Oncoproteomic profiling of AML: moving beyond genomics

Much of what is known about protein-signaling networks in cancer, or ‘oncoproteomics,’ has been indirectly derived from transcriptomic analyses[1],[2]. However, RNA regulation precludes a one-to-one correlation of mRNA abundance to protein abundance or activity. A corollary of this is that evaluation of RNA by itself is insufficient to fully appreciate pathogenic cellular signaling within the tumor ecosystem. Global proteomics and phosphoproteomics have emerged as powerful unbiased methodologies for detailing fundamental signaling networks of cancer cells and perturbations that sustain resistance against targeted therapies, contributing to the discovery of new therapeutic targets[3].

Similar to other cancers, the utility of mass spectrometry-based technologies has augmented our ability to categorize the underlying heterogeneity in acute myeloid leukemia (AML) – expanding our capacity to classify AML beyond genomic features alone. Efforts over the past decade have resulted in the creation of new datasets that have begun to characterize the AML proteome and phosphoproteome. A subset of these studies have been exploratory in nature [4–9] – leading to the generation of new hypotheses, while others have focused on examining particular aspects of a disease state (e.g. drug resistance) to identify new biomarkers¹⁰⁻¹⁷. These data provide a rich resource for further investigations aimed at mapping the ‘post-genomic’ landscape of AML (Table 1). Within this editorial, we discuss how integration and aggregation of such data with our current understanding of the AML genome and transcriptome holds the promise of refining our classification of leukemia cells – the genotype and phenotype – and yielding mechanistic insights that can inform the generation of improved therapeutic combinations.

Table 1.

Summary of recent publications that used global proteomics and phosphoproteomics to investigate AML in preclinical and clinical studies. Full datasets (with accession numbers) and information of patient cohorts are available on-line and can be accessed via the provided PMID numbers. HSPCs, hematopoietic stem and progenitor cells; MNCs, mononuclear cells; LSCs, leukemic stem cells.

