Inhibition of glycosphingolipid biosynthesis reverts multidrug resistance by differentially modulating ABC transporters in chronic myeloid leukemias

Multidrug resistance (MDR) in cancer arises from cross-resistance to structurally- and functionally-divergent chemotherapeutic drugs. In particular, MDR is characterized by increased expression and activity of ATP-binding cassette (ABC) superfamily transporters. Sphingolipids are substrates of ABC proteins in cell signaling, membrane biosynthesis, and inflammation, for example, and their products can favor cancer progression. Glucosylceramide (GlcCer) is a ubiquitous glycosphingolipid (GSL) generated by glucosylceramide synthase, a key regulatory enzyme encoded by the UDP-glucose ceramide glucosyltransferase (UGCG) gene. Stressed cells increase de novo biosynthesis of ceramides, which return to sub-toxic levels after UGCG mediates incorporation into GlcCer. Given that cancer cells seem to mobilize UGCG and have increased GSL content for ceramide clearance, which ultimately contributes to chemotherapy failure, here we investigated how inhibition of GSL biosynthesis affects the MDR phenotype of chronic myeloid leukemias. We found that MDR is associated with higher UGCG expression and with a complex GSL profile. UGCG inhibition with the ceramide analog d-threo-1-(3,4,-ethylenedioxy)phenyl-2-palmitoylamino-3-pyrrolidino-1-propanol (EtDO-P4) greatly reduced GSL and monosialotetrahexosylganglioside levels, and co-treatment with standard chemotherapeutics sensitized cells to mitochondrial membrane potential loss and apoptosis. ABC subfamily B member 1 (ABCB1) expression was reduced, and ABCC-mediated efflux activity was modulated by competition with nonglycosylated ceramides. Consistently, inhibition of ABCC-mediated transport reduced the efflux of exogenous C6-ceramide. Overall, UGCG inhibition impaired the malignant glycophenotype of MDR leukemias, which typically overcomes drug resistance through distinct mechanisms. This work sheds light on the involvement of GSL in chemotherapy failure, and its findings suggest that targeted GSL modulation could help manage MDR leukemias.

This work sheds light on the involvement of GSL in chemotherapy failure, and its findings suggest that targeted GSL modulation could help manage MDR leukemias.
As a multifactorial sum of diseases, cancer presents great challenges to the development of safe, successful therapies. Distinct mechanisms employed by transformed cells to avoid toxicity generated by chemotherapy often cross-talk, leading to an adapted phenotype comprising both intrinsic and acquired drug resistance. Multidrug resistance (MDR) 4 is the main hurdle to chemotherapy success, as stress-adapted molecular mechanisms, including reduced influx, increased efflux, and accelerated metabolism of xenobiotics, work in tandem reducing the effective concentration at the molecular target (1). Efflux transporters such as ATP-binding cassette (ABC) proteins actively detoxify cells and tissues from both xenobiotics and toxic metabolites, playing major roles in MDR. Regardless of the diversity of ABC subfamilies and isoforms, two proteins are mostly associated with MDR, ABCB1 (P-glycoprotein) and ABCC1 (multidrug resistance protein 1, MRP1) (1,2). ABCB1 and ABCC1 share similarities and differences when cellular localization and substrate specificity are considered. The latter is mostly located on the plasma membrane; the first is present on any cell membrane; and both were shown to associate with microdomains such as lipid rafts depending on cell subtype (3,4). Both actively extrude a variety of nonrelated chemotherapeutic drugs, but ABCC1 is able to transport substrates independently from and in conjugation with glutathione (GSH) or in cotransport as well (5), playing complementary roles for the MDR phenotype. ABCB1 interacts with liposoluble or amphipathic molecules that are prone to accumulate in the intramembrane space (6), and ABCC1 exhibits higher affinity for negativelycharged glucuronates, sulfates, GSH-conjugated compounds, and products from lipid metabolism (7).
Sphingolipids such as ceramides and its phosphorylated or glycosylated forms are directly involved in cell fate, assuming active parts on either cell proliferation or death (8). Ceramides, whether originating from sphingomyelin remodeling or synthesized de novo on the endoplasmic reticulum, are transferred to cis-Golgi, where they are employed as substrates to UDP-glucose ceramide glucosyltransferase (UGCG) to form glucosylceramide (GlcCer), the precursor to all glycosphingolipids (GSL). Endogenous ceramides have been directly linked to cancer treatment given that chemotherapeutic agents with unrelated mechanisms, for example paclitaxel, daunorubicin, etoposide (9 -11), and the tyrosine kinase inhibitors sorafenib and imatinib (12), increase ceramide contents, which drive the intrinsic pathway of apoptosis through caspase activation or caspase-and p53-independent mitotic catastrophe (11,13). Second to their structural role on the organization of lipid rafts (14), GSL relates to development of drug resistance considering that cancer cells often present increased UGCG expression, being able to incorporate ceramides on GSL (15). Concerning MDR, a close cross-talk of ABCB1 and GSL has been observed; ABCB1 and UGCG were coincidently overexpressed in drugresistant breast, ovary, cervical, and colon cancer and on chronicmyeloidleukemias (16,17);GlcCerandglobotriaosylceramide (Gb3) positively regulate ABCB1 expression, respectively, through NF-B and Wnt/␤-catenin (17,18); and this transporter is able to act as a flippase on the transfer of GlcCer from the cis-Golgi to trans-Golgi during GSL biosynthesis (19). Despite its capacity of translocating sphingolipids such as sphingosine 1-phosphate (20) and GlcCer on polarized cells (21) and its coexpression with UGCG on colon cancer (22), a similar relationship involving ABCC1 activity and GSL is not clear.
Considering the diversity of mechanisms MDR cancer cells resort to in order to avoid and adapt to chemotherapeutic stress and the prime involvement of UGCG on the generation of GSL (23), the fate of endogenous ceramides is critical for successful cancer chemotherapy on a molecular level. Several studies evaluated the expression of ABCB1 and reversal of drug sensitivity on solid tumors and its association with GSL; nevertheless, our work focused on leukemic cells that express both ABCB1 and ABCC1, extending to the functional evaluation of those proteins after UGCG inhibition, which finds little coverage from the literature. In this context, we report the distinct ways ABCB1 and ABCC1 expression and activity were modulated after impairment of GSL biosynthesis on clinically relevant models of drug-resistant chronic myeloid leukemias.

