Salinomycin disturbs Golgi function and specifically affects cells in epithelial-to-mesenchymal transition

Epithelial-to-mesenchymal transition (EMT) is a cell transdifferentiation program during which cells lose epithelial characteristics (e.g. adhesion and lack of motility) while they acquire a mesenchymal phenotype associated with cell migration (Nieto and Cano, 2012). Epithelial carcinoma cells that have passed through the EMT show properties of cancer stem cells (CSCs), which includes invasiveness, drug-resistance and the ability to form metastases at distant organs. Thereby the EMT contributes to cancer metastasis and relapse (Shibue and Weinberg, 2017). Hence, targeting the EMT program to selectively eliminate CSCs is a promising way to improve cancer therapy (Nimmakayala et al., 2019; Pattabiraman and Weinberg, 2014).

It has been demonstrated that the induction of the EMT in immortalized human mammary epithelial cells (HMLEs) results in the expression of stem cell markers and the acquisition of the breast CSC phenotype. Consequently, these cells have been established as an experimental cell model of CSCs (Gupta et al., 2009; Mani et al., 2008). High-throughput screening of a large compound library on the HMLE cell model identified several drugs selective towards EMT and CSCs. The most potent compound was salinomycin (Sal), a K⁺ ionophore widely used as a coccidiostat (Gupta et al., 2009). Many subsequent studies have proposed potential mechanisms for the selectivity of Sal against CSCs, as well as pursuing specific vulnerabilities of EMT and CSCs (Antoszczak, 2019; Kaushik et al., 2018; Pattabiraman and Weinberg, 2014; Wang et al., 2021).

To date, Sal has been shown to affect tumor cells through various mechanisms including activation of apoptosis, autophagy and elevation of intracellular reactive oxygen species, as well as mitochondrial membrane depolarization, and inhibition of multidrug resistance pumps. Moreover, several studies have used Sal to potentiate the efficacy of other commonly used anticancer chemotherapeutics (Antoszczak, 2019; Dewangan et al., 2017; Huczynski, 2012; Kaushik et al., 2018; Managò et al., 2015; Shi et al., 2015). Nevertheless, despite the enormous efforts in the past decade, the specific mechanism of EMT and CSC selectivity of Sal remains vague. Recently, Huang et al. showed that Sal accumulates in the endoplasmic reticulum (ER) and thus promotes the release of Ca²⁺ into the cytosol. This leads to the unfolded protein response (UPR) and activation of CHOP, which inhibits Wnt signaling, an effect already ascribed to Sal in chronic lymphocytic leukemia (CLL) (Huang et al., 2018; Lu et al., 2011).

It has been demonstrated that the ER-mediated UPR is permanently activated in EMT cells to support synthesis and secretion of large quantities of extracellular matrix (ECM) proteins. Hence, ER function is critical for maintenance of EMT cells, offering a new therapeutic approach (Feng et al., 2014). In addition to increased secretory output, the EMT process involves major cellular remodeling including changes in membrane lipidome characteristics and reduced membrane fluidity, which is needed for migration and invasion (Kalluri and Weinberg, 2009; Sampaio et al., 2011). Distribution of both the proteins and the lipids inside the cell is dependent on the functionally intertwined communication between the ER and the Golgi (Smirle et al., 2013). In addition, Mai et al. demonstrated that derivatives of Sal localize in lysosomes, causing accumulation of iron and activation of ferroptosis as the main mode of CSC death (Mai et al., 2017). Given that Sal affects most membrane-enclosed cellular compartments, especially those involved in protein production, modification and secretion, as well as the fact that some ionophores with CSC-selective abilities are Golgi inhibitors (Dinter and Berger, 1998; Kaushik et al., 2018), we were prompted to examine the significance of the Golgi in EMT cells and its sensitivity to Sal. In the present study, we demonstrate that Sal affects the Golgi morphology as well as the expression of ER-Golgi-related genes predominantly in EMT cells. Moreover, we show that Sal inhibits correct Golgi function, which leads to alterations in post-translational protein modifications that are specifically processed in the Golgi. These include reduced secretion of proteins and marked changes in the N-glycosylation profile of secreted proteins, primarily diminishing the amount of complex N-glycans. Given that our data undisputedly positions Sal as a Golgi-disturbing agent and because EMT cells demonstrate greater sensitivity to the Golgi perturbations, we propose that targeting of the Golgi could be a novel therapeutic path against EMT cells, and consequently against CSCs.

