Hypomethylating agents (HMA) azacitidine (AZA) and decitabine are approved low-intensity frontline therapy for AML pts, but patient selection between intensive versus low-intensity therapy remains mater of debate. To date, description of biomarker of response to HMA has brought few convincing findings. Most studies have focused on somatic mutations in candidate epigenetic genes such as DNMT3A, TET2, and IDH1/2 or tumor suppressor TP53 and led to controversies with none of this gene currently used to guide therapy decision in clinical practice. We performed an hypothesis-free, innovative, exome sequencing study to identify predictive biomarkers of outcome of AML pts treated with AZA.

From Jan 2007 to Dec 2016, 279 newly diagnosed non-M3 AML pts were consecutively treated with AZA in the regional cancer network ONCOMIP. The median age was 76 yrs (45-93 yrs). Median follow-up was 5.5 yrs. AML was secondary of MDS, MPN or therapy related in 141 pts (50.5%). Karyotype was adverse in 135 pts (49.1%). Median WBC count was 2.7 G/L. Overall, 54 pts obtained CR/CRi (19,4%) and median OS was 10.6 months. Among the 279 pts, we identified a discovery cohort of 49 pts with extreme phenotype consisting in 26 pts achieving CR/CRi versus 23 pts in failure despite at least 3 cycles of AZA, and a validation cohort of 175 pts with baseline DNA sample and evaluable for response. Adverse karyotypes were more frequent in the validation set (30% vs 53% respectively, p=.004), which was likely due to an enrichment of responders pts in the discovery cohort.

For exome sequencing of the discovery cohort, libraries were captured from baseline DNA BM samples with Sureselect all-exome V4 kit (Agilent) followed by paired-end, 150bp reads sequencing on a NextSeq sequencer (Illumina). Sequences were aligned with Nextgene and non-synonymous SNV were filtered according to 3 different patterns defined on gene function (nucleotide analogs metabolism/COSMIC census genes/AML genes), read depth, variant allele frequency and recurrence. An elastic-net regularization combining LASSO and Ridge penalties simultaneously (Zou & Hastie, 2005) was used to identify association between variants and clinical response. Using a resampling approach, bootstrap stability selection (BSS) were computed for each variant and only those with a BSS≥75% were selected. The exon capture baits targeted 98% of genes in the Consensus CDS database. A mean coverage depth of 85× per sample was achieved, with 92.5% of targets covered at ≥ 20× depth. From the 3 gene/SNV filtering patterns, bootstrapping of the data obtained with elastic-net regularization identified 4 candidate SNV in C11orf80, DZIP, ZNF543 and MECOM genes. Inspection of the exome-seq data for MECOM revealed that 11 of the 26 individuals in the responders sample had an heterozygous missense variant in the coding sequence of MECOM at position 169098992 (rs7622799; encoding p.Pro120Ser; MAF = 0.13 in the Exome Aggregation Consortium[ExAC]).

We screened the 4 candidates SNV by targeted sequencing in the validation set and identified MECOM rs7622799, in 34/175 pts (19.4%). Response rates did not differ between pts with rs7622799 (8.8% CR/CRi) and without rs7622799 (9.2% CR/CRi, p=1.00). We then assessed whether genotype rs7622799 predicted OS in time dependent analysis using a Piecewise Cox model and demonstrated that pts with rs7622799 had improved OS from 3 months after diagnosis compared with non-rs7622799 (HR 0.56 95%CI[0.34;0.93]; p=0.024). The 3 other SNVs were not associated with patient outcome.

We used an extreme phenotype study design, to discover that the missense variant rs7622799 of MECOM is correlated in time dependent analysis with the overall survival of older AML pts treated with AZA. MECOM is a complex locus gene which encodes for several isoforms namely EVI1, a very aggressive oncogene when overexpressed in AML, and MDS1-EVI1 (or PRDM3), which has epigenetic function as histone methyltransferase activity. Both isoforms directly interact with epigenetic modifiers (HDAC, DNMT3A). Further analyses are needed to best understand the functional impact of rs7622799. Although our findings could indicate that rs7622799 is associated with outcome of older AML pts in the setting of AZA, it remains unknown if it could be used to stratify pts between intensive and low-intensity therapy. Targeted sequencing of this SNV in a control cohort treated with intensive chemotherapy is ongoing


Fornecker:Takeda: Honoraria; Servier: Honoraria.

Author notes


Asterisk with author names denotes non-ASH members.