MDR chronic myeloid leukemias overexpress UGCG along with a complex GSL profile, which is reverted after treatment with a ceramide analog
De novo ceramide synthesis on Golgi increases during stress, and cancer cells are able to up-regulate ceramide glycosylation ultimately changing GSL contents on cell membranes. To determine whether selection with standard chemotherapeutics would alter these processes on human leukemias, the expression of UGCG and the profiles of GSL and GM1 were evaluated on K562 cells (drug-sensitive) and on MDR derivatives Lucena-1 (K562/VCR) and FEPS (K562/DNR) cells. Results on Fig. 1, A and B, indicate that the glucosyltransferase that catalyzes the addition of UDP-glucose to ceramide, UGCG, is upregulated on the MDR leukemia models, notably on the FEPS line. Next, the three cell lines were treated with a specific UGCG inhibitor, the ceramide analog EtDO-P4, and their effects on viability and GSL contents are depicted on Fig. 1, C-E. Based on these results, a sub-toxic concentration of this inhibitor was employed for further assays. MDR cells showed increased contents and more complex profiles of GSL after TLC when compared with the parental, and treatment with 1 M EtDO-P4 for 24 h significantly reduced GSL expression on all three leukemias. Because total extraction does not distinguish the GSL on plasma or intracellular membranes, log-phase growing viable cells were stained with cholera toxin (CHT-FITC). CHT specifically binds GM1 on the extracellular leaflet, and an increase in its fluorescence could be observed on the MDR cells Lucena-1 and FEPS after flow cytometry. In accordance, UGCG inhibition reduced GM1 contents in 55 and 75% on these cells, contrasting to a reduction of 35% on drug-sensitive K562 (Fig. 1, F and G).

Pharmacological UGCG inhibition induces cytotoxicity and decreases drug resistance
Disruption of the GSL biosynthesis machinery could lead to accumulation of sphingolipid mediators, altering cell signaling and leading to either survival or death depending on the intrinsic properties of each cell subtype. Thus, the effects of UGCG inhibition on cellular viability are depicted on Table 1. Results indicate that the MDR phenotype and the increased UGCG expression translated into a modest resistance to EtDO-P4, since only FEPS presented higher viability and IC 50 after a 72-h treatment ( Fig. 1C and Table 1).
Given that UGCG is the first enzyme on the GSL biosynthesis pathway, inhibition of its activity would impair cell responses to stress due to accumulation of nonglycosylated ceramides. In this context, the results in Table 1 suggest that GSL depletion synergizes with the cytotoxicity caused by chemotherapeutic drugs with diverse mechanisms of action, as sub-lethal cotreatment with EtDO-P4 reduced the IC 50 for vincristine (VCR) and daunorubicin (DNR) on Lucena-1 and the IC 50 for DNR and cisplatin (CDDP) on FEPS cells. This reduction did not manifest on K562, which, in line with earlier results, suggests that GSL depletion showed minimal effects on sensitive cells. Noteworthy, treatment with exogenous N-hexanoyl-D-erythrosphingosine (C 6 -ceramide, C6-cer) induced the opposite effect to

UGCG inhibition down-regulates ABC proteins on MDR leukemias
EtDO-P4 on these cells. The lower IC 50 values for C6-cer on Lucena-1 and FEPS than on K562 suggest that this sphingolipid induces collateral sensitivity, a hypersensitivity toward secondary agents that arises from the development of resistance toward an unrelated primary drug.

UGCG inhibition reduces mitochondrial membrane potential (⌬ m ) and induces apoptotic cell death
Among sphingolipids, ceramides are linked to apoptotic cell death through the intrinsic pathway due to direct or indirect interaction with the mitochondrial permeability transition pore. To investigate whether the mitochondria would be involved on the cell death observed after GSL depletion, the MDR leukemias Lucena-1 and FEPS were treated with EtDO-P4, DNR, or C6-cer for 24 h and incubated with rhodamine 123 (Rho123). This fluorescent dye accumulates within energized mitochondria, and this retention is progressively lost as ⌬ m is reduced. Fig. 2, A and B, indicates that only 2 M EtDO-P4 significantly reduced mitochondrial Rho123 fluorescence, a similar result obtained after C6-cer treatment. As expected, A, UGCG expression was analyzed by Western blotting as described under "Experimental procedures." Representative images are from four independent extractions. B, band densities were quantified, and the amount of UGCG was calculated as the density of the UGCG band divided by the density of the ␤-actin band for each cell line. Bars represent the mean UGCG to ␤-actin ratios ϩ S.D. n ϭ 4. C, black, violet, or red bars, respectively, represent the mean percentages of cellular viabilities ϩ S.E. for K562, Lucena-1, or FEPS, as measured by MTT assay. Normalized data were from three independent experiments, with each concentration evaluated in triplicate. D and E, total lipids from 5 ϫ 10 7 cells were extracted, purified, resolved by TLC, and developed by resorcinol/HCl staining as indicated under "Experimental procedures." The migration pattern of a GSL standard containing a mix of GM1, GM2, and GM3 is indicated on the left of the image (STD). Controls (CTR) were treated with diluent. Representative images are from two independent extractions. F, K562, Lucena-1, and FEPS were incubated for 1 h with 5 g⅐ml Ϫ1 CHT-FITC, and the presence of the prototype GM1 was assessed by the MFI of viable cells in a flow cytometer. Representative histograms for the MFI of whole populations of treated (dashed lines) and untreated populations (continuous lines) are shown. G, scatter plots for the MFI were obtained from each cell line as follows: black circles for K562; squares for Lucena-1, and inverted triangles for FEPS untreated cells, and white ones indicate EtDO-P4 treatment. n ϭ 4. *, p Ͻ 0.05; **, p Ͻ 0.01; and ***, p Ͻ 0.001.

