Proteomics in Acute Myeloid Leukemia: Current Applications in Precision Medicine and Targeted Immunotherapy

Abstract

Acute myeloid leukemia (AML) is a heterogeneous and aggressive malignancy with limited therapeutic options and high relapse rates. Despite advances in genomic profiling, many genetic aberrations remain untargetable, and current risk stratification models often fail to predict treatment responses. Proteomics offers a complementary approach by directly measuring protein abundance, post-translational modifications, and protein–protein interactions, providing mechanistic insights into drug resistance, disease progression, and therapeutic vulnerabilities. In this review, we explore the emerging role of proteomics in AML, focusing on its application in biomarker discovery, prediction of drug responses, and identification of novel therapeutic targets. Special attention is given to antigen discovery for immunotherapy, where surface and immunopeptidomics enable the identification of AML-specific antigens and neoepitopes. These insights are critical for the development of antigen-targeted therapies, including chimeric antigen receptor (CAR) and T cell receptor (TCR)-based immunotherapies. Integrating proteomics into a multiomics framework could provide actionable insights for guiding precision medicine and improving AML outcomes.

Introduction

Despite therapeutic advances, including the introduction of FMS-like tyrosine kinase 3 (FLT3) inhibitors and the BCL-2 inhibitor venetoclax, the prognosis of acute myeloid leukemia (AML) remains poor, with a 5-year overall survival of only 30%.^1-4

The backbone treatment for young and fit patients consists of intensive chemotherapy with cytarabine and anthracycline, followed by consolidation chemotherapy or allogeneic stem cell transplantation. ² In elderly patients or those with comorbidities, less intensive treatment with azacitidine with venetoclax or ivosidenib is currently the standard of care (Figure 1).^4,5 Relapse after treatment remains frequent, underscoring the limitations of current treatment strategies and the urgent need for novel approaches.^3-7

Figure 1.

Current AML treatment and potential advances by proteomics

Although previous genomic and transcriptomic studies have identified numerous genetic aberrations in AML, only a few are currently targeted, including FLT3, isocitrate dehydrogenase-1 (IDH1), and nucleophosmin-1 (NPM1).^5,6,8-10 Even when a targeted treatment is available, the response can be unpredictable, posing a challenge for precision medicine. ¹¹ As proteins are the main effectors of cellular function and mechanisms of drug resistance can occur at the protein level, protein-centric approaches are essential in the search for new targeted therapies.^12,13

AML is characterized by profound heterogeneity. ¹⁴ Current diagnostic classification and risk stratification systems rely primarily on cytogenetic and molecular profiling; however, these approaches often fail to fully capture the heterogeneity of AML. ² Crucial determinants of AML pathobiology often manifest at the protein level, underscoring the limitations of genomics and transcriptomics. ¹⁵ Furthermore, heterogeneity exists not only across different genetic and molecular subtypes of AML, but also within individual patients, where distinct subclones may coexist. ¹⁴ At the protein level, this diversity translates into the activation of distinct signaling pathways and variable antigen expression. ¹⁶

Leukemic stem cells (LSCs) are a cell subpopulation of interest. LSCs are a self-renewing, therapy-resistant subpopulation of AML cells that drive disease persistence and relapse. ¹⁷ Their eradication is essential to achieve durable remission and long-term survival.^17-22 Proteomics, including advances in single-cell proteomics, could aid the mapping of the clonal hierarchy in AML and the characterization of distinct subclones with unique properties and drug responses, such as LSCs (Figure 1).

The tumor antigen landscape of AML cells and LSCs remains poorly characterized, which hampers the development of antigen-targeted immunotherapies such as chimeric antigen receptor (CAR)- or T cell receptor (TCR)-based therapies.^16,18 Transcriptomic profiling alone is often insufficient for predicting (surface) antigen expression. Indeed, mRNA abundance does not always correlate with protein expression levels. ¹¹ Furthermore, post-translational modifications such as glycosylation and phosphorylation can alter the antigen, potentially generating AML-specific antigens that are invisible in transcriptomic analysis. ²³ Surfaceomics and immunopeptidomics could aid in the search for suitable antigen targets for cellular therapy in AML (Figure 1).

These considerations highlight the need for proteomics as an approach complementary to genomics and transcriptomics (Figure 2). Proteomics can assist in identifying new biomarkers and novel therapeutic targets to gain insight into the mechanisms of treatment response or resistance and to comprehensively map the AML surface and immunopeptidome, guiding the development of effective antigen-targeted immunotherapies (Figure 2).^12,15

Figure 2.

Application of proteomics as part of a multi-omics strategy in AML. Abbreviations: AML: acute myeloid leukemia, LSCs: leukemic stem cells

In this review, we highlight the emerging role of proteomics in AML and discuss how it can address key unmet needs. After outlining proteomic methodologies, including mass spectrometry (MS) and single-cell proteomics, we examined their application in biomarker discovery, drug response prediction, and antigen discovery. Our review differs from previous work^15,23-26 by focusing on the proteomics research most relevant to current clinical practice, with particular emphasis on the prediction of treatment response, immunotherapy-related target discovery, and immunopeptidomics, areas that remain challenging yet essential for advancing AML therapy.

