Abstract INTRODUCTION: Current dogma is that Acute Myelogenous Leukemia (AML) arises from combinations of cytogenetic events and mutations in genes frequently mutated (GFM), and current classification schema (ELN, WHO) are based on these. The mutation (MUT) frequency of GFM in AML varies from common (15%, FLT3, NPM1, DNMT3a, NRAS), to the uncommon (5-10%, TET2, IDH1 and 2, TP53, PTPN11), to the rare (1-5%, STAG2, EZH2, RAD21, GATA2, JAK2, KIT, MECOM, WT1). Most gain of function (GOF, underlined) mutations affect functional regulation without perturbing function. Most loss of function (LOF, bold italics) events cause absent or ineffective protein. We hypothesized that either over or under expression of the Wild-Type (WT) proteins from GFM would result in GOF/LOF ‘mutant-like’ phenotypes, with similar prognosis. This would greatly affect classification and would increase the number of cases that would benefit from therapy targeted at the GFM. Methods: We measured the levels of 438 proteins using Reverse Phase Protein Arrays (RPPA) in 805 fresh, untreated 95% blast-enriched AML samples. Protein expression was normalized against that of normal bone marrow-derived, non-G-CSF treated, CD34+ cells. Treatment data was known for 691 cases, with the mutation data availability varying depending on the GFM. The prognostic impact of each of the 17 GFM-related proteins listed above was assessed by splitting into quantiles (median, tertiles, etc.) and GFM proteins that were significantly prognostic for overall survival (OS) within their WT population were selected. Expression of each prognostic protein was categorized as High or Low based on this. Unbiased hierarchical clustering using these levels was used to recognize and optimize overall expression clustering. Results: Of the 17 GFM-related proteins assessed, 14 had significant OS prognostic impact (all but GATA2, TET2 and TP53). For GFMs causing LOF, Low protein levels were adverse for DNMT3a, EZH2 and WT1 as predicted, but were favorable for NPM1, RAD21 and STAG2. For GFMs that cause GOF, High expression was adverse for FLT3, MECOM, PTPN11 and NRAS, but was favorable for IDH1 and 2, JAK2, KIT. Importantly for 9 of 14 proteins the prognosis varied according to therapy: cases with Low DNMT3a, IDH1 and 2, EZH2, and WT1 or High NRAS performed poorly with AraC-based intensive chemotherapy (IC) compared to Hypomethylating Agent plus Venetoclax (VH), while High JAK2 and KIT or Low PTPN11 responded better with IC compared to VH. Clustering of the 14 GFM-related proteins revealed 9 unique proteomic signatures (C1-C9) and distinct outcomes (5yrs OS: 47-11% from C1 to C9, p0.001). A similar pattern was observed for Remission Duration. Overall, clusters had few biases in clinical and molecular features, but those with worse prognosis (C7-C9) had more adverse features (e.g. older, unfav. cyto., TP53 MUT). For the CoxPH models, we merged clusters by prognosis: C1-C2 (Gp1), C2-C5 (Gp2), C6-C9 (Gp3). In UV analysis for OS, all groups (Gp1-Gp3) were prognostic, along with other clinical, cytogenetic and molecular features (age, cyto. risk, 2nd AML, FLT3 and NPM1 MUT, etc.). In the MV analysis, Gp1 and 3 retained significance, as did age, cyto. risk and 2nd AML. Interestingly, C2, with high OS, had a high frequency of PTPN11 (20%) and RAS MUT (42%), but low levels of these proteins. In contrast C5-C7 patients had high levels of RAS and PTPN11 and the worst outcomes, supporting that protein level, and not the mutations have more prognostic impact. Conclusion: Levels of 14 of 17 GFM-related proteins were prognostic in AML, independent of mutation status, suggesting that under or overexpression of these proteins are replicating the biological consequences of the GOF/LOF mutations. Moreover, for 9 proteins the prognosis was different depending on the therapy (IC vs VH). Furthermore, GFM-related proteins formed recurrent protein expression signatures that were prognostic for OS and remission duration, independent of mutations or cytogenetic events. Importantly, the frequency of protein patterns with adverse prognosis far exceeded that of the corresponding mutation, suggesting that therapies directed against GFM-related proteins could be used in a larger proportion of the AML population. Proteomic testing using ELISA technology could rapidly classify patients in the clinic. As outcomes differed markedly for IC and VH therapy, such triaging might substantially improve outcome through therapy recommendation.
Magalhães et al. (Mon,) studied this question.