A proteomic approach to understand the clinical significance of acute myeloid leukemia-derived extracellular vesicles reflecting essential characteristics of leukemia.

Extracellular vesicle (EV) proteins from acute myeloid leukemia (AML) cell lines were analyzed using mass spectrometry. The analyses identified 2450 proteins, including 461 differentially expressed proteins (290 upregulated and 171 downregulated). CD53 and CD47 were upregulated and were selected as candidate biomarkers. The association between survival of patients with AML and the expression levels of CD53 and CD47 at diagnosis was analyzed using mRNA expression data from The Cancer Genome Atlas database. Patients with higher expression levels showed significantly inferior survival than those with lower expression levels. Enzyme-linked immunosorbent assay results of the expression levels of CD53 and CD47 from EVs in the bone marrow of patients with AML at diagnosis and at the time of complete remission with induction chemotherapy revealed that patients with downregulated CD53 and CD47 expression appeared to relapse less frequently. Network model analysis of EV proteins revealed several upregulated kinases, including LYN, CSNK2A1, SYK, CSK, and PTK2B. The potential cytotoxicity of several clinically applicable drugs that inhibit these kinases was tested in AML cell lines. The drugs lowered the viability of AML cells. The collective data suggest that AML-derived EVs could reflect essential leukemia biology.


Introduction
Acute myeloid leukemia (AML) is the most common form of acute leukemia in adults. The overall incidence of AML has been increasing gradually over the years (1)(2)(3). Despite many recent advances, the 7+3 chemotherapy regimen of cytarabine and anthracycline, originally developed in 1973, remains the basis of remission induction therapy. Morphologic analysis of marrow cells is still the gold standard for clinical diagnosis and response evaluation. The definition of complete remission (CR) for AML is based on the reduction of bone marrow (BM) blasts to 5% as determined by morphology (4). However, a considerable number of patients still experience relapse despite achieving morphological CR (5,6). To overcome this drawback, various techniques have been developed to detect minimal residual disease (MRD). These techniques typically include polymerase chain reaction, flow cytometry, or next-generation sequencing (7,8). However, there are still several limitations in the widespread use of these techniques for clinical practice. For example, flow cytometry requires specific and robust expertise to identify leukemia-associated immunophenotypes and employs an approach different from the normal one. Incorporating this type of analysis in clinical practice is also limited by the large amount of data that needs to be dynamically evaluated to detect and analyze immunophenotype switches. Additionally, the quality of the BM samples is imperative to all these analyses. If there is insufficient starting material, flow cytometry cannot be performed (7,9). Extracellular vesicles (EVs) express the properties of their parental cells, including proteins, RNAs, and DNAs (10). Cancer cell-derived EVs are involved in the overall etiology, including supporting tumor growth (11), inducing vessel formation (12) that contributes to the metabolic reprogramming of cancer cells, enabling sustained proliferation of cancer cells (13), and enhancing the capacity of tumors to become invasive (14,15). Considering EVs are representative of their parental tissues, they are excellent tools to understand how cancer cells adapt to their environment. This has led to the recent emergence of EVs as a focus in cancer research. Clinically, several studies have documented the potential uses of EVs in treating patients with cancer. One study reported that circulating EVs that were glypican-1-positive could be used as biomarkers for early detection and prognosis in patients with pancreatic cancer (16). Another study reported that EVs from drug-resistant breast cancer cells could be used as therapeutic targets to enhance patients' therapeutic responses (17).
The use of EVs as biomarkers of AML has been studied. The plasma levels of exosomes in patients newly diagnosed with AML (expressed as µg/protein/mL) were higher than those in normal controls (18). Plasma exosome levels of EVs were reduced after a course of remission induction therapy concomitant with the reduction of blasts in BM (19). A study of exosomes derived from primary AML cells and AML cell lines showed that these exosomes contained by guest on December 3, 2020 https://www.mcponline.org Downloaded from coding and non-coding RNAs relevant to AML pathogenesis that affected prognosis, response to therapy, and leukemic niche formation (20). Another study using a leukemic xenograft mouse model showed that microRNAs (miRNAs) in AML exosomes could serve as early biomarkers of relapse (21). In addition, proteomics analyses have demonstrated the involvement of AML-derived exosomes in leukemic transformation (22) and apoptosis inhibition (23).
Proteomic analysis is a useful technique in cancer research that can be both quantitative and qualitative when determining the interrelationships between proteins in cells. Proteomics analyzes the phenotype of any expressed protein, and it can identify the type of protein expressed under varying conditions. This approach can also be used to identify potential disease-specific biomarker candidates (24). The composition of proteins in the blood of patients with leukemia is different from that of healthy individuals. These changes can interfere with the activities of healthy blood cells in affected patients (25, 26). Therefore, the types of proteins that make up the blood of healthy individuals and patients with leukemia as well as the respective signaling molecules and networks may be different. In particular, circulating EVs are characterized by signal cascades that can promote cancer metastases and their migration to other organs and tissues (27). Thus, it is clinically important to identify novel differentially expressed proteins (DEPs) from EVs in AML cell that could be used as biomarkers (28). In this study, we validated the use of proteomic analysis in cancer-derived EVs to identify and investigate potential biomarkers and novel drug targets for AML.

