out with pairwise LogRank tests (p<0.05) for Overall Survival (OS) and Remission Duration (RD) adjusted with False Discovery Rate (FDR), followed by unbiased hierarchical clustering. Continuous variables were compared with Wilcoxon tests and categorial ones with Fisher’s Exact tests with simulated p-values (10000 replicates). Cox proportional hazards models (CoxPH) were generated for Uni-(UV) and Multi-variate (MV) analysis. Differentially expression analysis was performed with Wilcoxon tests with p-values adjusted with FDR (p<0.01) and a Log2-fold-change cutoff of 0.5.
RESULTS: We identified 30 prognostic proteins that split the patients into 2 clusters: C1 and C2. The biological processes related to those proteins were: cell cycle control and DNA damage response, T and B cell-related, epigenetics, metabolism, heatshock, adhesion and cytoskeleton organization, apoptosis, ribosomal activity, and other signaling transduction pathways, respectively with 9, 4, 3, 3, 2, 2, 2, 1 and 4 proteins. Clinical and molecular features were not biased towards any cluster, except for white blood cell count and blast percentage, which were higher in C2.
Outcomes differ greatly between clusters, with C1 patients having better OS and RD when treated with AraC (median OS: C1+AraC >120mo vs C1+HMA =8.5mo; p<0.001 and median RD: C1+AraC >120mo vs C1+HMA =8.5mo; p<0.001), compared to C2 which had poor outcomes independent of the therapeutic regimen (median OS: C2+AraC =16.8mo vs C2+HMA =8.9mo; p=0.55 and median RD: C2+AraC =5.5mo vs C2+HMA >72mo; p=0.20). A similar OS pattern was observed when clusters were filtered by caucasians, primary AML, intermediary cytogenetic (cyto.) risk, diploid karyotype (kar.), trisomy 8, patients <40yo, AraC therapy and FLT3-ITD mutants. Regarding RD, filtering for caucasians, males, primary AML, intermediary cyto. risk, diploid kar., patients <40yo and 41-55yo, AraC and VTX therapies and FLT3-ITD and NPM1 mutants showed a similar pattern.
Considering C1+AraC the reference cluster (ref), CoxPH models showed that all clusters were predictive for OS in the UV analysis (HR=1, 3.8, 8.5, 5.0; p=ref., 0.003, <0.001, 0.003), as well as age, secondary AML, complex kar, FLT3-ITD, RUNX1 and TP53 mutations. In the MV analysis, all clusters (HR=1, 3.7, 6.2, 4.4; p=ref., <0.001, <0.001, 0.002) and TP53 mutation were prognostic. Regarding RD, C1+AraC, C2+AraC and C1+HMA were prognostic in the UV model (HR=1, 19, 20, 4.8; p=ref., <0.001, <0.001, 0.20), along with age, and GATA2 and MLL mutations. In the MV model, C1+AraC, C2+AraC and C1+HMA were predictive for RD (HR=1, 21, 50, 22; p=ref., 0.006, 0.002, 0.30), as well as GATA2 mutation.
Differential expression analysis between C1 and C2 identified 39 proteins, among 429, which may represent novel therapeutic targets, such as CDKN1B, CBX7, ERN1, ETS1, NOTCH1.cle, PXN, S100A4, SRC.pY527 and ZAP70. In the case of S100A4, which is highly expressed in both clusters compared to normal bone marrow, inhibitors (e.g. niclosamide) are under clinical trial for solid malignancies.
CONCLUSION: Our proteomic-based strategy identified 2 unique protein profiles in AMoL patients: one that responded well to AraC and another that responded poorly either with AraC or HMA. Protein signatures were prognostic in UV/MV analyses, suggesting that stratification via proteomics could potentially guide therapy. We further propose that targeting highly active proteins (e.g. S100A4) with already existing drugs might improve AMoL outcomes of patients with poor responses to current therapies.
Disclosures: No relevant conflicts of interest to declare.