Proteomic Methodologies

Proteomics and the Role of Mass Spectrometry as a High-Throughput Technique

Mass spectrometry (MS)-based proteomics enables the direct large-scale analysis of protein expression in cells and tissues, including the surface and immunopeptidome (Figure 3).^8,27-30 Over the past decade, rapid advances in high-throughput MS technology have substantially increased the depth and accuracy of proteomic profiling. ³¹ Shotgun proteomics identifies thousands of proteins in complex mixtures. Targeted proteomics can be used to validate candidate markers detected by shotgun proteomics. ³² To achieve this large-scale identification, different acquisition strategies have been developed to achieve large-scale identification, each with distinct strengths and limitations. In data-dependent acquisition (DDA), the most abundant precursor ions detected in the first MS scan (MS1) are selected for fragmentation and subsequent MS analysis (MS2). While widely used, this approach inherently favors high-abundance peptides and may overlook peptides with lower abundance.^33,34 In contrast, data-independent acquisition (DIA) systematically fragments all ions within predefined mass-to-charge (m/z) windows, reducing abundance-driven bias. ³³ Although DIA generates complex spectra and requires sophisticated computational workflows, recent advances in bioinformatics have markedly improved data processing, thereby establishing DIA as a robust and comparatively unbiased acquisition strategy. ³³

Figure 3.

Proteomic methodologies and applications in AML. Abbreviations: LSCs: leukemic stem cells; CAR-T: chimeric antigen receptor T-cell therapy; TCR-T: T cell receptor T cell therapy; HLA: human leukocyte antigen

A proteomic workflow can be performed using label-free or label-based methods. Label-free methods have the advantage of higher proteome coverage and are currently nearly as accurate as label-based methods. ²⁴ Among labeling approaches, methods such as isobaric tagging with tandem mass tags (TMT) enable the multiplexed identification and relative quantification of thousands of proteins across multiple samples in a single experiment. ³⁵ However, a key limitation of TMT-based workflows is their reliance on DDA, which may introduce ratio compression and restrict the detection of low-abundance peptides.

Phosphoproteomics, which focuses on the identification of post-translational modifications, such as protein phosphorylation, employs selective phosphopeptide enrichment strategies prior to MS analysis to capture phosphorylation-dependent signaling. ³⁶ Emerging applications of proteomics, including structural proteomics using cross-linking MS, now provide insights into protein conformations and protein–protein interactions. ³⁷ Similarly, surfaceomics enables the direct examination of cell surface antigens through specialized extraction and enrichment of membrane proteins. ³⁰

A particularly promising advancement is single-cell proteomics, which allows proteome profiling at the individual-cell level and offers insights into cellular heterogeneity. Because this approach is particularly relevant for studying subclones in AML, we focus on this technique in the following section.

Single-Cell Proteomics

To analyze a small subpopulation such as LSCs, techniques with single-cell resolution or purified populations in small bulk analysis are essential.^12,38-40 Indeed, analysis performed on a composition of heterogeneous cell types will result in an average protein composition of the entire population, potentially masking features specific to LSCs. ⁴¹ Single-cell proteomics may address this limitation by analyzing data at the single-cell level.

Conventional approaches for single-cell protein analyses include antibody- and tag-based technologies.^12,42 After labeling the cells with an antibody or tag, the proteins of interest can be tracked. For example, mass cytometry, also designated as cytometry by time of flight or CyTOF, combines flow cytometry with MS. By using a panel of metal ions bound to antibodies, single cells can be identified based on their metal ion profile and subsequently analyzed by MS.^12,42 However, these antibody-based techniques are limited by the availability of validated antibodies and tags and can typically only quantify tens to hundreds of proteins per experiment.

Recent advances in MS have enabled the development of MS-based single-cell proteomics. ¹² Reducing the sample preparation volume to nanodrop-based sample preparation has reduced signal-to-noise ratios and facilitated the identification of single-cell proteomes. Alternatively, fluorescence-activated cell sorting (FACS) into low-protein-binding microwell-plates with working volumes <2 µL is also used. ¹² Single-cell proteomics by MS (SCoPE-MS) was introduced as an automated and quantitative single-cell proteomics method. ⁴³ To simultaneously identify and quantify peptides, tandem mass tags (TMT) were used. Each single-cell set was augmented with a set of approximately 200 carrier cells, which enhanced peptide identification and minimized protein loss. This approach enables the quantification of thousands of proteins in a single cell. ⁴³ More recently, a proof-of-concept study using primary AML cell cultures developed a semi-high-throughput single-cell MS proteomics workflow that integrated FACS-based single-cell sorting optimized for maximal quantitative accuracy within a limited timeframe. The resulting data analysis workflow successfully detected cellular heterogeneity and cell-specific protein signatures. ⁴⁴ A remaining limitation of this approach is its reliance on DDA, which may restrict the detection of low-abundance peptides.