A seminal study by Gupta et al. (2009) identified four highly selective compounds against EMT cells: abamectin, etoposide, nigericin and salinomycin (Sal). In addition, the screening process resulted in the discovery of an additional 32 selective compounds. We conducted a similar analysis on the same data set from Gupta et al., using the same algorithm but with a more permissive threshold. Our analysis revealed a total of 227 selective compounds, including several ionophores or ionophore-like molecules (see Materials and Methods for details; Table S1A). Among these, the most interesting compound was monensin (Mon), which was experimentally confirmed as a potent EMT-selective cytotoxic compound in a prostate cancer model (Vanneste et al., 2019) (Table S1B).

In the current study, we took the advantage of the same EMT models used by Gupta et al. (2009), in which non-EMT cells (HMLE-shGFP and HMLE-pBp; expressing shRNA against GFP and empty vector, respectively) are defined as expressing the epithelial marker E-cadherin and with CD24^high/CD44^low. Conversely, EMT cells (HMLE-shEcad and HMLE-Twist; expressing shRNA against E-cadherin and Twist, respectively) are defined as expressing N-cadherin as well as having CD24^low/CD44^high (Fig. S1A–C). We confirmed that Sal lowered the proportion of EMT cells in a mixed population experiment (Fig. 1A; Fig. S1D). Sal also showed a more pronounced effect towards EMT cells in proliferation assays (Fig. 1B,C). The observed selectivity diminished at higher concentrations of Sal (5 µM for the shEcad and shGFP cell pair, and 10 µM for the Twist and pBp cell pair), which were toxic for both EMT and non-EMT cells. Higher concentrations also resulted in immediate loss of mitochondrial potential, which was even more pronounced in HMLE-pBp (non-EMT) cells, possibly explaining the loss of selectivity (Fig. S1E). These experiments proved the selectivity of Sal towards EMT cells, although to a lesser extent than previously described (Gupta et al., 2009). Finally, Mon also displayed selective growth inhibition of EMT cells and enriched the percentage of cells with epithelial markers in a mixed population experiment (Fig. 1A–C; Fig. S1D). The effect was comparable to that of Sal. Conversely, paclitaxel (PTX; a negative control) enriched the EMT population, corroborating published data (Gupta et al., 2009) (Fig. 1A; Fig. S1F). Thus, our data clearly demonstrate that Mon exhibits EMT-selective inhibition not only in the prostate model, as reported previously (Vanneste et al., 2019), but also in the breast EMT cell model (HMLE). This confirmation of results in different models underscores the consistent efficacy of Mon as an EMT-selective agent.

To test selectivity more generally, particularly in breast cancer cell lines, we examined the influence of Sal and Mon in a broader range of cell lines distinguished by their epithelial or mesenchymal phenotype (Fig. 1D). We used the cell line MCF -7, which is non-metastatic and represents the luminal subtype of breast cancer, and the cell lines SUM 159 and MDA-MB -231, both basal B-cell lines known to be metastatic. Basal breast cancer cells are known to have a higher propensity to undergo EMT and express CSC markers (Feng et al., 2014; Zhao et al., 2016). In addition, we used HCT116 and HepG2 colon and liver cancer cells, respectively. Both cells exhibit an epithelial phenotype and were reported previously to be more resistant to treatment with Mon (Vanneste et al., 2019). The results confirmed that both Sal and Mon similarly affected the viability of these cell lines and confirmed that the cell lines with epithelial properties – MCF -7 and HCT116 – were the most resistant, whereas MDA-MB -231 and SUM 159 cells were significantly more sensitive. In contrast to the results of Vanneste et al., HepG2 cells showed higher sensitivity, comparable to that of SUM 159 cells.

The results reported previously (Vanneste et al., 2019) also suggest that Mon uptake correlates positively with cell sensitivity status. To test whether Sal selectively accumulates in HMLE EMT cells, we performed liquid chromatography tandem mass spectrometry (LC-MS/MS) analyses to quantify the amount of Sal in lysates of HMLE-pBp and HMLE-Twist cells exposed to 0.2 and 5 µM Sal for 24 h. Our results show no statistical difference in Sal uptake between EMT and non-EMT cells after treatment with 5 µM Sal (Fig. S1G). When incubated with 0.2 μM Sal, the intracellular concentration either fell below or was near the detection limit. This result led to considerable variability between measurements, making it difficult to draw definitive conclusions (data not shown).

Overall, our data confirmed that Sal and Mon, similar to nigericin (Nig), are EMT-selective agents. They are structurally similar polyether ionophores with preferences for monovalent ions. It is noteworthy that Mon and Nig are known Golgi-disturbing agents, a property not previously attributed to Sal.

