We have previously shown that the protein expression profile and especially the activation state of apoptosis and STP proteins are highly prognostic in AML. The development of RPPA permits a more comprehensive analysis of protein expression and phosphorylation (phos) patterns in the STP and apoptotic cascades. We have generated a screening array using protein derived from the leukemia enriched fraction of 94 primary AML samples (with 21 concurrent blood and marrow specimens), comprised equally of patients that were primary refractory (PrimRef), or that achieved complete remission that was continuous (CCR) or which relapsed after 6 to 24 months (REL). These slides have been probed with 22 total and 15 phos-specific antibodies (ABs) against apoptosis and STP proteins. Spot intensities were quantified using MicroVigene and the data for each protein standardized data by subtracting the mean expression levels across samples and dividing by the standard deviation. Unsupervised hierarchical clustering analysis on the proteins was performed. We found one small cluster of 5 proteins: Cyclin D1, β-catenin, phos-NMP, p53, and AMPK. The remaining proteins formed another cluster, with a subset of 8 proteins somewhat further separated. In several cases (mTOR, JNK, and PTEN) the total protein amounts clustered as nearest neighbors to the phos version of the same protein. Using perturbation bootstrap resampling to assess the significance of clusters we found 4 reproducible clusters and the structure suggests additional clusters in this data set. A Fisher exact test showed that these were significantly associated with response (p = 0.03) and with the prognostic category defined by cytogenetics (p = 0.04) but was not associated with the source (blood vs. marrow) of the sample (p = 0.67). Ten proteins were differentially expressed between cytogenetically defined prognostic groups: total and phospho AKT, PTEN and JNK, p-mTOR, p-STAT3, pp38 and PS6.p24.44 (all with P<0.02). No protein was prognostic for response across the entire population but several were prognostic for response and survival within specific clusters. Clustering of ratios of phos-/total AB revealed 2 patterns: one with higher phos- of kinases (“kinase high”); another with low kinase expression (“kinase low”) but higher phos- of apoptosis regulating proteins. In the “kinase high” cluster single or concurrent increased activation of pAkt308, pAkt473, pPKCa, p-mTor (2448), pErk1,2 (42/44), p-p38 over their corresponding total protein was associated with worse outcome (~15% CCR rate). Those with increased Bcl2/Bax and pJNK/JNK levels fared worse with only 10% CCR. A favorable outcome was associated with concurrent high pAkt308/Akt and pGSK3/GSK3 levels (50% CCR) or being in the “kinase low” group while having high pStat3(705)/Stat3 levels (58% CCRs). The identified clusters and their correlation with cytogenetic group and outcome mainly driven by the activated, phospho protein forms, strongly supports the idea that the activation state of STP and apoptotic regulators determines biological behavior and clinical outcome. Distinct pathway activation patterns can classify AML, provide prognostic guidance and may serve to triage patients to emerging targeted therapies aimed at these pathways.

Author notes

Corresponding author