Limitations of Proteomics

Despite advances in MS technology and bioinformatics, proteomics has not yet been integrated into routine clinical practice.^24-26 Many promising findings remain exploratory or translational and lack the extensive validation required for their clinical implementation. Current workflows are hindered by technical limitations in sensitivity and proteomic coverage, insufficient dataset standardization, and the need for robust computational frameworks for interpreting increasingly complex data. ¹⁵ In addition, sample preparation and MS analysis remain labor-intensive and dependent on specialized equipment, contributing to high costs and limited scalability. Methodological validation of the integration of proteomic profiling in disease stratification or drug prediction must be performed in large patient cohorts and, ideally, in primary patient materials, as cell lines may differ biologically from primary AML cells. ²⁴ Drug testing of ex vivo responses of primary AML cells requires validation in vivo. Collectively, these challenges impede the routine adoption of proteomics in clinical settings.

Nevertheless, ongoing improvements in MS instrumentation, automation, and clinical workflow designs are expected to enable faster and more comprehensive protein analyses. Continued research aimed at enhancing data integration and standardizing proteomic and multiomic datasets is essential for translating proteomics into reliable tools for clinical decision-making.

Applications of Proteomics in AML

Identification of Biomarkers and Potential Therapeutic Targets

Proteomic and phosphoproteomic profiles have emerged as powerful strategies for the identification of novel biomarkers and potential therapeutic targets. Deep-scale proteomic analysis of AML samples has revealed several previously unrecognized features. ⁴⁵ For example, unsupervised clustering of proteomic profiles from 44 AML samples identified distinct subgroups that correlated with known molecular covariates. Notably, several post-transcriptionally regulated proteins were detected, such as the KDM4A/B/C family of H3K9/27/36 demethylases, which showed increased protein expression in IDH-mutated AML, despite unchanged mRNA levels. Similarly, in NPM1-mutant AML, several nuclear importins were identified as being post-transcriptionally regulated and were found to interact with the cytoplasmic mutant of NPM1, but not with the wild type. ⁴⁵ These observations illustrate how proteomics can uncover regulatory mechanisms that are not captured by transcriptomics, although their functional and therapeutic relevance remains to be fully established. ⁴⁶

Multiomics approaches integrating genomics with proteomics and metabolomics have further refined biomarker discovery. Analysis of the bone marrow supernatants from 40 AML samples revealed several proteins and metabolites associated with high-risk mutations and relapse. ⁴⁷ Among these, thrombospondin-4 showed the strongest correlation with relapse rates. The metabolites linked to relapse were enriched in lipid metabolic pathways, including steroid hormone biosynthesis, linoleic acid metabolism, and serotonergic synapses. ⁴⁷

Phosphoproteomic profiling on AML blasts of eight patients and AML cell lines revealed that p21-activated kinase (PAK) is frequently upregulated and associated with adverse prognosis. A selective PAK inhibitor was sequentially tested and was found to significantly reduce the in vitro proliferation of AML cells, suggesting that PAK inhibition is a potential target for future research. ⁴⁸

LSCs exhibit distinct metabolic profiles, with oxidative phosphorylation as the dominant pathway. ¹⁹ In this context, IDH3 isoforms and several components of the electron transport chain complex I and IV have been found to be overexpressed in LSCs. ¹⁹ Also, the LIM domain transcription factor CRIP2, another protein known to promote oxidative phosphorylation, was highly overexpressed in LSCs and was associated with chemotherapy resistance. ¹⁹ In another study, SIRT3, a regulator of fatty acid oxidation required for oxidative phosphorylation, was implicated in LSC survival. ⁴⁹

In addition, increased amino acid metabolism has been observed in LSCs as a driver of oxidative phosphorylation and survival. Interestingly, LSCs of relapsed patients demonstrated “metabolic plasticity” by upregulating fatty acid metabolism to compensate for amino acid depletion. ⁵⁰ These adaptive mechanisms underscore the complexity of targeting LSC metabolism and suggest that single-pathway inhibition may be insufficient, strengthening the need for integrative metabolic targeting strategies.

Beyond metabolism, dysregulation of transcription factors (such as core-binding factors) and nuclear proteins are known to play a role in leukemogenesis. For example, S100A4 was significantly upregulated in AML nuclei compared to healthy controls in a proteomic analysis, a finding not predicted by transcriptomic data. Interestingly, S100A4 knockdown impairs the survival of AML cell lines, underscoring its potential as a therapeutic target. However, these findings were based on in vitro models, and further studies are required to assess their relevance in vivo. ⁵¹ Another study involving quantitative phosphoproteomic profiling of primary AML samples identified higher phosphorylation of DNA-dependent protein kinase (DNA-PK) in FLT3-mutant cells. Based on these observations, they tested whether DNA-PK inhibition could enhance the activity of FLT3 inhibitors. In vitro, this combination produced stronger antileukemic effects than either agent alone. Although the mechanistic basis of this interaction remains to be fully defined, this study illustrates how phosphoproteomic profiling can reveal signaling patterns that may help guide the development of potential combination strategies for AML. ⁵²