Table 1
Cytotoxicity of standard drugs before and after UGCG inhibition K562, Lucena-1, and FEPS cells were co-incubated for 72 h with a range of concentrations of either vincristine (VCR), daunorubicin (DNR), or cisplatin (CDDP) and 1 M EtDO-P4 and were compared with the ones treated with EtDO-P4 and C 6 -ceramide (C6-cer). Viability was evaluated by MTT assay, and results are reported as mean concentrations giving half-maximal inhibition (IC 50 ) Ϯ S.D. in micromolars (M). Results were obtained from three independent experiments, with each concentration evaluated in triplicate. The ¥ symbol denotes comparisons with untreated cells; † indicates comparison with the parental cell K562, n ϭ 3, and * is p Ͻ 0.05 and ** is p Ͻ 0.01.

Differential contribution of ABC transporters to the reversal of the MDR phenotype after UGCG inhibition
Results so far showed that depletion of GSL led to increased cell death, and this synergizes with chemotherapeutic drugs likely due to accumulation of nonglycosylated ceramides on MDR leukemias. ABCB1 and ABCC1 transporters are key drivers to MDR phenotype, actively extruding xenobiotics thus reducing cell death. In this context, the expression of these proteins was evaluated in conditions matching previous assays. Treatment with EtDO-P4 produced, as demonstrated on Fig. 3, distinct effects on expression of ABCB1 and ABCC1. Sub-toxic UGCG inhibition altered the expression of neither ABC protein, whereas treatment with a 2 M concentration of EtDO-P4 for 24 h reduced only ABCB1 expression on both cell lines. It should be pointed that despite 2 M EtDO-P4 -induced cell stress on diverse assays after 72 h, cells were still viable after 24 h (Fig. S1).
Regardless of not altering ABCB1 or ABCC1 expression after 1 M EtDO-P4, the ABC-mediated transport after pre-treatment with EtDO-P4 was investigated as well. For this, cells were then incubated with specific fluorescent substrates for ABCB1 and ABCC transporters, and dye retentions after free or inhibited efflux were analyzed by the median fluorescence intensity (MFI). Again, results in Fig. 4 indicate that GSL depletion did not impair ABCB1-mediated transport, because profiles of Rho123 MFI were similar irrespective of EtDO-P4 treatment. ABCC activity, however, was significantly modulated after treatment, because carboxyfluorescein (CF) MFI was higher after both free and inhibited efflux for Lucena-1 (Fig. 5, A and C) and for FEPS (Fig. 5, B and D). It is important to note that CF may be transported by ABCC subfamily members other than ABCC1, but not by ABCB1, given that ABCB1 inhibition with VP did not affect CF MFI, despite the MDR leukemias expressing both ABCB1 and ABCC1 (Fig. S2).

ABCC but not ABCB1 transports a bioactive cell-permeable ceramide analog
The roles of GSL and ABC transporters for the adaptation to cellular stresses are well-discussed, and pharmacological impairment of UGCG by EtDO-P4 efficiently sensitized MDR

UGCG inhibition down-regulates ABC proteins on MDR leukemias
cells due to modulation of ABC expression and activity. Notwithstanding the divergent effects on ABCB1 and on ABCC1, the results point to MDR leukemias being able to actively reduce ceramide levels by mechanisms apart from glycosylation. To examine this possibility, the efflux assays of ABCB1 and ABCC substrates were performed on untreated cells in the presence of C6-cer as competitive inhibitor and the resultant MFI and percentages of cells loaded with either Rho123 or CF after the efflux phases were evaluated. In agreement with previous experiments, one more time the ABCB1 efflux of Rho123 was not altered in the presence of increasing C6-cer concentrations (Fig. 6, A-D, and Fig. S3A). An opposite outcome was observed when C6-cer was co-incubated during the CF efflux assay, as results indicate CF retention in both Lucena-1 and FEPS. MFI was dose-dependently increased on both MDR cells, notably when 40 M C6-cer was employed as a competitor to ABCC-mediated efflux (Fig. 7, A-D). This profile could be better appreciated when observing the percentages of CFϩ cells, which reached 50% in the presence of 40 M C6-cer (Fig. S3B). These results suggest that ceramides formed either as a result of drug treatments or from a compromised sphingolipid glycosylation pathway could be transported by ABCC proteins.
To further delineate the roles of ABCB1 or ABCC1 when dealing with increased ceramide contents, the activity assay was once more performed on untreated cells, this time with a fluorescent nitrobenzoxadiazole-labeled ceramide derivative (C6-NBD-cer) as substrate. Results in Fig. 8 corroborate that ABCC transporters mediate the efflux of C6-NBD-cer, because of the fact that the MFI and percentages of C6-NBD-cerϩ cells significantly increased in the presence of MK-571 but not VP when compared with controls (Fig. 8, A-D, and Fig. S3C).

