Identification of the metabolic alterations associated with the multidrug resistant phenotype in cancer and their intercellular transfer mediated by extracellular vesicles

Multidrug resistance (MDR) is a serious obstacle to efficient cancer treatment. Overexpression of P-glycoprotein (P-gp) plays a significant role in MDR. Recent studies proved that targeting cellular metabolism could sensitize MDR cells. In addition, metabolic alterations could affect the extracellular vesicles (EVs) cargo and release. This study aimed to: i) identify metabolic alterations in P-gp overexpressing cells that could be involved in the development of MDR and, ii) identify a potential role for the EVs in the acquisition of the MDR. Two different pairs of MDR and their drug-sensitive counterpart cancer cell lines were used. Our results showed that MDR (P-gp overexpressing) cells have a different metabolic profile from their drug-sensitive counterparts, demonstrating decreases in the pentose phosphate pathway and oxidative phosphorylation rate; increases in glutathione metabolism and glycolysis; and alterations in the methionine/S-adenosylmethionine pathway. Remarkably, EVs from MDR cells were capable of stimulating a metabolic switch in the drug-sensitive cancer cells, towards a MDR phenotype. In conclusion, obtained results contribute to the growing knowledge about metabolic alterations in MDR cells and the role of EVs in the intercellular transfer of MDR. The specific metabolic alterations identified in this study may be further developed as targets for overcoming MDR.

KEGG pathway enrichment analysis indicated that the most significant active pathways enriched in the MDR cells were involved in metabolic processes. Following the identification of the DEPs, the most significant active pathways were analyzed using KEGG pathway enrichment analysis. Results indicated that the most significantly active pathways enriched in the MDR cells were those involved in metabolic processes. The glutathione metabolism (GSH), pentose phosphate pathway (PPP) and glycolysis were found to be the most enriched pathways identified in both cancer cell models (Table 1). In the chronic myeloid leukemia model (K562Dox) the following 4 DEPs were found to be involved in the GSH metabolism pathway (G6PD, PRDX2, IDH1 and 6PGD), 3 DEPs in the PPP pathway (G6PD, ALDOC and 6PGD) and 2 DEPs in the glycolysis pathway (ALDOC and PKM2). In the non-small cell lung cancer model the following 5 DEPs were found to be involved in the GSH metabolism pathway (G6PD, PRDX2, IDH1, MGST1 and 6PGD), 4 DEPs in the PPP pathway (G6PD, TALDO1, TKT and 6PGD) and 1 DEP involved in the glycolysis pathway (ALDH3A1) ( Table 1).
Other DEPs involved in MDR and metabolic processes were identified by bioinformatics analysis.
In addition to the DEPs referred above (Table 1) other DEPs also involved in metabolic process and MDR were Scientific RepoRts | 7:44541 | DOI: 10.1038/srep44541 analyzed in terms of normalized abundance between MDR and drug-sensitive counterpart cells (for both cancer cell models) ( Table 2).
The DEP with the highest fold change was P-gp, which was upregulated in both MDR cells with a fold change of 18.8 for K562Dox vs. K562 and 64.5 for NCI-H460/R vs. NCI-H460. In terms of metabolic processes, most of the DEPs involved in the PPP pathway were downregulated in the two MDR models (ALDOC, G6PD, 6PGD and TKT) and only TALDO1 was upregulated in the NCI-H460/R cells (Table 2). Regarding GSH metabolism, the enzymes responsible for the NADP + reduction to NADPH (G6PD, 6PGD and IDH1) were downregulated in the MDR cells but MGST1 was upregulated in NCI-H460/R cells. Regarding glycolysis, ALDOC was downregulated in MDR cells but the enzyme responsible for the pyruvate production (PKM2) was upregulated. Additionally, DEPs involved in the TCA cycle were also identified (ACLY and ACO2). ACLY was upregulated whereas ACO2 was downregulated in the MDR cancer cell models (Table 2).
Interestingly HYOU1 and NDRG1 (both involved in cellular response to hypoxia) and MTHFD1 (an enzyme involved in the methionine pathway) were upregulated in the MDR cells (Table 2).
Results obtained by quantitative label-free proteomic approach were validated by Western blot. The expression of P-gp, G6PD, 6PGD and IDH1 were further validated by Western blot. We focused the validation on DEPs involved in MDR, glutathione metabolism and pentose phosphate pathway. Consistent with the results obtained with the proteomic analysis, P-gp was found upregulated while G6PD, 6PGD and IDH1 were downregulated in the MDR cells (K562Dox and NCI-H460/R) when compared with their drug-sensitive counterpart cells (K562 and NCI-H460) (Fig. 2).

