Peripheral blood samples from CML patients were collected at the Erasto Gaertner Hospital (Curitiba, Brazil) following informed consent and approval from the local Ethics Committee (CAAE 08809419.0.0000.0098 and 53207021.5.0000.0098, issued on 23 May 2023). Plasma was first separated from the blood by centrifugation (400 g; 4 °C; 10 min) and carefully stored at −80 °C until use. The buffy coats were used to isolate DNA and RNA. Patients and samples were selected based on clinicopathological features (diagnosis, follow-up, prognosis, type and time of treatment, and presence/absence of resistance mutations).

The final cohort was composed of 5 groups related to the resistance-associated mutational burden of these patients (performed as indicated in the ‘Detection of ABL1 mutations associated with TKI resistance’ section). These groups were called: PTR (patients with relapse but without resistance mutations, all those patients changed their treatment for dasatinib, n = 4); T315I (relapsed patients harboring T315I resistance mutation, which samples was also collected before the onset of the resistance mutation; samples named P-T315I, n = 4); GTR (patients that had a good prognosis evaluation, being treated with imatinib and a rapid decline of the BCR::ABL1 transcript level, n = 4); TFR (patients who have responded well and started a discontinuation of the treatment, n = 4); and HC (samples from healthy blood donors, n = 5).

Two fractions of buffy coats were separated from peripheral blood samples; one fraction was used for total RNA extraction using the QIAamp® RNA Blood Mini Kit (Qiagen), while the other was used for DNA extraction with QIAamp® DNA Blood Mini Kit (Qiagen) following the manufacturer’s instructions. The extracted RNA and DNA were quantified in NanoDrop™ One (Thermo Scientific) and stored at −80 °C until further use.

Total RNA extracted from the buffy coat was analysed by probe-based RT-qPCR following the protocol described by Marin et al. (2023). Briefly, reverse transcription and PCR amplification were performed as one-step reactions to evaluate the expression ratio between the ABL1 endogenous gene and BCR::ABL1 target transcriptional gene. Quantifications were based on a commercial standard curve, with the final results presented as the ratio of ABL1 to BCR::ABL1, adjusted using a conversion factor and curve intersection values.

Specific Taqman™ SNP Genotyping Assays for each single nucleotide polymorphism (SNP) were used in qPCR reactions with TaqPath™ PromAmp™ Master Mix (Thermo Fisher Scientific). The PCR reactions were performed on a QuantStudio 5™ real-time PCR platform (Thermo Fisher Scientific) using the genotyping settings. The mutations that were characterized were: T315I (rs121913459), E255K (rs121913461), Y253H (rs121913461), E255V (rs121913449) and F359V (rs121913452), on all the analysed cohort just patients with the mutation T315I detected were allocated at the T315I group, there are no other patients with other mutations in any other groups.

EV isolation and characterization were performed in accordance with current MISEV guidelines (Welsh et al., 2024), using complementary physical and morphological approaches.

EVs were isolated from 500 μL of plasma samples from CML patients using size-exclusion chromatography columns for extracellular vesicles 70 nm series (Izon Science Limited) following the recommendations of the manufacturer, no protein depletion step was applied during EV isolation, which was performed solely by size-exclusion chromatography, a technique known to preserve EV integrity. Fractions 1 and 2 were selected for downstream analyses as they are enriched in small extracellular vesicles according to the column specifications and are commonly used to minimize possible contaminations by soluble plasma proteins and lipoproteins. These fractions showed particle size distributions compatible with small EVs as assessed by nanoparticle tracking analysis (NTA).

Samples from the same analytic group were pooled and diluted 1:12 in filtered PBS (using 0.2 μm filters) to a final volume of 1 mL prior to analysis and quantification using a NanoSight LM14C instrument (Malvern Instruments, Malvern, United Kingdom), samples from the same analytical group were pooled in order to obtain sufficient particle concentration for reliable size distribution and concentration measurements. Measurements were performed at an optimal concentration of 20–100 particles per frame. Camera level was adjusted to clearly visualize all particles while maintaining particle signal saturation below 20% (for cell line-derived EVs, camera levels ranged between 14 and 15). The detection threshold was optimized to capture the maximum number of particles, typically aiming for 10–100 red crosses, while limiting blue crosses to a maximum of 8. Autofocus was fine-tuned to eliminate indistinct particles from the field of view.

For each sample, three videos of 60-s were recorded under the following conditions: temperature at 25 °C, infusion rate set to 1,000 units, and syringe pump speed at 100 μL/s. Captured videos were analyzed using the NanoSight Analytical Software (NTA 3.4, Build 3.4.4), applying a detection threshold of 2 or 3 to minimize the inclusion of non-EV particles.