To evaluate the effects of Sal on intracellular compartments, we profiled the global gene expression in both EMT and non-EMT cells by RNA-Seq. We treated HMLE-Twist and HMLE-pBp cells for 24 h with a low concentration of Sal (0.2 µM), which showed the highest selectivity in both EMT models (Fig. 1A; Fig. S1D). The analysis of differentially expressed genes (≥2-fold change) revealed that: (1) a common set of 52 genes were upregulated in both cell lines, (2) one gene (NFE2L3) was upregulated in EMT but downregulated in non-EMT cells, and (3) three genes (IDI1, TM7SF2 and INSIG1) were downregulated in EMT cells, but upregulated in non-EMT cells (Fig. S2A, Table S2A). We further widened the analysis for the case (2), which yielded 64 genes (2+; ≤0 in non-EMT and ≥2-fold change in EMT cells, Fig. S2B, Table S2B). Both sets (1 and 2+) include multiple genes related to the ER, the Golgi and membrane function, suggesting that Sal affects these cellular compartments predominantly. The observed effect can be considered either as a general mechanism of Sal (1) or as specific to EMT cells (2+). Moreover, Gene Ontology (GO) analysis of these two gene sets showed enrichment of the secretory pathway-related genes (Table S2C,D, respectively).

We additionally performed GO analysis on full sets of differentially expressed genes in the three experimental cases (EMT versus non-EMT, EMT with Sal and non-EMT with Sal). The analysis demonstrated that Sal treatment enriched for genes related to the ER and the Golgi regardless of the cell type, which reinforces the notion that Sal enriches for genes involved in the secretory pathway (Fig. 2A). Additionally, a gene set enrichment analysis (GSEA) demonstrated that Sal significantly induced the expression of genes known to be upregulated by well-known Golgi stressors (monensin, nigericin, brefeldin A and 4MU-xyloside) or which activate the ER-related UPR (tunicamycin and thapsigargin) (Fig. 2B,C; Fig. S2C, Table S3). Sal also upregulated genes reported to be enriched upon the induction of the Golgi-specific UPR (Serebrenik et al., 2018) (Fig. 2B,C). By contrast, we observed no enrichment for the gene set upregulated by the treatment with the DNA-damaging chemotherapeutic doxorubicin (Fig. 2B,C). In addition, Sal also induced genes involved in glycosylation, a post-translational protein modification mainly executed inside the ER–Golgi compartments (Fig. 2C; Fig. S2C).

Importantly, the analysis also revealed the difference between EMT and non-EMT cells in response to Sal. Whereas GO enrichment analysis of EMT cells showed enrichment of processes related to the ER, Golgi, ion transport, extracellular matrix organization and regulation of cell adhesion, the GO categories with the highest enrichment in non-EMT cells were related to the metabolism of alcohol, cholesterol and lipids (Fig. 2D; Table S2E–G). Moreover, because genes upregulated in non-EMT cells and downregulated in EMT cells (set 3) are either directly involved in (IDI1 and TM7SF2) or regulate (INSIG1) cholesterol biosynthesis, we assessed the global effect of Sal and Mon on cellular lipid content. We stained cells with Nile Red, a dye which can differentiate between neutral lipids (cholesterol ester or triglycerides) found in lipid droplets (LDs) (Diaz et al., 2008). We found that both compounds increased green to yellow fluorescence, representing LDs with high cholesterol content, exclusively in non-EMT cells, agreeing with the gene expression data. On the other hand, there was no obvious change observed in EMT cells (Fig. S2D). In addition, EMT cells were reported to maintain low cholesterol to allow increased membrane fluidity necessary for migration and invasion (Sampaio et al., 2011). ABCA1, a cholesterol efflux transporter, is highly expressed in EMT cells and it has been demonstrated that it can be downregulated by anti-metastasis drugs including Sal (Zhao et al., 2016). Our RNA-seq data confirmed increased expression of ABCA1 in EMT cells (3.45-fold change) and Sal did decrease its expression in both cell lines. Taken together, observed data suggests that Sal targets cellular compartments involved in protein and lipid processing and secretion (the ER and the Golgi), which activate different gene responses in EMT and non-EMT cells.

Next, we decided to investigate the effect that Sal has on the secretory pathway, namely the Golgi and the ER, as well as the induction of UPR. We inspected the Golgi morphology by confocal microscopy, using the common Golgi markers GM130 and GIANTIN (also known as GOLGA2 and GOLGB1, respectively) that colocalize with cis-Golgi and cis- to medial-Golgi matrix membranes, respectively. Sal induced fragmentation of Golgi cisternae, which appeared as a dispersed area within cells that stained positive for both GM130 and GIANTIN (Fig. 3A,B). The extent of the Golgi perturbations in EMT cells was statistically significant when compared to mock-treated cells (Fig. 3C,D). On the other hand, the Golgi of non-EMT cells did not display perturbations in morphology after Sal treatment. Of note, the dispersal area of Golgi cisternae in untreated non-EMT cells was on average much larger than in EMT cells (P<0.001 for both markers, unpaired two-tailed t-test), but the variance within the populations was very large (Fig. 3C,D). More prominent Golgi fragmentation in EMT cells was also observed when cells were treated with Mon (Fig. 3E).