Authors	Study Title	Sample Information	PMID	Publication Year
Casado P., et al.	Kinase-substrate enrichment analysis provides insights into the heterogeneity of signaling pathway activation in leukemia cells	AML cell lines; Healthy donors HSPCs (CD34+; n = 5); Primary AML MNCs (n = 20)	23532336	2013
Schaab C., et al.	Global phosphoproteome analysis of human bone marrow reveals predictive phosphorylation markers for the treatment of acute myeloid leukemia with quizartinib	Primary AML blasts (CD34+, n = 21)	24247654	2014
Casado P., et al.	Proteomic and genomic integration identifies kinase and differentiation determinants of kinase inhibitor sensitivity in leukemia cells	Primary AML MNCs (n = 30)	29626197	2018
Hosseini M.M., et al.	Inhibition of interleukin-1 receptor-associated kinase-1 is a therapeutic strategy for acute myeloid leukemia subtypes	Primary AML MNCs (n = 9)	29743719	2018
de Boer B., et al.	Prospective Isolation and Characterization of Genetically and Functionally Distinct AML Subclones	Healthy donors HSPCs (CD34+; n = 6); Primary AML blasts (CD34+, n = 42)	30245083	2018
Alanazi B., et al.	Integrated nuclear proteomics and transcriptomics identifies S100A4 as a therapeutic target in acute myeloid leukemia	AML cell lines; Healthy donors HSPCs (CD34+; n = 5); Primary AML blasts (CD34+, n = 15)	31611628	2019
Raffel S., et al.	Quantitative proteomics reveals specific metabolic features of acute myeloid leukemia stem cells	Healthy donor HSPCs (CD34±, CD38±; n = 9); Primary AML LSCs or blasts (CD34±, CD38±; n = 14)	32556243	2020
Murray H.C., et al.	Quantitative phosphoproteomics uncovers synergy between DNA-PK and FLT3 inhibitors in acute myeloid leukemia	AML cell lines; Primary AML blasts (n = 7)	33067575	2020
Hernandez-Valladares M., et al.	Biological characteristics of aging in human acute myeloid leukemia cells: the possible importance of aldehyde dehydrogenase, the cytoskeleton and altered transcriptional regulation	Primary AML blasts (CD34+, n = 33)	33349623	2020
Schoof E.M., et al.	Quantitative single-cell proteomics as a tool to characterize cellular hierarchies	AML cell lines	34099695	2021
Joshi S.K., et al.	The AML microenvironment catalyzes a stepwise evolution to gilteritinib resistance	Gilteritinib sensitive/resistant cell lines (n = 39); Serial primary AML blasts (CD33+/34+; n = 11)	34171263	2021
Jayavelu A.K., et al.	The proteogenomic subtypes of acute myeloid leukemia	AML cell lines; Healthy donors (CD34+; n = 13); Primary AML blasts (CD45−; n = 252)	35245447	2022
Janssen M., et al.	Venetoclax synergizes with Gilteritinib in FLT3 wildtype high-risk Acute Myeloid Leukemia by suppressing MCL-1	Primary AML MNCs (n = 6)	35857899	2022
Kramer M.H., et al.	Proteomic and phosphoproteomic landscapes of acute myeloid leukemia	Healthy donors (CD34+; n = 6); Primary AML MNCs (n = 44)	35895896	2022
Gosline S.J.C., et al.	Proteomic and phosphoproteomic measurements enhance ability to predict ex vivo drug response in AML	Primary AML MNCs (n = 38)	35896960	2022
Emdal K.B., et al.	Phosphoproteomics of primary AML patient samples reveals rationale for AKT combination therapy and p53 context to overcome selinexor resistance	Selinexor sensitive/resistant AML cell lines (n = 4); Primary AML blasts (n = 20)	35947955	2022
Koschade S.E., et al.	Translatome proteomics identifies autophagy as a resistance mechanism to on-target FLT3 inhibitors in acute myeloid leukemia	AML cell lines (n = 2); Primary AML MNCs (n = 11)	35999260	2022

Casado et al. profiled the proteome and phosphoproteome of primary AML cells from 30 patients and the aggregation of these datasets with corresponding genomic, immunopheno-typic, and pharmacologic analyses was among the first studies to infer that cell differentiation state influences kinase signaling changes and drug sensitivity profiles[4]. The authors also showed that FLT3 mutation status alone was insufficient to predict response to the FDA-approved inhibitor midostaurin and that increased activation of PKCδ and GSK3A in AML cells, as revealed by phosphoproteomics, correlated with midostaurin response[4]. Early attempts to integrate proteomic, genomic, and/or transcriptomic datasets have expanded our ability to categorize the small sub-populations of leukemic stem cells (LSCs) that govern the underlying heterogeneity and complexity of AML[5] and our understanding of the nuclear proteome in the pathogenesis of AML[6].

More recently, Jayavelu et al. identified five AML subtypes with distinct biological features via proteomic characterization of 252 AML patient samples[7]. Integration of these data with corresponding genomic, cytogenetic, and transcriptomic analyses revealed that the mito-AML subtype was only captured within proteomic profiling. This subtype of AML is characterized by high expression of mitochondrial proteins involved with cellular oxidative phosphorylation and confers poor prognosis. While many effectors in AML remain ‘undruggable,’ the work of Jatavelu et al. reveals that proteomics-based technologies have the propensity to identify new members within a signaling network that can be targeted – effectors that otherwise would not be considered from a traditional pharmacologic standpoint but warrant investigation. Specifically, the authors discovered a metabolic vulnerability through proteomics and phosphoproteomics and show that pharmacological inhibition of effector proteins within the mitochondrial network may eradicate AML cells, underscoring that treatment of AML goes beyond traditional regimens established merely by genomic aberrations or transcriptomic changes. Jatavelu et al. are not alone in showing that metabolic changes are regulated post-transcriptionally and require examination of the proteome. Raffel et al. demonstrated that LSCs are dependent upon amino acid metabolism[9]. More broadly, both studies highlight the strength offered by proteomic technologies to survey a large range of effector proteins with great specificity as opposed to a single effector studied through traditional methodologies, which are often hindered by target limitations and cross-reactivity.