MDR cells presented higher levels of reduced glutathione (GSH) and lower levels of reactive oxygen species (ROS).
Since most of the proteins that had been identified as DEPs in MDR cells were related to metabolic processes such as GSH metabolism and PPP, the GSH (one of the main detoxification agents in the cell) and ROS levels were analyzed in both tumor models.
MDR cells (K562Dox and NCI-H460/R) presented significantly higher levels of GSH when compared to their drug-sensitive counterpart cells (K562 and NCI-H460) (Fig. 3A,B). In agreement with this higher capacity provided by the higher levels of GSH to respond to oxidative stress, MDR cells presented lower levels of ROS when compared to their drug-sensitive counterpart cells (Fig. 3C,D).
Changes in methionine/S-adenosylmethionine pathway were identified in the MDR cancer cells. As we found GSH significantly increased in MDR cells, we were also interested in studying the metabolic pathways involved in the biosynthesis of GSH. The main cellular pathway that generates precursor for the synthesis of GSH corresponds to the methionine/S-adenosylmethionine (SAMe) pathway. By UPLC-MS metabolomics 21 we have quantified metabolites belonging to this pathway in the MDR and their drug-sensitive counterpart cells (Table 3 and Table S3). Methionine was upregulated in both MDR cancer cells. On the contrary, S-adenosylhomocysteine (SAH) and spermidine were downregulated in both MDR cancer cell lines (K562Dox and NCI-H460/R) ( Table 3). In addition, no alterations in S-adenosylmethionine (SAMe) were found in both cancer cell models (Table 3). Some other metabolites were found altered (such as (MTA, dcSAMe, betaine, choline and serine) but in different manners between the two cancer models.
Furthermore, the methylation index (SAMe/SAH), which represents the methylation capacity of the cells, was higher in MDR cells. In the chronic myeloid leukemia model, K562 cells had a methylation index of 6.97, whereas the methylation index for K562Dox cells was 12.05. In the non-small cell lung cancer model, NCI-H460 cells had a methylation index of 2.43 whereas the methylation index for NCI-H460/R cells was 9.79. MDR cells (NCI-H460/R) showed an increase in the non-glycolytic acidification, glycolysis, glycolytic capacity and glycolytic reserve. In order to gain new insights into the metabolic phenotype of MDR cells, the extracellular acidification rate (ECAR), which reflects the rate of glycolysis, and the oxygen consumption rate (OCR), which reflects the rate of oxidative phosphorylation (OXPHOS), were measured in both MDR and sensitive cancer cells. Measurements were performed in both basal cellular state and after treatments with compounds capable of modulating glycolysis and OXPHOS.
To analyze the glycolytic function, a glycolysis stress test was performed. The ECAR was measured in cells treated sequentially with: their glucose-free assay media as a control (Fig. 4A, blue line A); then with glucose to allow the cells to enter into glycolysis (Fig. 4A, blue line B); next with olygomycin (an ATP coupler) in order to inhibit ATP synthesis by blocking the proton channel and shifting the energy production to glycolysis (Fig. 4A, blue line C); and finally with 2-deoxyglucose (2-DG) which is a glucose analog that inhibits glycolysis through competitive binding to glucose hexokinase (Fig. 4A, blue line D). The resulting data allowed calculating the following parameters: Non-glycolytic acidification (= last rate measurement prior to glucose injection); Glycolysis (= maximum rate measurement before oligomycin injection -last rate measurement before glucose injection); Glycolytic capacity (= maximum rate measurement after oligomycin injection -last rate measurement before glucose injection); Glycolytic reserve (= glycolytic capacity -glycolysis).
In terms of glycolytic function, the MDR cells (NCI-H460/R) showed a statistically significant increase in the non-glycolytic acidification, glycolysis, glycolytic capacity and glycolytic reserve, when compared to their drug sensitive counterpart cells (NCI-H460) (Fig.4A).
To analyze the mitochondrial respiration, a mitochondrial stress test was performed. The OCR was measured during cells' treatment with: corresponding assay media (Fig. 4B, blue line A); then with olygomycin ( Fig. 4B, blue line B); next with FCCP (electron transporter chain accelerator) that is an uncoupling agent which disrupts ATP synthesis leading to the collapse of the mitochondrial membrane potential and causing rapid consumption of energy and oxygen without the generation of ATP (Fig. 4B, blue line C); and finally with rotenone Scientific RepoRts | 7:44541 | DOI: 10.1038/srep44541 (complex I inhibitor) which shuts down mitochondrial respiration by preventing the transfer of electrons from complex I to coenzyme Q (Fig. 4B, blue line D). These modulators allowed to calculate the following parameters: Non-mitochondrial respiration (= minimum rate measurement after rotenone injection); Basal respiration (= last rate measurement before first injection -non-mitochondrial respiration rate); Maximal respiration (= maximum rate measurement after FCCP injection -non-mitochondrial respiration); Proton leak (= minimum rate measurement after olygomycin injection -non-mitochondrial respiration); ATP production (= last rate measurement before olygomycin injection -minimum rate measurement after olygomycin injection); Spare capacity (= maximal respiration -basal respiration).