Ethical approval and consent to participate
All procedures for the culture of primary human BM stromal cells (hBMSCs) and acquisition of BM blood samples and patients' medical records were approved by the internal review board of the Korea University Anam Hospital (IRB No. 2015AN0267). Informed consent was Genome Atlas (TCGA) public database. To validate the selected AML biomarkers, BM sera or plasma from a total of 17 patients with AML were collected at two different time points. The statistical analysis of the experiments conducted in this work is described in more detail in the statistical analysis of correlation of survival rates and AML-derived EV markers subsection.

Cell lines
HL-60, KG-1, THP-1, Kasumi-1, and MOLM-13 AML cell lines were selected for the study. 1% penicillin/streptomycin (Gibco). Exosome-depleted FBS was prepared by collecting the supernatant following ultracentrifugation of normal FBS at 100,000 ×g. treatment. All patients achieved CR following anthracycline-based remission induction chemotherapy. The baseline characteristics of these patients are shown in Table 1. All human sample collections were performed according to the guidelines of the Internal Review Board of the Korean University Anam Hospital. Informed consent was obtained from all participants.

Isolation of EVs
Isolation of EVs was performed using size-exclusion chromatography. Different columns were used for cell culture supernatants and human samples. The column used for the cell culture supernatants was packed with 10 mL Sepharose CL-2B (GE Healthcare Life Sciences, Pittsburgh, PA, USA) with a molecular weight separation range of 70 × 10 3 -40 × 10 6 (29). The eluted fractions (6,7,8,9, and 10; 0.5 mL each) were used in the subsequent steps of this experiment. The loaded samples were prepared by sequential centrifugation. The cell culture media were collected after 48-72 h of cell growth (10 × 10 6 cells in 50 mL) and centrifuged at 500 ×g for 10 min at 4°C, then at 5,000 ×g for 30 min at 4°C, and finally at 10,000 ×g for 30 min at 4°C. The supernatant was concentrated using an Amicon® Ultra 100 kDa filter with a molecular weight cut-off of 100 kDa (Merck Millipore, Temecula, CA, USA) according to the manufacturer's instructions. For the human samples, EVs were isolated using a chromatography-based method developed by our group (30).

EV sizing and evaluation using transmission electron microscopy (TEM) and dynamic light scattering (DLS)
TEM was performed using a model H-7500 transmission electron microscope (Hitachi, Tokyo, Japan). DLS was performed using a Zetasizer Nano S90 (Malvern, Worcestershire, UK). For TEM analysis, EVs were fixed using 2% paraformaldehyde, loaded on a 300-mesh formvar/carbon-coated electron microscopy grid (Electron Microscopy Sciences, Hatfield, PA, by guest on December 3, 2020 USA), and stained with 2% phosphotungstic acid (PTA). For the DLS measurements, size distribution data were collected from each sample suspended in phosphate-buffered saline.
Each EV population was measured three times.