Discussion
A number of studies successfully demonstrated associations linking glycosphingolipids and multidrug resistance in several solid tumor types through overexpression of UGCG (18), although few addressed how its inhibition would affect both the expression and efflux activity of ABCB1 and ABCC1 on nonpolarized cells. There is evidence that the transport of membrane lipid analogs varies in cellular localization; although ABCB1 translocates C6-NBD-sphingolipids across the apical plasma membrane, ABCC1 seems to transport those analogs to the basolateral plasma membrane on the polarized kidney cell LLC-PK1 (21). Observations that are more recent indicate that ABCB1 and ABCC1 differ in cellular localization, and ABCB1 was shown to transport ceramides along Golgi membranes as well (19). Although these observations suggest a similar aptitude to transport sphingolipids, the cellular and organ expressions of ABCB1 and ABCC1 likely take part on how diverse cells manage sphingolipid levels during proliferation, differentiation, apoptosis, and response to cellular stresses. Cells from epithelial origin may express ABC proteins on opposite sides of the cell membrane in a way that, for both blood-brain barrier cells and in the placenta, ABCB1 and ABCC1 are considered, respectively, apical and basolateral transporters (24,25). Our study proposes, given that peripheral leukemic

UGCG inhibition down-regulates ABC proteins on MDR leukemias
cells from hematopoietic origins do not present this organization, that ABCB1 and ABCC1 might play complementary yet diverse roles in dealing with xenobiotics and/or sphingolipid mediators.
In normal bone marrow, it has been demonstrated that GSL levels vary among the distinct stages of erythrocyte differentiation in a way that GM3 is increased on more differentiated cells, such as megakaryocytes (26). Likewise, the efflux activity mediated by ABC transporters is important for the development of hematopoietic progenitors, because ABCB1 and ABCC1 are present on cells with undifferentiated phenotypes on bone marrow or peripheral blood from human (27,28) and murine (29,30) origins. Chronic myeloid leukemias result from a reciprocal translocation of BCR and ABL genes among chromosomes 9 and 22 during the erythroblast stage, forming the Philadelphia chromosome. This abnormality contains the chimeric BCR-ABL oncogene, which encodes a protein with constitutive tyrosine kinase activity that sustains proliferative cell signaling and evasion of apoptosis (31). Sphingolipids contribute to the malignant transformation of erythroblasts in stabilizing the Bcr-Abl protein through the signaling pathway involving sphingosine kinase-1, sphingosine 1-phosphate, and its receptor (32). Furthermore, increased GM1 on cell membranes were described on chronic myelogenous leukemia cells among other phenotypes (33). Therapy is performed with tyrosine kinase inhibitors such as imatinib mesylate in combination with standard chemotherapeutic drugs for remission (34). 30% of patients display some degree of resistance to these drugs, which is closely associated with the expression of ABCB1 (35).
Despite originating from the highly-undifferentiated erythroblastic K562 cell line, Lucena-1 and FEPS MDR cells used in this work present a higher number of megakaryocytes as well as more differentiated profiles (36). Conversely, it was reported that K562 cells selected for resistance with minimal concentrations of vinblastine and epirubicin, drugs that share similarities in structure and in modes of action with VCR and DNR, showed increased ABCB1 expression with no correlation to specific markers of erythroid origin (37). Considering this, we used different assays to assess GSL and UGCG expression. Our results indicated that selection with the chemotherapeutic drugs VCR or DNR led to a complex and diversified GSL profile, despite that UGCG was only significantly increased on FEPS. In this context, treatment with the UGCG inhibitor EtDO-P4 dosedependently reduced viability on all three cells; nonetheless, only Lucena-1 and FEPS showed significant reductions in GM1 levels. A closer observation of the histograms of treated and

UGCG inhibition down-regulates ABC proteins on MDR leukemias
control cells suggests that MDR cells are more homogeneous in GM1 contents than the parental K562. In parallel, both MDR leukemias were sensitized to DNR, Lucena-1 to VCR, and FEPS to CDDP as well. This may relate to the specific mechanism of action exerted by each chemotherapeutic during selection, because small differences in expression and in the ceramide portions of sialylated glycolipids could be observed between Lucena-1 and FEPS. Doxorubicin and oxaliplatin, respectively, DNR and CDDP analogs, were demonstrated to increase ceramide production and UGCG expression on an assortment of tumor cells with variable degrees of drug resistance (38,39), thus strengthening the relationship between sphingolipids and therapy failure. Of note, work that employed shRNA or siRNA for UGCG knockdown observed similar sensitizations of MDR cells, considering fold change, to adriamycin and VCR (40), to doxorubicin (17), and to oxaliplatin (39), all analogous to drugs evaluated in this study.
Pharmacological UGCG inhibition could be increasing ceramide levels, which, along with the stress caused by chemotherapeutics, would explain the higher sensitivity of MDR cells. K562 would possibly shift ceramide glycosylation to galactosylceramide (GalCer) rather than GlcCer, a fact that was described to happen on U-937 and HL-60 human leukemic cells, when the EtDO-P4 analogs PPMP and PDMP actively protected those cells to DNR toxicity (41). In accordance with our results, MDR cells showed lower mitochondrial membrane potentials (⌬ m ) when GSL depletion was concomitant with DNR treatment, with toxic EtDO-P4 concentrations, or when C6-cer was added to the cultures. Such conditions would increase the contents of nonglycosylated ceramides, and those could interact with energized mitochondria through direct physical contact or by inducing conformational changes on Bax pro-apoptotic protein leading to the intrinsic pathway of apoptosis or caspase-independent cell death (42,43). Although our results would not discern those mechanisms, the participation of mitochondria in the apoptosis induced after EtDO-P4 is clear, considering that the ⌬ m loss observed after 24 h would translate into reduced viability, phosphatidylserine exposure, and DNA fragmentation after 72 h. It is important to highlight that most mitochondrial tracers are ABC substrates. Among others, Mitotracker Green (44), DiOC 2 (3) (45), 99m Tc-sestamibi (46), and JC-1 (47) fall in that category. Rho123 is considered ideal for rapid and acute ⌬m estimation at 1-10 M (48), and it was already employed in our cells with similar results to DiOC 2 (3) (44).
The equilibrium of ceramides and GlcCer, GalCer, or GSL may influence cell fate, and in extension, the success of therapeutic interventions, and a variety of drug-resistant cell lines