Regarding the mitochondrial respiration, the MDR cells showed a statistically significant decrease in the non-mitochondrial respiration, basal respiration, maximal respiration, proton leak, ATP production and spare capacity (Fig. 4B).
PPP inhibitor changed the metabolic phenotype of NCI-H460 cells towards the phenotype of their MDR counterparts (NCI-H460/R). As aforementioned, sensitive cells possess more rate limiting enzymes of the PPP than their corresponding MDR cells (Tables 1 and 2 and Fig. 2). In addition, they showed a decrease in glycolytic parameters (Fig. 4A) and an increase in mitochondrial respiration (Fig. 4B). In order to confirm that the PPP is involved in the MDR phenotype, NCI-H460 cells (drug-sensitive) were treated with a  sub-lethal concentration of dichloroacetate (DCA, a PPP inhibitor) and consequent alterations in glycolysis and OXPHOS were observed. After DCA treatment, the drug-sensitive cells acquired a metabolic phenotype similar to the one observed in the MDR cells, presenting an increase in glycolysis, glycolytic capacity and glycolytic reserve, and a decrease in their basal respiration, maximal respiration and ATP production by OXPHOS (Fig. 5).
The MDR cells (NCI-H460/R) treated with DCA showed no changes in glycolytic function and mitochondrial respiration (data not shown).

P-gp inhibitor changed the metabolic phenotype of MDR cells (NCI-H460/R) towards the phenotype of their corresponding sensitive cells (NCI-H460).
In order to confirm the possible influence of P-gp in the observed metabolic alterations of the MDR cells, the NCI-H460/R were treated with a sub-lethal concentration of a well-known P-gp inhibitor, verapamil, and the levels of glycolysis and OXPHOS were analyzed.  Table 2. DEPs involved in MDR and metabolic processes. a MDR cells/Drug-sensitive cells.

Figure 2. Validation by Western blot analysis of some DEPs initially identified by quantitative proteomics.
Representative blots were chosen from three independent experiments. Actin was used as a loading control.
Scientific RepoRts | 7:44541 | DOI: 10.1038/srep44541 After treatment with verapamil, the MDR cells acquired a metabolic phenotype more similar to the one observed in sensitive cells, presenting a decrease in glycolysis, glycolytic capacity, glycolytic reserve and non-glycolytic acidification, and an increase in their basal respiration (Fig. 6).

PPP inhibitor increased resistance of sensitive cells (NCI-H460) to doxorubicin.
To evaluate the possible contribution of PPP to the MDR phenotype, the NCI-H460 cells were pre-treated with a sub-lethal concentration of PPP inhibitor -DCA and then subjected to doxorubicin treatment (37.5 nM). The relative cell number was assessed and the ratio between sensitive cells treated with DCA and without DCA was calculated. The sensitive cells pre-treated with DCA were more resistant to doxorubicin compared to sensitive cells without DCA pre-treatment (Fig. 7A). The increase in doxorubicin resistance was also observed when cells were treated with higher concentrations of doxorubicin (75 nM and 150 nM) (data not shown).

P-gp inhibitor sensitized MDR cells (NCI-H460/R) to doxorubicin. To confirm the P-gp involvement
in the MDR phenotype, the NCI-H460/R cells were pre-treated with a sub-lethal concentration of P-gp inhibitorverapamil and then subjected to doxorubicin treatment (37.5 nM). The relative cell number was assessed and the ratio between MDR cells treated with verapamil and without verapamil was calculated. The MDR cells pre-treated with verapamil were more sensitive to doxorubicin compared to MDR cells without verapamil pre-treatment (Fig. 7B). The sensitization of MDR cells to doxorubicin was also observed with other concentrations of doxorubicin in a dose dependent manner (75 nM and 150 nM, data not shown).