Protein extraction and western blotting
To extract protein, cell/EV pellets were lysed using ProEX TM CETi Lysis Buffer (Translab, Seoul, South Korea) followed by sonication. Following sonication, nuclei and cell/EV membranes were separated by centrifugation at 10,000 ×g for 15 min at 4°C. The protein concentration was determined using the BCA protein assay kit (Pierce, Rockford, IL, USA). In total, 30 µg of each protein sample was separated using 10% SDS-PAGE. The resolved proteins were transferred onto a nitrocellulose membrane. After blocking with ProNA™ General-BLOCK solution (Translab) for 1 h, the membranes were probed overnight at 4°C

Tryptic digestion by filter-aided sample preparation (FASP)
In total, 30 µg of protein from EVs of AML or control cells was converted to peptides using the FASP method. Extracted EV proteins were resolved by reduction in 0.

High pH fractionation
To increase the number of peptides identified in each sample, high pH fractionation was used to separate peptides based on hydrophobicity. Samples were separated into 48 fractions   Table S1. Genomes pathways using DAVID software (32).

Network construction between kinases and functional node activity
A kinase-kinase interaction network was constructed by analyzing the biological function ontologies of kinase expressed in AML and control cells. Based on the biological process analyzed of kinases, STRING software was used to predict the possible interaction network between proteins (33). Expected protein-protein interaction (PPI) networks with a score of 0.4 for all kinases were plotted using Cytoscape software (34). Interaction networks had distinct biological processes between constituent genes, and each node was represented by color change and size, respectively, based on FC and p-value between AML and control cells.

ELISA
To identify changes in CD markers before and after treatment, EVs isolated from the BM of patients with AML were evaluated using a Human CD47 ELISA Kit (MyBioSource, San Diego, CA, USA) and an ExoTEST TM CD53-exosome ELISA Kit (HansaBioMed, Tallinn, Estonia). by guest on December 3, 2020 ELISAs were performed according to the manufacturer's instructions.

Cell viability assay
The following drugs were used: nilotinib (Sigma-Aldrich, St. Louis, MO, USA; catalog no.

Statistical analysis of correlation of survival rates and AML patient-derived EV markers
Median values and ranges or means and standard deviations for continuous variables and percentages for categorical values were used. First, we analyzed the correlation between survival and mRNA expression data of initial diagnosis for CD53 and CD47 from TCGA database (https://portal.gdc.cancer.gov/) (35). We set a cut-off value of fragments per kilobase of transcript per million (FPKM) of CD53 and CD47 with the highest hazard ratio (HR) and the lowest p-value to analyze the survival probability according to these markers. The survival rate of the divided group with these cut-off values was analyzed with Kaplan-Meier analysis using the log-rank test. The HR was estimated using multivariable Cox regression (36). Second, we analyzed the correlation between survival and the expression levels of CD53 and CD47 measured by ELISA in EVs from the BM of patients with AML before and after remission induction treatment. Relapse-free survival (RFS) was defined as the time from diagnosis to relapse or death and was calculated according to the Kaplan-Meier method using the logrank test. We used the R program version 3.6.1, and IBM SPSS version 25.0 software to analyze the data. A p-value <0.05 was considered to indicate a significant difference.

Validation of EVs derived from AML and control cells
The properties of EVs isolated by size-exclusive chromatography from supernatants of AML or control cell lines are shown in Fig. 1. TEM revealed that the size of EVs from AML and control cells was < 200 nm. They were visualized as cup-shaped vesicles under high magnification (Fig. 1A). The size distribution of these EVs measured by DLS indicated a range between 30 nm and 185 nm (Fig. 1B). Western blotting showed that the isolated EVs were positive for the exosome markers (CD9 and CD81) and negative for the endoplasmic reticulum marker calnexin (Fig. 1C).