To test the effect on Golgi morphology in other cells, while overcoming the cytotoxic effects observed at higher concentrations of Sal, a model cell line was used to allow monitoring of Golgi morphology over time. HeLa cells stably expressing the trans-Golgi marker GalT–GFP and H2B–mCherry were treated with Sal at a concentration of 5 µM, live-cell fluorescence microscopy was performed and Golgi morphology was compared with that of untreated cells. Remarkably, a significant fragmentation of the Golgi was observed within only 2 h after treatment with Sal, as shown in Movies 1 and 2.

We also examined the status of the compartments between the ER and the Golgi, by staining the cells with antibodies for ERGIC53 and β′COP (also known as LMAN1 and COPB2, respectively), markers of ER-Golgi intermediate compartment (ERGIC) and COPI retrograde vesicles, respectively. The microscopy confirmed that these compartments were also affected by Sal treatment and that EMT cells showed greater alterations, which were more evident on COPI vesicles (Fig. S3A).

Moreover, we also inspected Golgi morphology after treatment with tunicamycin, an early blocker of N-glycosylation that is a well-known inducer of the ER stress and the UPR, and has been shown to be selective against EMT cells (Feng et al., 2014). Tunicamycin led to a severe Golgi compaction only in EMT cells, whereas its effect on Golgi morphology in non-EMT cells was negligible (Fig. 3E). Nevertheless, it induced UPR in both cell lines (Fig. 3F,G). It has been previously shown that Sal can induce ER-stress and activate UPR (Xipell et al., 2016; Yoon et al., 2013), which we confirmed in our cell model. We observed the activation of the PERK but not IRE1 axis of UPR signaling cascade (Fig. 3F,G). In a recent publication, it was also shown that a fluorescent conjugate of Sal accumulates in the ER and leads to a massive release of the ER-stored Ca²⁺ into the cytosol in JIMT-1 cells (∼600% increase), which was proposed as its mechanism of action (Huang et al., 2018). This prompted us to examine changes in cytosolic Ca²⁺ in our experimental model. In contrast, we observed minimal or no increase of cytosolic calcium after treatment with both Sal and Mon (5 and 0.2 µM, respectively) in our EMT model (Fig. S3B–D). Surprisingly, even thapsigargin, an inhibitor of sarco/endoplasmic reticulum Ca²⁺ ATPase (SERCA), which leads to depletion of the ER Ca²⁺ storage, led to an increase of only 60%. Although our results differ from published data, we acknowledge that there might be factors contributing to the observed discrepancy. One possible explanation is that HMLE cells are not as susceptible to a strong release of Ca ²⁺ compared with other cell lines. A recent study by Lebeau et al. (2021) supports this notion and shows that Ca²⁺ release varies moderately after treatment with thapsigargin in different cell lines. Nevertheless, our collective results highlight the significant induction of stress in the Golgi by Sal, particularly in EMT cells, which showed increased sensitivity to the drug compared with non-EMT cells.

The Golgi serves as the central component of the secretory pathway, and it is the main site for the modification and sorting of the membrane and secreted proteins (Morré and Mollenhauer, 2009). Thus, we investigated the fate of membrane proteins after inducing the Golgi stress with Sal. Classical cadherins (such as E- and N- cadherin) have prodomain sequences (∼130 amino acids) that guard the adhesive domains against interaction before they are activated by proteolytic processing. These sequences are cleaved off by an endoprotease in the late Golgi (Koch et al., 2004). Sal, as well as Mon and nigericin, led to the appearance of an additional band for the EMT marker N-cadherin, with a higher molecular mass, in western blot analysis. The effect was more prominent with a lower concentration of Sal (0.2 µM, Fig. 4A and Fig. S4A) and was observed from 24 h through 72 h after treatment (Fig. S4B). The same effect was observed in breast tumor cell lines with high proportion of CSC-like cells (SUM 159; Fig. 4B). Conversely, paclitaxel did not disturb proteolytic cleavage of N-cadherin, corroborating the hypothesis that a consequence of Sal treatment is a Golgi stress-related effect. We confirmed incomplete proteolytic processing by using an N-cadherin prodomain specific antibody (Fig. 4C). Strikingly, although E-cadherin maturation also requires removal of its prodomain, the treatment with ionophores did not lead to an appearance of an additional protein band in non-EMT cells (Fig. 4A,B; Fig. S4A,B). Additionally, the E-cadherin expression increased after the treatment, especially with 5 µM Sal or 0.2 µM Mon (Fig. 4A,B).