UGCG inhibition down-regulates ABC proteins on MDR leukemias
present, increased cholesterol, sphingomyelin, and GSL compared with their respective sensitive counterparts (49). The MDR phenotype of cells used in this work is well-described, being representative of freshly-obtained cells from patients with chemotherapy-refractory chronic myeloid leukemia (50,51). Despite data indicating that GlcCer (17) and Gb3 (18) positively regulate ABCB1 expression, and considering the promiscuity and overlap in recognition of substrates (52), the relevance of GSL to ABC activity is controversial at best (53). As such, we observed that sub-toxic GSL depletion was not able to affect the expression of ABCB1 and ABCC1 on neither MDR cell; a 24-h incubation with 2 M EtDO-P4, however, showed reductions in the expression of ABCB1 exclusively. The GSL depletion induced by EtDO-P4 would reduce expression of GlcCer and Gb3, inhibiting nuclear translocation of NF-B or ␤-catenin and hence ABCB1 transcription. The effect of GSL depletion on efflux activity was the opposite, indicating that ABCC transport was partially hindered but not the one mediated by ABCB1.
Lucena-1 and FEPS are cross-resistant to a diversity of compounds with natural and synthetic origins because of, but not limited to, its high efflux activity mediated by the ABC transporters ABCB1 and ABCC1 and to imatinib mesylate as well, which has been demonstrated to increase ceramide and reduce sphingosine 1-phosphate levels on chronic myeloid leukemias (54). Resistance to that drug associates with the expression of ABCB1, however with no clear correlation to its efflux activity because it is considered a weak ABCB1 modulator (35), which suggests that these cells manage their ceramides in alternative forms, involving or not ABCC1 specifically. In this context, there is evidence that ABCB1 contributes to chemotherapeutic failure independently of the efflux of drugs, at least on the patient's leukemic cells (55), and the ABCC1 inhibitor MK-571 was shown to affect retrograde membrane transport and to reduce GlcCer formation (56), evidence that points to at least partial involvement of ABCC in regulating sphingolipid levels. Although our results concerning ABCC inhibition could not be directly associated with ABCC1 because CFDA, PRB, and MK-571 act as substrates and inhibitors to other members from the ABCC subfamily (57), to the best of our knowledge, neither Lucena-1 nor FEPS express ABCC2 or express in significantly lower levels than ABCC1 (58).
It is somewhat difficult to find suitable combinations of substrates and inhibitors to probe the efflux mediated by each subfamily member; therefore, we opted for the most studied and widely-employed ones. CFDA, as stressed in our work, may be transported by ABCC subfamily members other than ABCC1, but not by ABCB1. In the same context, a study performed by  Dogan et al. in 2004 (57) observed the same outcome on 11 leukemia cell lines with diverse degrees of drug resistance. That study considered CFDA or calcein as the best substrates to prime ABCC1 activity, and given that calcein is a well-known ABCB1 substrate (59), our choice of substrates is in accordance with the literature. The same logic applies to the inhibitors. VP, shown to present ABCC1 modulation properties (47), was selective to ABCB1 when combined with Rho123 as substrate and did not affect CFDA efflux. In contrast, MK-571 is a potent, specific inhibitor of ABCC subfamily proteins that has no effect on ABCB1 (57). It is important to notice that we employed PRB as opposed to the more specific MK-571 ABCC inhibitor during the efflux assays on EtDO-P4 pre-treated cells because it could interfere with ceramide levels (56). As such, when combined with the aforementioned substrates, VP (for ABCB1) and PRB or MK-571 (for ABCC) minimize the possibility of cross-interference.

UGCG inhibition down-regulates ABC proteins on MDR leukemias
When C6-cer, an exogenous model used to study ceramides produced after UGCG inhibition, was evaluated as a competitor to either ABCB1 or ABCC transport, results reprised the ones after EtDO-P4 treatment. This sphingolipid impaired the ABCC-mediated efflux from Lucena-1 and FEPS, increasing dye retention and percentages of CFϩ cells after the challenge. Once more, no significant changes in either ABCB1-mediated Rho123 efflux or Rho123ϩ cells were observed. Finally, when a fluorescent derivative of C6-cer was employed as the sole sub-strate, once again the MFI and the percentages of C6-NBDcerϩ were significantly higher when ABCC activity was inhibited, with a marginal increase in the presence of an ABCB1 inhibitor. It is important to disclose that the efflux assays with C6-cer were performed in shorter (1 h) times than the viability assays (72 h) and with a higher cell count as well (2 ϫ 10 5 cells on the first; 2 ϫ 10 4 cells⅐ml Ϫ1 on the latter).
Noteworthy, the effects of C6-cer and EtDO-P4 on MDR cells seemed inversely correlated to K562. Lucena-1 and FEPS showed greater decreases in GM1 contents, moderate resistance to EtDO-P4 cytotoxicity, and sensitivity to C6-cerinduced viability loss. While sharing a common precursor, the chemotherapy employed during selection resulted in distinctive phenotypes, in a way that microarray analysis showed 130 genes with altered expression compared with K562 and Lucena-1, 932 between K562 and FEPS, and 1211 between the two MDR lines (58), with ABCB1 being the most overexpressed gene in these cells. Considering this, it was beyond our focus to thoroughly investigate the specific way each cell line would respond to UGCG inhibition; nonetheless, our results point to the possibility of exogenous C6-cer as well as the combination of EtDO-P4 and standard drugs inducing collateral sensitivity, a hypersensitivity toward secondary agents that arises from the development of resistance toward an unrelated primary drug. Those agents passively enter the cell and, after being extruded by ABCB1 or ABCC1, repeat this futile cycle, increasing ATP