EVs from MDR cells were able to transfer their metabolic phenotype to the sensitive cells. Next, we wanted to verify if MDR phenotype could be transferred to the drug-sensitive cells via EVs
shed by the MDR cells. Therefore, drug-sensitive cells (NCI-H460) were treated with EVs isolated from the counterpart pairs of non-small cell lung cancer cells (NCI-H460 and NCI-H460/R) and chronic myeloid leukemia cells (K562 and K562Dox) (Fig. S1). After 15 h incubation, the drug-sensitive cells treated with the EVs shed by MDR cells (NCI-H460/R - Fig. 8A and K562Dox - Fig. 8B) acquired a metabolic phenotype more similar to the MDR cellular phenotype, i.e., an increase in glycolysis and in glycolytic capacity.

Discussion
Multidrug resistance (MDR) is a major problem in cancer treatment, responsible for chemotherapy failure 4 . MDR cancer cells usually have multiple mechanisms of resistance 14 . One of the most frequent MDR mechanisms is the overexpression of ABC transporters, such as P-gp 2 . Recent findings pointed that altered metabolic pathways help cancer cells to proliferate, adapt their metabolism to nutrient limited conditions, and importantly develop drug resistant phenotypes 22 . Therefore, targeting cellular metabolism could chemosensitize MDR cells 9,23,24 . Understanding the metabolic adaptations of MDR cancer cells is important for the identification of new approaches to counteract this phenotype.
In the present study, we have identified a complex network of metabolic alterations associated with the MDR phenotype that could lead to the identification of more efficient therapeutic MDR circumvention strategies. In addition, we have shown that EVs released by MDR cells are capable of stimulating a metabolic alteration (towards a MDR phenotype) in recipient drug-sensitive cancer cells.
We confirmed that comparative proteomic approach is a powerful tool to investigate MDR mechanisms in cancer cells. MDR and their corresponding drug-sensitive cell lines from two distinct models (chronic myeloid leukemia and non-small cell lung cancer) were subjected to a label-free LC-MS quantitative proteomics. The obtained data allowed a comparison between the proteome from MDR cells and the one obtained from their drug-sensitive counterparts. In terms of biological processes, most of the identified DEPs (differentially expressed proteins between drug-sensitive and MDR cells) were involved in metabolic processes and the most active pathways enriched were those involved in cellular metabolism (GSH, PPP and Glycolysis). Therefore, we showed that the two MDR cell models acquired a similar metabolic profile but significantly distinct from the metabolic profile observed in their corresponding drug-sensitive cells.
Interestingly, most of the DEPs involved in the PPP (including the rate limiting enzyme, G6PD) were downregulated in the MDR cells. PPP is a pivotal biosynthetic pathway branched to glycolysis and one of the main antioxidant cellular defense systems 25 . PPP has a central role as a source of nucleic acid precursors and provides reducing power and ribose phosphate to the cell 26 . Changes in the PPP activity can affect response to anticancer drugs, however the specific role of PPP in MDR phenotype is still unclear. Other authors implied that PPP is more active in MDR tumors 27,28 . Moreover, evidence suggested that elevated levels of NADPH and GSH, together with an active PPP, play an important role in MDR 29,30 . Our results together with other recent studies 31,32 contradict the above-mentioned reports, showing that PPP enzymes are downregulated in MDR cells. These results are controversial since one of the roles of PPP is to provide reducing power to the GSH metabolism and high levels of GSH in tumors have been linked to the development of MDR 33 . In addition, we found increased GSH levels in both MDR models, as previously published by other authors 34,35 . Moreover, since GSH is a detoxification agent 36 , the levels of ROS in the MDR cells were decreased. In all, our results suggest that MDR cells are capable of maintaining high GSH levels even when PPP is downregulated. Therefore, we assumed that MDR cells upregulated another source of GSH.