In addition to altered processing, we observed higher mobility of membrane proteins N-cadherin, EGFR (epidermal growth factor receptor) and TFRC (transferrin receptor) in all treated cell lines (Fig. 4A,B). We hypothesized that Sal also alters glycosylation, as the ER and the Golgi are the major sites of protein glycosylation, a most abundant post-translational modification of membrane and secreted proteins (Varki et al., 2009). Hence, to examine the effect of reduced glycosylation of N-cadherin, we treated protein lysates with PNGase F, which removes glycans attached to proteins at asparagine residues (N-glycosylation). Given that PNGase F removes all N-linked sugars, the mobility of N-cadherin was even greater than after Sal treatment, suggesting that Sal hinders N-glycosylation (Fig. 4D).

We also examined the effect of the Golgi stressors on protein secretion. We specifically measured the amount of interleukin-8 (IL-8), a cytokine secreted by both cell types (non-EMT cells secrete a higher amount of IL-8 than EMT cells; Low-Marchelli et al., 2013). Both Sal and Mon decreased the secretion of IL-8 by 40% or more in both cell lines (Fig. 4E).

Next, we examined the fate of E- and N-cadherins after treatment with the competitive inhibitor of O-glycosylation 1-benzyl-2-acetamido-2-deoxy-(α)-D-galactopyranoside [GalNAc(α)-O-bn], a known Golgi stressor (Miyata et al., 2013), alone or in combination with Sal. As expected, treatment with GalNAc(α)-O-bn prevented O- linked glycosylation of N-cadherin and therefore decreased the molecular mass of the mature protein (Fig. S4C). Exposure to Sal and GalNAc(α)-O-bn resulted in both modifications, and the mobility of both the N-cadherin prodomain and the mature protein was even greater than after treatment with Sal. On the other hand, neither the mobility nor the processing of E-cadherin was affected by GalNAc(α)-O-bn in non-EMT cells. However, the combination treatment reduced the expression and decreased the molecular mass of the mature protein to some extent. It is also noteworthy that treatment with GalNAc(α)-O-bn in combination with Sal or Mon additionally decreased the proportion of EMT cells in a mixed population experiment compared with Sal alone and showed a more pronounced effect on EMT cells in proliferation assays, whereas treatment with GalNAc(α)-O-bn alone had no effect (Fig. S4D).

In addition, we tested the effect of another compound known to disrupt Golgi function and organization by inhibiting protein transport from ER to Golgi, brefeldin A (BFA), together with the ER stressors tunicamycin and thapsigargin on N-cadherin and E-cadherin processing (in EMT and non-EMT cells, respectively) (Fig. S4E). Not surprisingly, tunicamycin inhibited N-linked glycosylation and therefore decreased the molecular mass of mature proteins in both EMT (N-cadherin in HMLE-Twist) and non-EMT cells (E-cadherin in HMLE-pBP and MCF -7 cells). Thapsigargin led to the appearance of an additional band with a higher molecular mass, indicating inhibition of prodomain processing in both non-EMT and EMT cells. On the other hand, surprisingly, BFA also resulted in an additional band, but only in non-EMT cells (i.e. the prodomain of E-cadherin), with no effect on glycosylation, whereas it inhibited only the glycosylation of N-cadherin, which was observed as a protein with a smaller molecular mass in EMT cells. A similar effect of thapsigargin and BFA in MCF-7 cells has been described previously (Geng et al., 2012), but their different effects in EMT cells had not been described previously. These results suggest that the same treatment modality has different effects in different cell models.

Our experiments on dose-dependent cytotoxicity against EMT and non-EMT cells revealed that BFA exhibited a significant decrease in the proportion of EMT cells in the HMLE cell model during a mixed population experiment. This selective effect on EMT cells was further confirmed in proliferation assays (Fig. S4F,G), supporting its selectivity. Previous studies have also demonstrated the ability of BFA to reduce stem cell potential and migration in breast cancer cells (Tseng et al., 2014). However, when we tested dose selectivity on a broader range of cell lines, the pattern observed for BFA differed from that observed for Sal or Mon. SUM 159 cells showed the highest sensitivity, whereas MCF-7 and MDA-MB -231 cells showed comparatively lower sensitivity. This discrepancy in the activity of BFA, Sal, and Mon in EMT and non-EMT cells might be due to differences in mechanisms and sites of action; BFA inhibits the transport of secreted and membrane proteins from the ER to the Golgi, whereas Mon acts in the medial to trans cisternae of Golgi. In addition, cell type-specific sensitivity to BFA has been previously demonstrated (Dinter and Berger, 1998; Rosa et al., 1992).