The same sample’ pools that were used in the NTA analysis were visualized by Electron microscopy (pooled samples from each analytical group were used to allow representative visualization of vesicular structures). Each pool of EVs were loaded separately on 300-mesh Formvar-coated copper grids (Electron Microscopy Science, Washington, DC, United States of America) and allowed to adsorb for 1 h. The grids were then washed and prepared for the analysis as described by Souza et al. (2024). The analysis was carried out at the microscopy facility of the Carlos Chagas Institute (Oswaldo Cruz Foundation, Fiocruz- Paraná), performed at an electron microscope with acceleration voltage of 100 kV in a JEOL JEM-1400 Plus transmission (JEOL Ltd., Tokyo, Japan).

Protein contents were isolated from extracellular vesicles with a lysis buffer provided in the Total Exosome RNA and Protein Isolation Kit (ThermoFisher Scientific). Proteins were quantified using Qubit™ Protein and Protein Broad Range (BR) Assay Kits (ThermoFisher Scientific) in a Qubit fluorometer (ThermoFisher Scientific) and were sent to the mass spectrometry facility (Oswaldo Cruz Foundation, Fiocruz- Paraná).

The Multiple Affinity Removal Spin Cartridge Human-14 kit (Agilent) was used to deplete 14 plasma high-abundant proteins, including albumin, aiming to decrease sample complexity and improve detection sensitivity, high-abundance plasma protein depletion was carried out only after EV isolation, characterization by NTA and TEM, and vesicle lysis, and therefore could not affect EV recovery or structural integrity. The depletion process consisted of 3 stages: sample preparation, depletion by column affinity, and sample concentration. The Evs samples were centrifuged and applied to a 0.22 µm spin filter to remove debris and impurities that could clog the column. Then, 10 µL of the filtered sample was transferred to a new tube, and the depletion was realized following the kit standard pipeline, using a column affinity, resulting in 1 mL of depleted sample. After depletion, 500 µL were concentrated by Amicon ultra – 3 kDa, resulting in 200 µL of depleted and concentrated samples. This process was performed for all 28 plasma samples. The protein concentration was measured by UV 280 nm (ThermoFisher Scientific) before and after depletion and after the concentration step. The absorbance value was corrected using the molar absorption coefficient. Next, the samples were prepared for the NanoLC-MS/MS analysis.

Digestion of depleted Evs was carried out with 40 µg of proteins after sample drying in SpeedVac Vacuum Concentrator (ThermoFisher Scientific). Protein denaturation and reduction were achieved by sample incubation with 40 µL of urea 8 M in ammonium bicarbonate (ABC) 50 mM and DTT 25 mM for 1 h at 37 °C and 800 rpm followed by alkylation with 5.45 µL iodoacetamide 500 mM for 1 h at room temperature 25 °C and 800 rpm in the dark. Urea 8 M was diluted to 0.8 M with ammonium bicarbonate 50 mM prior to digestion with Sequencing Grade Trypsin (Promega) at 1:50 enzyme:protein m/m for 18 h at 37 °C. Before injection, samples were desalinated with an in-house stage-tip, and the peptides were quantified in NanoDrop (ThermoFisher Scientific) at 280 nm absorbance. Sample concentrations were normalized prior to injection.

The digested samples (0.5 µg) were separated by online nanoscale capillary liquid chromatography and analyzed by nanoelectrospray tandem mass-spectrometry (nLC-MS/MS). The chromatography was performed on an Ultimate 3,000 nanoLC (ThermoFisher Scientific) followed by nanoelectrospray ionization, and MS/MS on an Orbitrap Fusion Lumos (Thermo Fisher Scientific). The chromatographic conditions were as follows: mobile phase A 0.1% formic acid, mobile phase B 0.1% formic acid, and 95% acetonitrile. The flow of 250 nL/min, with a 60 min linear gradient from 5% to 40% B. The separation was carried out on an in-house C18 packed emitter with 15 cm length, 75 μm Internal diameter, packed with 3.0 μm C18 particles (Dr. Maisch - ReproSil-Pur). MS and MS/MS scan parameters were as follows: MS1 acquisition in the Orbitrap analyzer with a resolution of 120,000, m/z window of 300–1,500, positive profile mode with a maximum injection time of 50 ms. MS2 analysis was performed in data-dependent acquisition (DDA) mode of ions with 2-7 charges, 2 s per cycle where the most intense ions were subjected to high energy collisional dissociation (HCD) fragmentation at 30% normalized collision energy, followed by acquisition in the Orbitrap analyzer with a resolution of 15,000 in centroid mode. A dynamic exclusion list of 60 s and the internal mass calibration for the MS1 scans were applied. The nESI voltage was 2.3 kV, and the ion transfer capillary temperature was 275 °C.