Comparative analysis of the proteomic profiles of (pediatric) AML cells, LSCs, and normal hematopoietic stem and progenitor cells revealed several differentially expressed proteins, including HSPE1, SRSF1, and NUP210. ⁵³ Signaling through the EIF2 pathway was significantly downregulated in LSCs, compatible with protein synthesis perturbations, suggesting that LSCs adopt a low-translation and stress-tolerant state reminiscent of quiescent stem cells. These findings encourage further exploration of LSC biomarkers. ⁵³

Finally, in addition to its role in identifying novel therapeutic targets, proteomics has demonstrated value in prognostication. Large-scale proteomic profiling of peripheral blood samples from over 550 AML patients led to the development of the “leukemia inflammatory risk score”, an 8-protein signature with prognostic value. Among these eight proteins, the soluble oncostatin M receptor has emerged as an independent prognostic biomarker, demonstrating the clinical utility of proteomics to fine-tune AML risk classification beyond conventional genetic or molecular risk profiling. ⁵⁴

Prediction of Treatment Response or Resistance

Intensive chemotherapy with cytarabine and anthracycline remains the standard treatment for young patients with AML (Figure 1). ² AML in elderly patients is characterized by an increase in chemotherapy-resistant diseases. Although this resistance is partly attributable to a higher prevalence of adverse-risk genetic abnormalities, proteomic and phosphoproteomic studies have indicated that alterations at the protein level also play a significant role. Exploratory findings suggested that elevated levels of aldehyde dehydrogenase-2, altered expression of cytoskeletal proteins, and altered transcriptional regulation contribute to resistance. ⁵⁵

Venetoclax-based regimens have become the standard of care for older AML patients or those who are unfit for intensive treatment (Figure 1). ⁴ Current risk-stratification guidelines for patients receiving hypomethylating agents combined with venetoclax rely primarily on mutational profiling, with TP53, FLT3-ITD, NRAS, and KRAS being the most informative. ⁵⁶ Relapse or resistance to venetoclax-based treatments remains an unmet need. Proteomic research aimed at understanding the mechanisms underlying venetoclax resistance is therefore of considerable clinical importance. In this context, a multi-omics analysis of 210 patients from the larger ‘Beat AML’ cohort integrated multi-omics data with drug sensitivity data to build a predictive model. This study identified four proteogenomic subtypes that extend beyond traditional genetic classifications and accurately reflect the functional biology of AML. These subtypes were strongly associated with distinct drug-response patterns, enabling the development of predictive models for 46 therapeutic agents, with venetoclax being among the drugs for which treatment response could be predicted most accurately. ⁴⁶ Although promising, this model is not yet applicable in clinical practice, but could serve as a foundation for further research.

Proteomics-based profiling of 810 newly diagnosed AML patients identified 109 prognostic proteins that could be used to stratify patients into five predictive proteomic subgroups. Subsequent analysis of these prognostic proteins suggested that a subset of 14 proteins may have the potential to guide or predict the response to venetoclax-based treatment or standard chemotherapy; however, further validation is required. ⁵⁷ Another study of 252 AML samples also identified five distinct proteomic subgroups, including a “mito-AML” cluster characterized by high expression of mitochondrial proteins. This subgroup showed a poor response to standard chemotherapy but increased sensitivity to BCL-2 inhibition with venetoclax, highlighting the translational potential of proteomic profiling to guide therapy selection. ³⁴ Further exploratory proteomic studies revealed that the mitochondrial proteins electron transfer flavoprotein subunit α and β (ETFA and ETFB) were upregulated in venetoclax-resistant AML despite unchanged transcript levels. Silencing of these proteins induced mitochondrial stress and apoptosis and sensitized AML cells to venetoclax, again demonstrating the unique insights provided by proteomic analysis with translational potential. ⁵⁸

Venetoclax resistance has also been associated with enrichment of proteins involved in the FLT3 pathway. Interestingly, combining venetoclax with the clinically approved FLT3 inhibitor gilteritinib showed synergistic effects, even in FLT3 wild-type AML, through the suppression of the anti-apoptotic protein myeloid cell leukemia-1. ⁵⁹ This showcases that proteomics can inform rational drug combinations. These findings have already informed clinical development; triplet therapy with the hypomethylating agents azacitidine, venetoclax, and FLT3 inhibitors is currently being evaluated in phase 3 trials, following the encouraging results of phase 2 trials. ⁶⁰

(Phospho)proteomic studies have also identified the mechanisms of response to FLT3 inhibitors. FLT3 inhibitors are widely used in patients with AML carrying mutations in the FLT3 gene. ⁶¹ These mutations include tyrosine kinase domain (TKD) mutations and internal tandem duplications (ITD). FLT3-TKD mutations are associated with increased phosphorylation of Src family kinases (FGR and HCK) and related signaling proteins. Downstream inhibition of these kinases may contribute to the clinical activity of the FLT3 inhibitor midostaurin. ⁴⁵

Several translational studies have suggested that the incorporation of proteomic data improves the prediction of therapeutic responses to FLT3 inhibitors beyond what can be achieved with mutational profiling alone.^11,46 An integrated proteomic and phosphoproteomic analysis of 38 patient samples from the Beat AML cohort was performed to investigate whether proteomic signatures could enhance the prediction of ex vivo responses to 26 drugs. The findings demonstrated the added value of proteomics signatures within the predictive model, showing that a broad, data-driven multi-omics approach was more effective than a targeted, pathway-focused strategy. The model was further explored and validated for quizartinib (an FLT3 inhibitor) and trametinib (a RAS/MEK inhibitor) in AML cell lines. ¹¹ Although the preliminary data are promising, further validation of these models is required before they can be broadly applied. ¹¹