Protein identification and quantification
To analyze the EVs from HL-60, KG-1, and THP-1 AML cell lines and control cells (HDFa and hMSCs), quantitative and comparative proteomic analyses were performed using LC-MS/MS. To determine the proteomic profiles of the EVs from AML cells, LC-MS/MS was performed following FASP digestion and subsequent fractionation. Proteins from the isolated EVs were digested and fractionated to form peptides for MS examination. To improve protein confidence, the analyzed protein levels were tested against the 1% FDR threshold and identified with at least one unique peptide ( Fig. 2A). After triplicate runs, the acquired independent MS/MS spectra were searched using the SEQUEST HT algorithm in Proteome Discoverer 2.4 against the Human UniProtKB database.
The overlapping proteins in all three replicate proteomic profiles were plotted in a Venn diagram. In total, 1980, 1972, and 1819 proteins from subjects 1, 2, and 3, respectively, were identified. A total of 2450 unique proteins were identified by LC-MS/MS analysis from all three independent replicates (Fig. 2B).
Based on the EV data from the five cell types, the significance level was determined by performing ANOVA and post hoc Bonferroni's tests to select DEPs. All overlapping DEPs between the control and AML groups were selected with a cut-off value established at a fold change of more than 1.5 and less than 0.67 and significance at 95% confidence. The heat map was represented by the expression levels of the DEPs (Fig. 2C). In total, 461 proteins were differentially expressed in AML cell-derived EVs. Of the 461 DEPs, 290 were upregulated and 171 were downregulated. Hierarchical clustering of the DEPs was performed as shown in Fig. 2C. Functional analysis of DEPs was conducted using DAVID software, which performs GO analyses for biological processes. Analysis of the biological processes associated with the upregulated proteins revealed an enrichment of the pathways associated with extracellular matrix organization, cell adhesion, and positive regulation of exosomal secretion (Fig. 2D).
The selected DEPs were also confirmed to be related to the major pathways of leukocyte migration, regulation of immune response, DNA repair, DNA damage response receptormediated endocytosis, positive regulation of T-cell proliferation, and positive regulation of telomere maintenance via telomerase. Pathways associated with translational initiation, nuclear-transcribed mRNA catabolic process, nonsense-mediated decay, viral transcription, regulation of cell shape, protein stabilization, and tumor necrosis factor-mediated signaling pathways were represented in the downregulated proteins (Fig. 2E). The highly expressed proteins identified were interrelated, which was supported by the significantly high PPIs characteristic of AML biological processes. Proteins generally had several GO annotations (annotated proteins p < 0.01).
The biological processes of the DEPs in AML and control cell line-derived EVs are listed in Table 2. Among the upregulated proteins, proteins were selected if expression has been reported in AML cells, hematopoietic stem cells, or leukocytes, except those already used clinically, and the proteins were surface-exposed and capable of detection by common clinically used methods, such as ELISA and flow cytometry. The CD4, CD53, CD33, and CD47 by guest on December 3, 2020 surface proteins were selected as potential biomarker candidates for further analysis (37)(38)(39)(40)(41).