The spectra identification was carried out in MaxQuant (v. 2.2.0.0) as follows: Oxidation on methionine and acetylation on protein N-terminal were set as variable modifications, and carbamidomethylating of cysteine was selected as fixed modifications. The analysis was performed using the specific digestion mode and trypsin defined as protease. The search tolerance was set to 20 ppm for both MS and MSMS. The search was conducted using a peptide length of at least seven amino acids and a tolerance of two missed cleavages in the in silico digestion.

The Homo sapiens dataset was downloaded from UniProt (UP000005640; 81,837 entries) in January 2023 and used as the reference proteome, to which common contaminants (human keratins, BSA, and porcine trypsin) were added. The reverse dataset was employed as a decoy for FDR estimation (1% FDR for both Peptide Spectrum Match and protein assignment was accepted). The ‘match between runs’ and ‘label-free quantification’ (LFQ) options were enabled.

The mass spectrometry proteomics data has been deposited to the ProteomeXchange Consortium via the PRIDE25 (Project accession: PXD071200) partner repository.

All data processing was conducted at the Perseus software (v. 2.0.11) (Tyanova et al., 2016) using the Label-free quantification intensity (LFQ-intensity) to estimate protein expression levels. Initially, proteins indicated as potential contaminants, reverse hits, or only identified by site were excluded. To ensure data robustness, a filter was applied where only proteins detected in ≥70% of samples within at least one analytical group were retained for statistical analysis. LFQ-intensity values were then log₂-transformed, and missing values were imputed from a normal distribution (width = 0.3; downshift = 1.8), to model low-abundance proteins below the detection limit and minimize bias introduced by stochastic MS sampling. Differential expression analyses were performed on continuous log₂-transformed LFQ intensity values and did not rely on presence/absence-based protein detection.

Two statistical approaches were used to identify differentially expressed proteins (DEPs): Student's t-test (P-value <0.05) for two-group comparisons, and ANOVA followed by Tukey’s post-hoc test (FDR-value <0.05) for multiple-group comparisons. Principal component analysis (PCA) was performed in Perseus. Paired t-tests (P-value <0.05) were carried out using R software (v. 4.4.2) (https://www.R-project.org/).

Biological processes and pathways significantly enriched in the lists of differentially expressed proteins (DEPs) were identified using the REACTOME and Hallmark gene sets from the Molecular Signature Database (MSigDB; v. 2024.1) (Liberzon et al., 2015; Subramanian et al., 2005). Additionally, Gene Ontology (GO) terms related to cellular components and molecular functions were evaluated in the total proteome to verify whether the observed enrichment was consistent with known EV proteome characteristics. Gene Set Variation Analysis (GSVA) was performed using the GSVA R package (v. 2.0.7) (Hänzelmann et al., 2013) to calculate the activation scores for each sample based on the significantly enriched pathways. The kernel used for non-parametric estimation of the empirical cumulative distribution function (ECDF) across samples was set to ‘auto’ for optimal selection based on input data characteristics.

Protein–protein interaction (PPI) data was obtained from the STRING database (v. 12.0), incorporating experimentally validated interactions, co-expression data, and data from partner databases (Szklarczyk et al., 2021). Interactions were categorized as medium-confidence (score 0.4–0.7) or high-confidence (score >0.7). The resulting networks were visualized using Cytoscape software (v. 3.10.3) (Shannon et al., 2003).

The total proteome obtained after the filtering stages was compared to the Vesiclepedia (Kalra et al., 2012) and ExoCarta (Keerthikumar et al., 2016) annotated proteomes to verify if our data was enriched in previously EV-associated proteins. The 100 most often identified proteins in both databases were considered as EV-markers in this analysis.

All the selected patients were diagnosed with CML, and samples were selected using the criteria of a time collection by 20 months (±6) after the initiation of treatment (patients with resistance mutations were selected based on the mutation point). In this way, the selected groups are demonstrated at Table 1 and the clinical information with the BCR::ABL1 quantification and mutational profile are demonstrated in the Supplementary Table S1.