In another study, phosphoproteomic profiling of primary AML samples identified distinct phosphorylation patterns, particularly involving the MAPK and STAT pathways, which differentiated FLT3-ITD from FLT3 wild-type AML and were associated with differential sensitivity to FLT3 inhibitors. Ex vivo drug testing of 19 AML samples further showed increased MAPK1/MAPK3 and EGFR activity in gilteritinib-resistant samples, despite similar FLT3 activity, indicating the activation of alternative signaling pathways in gilteritinib resistance. ⁶²

Phosphoproteomic analysis of 47 AML samples was performed to identify signaling features predictive of response to midostaurin combined with intensive chemotherapy. Using MS–based phosphoproteomics, a 29-phosphomarker model was developed to accurately distinguish responders from non-responders. This study highlights how phosphoproteomic profiling can capture functional signaling states that are not reflected by genomic data alone and underscores the translational potential to refine treatment stratification. ⁶³ While this model shows translational promise, external validation is required before its clinical implementation.

Phosphoproteomic analysis of AML bone marrow samples was performed to identify the signaling features associated with the response to the FLT3-ITD inhibitor, quizartinib. A five-site phosphorylation signature predicted the response to quizartinib, which explains why approximately 30% of patients with FLT3-ITD-mutations fail to benefit from treatment. ⁶⁴

Further proteomic analyses revealed the presence of additional resistance pathways. An exploratory multiomics study investigating gilteritinib resistance reported that early resistance was linked to aurora kinase B activity and metabolic reprogramming, whereas late resistance involved the expansion of pre-existing NRAS-mutant subclones and continued metabolic adaptation. ⁶⁵ In FLT3-ITD–mutated AML, increased expression of autophagy-related proteins was observed following FLT3 inhibitor treatment, suggesting that autophagy is another resistance mechanism. ⁶⁶ The PTK2B–LPXN signaling cascade, which regulates cell adhesion and migration and is frequently expressed in LSCs, was found to be upregulated early during acquired resistance to midostaurin. ⁶⁷ Pim kinase, a downstream effector of FLT3-ITD, has also been implicated in resistance development. ⁶⁸ A large-scale proteomic and transcriptomic study identified RSK2 serine/threonine kinase as a novel Pim2 target. RSK2 depletion increases BAX expression and induces apoptosis in FLT3-ITD-mutant AML cells. These findings remain preclinical but highlight the emerging vulnerabilities to be investigated in further research. ⁶⁸

In KMT2A-rearranged AML, phosphoproteomic profiling of 74 AML samples identified two biologically distinct subgroups. Group A showed increased activity of DOT1-like histone lysine methyltransferase (DOT1L) and transcription elongation factor complexes, elevated HOXA expression, enhanced CDK1 activity, and higher phosphorylation of proteins involved in RNA metabolism and DNA-damage responses than group B. Functionally, this group was more sensitive to genotoxic agents, mitotic kinase inhibitors, and inosine-5-monophosphate dehydrogenase inhibitors. ⁶⁹ Although promising in the context of targeted approaches, these findings remain exploratory and have not yet been evaluated in the context of menin inhibition, a new therapeutic strategy for KMT2A-rearranged AML. ⁶

Identification of Cell Surface Proteins for Antigen-Targeted Immunotherapies

The development of effective antigen-targeted immunotherapies, such as monoclonal antibodies or CAR-T cell therapies for AML is challenging because of the scarcity of suitable target antigens. ⁴⁵ An ideal candidate antigen should be highly expressed on the surface of leukemic blasts and LSCs but absent or minimally expressed in normal cells or tissues.^16,18,70 However, many currently explored antigens (e.g. CD33, CD123) are also present in normal hematopoietic cells, creating a substantial risk of target/off-tumor toxicity. ¹⁶ Antigen expression is heterogeneous across AML subtypes and even within individual patients owing to subclonal diversity and clonal evolution under therapeutic pressure. This variability makes the identification of robust and durable immunotherapeutic targets difficult. ¹⁶

Proteomics offers a strategy for systematically identifying surface antigens expressed on leukemic cells, including leukemic cell subpopulations, thereby supporting the discovery of novel immunotherapeutic targets (Table 1). ¹⁶ For example, CD180 and MRC1/CD206 have been identified by proteomic analysis as being preferentially expressed in AML blasts, indicating their utility as surface biomarkers or potential immunotherapeutic targets. ⁴⁵ Quantitative proteomics of 42 AML samples revealed that 50 plasma membrane proteins were upregulated in AML cells compared to those in normal hematopoietic cells, enabling the prospective isolation of distinct subclones. Several of these proteins, including CD25, CD82, CD97, CD123, FLT3, IL1RAP, and TIM3, have been proposed as candidates for inclusion in diagnostic flow cytometry panels to define leukemia-associated phenotypes and as potential therapeutic targets. ⁷¹

Table 1.