UGCG inhibition down-regulates ABC proteins on MDR leukemias
consumption. Replenishment of ATP increases oxidative stress and the demand for GSH, a prime ABCC1 substrate (52) and the main peptide involved in defense to oxidative stress. This ultimately leads cells to a form of synthetic lethality due to depletion of these molecules (60,61). This possibility is feasible once we consider that treatment with C 2 -ceramide reduced levels of this peptide on a model of epidermal tumor (62) and induced ⌬ m loss and apoptosis on models of prostate and colon cancer (43).
Our results propose that, mechanistically, EtDO-P4 would decrease GlcCer and/or Gb3 contents, leading to reduced ABCB1 expression and increased drug accumulation. Because its activity was not affected after glycosphingolipid depletion nor after exogenous ceramides, on this proposal ABCB1 would mostly mediate drug efflux. Nonglycosylated ceramides, resulting from inhibition of UGCG or induced by chemotherapeutics themselves, would compete for ABCC1 (or, in a safer assumption, ABCC subfamily members) transport, which might be already dealing with a higher chemotherapeutic influx. The combination would exhaust or at least saturate cell detoxification through efflux, depleting cells of GSH and ATP, and further increase drug content and lead MDR cells to ⌬ m loss and then apoptosis. This proposal does not exclude the participation of alternative mechanisms; ultimately, what we observed is merely a reflection of the many ways MDR cells try to circumvent the stress triggered by GSL depletion.
In a broader scope, high levels of plasma circulating ceramides can result from a diversity of inflammatory-associated conditions such as diabetes, obesity, and cardiovascular dis-eases, possibly due to changes in remodeling of membrane sphingolipids (63). Concerning cancer, the first study to demonstrate clinical relevance of ceramides to breast cancer reported higher ceramide levels in the neoplastic tissue than in peritumor region or in the plasma of primary breast cancer patients, which is associated with better prognostics (64). Strikingly, a few studies related increased long-chain ceramide levels on the plasma from patients with advanced ovarian (65) and pancreatic cancer (66), which is related to higher malignancy. Therefore, participation of ABC transporters on ceramide homeostasis and in an efflux mechanism to extracellular plasma acceptors, such as apolipoproteins, apart from ceramide glycosylation or conversion to sphingomyelin had already been proposed but not demonstrated (67). Given the ubiquitous expression of ABCC1, its active efflux of inflammatory lipid mediators such as leukotriene C4 and sphingosine 1-phosphate, and its association with poor prognostics in cancer (20,52), ABCC1 emerges as candidate for this role.
In this work, our results extend the relationship between sphingolipid glycosylation and acquired resistance mechanisms on models of human MDR neoplastic cells. Further studies addressing the modulation of sphingolipid levels and their cellular fates could translate to the development of strategies directed to MDR hematologic neoplasias or to the proposal of an adjuvant, off-label use of drugs currently employed against Gaucher's disease, which accumulates GlcCer in multiple organs as in the bone marrow. Comprehension of the distinct roles of ABC proteins on the clearance of ceramides could ulti-

UGCG inhibition down-regulates ABC proteins on MDR leukemias
mately return therapeutic possibilities to patients with chemotherapy-refractory leukemias.

Cell lines
The chronic myeloid leukemia cell lines K562, Lucena-1 (K562/VCR), and FEPS (K562/DNR) were cultured in RPMI 1640 medium (Sigma-Aldrich) supplemented with 25 mM HEPES and 2 g⅐liter Ϫ1 sodium bicarbonate adjusted to pH 7.4, 100 units of penicillin and 100 g⅐ml Ϫ1 streptomycin (all obtained from Sigma-Aldrich), and with 10% fetal bovine serum (FBS) (Thermo Fisher Scientific, Waltham, MA) inactivated at 56°C for 1 h prior to use. Dr. Vivian M. Rumjanek kindly donated the MDR cells Lucena-1 and FEPS. Briefly, K562 cells were exposed to increasing concentrations of the chemotherapeutic drugs vincristine sulfate (VCR) and daunorubicin hydrochloride (DNR) (both from Sigma-Aldrich), as described before (68,69). For subcultures, 2 ϫ 10 4 cells⅐ml Ϫ1 were harvested every 3 days, and complete RPMI 1640 medium was added and then maintained at 37°C in 5% CO 2 . Lucena-1 and FEPS were cultured, respectively, in the presence of 60 nM VCR and 500 nM DNR in order to maintain the MDR phenotypes. Prior to all experiments, the MDR Lucena-1 and FEPS were cultured free of drugs to avoid additive effects.