In line with Jayavelu et al., Kramer et al. developed a proteome and phosphoproteome database using a cohort of 44 AML patients[8]. While their dataset recapitulated many of the well-known features of AML, it refined our conceptualization of the regulatory processes sustaining AML including the importance of post-transcriptional protein modifications.

The authors showed that presence of IDH1/2 mutations resulted in elevated levels of 2-oxoglutarate-dependent histone demethylases KDM4 A/B/C, despite no observed changes at the mRNA level. They further reported that mutant NPM1 is associated with an increased network of nuclear importins, not evident at the mRNA level. While genomic aberrations are merely suggestive of the underlying cellular changes, in the aforementioned examples, proteomics resolved the biological underpinnings concordant with the observed disease phenotype in patients. Use of proteomics also enabled the identification of cell surface markers, CD180 and MRC1/CD206, expressed on AML blasts of many patients but not on healthy CD34 stem cells. Orthogonal validation and integration of the proteomic and phosphoproteomic findings strengthened their overall conclusions, providing a valuable dataset for the AML research community.

Alongside these advances has been the use of proteomics and phosphoproteomics to chart the evolution of drug resistance in the setting of AML [10–14]. Most studies, including our own, have focused on drug resistance within a sub-type of AML that is driven by the activity of receptor tyrosine kinase FLT3, which is altered in over 30% of cases [10–12,15,16]. Prior studies have primarily described genetic mechanisms of resistance to FLT3 inhibitors[17]. A detailed understanding of the proteome and phosphoproteome in this setting offers a unique opportunity to discover and characterize non-genetic mechanisms of resistance that are equally important to consider within the tumor ecosystem.

As demonstrated by Koschade et al., the combination of proteomics and phosphoproteomics led to the identification of the autophagy network – AKT/mTORC1/ULK1/ATG3 – as a primary node to FLT3 inhibitor resistance[12]. The authors employed a recently developed proteomics technology, termed translatome proteomics, which measures changes in the nascent cellular proteome and paired this approach with phosphoproteomics. The aggregate of proteomic analyses performed on cell lines, patient-derived xenografts, and primary AML cells ex vivo identified autophagy as a key pathway modulated upon FLT3 inhibition. Pharmacological or genetic inhibition of autophagy restored FLT3 inhibitor sensitivity and led to extended overall survival. It is important to note that the essential finding of autophagy hinges on the changes detected in the cellular proteome – changes that are unlikely to be detected with RNA-seq and which require the depth and quantitative characterization offered by translatome proteomics. This study lends additional support to the notion that transcriptomics is not a viable surrogate for investigating non-genetic mechanisms of resistance that manifest at the protein level.

Recognizing that growth factors present within the bone marrow microenvironment[17] could promote the survival of residual leukemic cells following treatment with FLT3 inhiitors, our laboratory developed a biphasic cell culture-based model of resistance and characterized this model with multiple orthogonal-omics approaches[11] (Figure 1A). Proteomic and phospho-proteomic analyses of our resistant cell cultures were suggestive of a stepwise, dynamic process of FLT3 inhibitor resistance where initial microenvironmental ligand support led to an early phase of resistance characterized by activation of alternative signaling pathways (AURKB, CDC7) and changes in cell state (slower cell cycling and metabolic reprogramming) that progressed to a late resistance phase coincident with the acquisition of mutations in the RAS pathway[16] following ligand withdrawal. A hallmark of early resistance to FLT3 inhibitors was a slowing of the cell cycle mediated by AURKB, a key finding that was only identified via proteomics and phosphoproteomics. Further corroboration of this ‘discovery’ dataset using additional genetic, pharmacologic, and metabolomic analyses led us to refine our hypothesis and develop a targeted proteomics panel to validate our model in cells from patients undergoing treatment with a FLT3 inhibitor. Residual cells from these patients, displayed a reduced cell cycle and a transition to lipid metabolism. Such cells also became dependent on AURKB signaling and were exquisitely sensitive to the combination of AURKB and FLT3 inhibitors ex vivo, confirming our model as predictive of the evolving biology of residual cells. Taken together, our study displayed the dual advantage offered by employing global proteomics and phosphoproteomics – hypothesis-generation and the ability to prioritize proteomic signatures capable of predicting drug response in patient samples by building targeted panels (Figure 1A). Importantly, the development of a targeted panel serves both as an opportunity to distill the growing omics data and enables one to make the most of clinical samples that are oftened limited in quantity.