Surface Proteins in AML Blasts and LSCs as Potential Targets for Immunotherapy

Surface antigen	Expression	Study design	Reference
CD180	-Expression: bulk AML	Deep-scale proteomics and phosphoproteomics on 44 representative AML samples from the LAML TCGA dataset	Kramer et al. 45
MRC1/CD206	-LSC expression: not reported
	-Other expression:
	-CD180: B cells, monocytes and macrophages
	-MRC1/CD206: macrophages and hepatic stellate cells
MILR1, CTSG, CD180, CD33, CD123 (IL3RA), LAIR1, CD47, PTPRC, FLT3, CD37, CLEC12A, CD38, KIT, ITGA4, VSIR, CD74, LY75, TLR2, PTK7, IL1RAP, ITGB7, STAB1, ENG, IL17RA, CSF2RA,SEMA4D, NOTCH2,SLC39A6, IGF1R, SORT1, ALCAM, SIRPA	Pan-AML markers	Surface proteome analysis of 100 primary human AML specimens combined with RNAseq data from the Leucegene collection (Leucegene AML surfaceome atlas)	Bordeleau et al. 30
	-Expression: pan AML
	-LSC expression: ITGA4, FLT3, IL3RA, and IL1RAP
	-Other expression:
	-MILR1, CTSG, CD180: limited expression in other organs
	-Targeted by therapeutic antibodies:
	-AML: CD33, CD123 (IL3RA), LAIR1, CD47, PTPRC, FLT3, CD37, CLEC12A, CD38, KIT
	-Other disease: ITGA4, VSIR, CD74, LY75, TLR2, PTK7, IL1RAP, ITGB7, STAB1, ENG, IL17RA, CSF2RA, SEMA4D, NOTCH2, SLC39A6, IGF1R, SORT1, ALCAM, SIRPA
CD34, SLC38A1, NPR3, ADGRG1, CD7, CD96, F2R, ADRA2A, CALCRL, CD109, MPIG6B, PROM1, SV2A, TMIGD2, TNFRSF4	LSC markers
	-Expression: LSC
	-Pan AML: CD34, SLC38A1, NPR3
	-Other expression: diverse
IL3RA (CD123) CD99 CD44 IL1RAP	-Expression: enrichment in LSC	Proteomics and RNA sequencing in 20 AML bone marrow samples	Raffel et al. 19
	-Other expression:
	CD44 and IL1RAP >> IL2RA and CD19: HPSC
Activated conformation of integrin β2	-Expression: bulk AML	Structural surfaceomics combining cross-linking MS with glycoprotein surface capture in AML cell lines and primary samples	Mandal et al. 37
	-LSC expression: no data
	-Other expression: not present on healthy HPSC
ADGRE2 CCR1 CD70 LILRB2	- Expression: AML bulk (>75% of AML cells)	Integrated dataset of AML surface proteins (surfaceomics & previous data). Twenty four candidate targets were analyzed by flow cytometry on 30 primary AML samples. Combinatorial pairing of candidate targets to minimize off-target toxicity.	Perna et al. 70
	- LSC expression:
	CD82, TNFRSF1B, ADGRE2, ITGB5, CCR1, CD96, PTPRJ, CD70, and LILRB2
	-Other expression:
CD33+ADGRE2	-Only low level (<5%) expression in HPSC: TNFRSF1B, ADGRE2, CCR1, CD96, CD70, LILRB2
CLEC12A+CCR1	-Only low-level expression (<5%) in T cells: ADGRE2, CCR1, CD70, LILRB2
CD33+CD70	-Expression on normal HSCP: CD33 and CLEC12A
LILRB2+CLEC12A	-Combinatorial pairing:
	-Near 100% FACS positivity in all 30 AML specimens of the 4 combinations.
CD148 ITGA4 Integrin beta 7	-Expression: bulk AML	Surfaceomics by cell surface capture technology on peripheral blood and bone marrow samples.	Köhnke et al. 72
	-LSC expression: not reported
	-Other expression
	CD148: mature lymphocytes, monocytes and granulocytes
	ITGA4: HSPCs
	Integrin beta-7: -

Abbreviations: AML: acute myeloid leukemia; HSPCs: hematopoietic stem and progenitor cells; MS: mass spectrometry; LSC: leukemic stem cells; RNAseq: RNA sequencing). Bold: high expression. Underlined: immature cell marker.