Evaluation of the selected protein biomarkers
The differential expression of CD4, CD53, CD33, and CD47 was verified by western blotting (Fig. 3). CD4, CD33, CD53, and CD47 were expressed in EVs from AML cell lines but not in EVs from the control cell lines. CD53 and CD47, which were expressed in all three AML cell lines, were selected as the final candidates to investigate whether these enriched proteins could be potential biomarkers in patients with AML. Two assessment approaches were used.
First, we investigated the survival of patients with AML using information from the TCGA database. The survival and mRNA expression data of CD53 and CD47 at initial diagnosis were obtained and a total of 187 patients with AML were analyzed. In the case of CD53, the effective cut-off value of FPKM was 120 (minimum p = 0.015). We divided the patients into two groups according to the FPKM value. If CD53 expression was < 120, the patients were classified into  Table 3).
In the case of CD47, the effective cut-off value of FPKM was 13 (minimum p = 0.018), and patients were divided into two groups, as was the case with CD53. The survival probability of Group 2 was significantly lower than that of Group 1 (Group 1: n = 84, Group 2: n = 103, HR 1.54, CI 1.01-2.34, p = 0.041) (Fig. 4B and Table 3). Next, we analyzed the survival probability by combining two markers and reclassifying the patients into three groups: Group 1 was 2.57, p = 0.033) (Fig. 4C and Table 3).
Second, the expression levels of selected markers (CD53 and CD47) in EVs from the BM of patients with AML at the time of diagnosis and after treatment were measured by ELISA.
We analyzed the correlation between the differences in expression levels of the two markers (before and after chemotherapy) and RFS to understand the prognosis after treatment. All patients achieved CR following anthracycline-based remission induction chemotherapy. The implications of changes in the expression levels of CD53 or CD47 between the time of diagnosis and after treatment were determined. If the difference in expression levels of CD53 or CD47 between the two time points was greater than the median, the patient was classified into Group 1. If not, they were put into Group 2 (Fig. 5A, 5B, and 5C). If the expression levels of CD53 decreased after treatment, RFS tended to be longer, but there was no significant difference (Group 1; n = 8 vs. Group 2; n = 9, p = 0.235) (Fig. 5D). In the case of CD47, RFS was significantly longer in the group in which CD47 was significantly decreased after treatment (Group 1; n= 8 vs. Group 2; n = 9, p = 0.047) (Fig. 5E). When these two markers were combined, we were able to further subdivide the categories for prognosis. According to changes in the expression levels of CD53 and CD47, patients were divided into three groups.
If patients had decreased values for both CD53 and CD47, they were categorized as Group 1 (n = 5). If patients did not show any decrease in either CD53 or CD47, they were categorized as Group 2 (n = 6). Patients who could not be categorized as either Group 1 or 2 were categorized as Group 3 (n = 6). RFS was significantly better when both CD53 and CD47 were used together than when either marker was used alone (Group 1; n = 5 vs. Group 2; n = 6, p = 0.035) (Fig. 5F).

Evaluation of identified kinases and their interactions
Kinases maintain various cellular functions and interact closely with each other to establish a network to regulate biological activity. Compared to the control group, kinases with increased by guest on December 3, 2020 expression in AML regulated cell signaling in cancer related to the origin and dissemination disease-specific pathology. A total of 56 kinase proteins were identified in the AML and control cell types. Of these, 39 had close interactions with each other (Fig. 6A), with important roles in a variety of important biological signaling pathways, including immune responses, signaling, activation of protein kinase activity (PKA), glycolysis, and cellular processes associated with the mitogen-activated protein kinase (MAPK) cascade ( Figure 6A). The FC of AML cell-derived EV normalized to the control group was calculated, and the p-value was obtained by performing a one-way ANOVA, followed by Bonferroni's multiple comparison test (Supplemental Table S2). Lyn, SYK, CSK, CSNK2A1, and PTK2B were identified as kinase proteins whose FC was >1.5 times (p<0.05). These kinases are involved in the immune response and signal transduction pathways.
Among the clinically available drugs, nilotinib, acalabrutinib, and fostamatinib block Lyn, CSNK2A1, and SYK proteins, which were the most represented in our analysis of AML cellderived EVs. These drugs were used to treat AML cell lines for 24, 48, or 72 h. We also compiled literature data on blood concentration levels for nilotinib, acalabrutinib, and fostamatinib in humans so that we could evaluate the clinical suitability. When used in a clinical setting, the maximum serum concentration (Cmax) was in the range of 1,000-4,000 nM for nilotinib (42), 600-2,000 nM for acalabrutinib (43, 44), and 668-1,020 nM for fostamatinib (45).
All three drugs had an inhibitory effect on the proliferation of the HL-60, KG-1, THP-1, Kasumi-1, and MOLM-13 AML cells at concentrations lower than the reported Cmax concentrations (Fig.   6B). Inhibition of proliferation was the strongest for fostamatinib. by guest on December 3, 2020