Extraction, purification, and analysis of glycosphingolipids
Total GSLs were obtained by previously described procedures (70, 71). K562, Lucena-1, and FEPS (2 ϫ 10 4 cells⅐ml Ϫ1 ) were seeded in complete RPMI 1640 medium for 72 h in 150-mm Petri plates in the presence of a sub-toxic concentration of the UGCG inhibitor EtDO-P4 (Glixx Laboratories, Hopkinton, MA). Controls were treated with the diluent, 0.1% absolute ethanol (Sigma-Aldrich Brazil, Duque de Caxias, RJ, Brazil). Equal amounts of 5 ϫ 10 7 cells were harvested and washed in PBS, and precipitates were extracted with 1 ml of isopropyl alcohol/hexane/water 55:25:20 (v/v/v, lower phase) twice. The extracts were dried under a N 2 stream and then saponified in 2 ml of 0.1 M NaOH/methanol to degrade glycerophospholipids. After extraction with hexane, total GSLs were isolated using C 18 cartridges (Analytichem International/Varian, Harbor City, CA) as described, and analyzed by TLC followed by staining with a solution of 6% resorcinol and 1% CuSO 4 in concentrated HCl for the analysis of sialylated GSL. A commercial GSL standard containing a mix of GM1, GM2, and GM3 was chromatographed as reference (Matreya LLC, State College, PA). All solutions were prepared with HPLC-grade reagents from Sigma-Aldrich, Tedia (Fairfield, OH), or SK Chemicals (Seongnam-si, Gyeonggi-do, South Korea).

Western blot analysis
K562 and their MDR counterparts Lucena-1 and FEPS were cultured as described before for 48 h, then lysed in lysis buffer (1% Triton X-100; 150 mM NaCl; 25 mM Tris, pH 7.4; 5 mM EDTA; 0.5% sodium deoxycholate; 0.1% SDS; 5 mM tetrasodium pyrophosphate; 50 mM sodium fluoride; 1 mM sodium orthovanadate) containing 1:200 protease inhibitor mixture (Sigma-Aldrich), and centrifuged at 21,000 ϫ g for 10 min at 4°C. 60 g of protein was subjected to SDS-PAGE, followed by transfer to polyvinylidene difluoride membranes (Millipore, Burlington, MA). After blocking with 5% skim milk for 30 min, membranes were rinsed with T-TBS and stained with 1:250 anti-UGCG (clone 1E5; Santa Cruz Biotechnology, Dallas, TX) and 1:1000 anti-␤-actin (clone AC-74; Sigma-Aldrich) antibodies overnight at 4°C. Membranes were rinsed and incubated with 1:1000 anti-mouse IgG-horseradish peroxidase (Cell Signaling Technology, Danvers, MA) for 2 h at room temperature. After rinsing with T-TBS, membranes were developed by use of Western Lightning Chemiluminescence Reagent Plus (PerkinElmer Life Sciences). Densitometry analyses were performed using ImageJ 1.52d software (National Institutes of Health, Bethesda, MD). The relative expression of UGCG was calculated as the density of the UGCG band divided by the density of the ␤-actin band for each cell line.

Assessment of GM1 expression on plasma membranes
The GM1 was assessed on cell surfaces by flow cytometry. K562, Lucena-1, and FEPS cells were cultured for 24 h as described before. Cell density was adjusted to 2 ϫ 10 5 per well and then incubated with 5 g⅐ml Ϫ1 CHT-FITC (Sigma-Aldrich) for 30 min at 4°C in a light-protected environment. Cell suspensions were centrifuged, resuspended in cold PBS, and then analyzed by flow cytometry. Five identical polypeptide B subunits from CHT specifically bind to the GM1 on cell surfaces, and considering that the GSL biosynthesis occurs in a stepwise fashion, the GM1 MFI can relate to the global cell GSL composition (72).

In vitro cell viability
Cell viability was determined with the tetrazolium salt 3-(4,5dimethylthiazol-2-yl)-2,5-diphenyltetrazolium (MTT; Sigma-Aldrich). Briefly, 2 ϫ 10 4 cells⅐ml Ϫ1 were treated with a range of concentrations of either EtDO-P4, VCR, DNR, cisplatin (CDDP, Incel; Darrow Laboratórios S/A, Rio de Janeiro, RJ, Brazil), or N-hexanoyl-D-erythro-sphingosine (C6-cer; Avanti Polar Lipids, Alabaster, AL) for 72 h at 37°C in 5% CO 2 . Alternatively, cells were co-incubated with 1 M EtDO-P4 and a range of concentrations of VCR, DNR, or CDDP. Negative controls were prepared with the respective diluents (0.1% absolute ethanol for EtDO-P4 or C6-cer; 0.1% DMSO for VCR and DNR; and RPMI 1640 medium for CDDP). Then, MTT was added to each well to a final concentration of 0.5 mg⅐ml Ϫ1 , and plates were incubated at 37°C in 5% CO 2 for 3 h in a light-protected environment. After centrifugation, 200 l of DMSO was added to dissolve the dark-blue formazan crystals formed after MTT reduction. Absorbance was measured on a Beckman Coulter AD340S spectrophotometer microplate reader (Beckman Coulter, Brea, CA) at 570 nm. The concentrations giving halfmaximal inhibition (IC 50 ) were calculated by nonlinear regression using the GraphPad Prism version 7.0 software (GraphPad Software, San Diego, CA).