Figure 1. Proteomics & phosphoproteomics can unveil the dynamic nature of AML.

a. An integrative approach that combined genomic, metabolomic, phosphoproteomic, proteomic analyses with small-molecule inhibitor and genome-wide CRISPR screens revealed a stepwise process for FLT3 inhibitor resistance. This discovery dataset led to the creation of a targeted proteomics panel that was used to investigate FLT3 inhibitor resistance in a subset of ex vivo serial patient samples. Our cell culture model recapitulated the underlying biology of patients resistant to FLT3 inhibitors. b. The integration and aggregation of the amassing - omics data with ex vivo small-molecule inhibitor screening holds the potential to individualize clinical management for patients with cancer.

In addition to examining signaling networks that sustain leukemic growth or confer resistance, proteomic and phosphoproteomic methodologies have also been extended to create and validate a framework that can predict drug response to targeted inhibitors in primary AML patient samples. Casado et al. were among the first to develop a computational approach that linked proteomic signatures with drug response to targeted inhibitors by using primary AML cells from 20 patients[18]. Similarly, we recently evaluated a pilot set of 38 AML patient samples possessing genomic, proteomic, and functional drug response data[19]. Initial analyses showed that mRNA abundance and protein levels were weakly correlated in patient samples, underscoring potential disconnects between RNA expression and protein biology. Combined regression analyses and cross-validation were used to determine the best signature for each drug or drug family, which ultimately were interpreted using data from external sources and validated in cell culture models. In general, linear regression modeling, using both RNA and proteomic data, produced the best overall predictions of drug response in patient samples, although in some cases proteomic data alone was sufficient to produce superior results. The selection of top performing data features could vary based upon the drug or drug family, underscoring the need to use a broad strategy to select the best features to evaluate for each drug. Incorporation of tools such as OmniIntegrator can enable a better understanding of what causes drug resistance in a subset of patients, and potentially assist in understanding the effects of drug combinations, which are becoming increasingly common in the clinical management of AML.

Preponderant genomic sequencing and clinical evidence have revealed the remarkable heterogeneity that underpins leukemia [20,21]. Such efforts have brought to the forefront new oncogenic alterations and genetic mechanisms of drug resistance. More recently, advances in proteomic and phosphoproteomic technologies (e.g. single cell omics) have begun to unmask the plethora of interconnected signaling networks that sustain cancer growth or mediate non-genetic drug resistance [3,22]. However, the use of proteomics and phosphoproteomics to interrogate signaling pathways that enable leukemia progression or resistance is largely in its infancy. As a first step, efforts to harmonize the existing AML proteomic datasets that we discuss above may pave the way for future investigations. We believe that the comprehensive cataloging of such data with genomics, transcriptomics, metabolomics, and ex vivo functional inhibitor sensitivity may potentially expand our taxonomy of leukemia cells and ability to deliver informed personalized therapy [23] (Figure 1B). Beyond the leukemic cell, aggregate analyses could be also considered to further characterize the ‘soil’ or microenvironment that sustains the cancer cell, providing a wholistic view of leukemogenesis and mechanisms governing drug resistance.