The methionine cycle is a key pathway for many methylation reactions (methylation of DNA, histones and non-histone proteins, such as transcription factors) and could be the source of cysteine residues necessary for GSH synthesis 37 . Growing evidence links aberrant regulation of methylation to tumorigenesis 38 . The epigenetic mechanisms underlying drug resistance have not been fully elucidated, although some studies have suggested the contribution of an altered chromatin state to drug resistance 39 . Our results showed considerable differences between MDR and their drug-sensitive cells, in the amount of several metabolites of the methionine/SAMe  pathway. Methionine levels were significantly increased in the MDR cells, which could lead to the GSH increase observed in those cells. In addition, the methylation capacity of the cells was increased in both MDR models and this could be attributed to the metabolic alterations in MDR cells during development of their resistant phenotype. In agreement with our results, other authors have described that treatment with a methylation inhibitor reversed drug resistance indicating that the development of some cases of drug resistance could be methylation-dependent 40 . Also, MTA, dcSAMe, betaine, choline and serine are altered in MDR cells, although not in a similar manner. These differentially altered metabolites indicate that among these two MDR models there are also metabolic differences, which enhance the idea of the complex metabolism associated to MDR. The energy dependence of drug transport in MDR cells suggests that during development of MDR, cells undergo alterations in the energy utilization pathways. The main cell energy carrier, ATP, is produced in two metabolic pathways: glycolysis and/or oxidative phosphorylation (OXPHOS) of metabolic fuels 22 . One of the fundamental hallmarks of cancer cells is the shift in the balance between these two energy-production pathways, in favor to glycolysis, known as the Warburg effect [41][42][43] . For this reason, glycolysis inhibition has attracted significant interest as a possible way to sensitize cancer cells to chemotherapy. Indeed, several studies in different cancer models have demonstrated an efficient suppression of MDR by glycolytic inhibitors 8,23,24,44,45 . Moreover, during the course of this work another study has suggested the possibility of an accelerated process of glycolysis in MDR cells to increase their energy supply 46 . To our knowledge, a comparison between MDR and drug-sensitive cells in terms of metabolic energy supply, using different modulators of glycolysis and OXPHOS, has never been performed. The results presented herein showed a statistically significant increase in glycolysis and in the glycolytic capacity of the MDR cells (NCI-H460/R), together with a decrease in the mitochondrial basal respiration and in the maximal capacity of the cells to perform OXPHOS compared to the drug-sensitive counterparts (NCI-H460). Therefore, our findings indicate that MDR cells, when acquiring the resistant phenotype, enhance their switch in the cellular energy supply pathways from OXPHOS to glycolysis. This altered metabolic activity in MDR cells could be crucial for: (i) supporting uncontrolled proliferation, since glycolysis provides the intermediates necessary for biosynthetic pathways; (ii) allowing the use of the most abundant extracellular nutrient (glucose) to produce abundant ATP. Even though the yield of ATP per glucose consumed is low, if the glycolytic flux is high enough, the percentage of cellular ATP produced by glycolysis can exceed the one produced by OXPHOS 47,48 . Therefore, this metabolic phenotype could be beneficial for MDR cells at both levels of bioenergetics and biosynthesis.
Although MDR cells perform less OXPHOS, the application of different OXPHOS modulators (oligomycin, FCCP and rotenone) affected more the drug-sensitive than their MDR counterpart cells. The OCR levels in the MDR cells have been less affected than the OCR levels in drug-sensitive cells. This suggests that MDR cells are more capable of sustaining the acquired metabolic phenotype in order to maintain the higher energy demand. In accordance, it was reported that the relationship between drug resistance and glycolysis may partially be due to the radical scavenging potential of the glycolytic intermediates, and the link between them and the cellular redox status 49 .
Furthermore, the increase in glycolysis could be responsible for the decrease in the PPP observed in the MDR cells. Since some of the glycolysis and PPP intermediates are the same, it is possible that MDR cells are directing these intermediates mostly to glycolysis, in order to sustain the energy for the requested biosynthetic pathways. In fact, we demonstrated that when the PPP pathway was inhibited (with DCA 50 ), drug-sensitive cells (NCI-H460) acquired a metabolic phenotype similar to that observed in MDR cells by increasing glycolysis and decreasing OXPHOS. Also, PPP inhibition increased resistance of sensitive cells to doxorubicin, when compared to sensitive cells without PPP inhibition. These results support the hypothesis that there is a competition between glycolysis and PPP activity in the MDR phenotype and also that modulation of the PPP and glycolytic pathways is capable of altering the sensitivity of cells to drugs. Additionally, in multicellular tumor spheroids and in doxorubicin-resistant human breast adenocarcinoma cells, the inhibition of glycolysis raised intracellular ROS, downregulated P-gp and reverted the MDR phenotype 8,23 . In agreement with these data, our work showed that treatment of MDR cells (NCI-H460/R) with a P-gp inhibitor (verapamil) switched their metabolic phenotype to that characteristic of sensitive cells (decrease in glycolysis and an increase in OXPHOS) and also promoted sensitization of the MDR cells to doxorubicin.