Surface analyses have expanded the repertoire of potential candidate antigens. Biotinylation of glycosylation sites or lysine residues followed by MS identified CD148, ITGA4, and integrin β7 as promising targets. Among these, integrin β7 appeared to be leukemia-specific with minimal expression on normal hematopoietic cells, making it a particularly attractive candidate for targeted immunotherapy. ⁷² A large multi-omics study integrating (single-cell) RNA sequencing and surface proteomics of 100 AML samples led to the creation of the Leucegene AML Surfaceome Atlas (LASA).^30,73 Surface proteins were highlighted by the surface protein annotation tool (“SPAT”), an analysis that evaluates the likelihood of proteins being surface-localized versus contaminants. ⁷⁴ As listed in Table 1, this study identified 32 pan-AML antigens (CD47, CD37, ITGA4, VSIR and CD74) and 15 candidate LSC markers. Of these 15 LSC antigens, three (CD34, NPR3, and SLC38A1) were universally expressed but were also present on normal hematopoietic stem and progenitor cells, underscoring the difficulty of identifying truly LSC-specific targets. ³⁰

In-depth quantitative multiplex proteomic profiling of FACS-sorted bone marrow–derived AML and healthy donor samples was conducted to identify LSC-specific proteins with minimal or no expression in healthy HSCs. This analysis, combined with transcriptomic data, confirmed that IL3RA (CD123) and CD99 are potential target molecules in LSCs. ¹⁹

Innovative approaches such as structural surfaceomics, which combine cross-linking MS with cell surface capture technology, have enabled the identification of conformationally specific antigens. Cross-linking MS uses chemical crosslinkers to connect nearby amino acids in proteins to preserve native protein confirmation. This is followed by biotinylation of cell surface proteins and subsequent MS/MS analysis. This technique revealed the activated conformation of integrin β2 as a promising immunotherapeutic target in AML, which has a widespread AML-specific expression pattern while being absent on healthy hematopoietic cells. ³⁷

Combinatorial targeting strategies using logic-gated CAR-T cells have been proposed to address antigen heterogeneity and immune escape. Integrated proteomic and transcriptomic analyses identified 24 candidate antigens, among which ADGRE2, CCR1, CD70, and LILRB2 were selected for combinatorial targeting. ⁷⁰ Combining these antigens with established targets such as CD33 or CLEC12A/CLL-1 results in maximal AML targeting while sparing normal cells. ⁷⁰ Together, these studies underscore the value of proteomic profiling in uncovering AML-specific surface antigens and LSC markers. By enabling antigen selection and supporting combinatorial strategies, proteomics could contribute to the development of safer and more effective antigen-targeted immunotherapies for AML and help unlock the full therapeutic potential of CAR T-cell therapy for AML. However, the translation of proteomic findings into clinically actionable cell therapies remains challenging. AML is highly heterogeneous, with subclones, clonal evolution, and phenotypic switching contributing to the dynamic antigen expression over time. ⁷⁵ Loss of the target antigen over time may lead to relapse. ⁷⁶ LSCs remain incompletely characterized at the proteomic level. ¹⁴ Because LSCs represent a rare population, it is often difficult to predict which antigens are reliably expressed in these cells. However, their persistence is associated with a risk of relapse, making accurate targeting of LSCs essential. In addition, off-tumor toxicity is a major barrier. Many candidate antigens are shared by healthy hematopoietic cells, and targeting them can lead to profound cytopenia in combination with a poor marrow reserve. ⁷⁷ This highlights the need for antigen combinations, safety switches, and more refined approaches for antigen discovery that can distinguish malignant from healthy hematopoietic cells with greater precision.

Identification of HLA-Presented Tumor-specific Antigens by Immunopeptidomics

Unlike “extracellular” antigens targeted by monoclonal antibodies or CAR-T cells, most tumor antigens in AML are intracellular proteins presented on the cell surface as peptides bound to major histocompatibility complex (MHC)/human leukocyte antigen (HLA) molecules. These MHC-presented antigens serve as targets for TCR-based immunotherapies such as TCR-engineered T cells. Immunopeptidomics refers to a proteomic strategy used to map peptides presented by MHC molecules on the AML/LSC surface. ⁷⁸ For therapeutic purposes, tumor-specific antigens (TSAs), which are peptides uniquely presented by malignant cells, are the most desirable. ²⁹ In this context, neo-epitopes arising from tumor-specific genetic alterations are a key examples of TSAs. Importantly, the detection of a peptide by immunopeptidomics demonstrates only its presentation but does not imply that the peptide is immunogenic or therapeutically actionable. Peptide identification confirms that a peptide–HLA complex exists on the cell surface, but immunogenicity requires evidence that the peptide can elicit a T-cell response. Therapeutic feasibility for TCR-based strategies represents an additional and more stringent layer, requiring robust endogenous peptide presentation, tumor specificity, availability of safe and high-affinity TCRs, and the absence of off-target or on-target/off-tumor toxicity.

Despite AML’s relatively low mutational burden, recurrent mutations, such as those in NPM1, are present in a substantial subset of patients. ²⁹ Immunopeptidomic analyses have identified MHC class I and II neoepitopes derived from NPM1 mutations.^29,79-81 For example, two MHC class I epitopes (AVEEVSLRK and CLAVEEVSL) are consistently detected in NPM1-mutated AML cases and show strong predicted binding to common HLA allotypes, including A*02:01, A*03:01, and A*11:01. ⁸⁰ Additional neoantigens have been identified in FLT3-mutated AML (e.g., an HLA-A*01:01–restricted FLT3-ITD neoepitope) and in AML with other recurrent fusion genes, such as PML-RARA, DEK-CAN, and ETV6-AML1. ²⁹ However, some mutated regions remain as “dark spots” in the immunopeptidome, lacking detectable MHC-presented peptides. ²⁹