Furthermore, we investigated the effect of bafilomycin A, another stressor known to induce acidification of lysosomes, endosomes and secretory vesicles, and which is commonly used to detect Golgi-associated post-translational changes in proteins. In contrast to previous findings, we did not observe selectivity in the HMLE cell model (Fig. S4H,I), and there was a divergent effect on other cell lines. These results indicate a significant difference compared to the effects of Sal, Mon and BFA, potentially due to distinct mechanisms of action (Dinter and Berger, 1998).

Overall, our experiments showed that Sal interferes with post-translational modification of proteins, glycosylation and protein secretion, aligning with the effects observed with Mon and to some extent with BFA, thus confirming Sal as a Golgi stressor.

We decided to examine the N-glycome of secreted proteins for two reasons. First, we hypothesized that Sal also functions as a glycosylation inhibitor considering the observed effects on the Golgi, similar to Mon, an established inhibitor of glycosylation (Dinter and Berger, 1998). Second, as EMT cells were shown to depend profoundly on the increased protein secretion (Feng et al., 2014), we also hypothesized that not only diminished secretion but also a change in glycosylation of secreted proteins negatively affects EMT cell maintenance and thus represents the underlying mechanism for the selectivity of these compounds. Analysis of the secretome, in comparison to the complete proteome, allowed us to overcome possible masking by uncompleted glycan structures of proteins undergoing the process of glycosylation in cellular compartments.

The major glycan structural features in EMT and non-EMT cells were mostly similar, with both having a similar distribution of N-glycan types (oligomannose, hybrid and complex), and similar relative amounts of fucosylated structures and structures with higher branching (Fig. 5A–F). Moreover, the secreted N-glycomes of both EMT and non-EMT cells comprised primarily complex-type N-glycans (>80%). We also found one paucimannose structure (peak 3). However, the relative ratios of specific glycan structures were dramatically different, resulting in N-glycome profiles specific for each cell line (Fig. S5). This difference was expected because of different expression profiles between EMT and non-EMT cells and thus the different composition of secretomes. Besides, it was demonstrated that breast cell lines exhibit specific N-glycosylation signatures of their secretomes relative to their tumorigenic level and breast cancer subtype (Lee et al., 2014).

After chromatogram integration, we defined 33 distinct chromatographic peaks in EMT and 30 peaks in non-EMT cells (Fig. S5; Tables S4 and S5). The N-glycan structures present in each peak were determined by hydrophilic interaction ultrahigh performance liquid chromatography fluorescence tandem MS (HILIC-UPLC-FLR-MS/MS) approach. Many peaks contained more than one glycan structure, but in most cases, one single structure was the most abundant (i.e. the major structure; Tables S4–S6). The glycan profile of EMT cells was dominated by three peaks (14, 16 and 20), which represented 39.3% of the total chromatogram. On the other hand, the glycan profile of non-EMT cells was dominated by four peaks (16, 17, 19 and 20) representing similarly 39.2% of the total glycome profile. Major glycans found in peaks 16 and 20 from HMLE-Twist cells matched the major glycans found in peaks 16, 17 and 19, 20 from HMLE-pBp cells (FA2G2S1 and FA2G2S2, respectively; Fig. S5D, Tables S4 and S5). Furthermore, EMT cells lacked terminal fucosylation reflecting the fucosyl transferases' expression. In addition, EMT cells had reduced total relative amount of sialylated glycans in comparison to non-EMT cells. Besides, sulfated or phosphorylated N-glycans (S/P) were found only in EMT cells, which had four such structures forming individual peaks 17, 21, 28 and 32 (Fig. 5A–F, Fig. S5, and Tables S4 and S5).

Treatment with either Sal or Mon had a drastic impact on the N-glycosylation of secreted proteins in both cell lines. Strikingly, the effect of these two ionophores was extremely similar. Detailed analysis showed that each observed change was in the same direction and of similar potency when compared to mock-treated samples (Fig. 5A–F, Fig. S5B–D, and Tables S4 and S5). Thus, as treatments with both ionophores produced N-glycome profiles that were hardly distinguishable, we will mainly discuss treatment versus control regarding glycosylation. The differential effect of the treatment between cell lines was marginal (Fig. 5A; Fig. S5D). Treatment did not completely abolish N-glycosylation of secreted proteins. Rather, it mostly blocked the processing of larger and more complex N-glycan structures as the relative amount of these glycans was significantly reduced in both cell lines. In turn, this led to a relative increase of simple, oligomannose-type N-glycans structures, reflecting the semi-stalled glycosylation machinery (Fig. 5A; Fig. S5D). Moreover, an increase in oligomannose structures after treatment is indicative of Golgi- rather than ER-specific action, because mannoses are formed at the ER. Therefore, a decrease in oligomannose structures would be expected if the treatment is ER-specific. A similar phenotype was observed on all structures that appeared to have increased N-glycan complexity. For instance, we observed decreased total sialylation, with tri-/tetra-antennary sialylated structures especially affected, and decreased branching and decreased fucosylation of the N-glycome (Fig. 5B–E). These findings demonstrated that Sal and Mon disturb N-glycosylation at the stage of building complex glycans, which takes place within the Golgi.