Several studies have associated metabolic alterations with the shedding and cargo of EVs in different types of cells (such as increase lactate production 18 , inhibition of glutamine metabolism 19 and the presence of several enzymes involved in glucose and glutamine metabolism 20 ). The importance of EVs in the transfer of the MDR phenotype has been recently described 13,14,51 . Therefore, the acquisition by MDR cells of a different metabolic phenotype, could be responsible for the differences in: (i) the type of EVs population released by MDR cells; (ii) their cargo and (iii) the capacity to transfer phenotypes to receiving cells. In fact, we have recently shown that MDR cells shed a different population of EVs (more microvesicle-like EVs and less exosomes) when compared to the EVs shed by the drug-sensitive counterpart cells 17,52 . Additionally, our results showed that drug-sensitive cells (NCI-H460) co-incubated with EVs from MDR cells (NCI-H460/R and K562Dox), acquired a metabolic phenotype (increase in glycolysis) similar to the one observed in MDR cells. These results showed that, independently of the origin of the MDR cells (leukemia or lung cancer), their EVs are capable of inducing such alterations. However, surprisingly, our results showed that the EVs from the sensitive leukemic cell line were also capable of inducing alterations in the glycolytic capacity of the recipient drug-sensitive cells. The reason for this is unknown, but is possibly due to the exposure of cells to high levels of EVs, therefore not related to MDR. Furthermore, a recent study has described that glutamine metabolism was altered in cancer cells following incubation with large EVs (microvesicles), an effect that was not observed upon incubation with exosomes 20 . This further highlights the importance of verifying the type of EVs released by donor cells, since different EVs may transfer different phenotypes to receiving cells.

Conclusions
In conclusion, P-gp overexpressing MDR cells may employ various protective metabolic strategies to survive (schematically represented in Fig. 9). These include (i) alterations in the GSH metabolism, (ii) increasing the methylation index influencing epigenetic regulation, (iii) increasing the rates of glycolysis and (iv) changing the phenotype of the surrounding drug-sensitive cells, by EVs mediated transfer of new features. Our results indicate that the development of MDR is a complex phenomenon that involves several simultaneous metabolic alterations.
We clarified for the first time the complex metabolic network of the various metabolic alterations associated with MDR in cancer cells. In addition, that MDR cancer cells are more capable than drug-sensitive cells of sustaining its specific metabolic profile. Specifically, we showed for the first time that MDR cancer cells, besides increased glycolysis also have increased glycolytic capacity and glycolytic reserve, while their OXPHOS rate is decreased. In addition, our work found differences between MDR and drug-sensitive cells in the amount of several metabolites of the methionine/SAMe pathway which regulates DNA and protein methylation, as well as GSH production. Finally, we demonstrated for the first time the transfer of metabolic information from MDR to drug-sensitive cancer cells through a specific population of EVs.

Materials and Methods
Cell culture. The chronic myeloid leukemia cell line K562 was from European Collection of Cell Cultures (ECACC) and its P-gp overexpressing counterpart cell line K562Dox was a kind gift of Dr. J.P. Marie (Paris, France) 53,54 . The non-small cell lung cancer cell line NCI-H460 and its drug-resistant P-gp overexpressing counterpart cell line NCI-H460/R were a kind gift from Dr. M. Pešić (Belgrade, Serbia) 55,56 . All cell lines were genotyped and routinely monitored for mycoplasma contamination by PCR (VenorGeM ® Advance Mycoplasma Detection Kit, Minerva). All cells were routinely grown in RPMI-1640 (with Ultraglutamine I and 25 mM HEPES) medium (Lonza), supplemented with 10% fetal bovine serum (FBS, PAA) at 37 °C in a humidified incubator with 5% CO 2 in air. Cell number and viability were analyzed with trypan blue exclusion assay. All experiments were carried out with exponentially growing cells having over 90% viability.

Sample preparation and mass spectrometry using LC/MS/MS. Pellets from cells (K562, K562Dox,
NCI-H460 and NCI-H460/R) were processed and 10 μ g of proteins were analyzed by mass spectrometry using LC/MS/MS as previously described 17 . Biological replicates (from 4 independent experiments) were analyzed for each sample type.
Label-Free LC-MS quantitative profiling. Label-free LC-MS analysis was carried out using Progenesis QI for proteomics v4.1 software (NonLinear Dynamics, UK), essentially as recommended by the manufacturer (see www.nonlinear.com for further background to alignment, normalization, calculation of peptide abundance, etc.).
This software extracts quantitative information from MS1 data by aligning the data based on the LC retention time of each sample to a reference file (sample run that yielded most peptide ions); this allows for any drift in retention time, giving an adjusted retention time for all runs in the analysis.