Canonical tumor antigens are encoded within the open reading frames of protein-coding genes. Non-canonical antigens arise from the aberrant translation of small open reading frames in non-protein-coding regions such as untranslated regions, introns, long non-coding RNAs, and endogenous retroelements. ⁸² These abnormal translation events can give rise to tumor-specific, immunogenic MHC-associated peptides. ⁸³ Such cryptic peptides may originate from tumor-specific alterations in MHC antigen processing and presentation pathways, DNA methylation, RNA editing, aberrant protein synthesis and proteasomal splicing. ²⁹ Spliceosome mutations, which are common in AML, may generate neoepitopes. ⁸⁴ Additionally, post-translational modifications (e.g., phosphorylation) can also create neoantigens. ²⁹ A study investigating both canonical and non-canonical MHC-presented antigens identified 13 tumor-specific neoantigens, over half of which were derived from non-coding regions. ⁸³ One notable example is the non-canonical HLA-A*02:01-restricted peptide ILPSYQLFL, derived from the pseudogene RP11-9L18.2. ⁸³

In another analysis of MHC class I immunopeptidomes from 19 primary AML samples, 58 TSAs were identified, predominantly originating from non-coding genomic regions and often involving intron retention and epigenetic alterations. ⁸² None of the TSAs were derived from mutations. AML-associated spliceosome dysfunction contributes to intron retention and epigenetic dysregulation, which appears to influence TSA expression. For instance, ZNF445, a regulator of methylation-dependent genomic imprinting, correlated with TSA abundance, suggesting a role of epigenetic dysregulation in the development of TSAs. RNA-sequencing analysis of the Leucegene AML cohort suggested that over 90% of the patients harbored at least one TSA. ⁸²

In a study mapping the MHC class I ligandome of 15 patients with AML and 35 healthy donors, 132 potential antigens were identified, defined by AML exclusivity and prevalence in >20% of patients. Among these, FAS-associated factor 1 was present in 8 of 15 patients via six different HLA ligands. ⁸¹ However, many of the identified antigens were also found in healthy donors, with only a few, such as FLT3, PASD1, HOXA9, AURKA, and CCNA1, demonstrating AML specificity. ⁸¹ Several of these antigens induced AML-specific CD8⁺ T cell responses.

Finally, a large-scale immunopeptidomics study of 47 AML samples identified 433 AML-exclusive MHC class I antigens, 188 of which were also present in LSCs. ⁸⁵ MHC class II–restricted peptides were also characterized, revealing 311 AML-exclusive antigens. In total, eight of those antigens were found on bulk AML cells as well as LSCs. Among these, CCL23 and RRS1 frequently showed AML-associated presentation. ⁸⁵ Interestingly, although NPM1-and IDH2-derived neoepitopes were detected in AML bulk cells, they were absent in LSCs in this study, highlighting the need for further characterization of the LSC immunopeptidome. ⁸⁵

Conclusion

In AML, proteomics is emerging as a powerful tool, alongside genomics and transcriptomics. It plays a critical role in the search for new biomarkers, drug resistance mechanisms, and potential targets for CAR- or TCR-based immunotherapy. Unlike transcriptomic approaches, proteomics directly captures protein abundance and post-translational modifications, revealing functional states and regulatory networks that are often invisible at the mRNA level. ⁵² Proteomic and phosphoproteomic analyses have elucidated non-genetic mechanisms of resistance to targeted therapies, such as in FLT3-mutated AML. ¹¹

Despite significant therapeutic advances, AML treatment remains challenging owing to the notorious biological heterogeneity, high relapse rates, and an overall limited number of targetable therapies. Proteomic approaches offer opportunities to address these limitations. Proteomics can advance biomarker discovery, improve risk stratification, and enhance the prediction of therapeutic responses. Surfaceomics and immunopeptidomics further expand the landscape of targetable antigens for CAR- and TCR-based immunotherapies, whereas single-cell proteomics enables the resolution of clonal heterogeneity, characterization of LSC populations, and tracking of disease evolution over time. Together, these complementary strategies underscore the potential of proteomics to bridge the current gaps in AML treatment by informing therapeutic decision-making and guiding the development of novel targeted approaches.

Despite its promise, this field faces challenges, including the need for specialized equipment and computational frameworks to interpret complex data. ¹⁵ Advances in MS technology will hopefully allow faster analysis of a larger number of proteins at a lower cost, allowing broader clinical applications. ²³ As proteomics continues to evolve, its integration into multi-omics platforms and clinical workflow holds significant potential for enhancing diagnostic precision, predicting therapeutic responses, discovering biomarkers, and uncovering actionable targets, ultimately contributing to more effective and individualized treatment paradigms in AML.

ORCID iDs

Eva De Backer https://orcid.org/0000-0002-5053-1315

Florian Van Oers https://orcid.org/0009-0001-5176-3214

Sébastien Anguille https://orcid.org/0000-0002-1951-6715

Data Availability Statement

Data sharing is not applicable to this article as no datasets were generated or analysed during the current study.*