The EV isolation protocol resulted in 5 EV size fractions: Fraction 1 is enriched in smaller extracellular vesicles, while fraction 5 is enriched in larger vesicles. The fractions 1 and 2 (smaller EV fractions) from each sample, were mixed and the size distribution confirmed by Nanosight tracker (NTA) and Electron microscopy, where the samples within the same group were pooled and analysed as an unique sample (Figure 1; Supplementary Table S1), demonstrating the presence of EVs in each of these pools. For MS analyses, patients were analysed individually were each sample was composed by the mix between fractions 1 and 2, samples were prepared for MS analyses, to assess the differentially expressed proteins in each sample from the analytical groups. NTA demonstrated particle size distributions consistent with small extracellular vesicles across all pooled samples. TEM analysis further confirmed the presence of membrane-bound vesicular structures with sizes compatible with EVs.

After EV isolation, their protein contents were released, prepared and analyzed by a high-throughput label-free mass spectrometry approach, resulting in the identification of 598 proteins across samples at FDR 1% (mean 214.96 proteins per sample) (Figure 2A; Supplementary Table S2). To obtain a more representative proteome, we filtered the identified proteins to retain only those detected in a minimum of 70% of the samples within at least one analytic group, resulting in 257 proteins selected for downstream analysis (Supplementary Table S2). To assess whether the identified proteomic profile was consistent with the canonical content of human extracellular vesicles, we compared our protein list with the annotated proteome found in Vesiclepedia and ExoCarta databases. As shown in Figure 2B, only 10 proteins found here were not previously annotated in extracellular vesicles (Supplementary Table S3). Apolipoproteins represented 12 out of the 258 (4.6%) total identified proteins after filtering (Supplementary Table S2), which indicates a small percentage of contamination with plasma proteins.

Notably, the identified proteome included classic EV-associated proteins such as the chaperones HSP90AA1, HSP90AB1, HSPA5, and HSPA8 as well as A2M, C3, ANXA1, ANXA2 (Figure 2C). Additionally, we performed functional enrichment analysis using GO terms, identifying cellular compartments and molecular functions that were consistent with those expected for the EV proteome (Figure 2D), such as blood microparticles, vesicle lumen and secretory granules and vesicles.

To determine whether the total proteome was able to segregate the groups of samples, a PCA and an unsupervised hierarchical clustering (Supplementary Figure S1A,B) were performed. No clear separation of the groups could be observed, which highlights the subtle differences between the groups of patients and the challenges to identify differentially expressed proteins across groups to better capture the proteomic signatures associated with distinct clinical and biological behaviors in CML.

To gain a broader understanding of how the EV-proteome may impact the response to treatment in CML patients, the EV-proteomes of the good response (GTR), poor response (PTR) and healthy controls (HC) groups were compared (FDR <0.05), which led to the identification of 42 DEPs (Figure 3A; Supplementary Table S2). In the PCA analysis, PTR samples segregated from the HC and GTR samples, which clustered closely together (Figure 3B). The distribution of DEPs across comparisons is illustrated in Figure 3C. Most DEPs showed a decreased expression in the PTR group when compared to the HC samples. Additionally, DEPs such as DEFA3, FERMT3, TPM4, HSPA5, SLC4A1, and NME1 were downregulated in CML x HC samples independently of the treatment response level.

The functional enrichment and GSVA analyses (FDR value <0.05; Supplementary Table S4), allowed the identification of significantly enriched cancer-related pathways, including cell cycle, autophagy, glycolysis, aggrephagy, hypoxia, epithelial-to-mesenchymal transition (EMT), and cytokine signaling in the immune system. Key proteins involved in these processes included the tubulin isoforms TUBA1B, TUBB1, and TUBB4B, as well as the tyrosine 3-monooxygenases YWHAE, YWHAG, and YWHAZ, along with FLNA, VIM, HSPA5, and LDHA (Figure 3D). Other processes included Rho-GTPases, mTORC1, and ALK signaling pathways (MYH9, MYL6, VCL, TPM3, and TPM4), chaperone-mediated autophagy (EEF1A1, HSP90AB1, and VIM), and oxidative phosphorylation (LDHA, LDHB, ATP5A1, ATP5B). Notably, all these processes showed higher GSVA scores in HC samples compared to CML samples, with the lowest GSVA scores observed in the PTR samples, suggesting a decline in the expression of these EV proteins in samples with a poor response to leukemia treatment.

A protein-protein interaction (PPI) analysis demonstrated a significant PPI enrichment (PPI enrichment score = 1.11e-16) associated with the 42 DEPs. The most interconnected DEPs in the PPI were VCL, TMP3, TPM4, TLN1, HSP90AB1, and MYH9 (Figure 3E; Supplementary Table S5).