Results were filtered, based on statistical analysis. The Progenesis peptide quantification algorithm calculates peptide abundance as the sum of the peak areas within its isotope boundaries. Each abundance value is then transformed to a normalized abundance value by applying a global scaling factor. Protein abundance was calculated as the sum of the abundances of all peptide ions, which have been identified as coming from the same protein. Any peptides with an ANOVA score of p > 0.05 were eliminated. The MS2 data for the remaining peptides was exported and the resulting MGF file used to search the NewUniProtSwissprot database (updated in January 2014) on the MASCOT server (www.matrixscience.com) for protein identifications. The search parameters used were as follows: (1) species, Homo sapiens; (2) allowed number of missed cleavages, 2; (3) fixed modification, carboxylmethyl; (4) variable modifications, methionine oxidation; (5) peptide mass tolerance ± 20 ppm; (6) MS/ MS tolerance ± 0.6 Da; and (7) peptide charge + 1, + 2 and + 3. Only peptides with ion scores of 40 and above were considered and re-imported back into Progenesis QI software for further analysis. Peptide identifications were imported into the Progenesis software and assigned to the matching features. Protein scores were based on ANOVA values with a cut off of p < 0.05. Differentially expressed proteins (DEPs) between drug-sensitive and MDR cells with ≥ 2 peptides matched and a ≥ 1.5 fold differences in abundance were considered as significant. Peptide conflicts occur when a peptide is identified as present in more than one protein. These were resolved by assigning the peptide to the protein with the greater number of hits, a greater Mascot score or a lower mass error; when conflicts could not be clearly resolved, the peptide was excluded from the analysis.
Bioinformatics analysis of the detected DEPs. The two data sets (DEPs from K562 vs. K562Dox cells and NCI-H460 vs. NCI-H460/R cells) were analyzed using bioinformatics methods. Database for Annotation, Visualization, and Integrated Discovery (DAVID 6.7) was used to identify protein and molecular pathway modifications. UniProt accession numbers were obtained by Progenesis software. To understand the high-level functions and utilization of the biological systems from molecular-level information, the Kyoto Encyclopedia of Genes and  (2), alterations in the GSH metabolism (3), decreasing rates of PPP (4) and OXPHOS (5) and finally changing the phenotype of the surrounding drug-sensitive cells by EVs-mediated transfer of new features (6). Filled lines and arrows: increased pathways; Dashed lines and arrows: decreased pathways; Bold and bigger fonts: increased metabolic processes; Smaller fonts: decreased metabolic processes.
Genomes (KEGG) pathway was used. Distributions in subcellular locations, biological processes and molecular functions were assigned to each protein based on Gene Ontology (GO) categories. The significantly (p < 0.05) enriched categories are presented. The following secondary antibodies were then used: anti-mouse IgG-HRP; anti-rabbit IgG-HRP or anti-goat IgG-HRP (all diluted 1:2000; Santa Cruz Biotechnology). Signal was detected using the ECL Western blot Detection Reagents (GE Healthcare, UK), the Amersham Hyperfilm ECL (GE Healthcare, UK), and the Kodak GBX developer and fixer (Sigma, EUA) 57 . The intensity of the bands obtained in each film was further analyzed using the software Quantity One -1D Analysis (Bio-Rad, USA).

Analysis of Protein Expression by Western
GSH/GSSG-Glo TM assay. To measure GSH levels, the GSH/GSSG-Glo TM Assay (Promega, USA) was used following manufacturer's instructions. Briefly, 5000 cells/well were plated in white 96-opaque well plates. NCI-H460 and NCI-H460/R cells were plated in 100 μ l of RPMI-1640; K562 and K562Dox cells were plated in 20 μ l of Hanks Balanced Salt Solution (HBSS). Total glutathione lysis reagent or oxidized glutathione lysis reagent were added to the cells. After incubation at room temperature a luciferin generation reagent was added to all samples and the plates were incubated for 30 min at room temperature. Luminescence was quantified using a microplate reader (Biotek Instruments Inc. Synergy MX. USA). Positive control (cells treated with 5 μ M staurosporine for 4 h) and wells with no cells were also included as controls. GSH levels were calculated as follows: GSH levels = total glutathione levels -oxidized glutathione levels. Semi-quantitative analysis of metabolites of the methionine/S-adenosylmethionine pathway. Total cell pellets (4 independent preparations for each cell line) were lysed in 500 μ L of a mixture of ice-cold water/metanol/10 mM acetic acid (49/50/1 v/v/v%) with a tissue homogenizer (Precellys) in 1 × 20 second cycles at 6000 rpm. Subsequently, 400 μ L of the homogenate was transferred to a new aliquot and shaken at 1400 rpm for 30 minutes at 4 °C. Next the aliquots were centrifuged for 15 min at 14000 rpm at 4 °C. 75 μ L of the supernatant was transferred to a fresh aliquot and placed at − 80 °C for 20′ . The chilled supernatants were evaporated with a speedvac in approximately 3 h. The resulting pellets were resuspended in 100 μ L water/acetonitrile (MeCN)/formic acid (40/60/0.1 v/v/v%, resuspension solution).

