Background. Acute Myeloid Leukemia is a heterogeneous disease, characterized by the uncontrolled proliferation of hematopoietic precursor cells. AML accounts for 15% of acute leukemias in children, with a cure rate of 55-60% as overall survival (Pession A Blood 2013). Complete remission (CR) is achieved in 80-85% of children with AML. Current protocol defines CR by morphological evaluation (blasts <5%) and cytofluorimetric analyses (blasts <0,1%), after two cycles of high-dose chemotherapy. Around 10-15% of children and 30% of adults do not achieve remission and are defined as Primary Induction Failure (PIF). The majority of these cases are addressed to other form of high-dose chemotherapy blocks, experimental therapies or directly to hematopoietic stem cell transplantation (HSCT) as salvage treatment, showing a very poor prognosis (Quarello P et al. British Journal of Haematology 2015). Strong efforts to generate biological data in order to define this subtype of AML are needed for early identification and for target therapy design.Aim. Our study aims to characterize a specific gene-expression profile in children with PIF-AML in order to early detect peculiar genetic aberrations and to identify potential targets to new compounds.Materials and Methods. We analyzed diagnostic samples from 4 children with PIF-AML, diagnosed and treated in our Center, from 2013 to 2016. All cases were female and 3 had more than 10 years of age. One presented with a monosomy of chromosome 11; two with complex karyotype (one with NUP98-NSD1 rearrangement) and one with no cytogenetic and molecular markers. After two cycles of Induction phase, one case showed complete blast regeneration (>60%); in the other three cases we detected, by flow-cytometry, a blast rate between 10-20%. In the attempt to compare these cases with those presenting with very high-risk (VHR) features but achieving CR after induction (Primary Induction Responder, PIR), we included in our analyses 8 children with VHR-PIR-AML, diagnosed and treated in our Center: 2 with FLT3-ITD and NUP98-NSD1 fusion genes; 2 with FLT3-ITD and complex karyotype; 2 with DEK-CAN fusion genes (1 with FLT3-ITD); 2 with NUP98 rearrangements and complex karyotype. We performed whole transcriptome analyses using Human Transcriptome Array 2.0 (Affymetrix). Then we compared our analyses with a recently published data set, including a large population of children with AML (Bisio V Leukemia 2017). Based on transcriptomic findings, we determined the expression of specific genes, by Real-time PCR using gene-specific probes (UPL probes by Roche Merck, Germany), applying a QuantStudio7 Flex System technology (Applied Biosystem, California, USA) and the relative quantification method (the comparative 2-ΔΔCt method). We applied a t-student test, for statistical analyses among different subgroups.Results. A direct comparison of PIF-AML vs PIR-AML subgroups revealed 919 protein-coding and 8 pri-miRNA overexpressed genes (fold-change >3, FDR adjusted p value <0.05) and 18 protein coding and 43 pri-miRNA down-regulated genes (fold-change >-3, FDR adjusted p value <0.05). Marked differences were observed in cancer-related genes such as KIT, PTEN, FLT3 and POU4F1, and in cell-cycle related genes such as cyclin transcripts (CNNG1, CCNA1, CCND2, CCNJ, CCNC, CCNY). Moreover some genes that discriminate PIF from PIR subgroups are also differentially expressed between cases with NUP98-negative and NUP98-positive AML (PDGFD, CCNA1). The latter group is considered as a VHR-AML subtype. Based on these findings, we selected several genes including HOX-B3, HOX-B5, CEBP-D, CCNA1, PDGF-D and performed a Real-Time PCR with specific-gene probes. Comparing gene expression data of our 4 PIF-AML cases with 8 VHR-PIR-AML cases, we observed a strong statistically significant different expression in CCNA1 (p<0.0001) and PDGF-D (p<0.0001).Conclusions. Our findings defined a characteristic profile of PIF-AML, highlighting the overexpression of cell-cycle related genes (CCNG1, CCNA1) and several potential targets (KIT, FLT3, PDGF-D) to be considered for an alternative treatment in this very poor-prognostic subgroup of patients. In conclusion our results suggest that PIF-AML samples bear a specific gene expression profile, not overlapping with those of known molecular and karyotype AML subgroups. However, a larger population study is needed to confirm our preliminary data. DisclosuresNo relevant conflicts of interest to declare.