Sal has been in the spotlight of cancer research ever since its EMT- and CSC-selective capabilities were revealed (Gupta et al., 2009). Although numerous effects of Sal in different cellular models have been reported, the mechanism of its selectivity against EMT cells remained elusive. Our work, for the first time, establishes Sal as a Golgi-disturbing agent, positioning it along with the established Golgi inhibitors monensin and nigericin. The EMT and CSC selectivity of Mon was also recently experimentally confirmed in a prostate cancer model (Vanneste et al., 2019). In our experiments, both Sal and Mon led to marked morphological perturbations of the Golgi accompanied by hindered post-translational protein modifications, elevated ER stress and an UPR, diminished secretion and highly altered N-glycosylation of secreted proteins. These changes induce the expression of ER-, Golgi- and membrane-related genes, pointing to problems in the secretory pathway.

Sal targeted the Golgi in both EMT and non-EMT cells. However, the individual responses were largely different and resulted in different outcomes in EMT and non-EMT cells. Importantly, the effects of Sal are concentration-dependent, and selectivity is achieved at low concentrations (0.2 µM). Our results showed that Sal caused a strong fragmentation of Golgi cisternae in HMLE-Twist (EMT) cells, whereas it had no similar effect in their non-EMT counterparts. However, the Golgi of EMT cells was on average much more condensed than that of non-EMT cells. Although this might be a specific feature of this particular cell model (HMLE), EMT has been shown to induce the compaction of Golgi by inducing high levels of a scaffold protein PAQR11 in EMT-driven lung adenocarcinoma models (Tan et al., 2017). However, systematic evaluation has yet to provide evidence for the universality of this phenomenon.

We also demonstrated that Sal specifically targets the Golgi and, like Mon and Nig, primarily affects the trans-Golgi compartment. For example, we observed incomplete proteolytic removal of the N-cadherin prodomain in EMT cells after Sal and Mon treatment. Activation of cadherins requires proteolytic removal of their prodomain in the late Golgi compartments (trans-Golgi network; TGN) by proprotein convertases, most likely furin (Koch et al., 2004; Maret et al., 2010). However, the removal of E-cadherin prodomain in non-EMT cells (HMLE-pBp, HMLE-shGFP or MCF-7), which requires the same proprotein convertases, was not inhibited by Sal. On the other hand, BFA, which inhibits vesicle formation between the ER and the Golgi, also impaired proteolytic cleavage, but only in non-EMT cells, and also showed selectivity toward HMLE EMT model cells. It has been shown that compartmentalization of enzymes located at the Golgi can vary between different cell types, leading to different effects in different cell models (Rosa et al., 1992). Consequently, BFA might not exert the same effect on Golgi function that Sal does.

In addition, we primarily detected diminished relative amounts of complex N-glycan structures after the treatment with Sal and Mon. All glycosyltransferases that introduce and process components of complex N-glycans (e.g. MGATs for branching, STGALs for sialylation and FUTs for fucosylation) are found within cisternae of the Golgi, mostly in the TGN (Varki et al., 2009). Furthermore, we observed the activation of ER-related UPR, which has already established as the mechanism behind the selectivity towards EMT cells (Feng et al., 2014). The activation of UPR in our experimental setting is likely a downstream effect of inducing the Golgi stress through the excessive accumulation of properly folded proteins. It was previously shown that the inhibition of N-glycosylation inside the Golgi leads to the accumulation of improperly glycosylated proteins within the Golgi, the diminished secretion of proteins and the induction of the ER-stress-related UPR (Xu et al., 2010). In our study, Sal activated the PERK and not the IRE1 branch of UPR, positioning it in a cluster of UPR-activating compounds together with Mon and BFA, and not the cluster containing tunicamycin and thapsigargin (Shinjo et al., 2013).