ROS
Concentrations of methionine, MTA, SAMe, SAH, spermidine and spermine were determined with a semi-quantitative method. Calibration curves for these compounds were obtained, by measuring serial dilutions of a pooled standard mixture in resuspension solution. The concentrations for all compounds in the dilutions ranged from 100 μ M to 0.025 μ M. For the standard mixtures, separate 10 mM stocks of the standards were made. These were then pooled and further diluted in resuspension solution in order to obtain the final concentrations as used for the calibration curve.
In order to obtain a good separation of the analytes the following gradient was used: from 5% A to 50% A in 2.4 minutes in curved gradient (#8, as defined by Waters), from 50% A to 99.9% A in 0.2 minutes constant at 99.9% A for 1.2 minutes, back to 5% A in 0.2 minutes. The flow rate was 0.250 mL/min and the injection volume was 2 μ L. All samples were injected randomly. After every 10 injections a QC sample was injected. All samples were injected in duplicate.
Analytes were measured in enhanced duty cycle (EDC) mode, optimized for the mass of the analyte in question. MTA was measured in scan function 1 (EDC at 298), choline was measured in scan function 2 (EDC at 104), methionine was measured in scan function 3 (EDC at 150), SAH was measured in scan function 8 (EDC at 385), SAMe was measured in scan function 10 (EDC at 399), spermidine was measured in scan function 12 (EDC at 146), spermine was measured in scan function 13 (EDC at 203). The cone voltage was between 20 and 25 depending on the analyte. A 2 ng/mL leucine-enkephalin solution in water/acetonitrile/formic acid (49.9/50/0.1%v/v/v) was infused at 10 μ L/min and used for a lock mass which was measured each 47 seconds for 0.5 seconds. Spectral peaks were automatically corrected for deviations in the lock mass. Extracellular flux assay using Seahorse XF-24 Analyser. ECAR, reflecting the rate of glycolysis, and OCR, reflecting the rate of OXPHOS, were measured using a Seahorse Bioscience (Copenhagen, Denmark) XF24 analyzer. Briefly, 2 × 10 4 cells/well for glycolysis stress test and 3 × 10 4 cells/well for mitochondrial stress test were plated in Seahorse XF24 plates in 200 μ l of RPMI-1640 and incubated for 20-24 h at 37 °C in a humidified incubator with 5% CO 2 in air and 1 h hour prior to the XF assay in a humidified incubator without CO 2 with the corresponding assay media. In the experiments where the cells were treated with DCA, the treatment was performed for 24 h with a concentration of 5 mM. The verapamil treatment was performed for 15 h with 2 μ M. NCI-H460/R treated with DCA and NCI-H460 treated with verapamil were used as controls.
Steady-state (baseline) oxygen consumption rates and extracellular acidification rates were measured. Non-glycolytic acidification, glycolysis, glycolytic capacity and glycolytic reserve were measured, through ECAR, by injecting: glucose, oligomycin and 2-deoxyglucose. Basal respiration, proton leak, spare capacity, maximal respiration, non-mitochondrial respiration and ATP production were measured, through OCR, after the sequential injection of oligomycin, FCCP and rotenone. All measurements were normalized to protein quantity with crystal violet.
Sulforhodamine B assay. The Sulforhodamine B (SRB) assay was performed to measure cell growth indirectly, by measuring total protein content. The NCI-H460 and NCI-H460/R cells were seeded in 96-well plates (5000 cells/well), and pre-treated with 5 mM DCA for 24 h or with 2 μ M verapamil for 15 h, respectively. Cells were then treated in duplicate with 3 serial dilutions of doxorubicin (37.5 nM, 75 nM and 150 nM) for 48 h. Cells were fixed with 10% (w/v) ice cold trichloroacetic acid (TCA) and then washed with distilled water. After staining proteins with 0.4% (w/v) SRB, cells were washed with 1% (v/v) acetic acid, the bound SRB was solubilized with 10 mM Tris base and absorbance was measured at 510 nm in a multiplate reader (Synergy Mx, Biotek Instruments Inc.). The ratio between cells with a pre-treatment and cells without the pre-treatment was calculated. NCI-H460 treated with DCA only and NCI-H460/R treated with verapamil only were used as controls.
Co-culture of H460 cells with EVs. EVs were isolated from the culture media of drug-sensitive or MDR cells by various centrifugation steps as previously described 17 . EVs pellets were re-suspended in 100 μ l of PBS and frozen at − 80 °C. NCI-H460 cells were treated with 18 × 10 8 EVs (quantified by nanoparticle tracking analysis as in Supplementary Information) for 15 hours and were subsequently metabolically profiled using Seahorse XF-24 Analyzer.