Discussion
We isolated and highly-purified AML cell-derived EVs and identified specifically enriched proteins in using MS. CD53 and CD47 were selected from the upregulated proteins in AML cell-derived EVs. Their enhanced expression at diagnosis of AML patients was associated with reduced survival. Their reduced expression after treatment might be related to RFS. Kinase interaction networks provide a systematic understanding of the biological context, function, and regulation in cells. In particular, kinases with greater FCs in AML cell-derived EVs than in control cells are involved in the signaling network related to cancer and are often a driving force of disease. Therefore, treatment with inhibitors that target these kinases can block many factors related to cancer progression. We also selected drugs (nilotinib, acalabrutinib, and fostamatinib) that block LYN, CSNK2A1, and SYK proteins, which were the most enriched in the EVs derived from the AML cells, to evaluate their potential clinical applications. All three drugs reduced the viability of the AML cells at clinically relevant concentrations.
Proteomics is a protein analysis tool that can be used to interpret how genes function in their environment. Therefore, proteomics could become a powerful tool to evaluate and identify biomarkers based on the analysis of dynamic protein changes in patients with cancer exposed to various treatment environments, including surgery, chemotherapy, and radiation therapy (46). However, cell-based proteomics has limitations in clinical applications because of the huge amount of data generated as a result of the heterogeneity of cancer cells and the dynamic changes in their environment (47). EV-based proteomics could be one of the solutions to these limitations. The total amount of EVs is higher in patients with cancer than in healthy controls (48). In addition, EV sorting based on proteins, RNAs, and DNAs from the original cells can be performed (49). As EVs can be obtained from any bodily fluid and the total amount of information they possess is less than that of their originating cells, they can be tested in a clinically relevant manner (50). In previous studies, proteomic analysis has been used to suggest biomarker candidates from cancer-specific EVs generated from a variety of by guest on December 3, 2020 solid tumors, including breast, colon, and lung cancers (51). In this study, we established that a similar approach using AML cell-derived EVs can be applied to novel drug and biomarker identification.
Although the precise function of EVs remains unknown, they reflect the phenotypic state of the cells that generate them and contain all the known molecular constituents of these cells, including proteins, RNAs, and DNAs (52). Thus, EVs are a potential source of information to identify new biomarkers and therapeutic targets for various cancers. Previous studies reported that glypican-1-positive circulating exosomes might be diagnostic and prognostic biomarkers for the early detection of pancreatic cancer (16). Another study reported that exosomes from drug-resistant breast cancer cells contain miRNAs that could transmit chemo-resistance, making them a potential therapeutic target, which may enhance a patient's response to therapy (17). Thus, there have been few studies focusing on the roles of AML-derived EVs and their potential applications in clinical diagnostics, prognosis, and therapy. In this study, we demonstrated that the EVs derived from AML cells could be a useful platform to develop biomarkers and identify novel drug targets in AML.
We confirmed the positive AML cell-derived EV biomarker by referring to previous studies on EVs including microvesicles and exosomes derived from AML. The positive AML cellderived EV biomarkers identified in our experimental data were CD13, CD33, CD34, NPM1, and TGFβ1, as previously reported (18,20,(53)(54)(55). Presently, CD33 and NPM1 were significantly increased in all AML groups compared to the control group (p < 0.05). However, compared to the control group, CD34 expression was increased only in EVs derived from THP1 cells. The expression was decreased in other AML cell-derived EVs, and CD13 and TGFβ1 were decreased in all AML cell groups compared to control cells. These discrepancies may result from different experimental conditions, including EV preparation, analytical equipment, and statistical methods. The AML cell-derived EV markers selected in our study included positive markers that were previously identified in AML cells or hematopoietic stem cells in previous studies (18,20,(37)(38)(39)(40)(41)(53)(54)(55). by guest on December 3, 2020 Among the DEPs in AML cell-derived EVs, 62 proteins were previously identified in whole cell studies (56-59). This discrepancy strongly suggests the benefit of EV research, as EVs reflect the dynamic changes of cell states. Interestingly, the EV proteins proposed as candidate biomarkers, including CD47, CD33, CSK, LYN, and SYK, agreed well with results from whole cells. However, some candidate biomarkers, such as CSNK2A1, have shown opposite protein expression patterns in cells and EVs (56).
Among the selected markers, CD53 and CD47 were significantly upregulated in AML cellderived EVs, and were associated with the survival of patients with AML. The TCGA database is a comprehensive atlas of cancer genomic profiles and provides data for genome, transcriptome, and proteome including clinical metadata (60). We used the TCGA database to validate the levels of CD53 and CD47, which were selected as biomarkers in this study.