Apoptosis assay
The annexin V/propidium iodide (PI) assay was performed for apoptosis detection. Lucena-1 and FEPS cells were incubated in 24-well plates in the conditions described above and

Assessment of mitochondrial membrane potential
Changes in the mitochondrial membrane potential (⌬ m ) were probed using Rho123 (Sigma-Aldrich), which specifically stains energized mitochondria in cultured cells (73). Lucena-1 and FEPS cells were cultured for 24 h with DNR, EtDO-P4, or C6-cer as described before. Then the cell density was adjusted to 2 ϫ 10 5 per sample and incubated with 2.5 M Rho123 for 30 min at 37°C in 5% CO 2 in a light-protected environment. Cell suspensions were centrifuged, resuspended in cold PBS, and then analyzed by flow cytometry.

Immunophenotypes of MDR cells
The expression of the ABC transporters ABCB1 and ABCC1 was evaluated by flow cytometry. Lucena-1 and FEPS were treated with EtDO-P4 for 24 h as described before, and cell density was adjusted to 2 ϫ 10 5 per sample at room temperature. The cells were then permeabilized and fixed with BD FACS Lysing solution (BD Biosciences) for 10 min, and nonspecific binding was blocked with PBS 10% FBS for 20 min. Following incubation with 1:50 anti-ABCB1 (clone D-11) or anti-ABCC1 1:50 (clone QCRL-1) human primary antibodies (both from Santa Cruz Biotechnology, Dallas, TX) for 30 min at 4°C, cells were washed with cold PBS and stained with 1:1000 Alexa 488 -conjugated mouse anti-human IgG secondary antibody (Thermo Fisher Scientific) for 30 min at 4°C in a light-protected environment. Cell suspensions were centrifuged, resuspended in cold PBS, and then analyzed on a flow cytometer.

ABC-mediated efflux assays
The ABCB1 and ABCC transport assays were performed, respectively, with the use of the Rho123 and CFDA dyes (Sigma-Aldrich). Rho123 and CFDA are able to passively distribute into the cell. Rho123 is fluorescent and is actively extruded by ABCB1; CFDA is hydrolyzed in the cytosol and originates the fluorescent substrate CF, which is transported to the extracellular milieu by ABCC subfamily members, notably ABCC1 (74). Briefly, assays were performed in two 30-min steps, sufficient for the accumulation and efflux of dyes, at 37°C in 5% CO 2 in a light-protected environment. 2 ϫ 10 4 cells⅐ml Ϫ1 were treated for 24 h with 1 M EtDO-P4 as described prior to the assays. Then, 2 ϫ 10 5 Lucena-1 or FEPS cells were incubated in 96-well plates with 250 nM Rho123 or 500 nM CFDA diluted in RPMI 1640 medium to allow accumulation of the dyes within cells. Following that, the cells were centrifuged at 200 ϫ g for 7 min and resuspended in fresh RPMI 1640 medium to allow efflux of the dyes (free efflux). In parallel, cells were incubated with the ABCB1 inhibitor VP (75) or the ABCC inhibitors PRB or MK-571 (57) (inhibited efflux), respectively, at the concentrations of 10 M, 1.25 mM, and 25 M. As negative control, cells were exposed to medium only. Next, cells were again centrifuged, resuspended in cold PBS, and maintained on ice until acquisition by flow cytometry. Alternatively, C6-cer was employed as a competitive inhibitor during both steps of the assay. Fluorescence histograms were divided into two distinct areas: Rho123 or CF-negative on the left, accounting for 95% of control cells with low MFI for Rho123 or CF, and Rho123 (Rho123ϩ) or CF-positive (CFϩ) on the right, corresponding to cells still loaded with dyes after efflux phase. A vertical line on each graph indicated those regions.

C 6 -ceramide efflux assays
Assays were performed in two 30-min steps, in similar conditions to the ABC-mediated efflux assays. 2 ϫ 10 5 Lucena-1 or FEPS cells were incubated in 96-well plates with 1 M nitrobenzoxadiazole-labeled C 6 -ceramide (C6-NBD-cer) (Avanti Polar Lipids) diluted in 0.1% absolute ethanol to allow accumulation of the fluorescent sphingolipid within cells. In parallel, cells were incubated with the ABCB1 inhibitor VP or the ABCC inhibitor MK-571 (inhibited efflux). As negative control, cells were exposed to medium only and treated as described before for acquisition by flow cytometry. Fluorescence histograms were divided into two distinct areas: C6-NBD-cer-negative on the left, accounting for 95% of control cells with low MFI, and C6-NBD-cer-positive (C6-NBD-cerϩ) on the right, corresponding to cells still loaded with the fluorescent sphingolipid after efflux phase. A vertical line on each graph indicated those regions.

Flow cytometry
The MFI from 15,000 viable cells, gated in accordance with forward and side scatter parameters, representative of cell size and granularity, were acquired using the FL1-H filter on a BD FACSCalibur flow cytometer (BD Biosciences). All post-analyses were performed on Summit version 4.3 software (Dako Colorado, Inc., Fort Collins, CO).

Statistical analysis
Statistical analyses were performed using GraphPad Prism version 7.0 software. For two paired comparisons, statistical significance was calculated by parametric Student's t tests for normally distributed data, according to the D'Agostino-Pearson test. Otherwise, the nonparametric Wilcoxon test was employed. For more than two comparisons, unpaired one-way ANOVA or Kruskal-Wallis tests were used, respectively, for parametric and nonparametric data. Otherwise, paired oneway ANOVA and the Friedman test were employed for parametric and nonparametric data, respectively. Bonferroni's post-test was used for parametric data, and the Dunn's post-test was employed for nonparametric data. Null hypotheses were rejected when p values were lower than 0.05, and significances were represented by * for p Ͻ 0.05, ** for p Ͻ 0.01, and *** for p Ͻ 0.001.