Following the initial assessment of proteome profiles in EVs from HC samples and CML samples with distinct treatment responses, we extended the analysis to include the T315I and TFR groups. This approach deepens the evaluation of proteomic alterations across distinct prognostic profiles, capturing the potential impact of the T315I mutation and proteomic particularities associated with treatment-free remission events.

We identified 50 DEPs across groups (FDR <0.05; Supplementary Table S2), with the majority observed in the PTR samples compared to the other groups (Figure 4A). Consistently, the PCA analysis showed a clear separation of PTR samples, while the remaining groups clustered together (Figure 4B). A cluster of 23 DEPs, including FLNA, HSPA5, HSPA8, MDH2, PKM, and STOM, presented a significant lower-expression in the PTR group compared to the T315I patients, while a 10-DEP cluster, including DEPs such as C1R, VIM, CKB, TUBA1B and TUBB4B, presented lower-expression in PTR compared to all other sample groups simultaneously. Also, a cluster composed by FABP5 and IGHM presented lower expression in the GTR samples compared to the TFR and T315I patients (Figure 4C).

To better elucidate if proteomic alterations are associated with the treatment remission profile in chronic myeloid leukemia, we compared the proteomes of TFR, good (GTR) and poor treatment (PTR) response groups.

The comparison of TFR and GTR revealed 29 DEPs (Figure 5A; Supplementary Table S2). IGKV1-5, CFP, C3, C4B, and CPN2 were enriched in the innate immune-system, complement cascade, and neutrophil degranulation processes; DSG1 and DSP, which were associated with apoptotic execution phase and cleavage of adhesion proteins, and F10, F11, ARG1, and SERPINA6, which were enriched in xenobiotic metabolism (Figures 5B,C; Supplementary Table S4). The TFR x GTR PPI network (PPI enrichment score <1.0e-16) presented C3, SERPINC1, F9, and F11 as the more interconnected nodes (Figure 5D; Supplementary Table S5).

By comparing TFR and PTR samples we identified 37 DEPs, with prevalence of over-expressed DEPs in the TFR group (Figure 6A; Supplementary Table S2). These proteins were enriched in processes such as MAPK2 and MAPK activation (VCL, TLN1, and RAP1B), hypoxia and glycolysis (PGK1, TPI1, PPIA, and PKM), aggrephagy and autophagy (TUBAB1, TUBB4A, VIM, HSPA8, HSP90AB1), cell cycle (YWHAZ, H4C1, H2AC14, H3-3B), and signaling by WNT, MTORC1, and ALK (PGK1, TPI1, PPIA, YWHAZ, CLTC, TPM4). The TFR x PTR PPI network (PPI enrichment score = 6.25e-11) included chaperones such as HSP90AB1, HSPA8, and HSPA5, besides H4C6, VCL, TPM4, and TUBA1B as its more interconnected nodes (Figure 6D; Supplementary Table S5).

DNA and histone methylation processes and chromatin modification (H3-3B, H2BC15, H4C1, H2AC14), KRAS and PI3K-AKT-mTOR signaling (F13A1, CFHR2, PPBP, CLTC, CFL1), as well as apical junctions and GAP-junction trafficking (MYH9, VCL, TUBB4A, TUBA1B, TUBB1) (Figures 6B,C; Supplementary Table S4) were enriched in TFR x PTR. The TFR x PTR PPI network (PPI enrichment score = 5.77e-15) included chaperones such as HSP90AB1 and HSPA8, besides FLNA, VCL, TLN1, and HIST1H4A as its more interconnected nodes (Figure 6D; Supplementary Table S5).

In order to determine the impact of the T315I mutation on the EV-proteome, we compared the proteome of paired patients before and after acquisition of the mutation. As shown in Figure 7A, we identified 17 DEPs (Supplementary Table S2). APOC1, COL1A1, DEFA3, and SSC5D presented the most conspicuous expression increase after mutation acquisition, while LBP, C2, CBP2, and F11 the most evident expression decrease after mutation (Figure 7B).

These DEPs were enriched in processes such as scavenging receptor functioning (COL1A1, APOE, SSC5D), NR1H2 and NR1H3 signaling (APOE, APOC1), plasma lipoprotein clearance (APOE, APOC1), and complement cascade (C1QB, C1R, C2, CPB2,C4BPB), among others (Figures 7C,D; Supplementary Table S4). The PPI network associated with these DEPs presented a significant PPI enrichment score (PPI enrichment score = 6.97e-07) (Figure 7E; Supplementary Table S5).