Transcriptome data analysis revealed that higher levels of CD53 and CD47 at diagnosis were associated with lower survival rates. In addition, we further measured the levels of CD53 and CD47 at diagnosis and after treatment using ELISA. The analysis revealed a lower risk of recurrence if values of CD53 and CD47 in EVs were lower after treatment. Collectively, the results of this study suggest that proteomic approaches using AML cell-derived EVs could be useful platforms for biomarker research to predict patient survival and measure MRD.
One of the most interesting observations in this study is that drugs selected based on protein expression in AML cell-derived EVs reduced the viability of parental AML cells. The network model analysis of DEPs demonstrated a prominent increase in Lyn, CSNK2A1, SYK, CSK, and PTK2B kinase protein expression. Lyn interacts with and phosphorylates tyrosine residues in SYK and BTK kinases (61). Based on this information, tyrosine kinase inhibitors targeting these kinases (nilotinib, acalabrutinib, and fostamatinib) were selected for evaluation.
In general, nilotinib is used for chronic myeloid leukemia or Bcr-Abl-expressing hematological malignancies, acalabrutinib is used for chronic lymphocytic leukemia or B cell lymphoma, and fostamatinib is used in the treatment of autoimmune diseases. The effect of these drugs in the treatment of AML is unclear. However, several studies have suggested the potential of nilotinib by guest on December 3, 2020 as a therapeutic agent for AML that features BCR-ABL1 transcription (62)(63)(64). The potential of fostamatinib as a therapeutic agent in FLT3-ITD-positive AML has been supported by evidence of the importance of SYK in the regulation of FLT3 (59). However, these studies were conducted using leukemia cell-based gene analysis, which selected the drugs by simple genotyping and not by functional analysis of that gene. In this study, drugs that could affect the original AML cell lines were selected by analyzing the proteome of EVs. In our opinion, the success of these drugs against parental cell lines suggests that EVs might be a key to understanding the crucial oncological features of AML.
There are several limitations to this study. First, there are still various challenges to analyze biomarkers with EVs (65, 66). The absence of a standardized procedure for EV isolation is one challenge. Since the EV isolation method affects the physical and molecular properties of the isolated EV, standardization is needed to enable comparison between studies and to improve the reproducibility of results. Collection and utilization at the clinical stage is another challenge. Since various treatments have been applied to patients, it is very difficult to determine biomarkers using EVs collected from patients. In addition, the characteristics of EVs collected from patients are also affected by the treatment applied to the patient. To avoid these challenges, we first compared EVs from cell lines. The second limitation is that EVs derived from AML and control cells line-derived EV proteins were identified and quantified by tagging each sample with TMT. Protein identification was then performed using the reporter ion quantification method based on the TMT for each sample. TMT labeling may produce lesser protein profiling information than label-free quantification methods (67). The label-free approach has been reported to have a higher number of confidently identified proteins than the TMT approach, and generates more information for protein profiling using at least two peptides for identification. In addition, the protein profile information may vary between samples in the label-free approach. The label-free approach could be useful to generate the protein profile for AML cell line-derived EVs on their own. However, the TMT approach improves the ability of researchers to compare different samples under different conditions by guest on December 3, 2020 when they are collected and analyzed together. Therefore, the TMT quantitative analysis approach can reduce variation between samples and allow for precise quantification between groups. In addition, each experiment was reproducibly identified, and a statistical analysis of the list of proteins and quantification values is possible. Third, it would be better to compare the effects of enriched proteins in EVs derived from AML cells on the RFS as opposed to the preexisting cytogenetic and molecular risks of AML. However, we could not confirm this because of the small sample size in this study. Fourth, this study validated the use of AMLderived EVs from BM blood as a biomarker for determining the treatment direction, but it is necessary to examine whether the same findings would be obtained using other bodily fluids, including peripheral blood. Fifth, it is not certain whether the upregulated proteins in AML cellderived EVs simply mirror the original AML cells or whether they are enriched in EVs depending on the dynamics of the disease in this study alone. Further research is needed.
This study showed that AML cell-derived EVs could be used to identify biomarkers to predict the survival and therapeutic response and to determine future therapeutic directions for patients when their protein profile is analyzed by MS. In conclusion, AML cell-derived EVs represent key protein cargo targets that carry important protein information from their parental AML cells. Thus, EVs derived from AML can be used as representatives of original AML cells and might be useful to investigate the biology of AML and determine the clinical value of certain observations.

Data Availability
Mass spectrometry global proteomics datasets are available via ProteomeXchange with identifier, PXD022758 and 10.6019/PXD022758. All MS/MS spectra, chromatograms, identification, and quantification can be viewed using PRIDE Inspector. Annotated spectra for the proteomic experiment are viewable on MS Viewer. Information on peptides and proteins identified using the LC-MS/MS data can be found in Supplemental Table S1.

Disclosure of Conflicts of Interest
The authors declare that they have no competing interests.      by guest on December 3, 2020

Figure 3. Validation of selected EV protein biomarkers
Among the top-ranked upregulated DEPs, surface antigens CD33, CD4, CD53, and CD47 were selected as AML protein biomarker candidates for further analysis. CD33, CD4, CD53, and CD47 were expressed in EVs from AML cell lines but not in EVs from control cell lines.
CD53 and CD47, upregulated in all three AML cell lines, were selected as the final candidates for use as biomarkers.
by guest on December 3, 2020 with AML. The patients were divided into two groups according to an effective cut-off value. If the CD53 or CD47 expression was lower than the cut-off value, patients were classified into Group 1. If not, they were classified into Group 2. In case of CD53, Group 2 showed decreased survival probability with borderline significance relative to Group 1. (B) In the case of CD47, the survival probability of Group 2 was significantly lower than that of Group 1. (C) Next, we analyzed the survival probability by combining two markers and reclassifying patients into three groups. Group 1 was composed of patients with AML referred to Group 1 with regard to both markers. Group 2 was composed of patients with AML referred to Group 2 with regard to both markers. Patients with AML not belonging to Group 1 or 2 were classified into Group 3.
Group 2 showed the lowest survival probability relative to Group 1 or Group 3. (D) Kaplan-Meier survival analysis with log-rank test for overall survival of patients with AML according to CD53 and CD47 is shown in Table 3. there was no significant difference (Group 1; n = 8 vs. Group 2; n = 9, p = 0.235). (E) In the case of CD47, RFS was significantly longer in Group 1 in which CD47 was significantly decreased after treatment (Group 1; n = 8 vs. Group 2; n = 9, p = 0.047). (F) When these two markers were combined for analysis, the prognosis could be further subdivided. According to changes in the expression levels of CD53 and CD47, patients were classified into three groups.
If the patients showed decreased values compared to the median CD53 and CD47 levels, they were classified into Group 1 (n = 5), and if the patients did not show a decrease in CD53 and CD47 levels, they were classified into Group 2 (n = 6). Patients who could not be categorized to either Group 1 or Group 2 were classified into Group 3 (n = 6). RFS was better predicted when CD53 and CD47 were used together than when CD53 or CD47 was used alone (Group 1; n = 5 vs. Group 2; n = 6, p = 0.035).
by guest on December 3, 2020