Maintenance of acidic compartments of the secretory pathway is crucial for proper protein processing, glycosylation, and sorting, as well as the organelle morphology (Rivinoja et al., 2012; Weisz, 2003). We hypothesize that Sal leads to alkalinization of the Golgi by an increased release of protons from the organelle in exchange for K⁺ ions, similar to the effect of Mon and Nig (Fig. 5G). However, Sal and Mon affect the Golgi differently with regard to the EMT status of the cell and/or at different points within this organelle. This, in turn, selectively affects the viability of EMT cells, most probably by abrogating their proper maintenance.

It has been previously shown that increased sensitivity of EMT-like cells to Mon is associated with increased uptake of Mon compared to that in resistant cells (Vanneste et al., 2019). Our results clearly show a pronounced effect of lower concentrations of Sal on protein processing in EMT cells, which was lost at higher concentrations, suggesting more severe effects on the cells and possibly causing the loss of selectivity. However, the data from this study do not rule out selective uptake of Sal as suggested by Vanneste et al. (2019) for Mon, possibly due to different cell lines and methods used to assess this issue. For example, in the published study on uptake of Mon, a radiolabeled compound was used to measure uptake, and data were measured at different time points. In contrast, the data in this study were obtained at a single time point and effects were undetectable after treatment of the cells with a Sal concentration of 0.2 µM, which was important for selectivity. Therefore, further optimization of the measurements for Sal concentration should be sought in order to draw clear conclusions about the selective uptake of Sal. In addition, it is conceivable that Mon and Sal have different cell penetration abilities, and although BFA has some selectivity, its effect on the function of the Golgi might not be comparable to that of Sal.

Furthermore, Mon was shown to exert its best ionophoric activity in the cholesterol-rich membranes (trans-Golgi) as compared to the low-cholesterol membranes (cis-Golgi) (Dinter and Berger, 1998; Orci et al., 1981). It has been shown that the variable lipid/cholesterol environments within organelles regulate traffic of membrane proteins (Lippincott-Schwartz and Phair, 2010). Although critical for the maintenance of normal cellular function, the relationship between cholesterol homeostasis in organelles and EMT is not well understood. Therefore, it remains an open question whether the observed different outcomes between EMT and non-EMT cells are the result of the different degree of alkalinization and/or cholesterol distribution as well as the different Golgi morphology.

Changes in glycosylation are of major importance for understanding the significance of the EMT during carcinogenesis (Taniguchi and Kizuka, 2015). Interestingly, the effects on N-glycosylation were very similar in EMT and non-EMT cells. Although we only measured the effects on N-glycosylation of secreted proteins, we can assume that the changes in glycan structures also extend to nonsecreted and membrane proteins, as well as to additional glycosylation classes processed within the Golgi (e.g. O-linked and lipid glycosylation, glucosaminoglycans), which might be the reason for the different sensitivity of EMT and non-EMT cells.

In addition, our study builds on the seminal research that highlighted ER stress as a vulnerability of cancer cells undergoing EMT due to the excessive synthesis and secretion of ECM proteins by the EMT cells (Feng et al., 2014). Our analysis of gene expression data after treatment with Sal confirmed the enrichment of ECM-related genes specifically in EMT cells, suggesting a role for the Golgi in regulating ECM organization and cell adhesion. The Golgi plays a critical role in coordinating post-translational modifications and proteolytic processing of ECM components, which might have far-reaching effects on ECM function (Hellicar et al., 2022). However, the precise impact of the Golgi on ECM remodeling in cancer has not been fully elucidated, and further studies are needed. Despite the gaps in our knowledge, the consistent findings of an enrichment of ECM-related genes in EMT cells and their response to Sal treatment support the importance of the Golgi in controlling ECM production and function.

Described observations led us to conclude that EMT cells are sensitive to Golgi-disturbing agents, first, because they are highly dependent on the function of the Golgi and, second, because significant differences between the structure and function of the Golgi in epithelial and mesenchymal cells might be the reason for the different outcomes. Our results, most of which were obtained from a specific EMT model, did not reveal explicit targets that clearly prove that the Golgi is critical for EMT selectivity. Moreover, the ionophores used in this study target multiple aspects of the Golgi function and are known to have various toxic effects on the whole organism (Boehmerle et al., 2014). Nevertheless, we believe that this work should stimulate further studies focused on elucidating the mechanisms that regulate the function of the Golgi in EMT and that could lead to therapeutic targeting of individual components. Elucidation of the specific Golgi-related processes would lead to the development of more precise inhibitors with better efficacy and lower toxicity. For example, a recent study identified two novel regulators of Golgi secretion as potential drug targets in lung cancer (Bajaj et al., 2020). Similarly, Eisenberg-Lerner et al. have shown that disruption of Golgi homeostasis triggers multiple myeloma cell death in vitro and in vivo, providing a therapeutic strategy for this type of cancer cells, which are highly involved in protein glycosylation and secretion (Eisenberg-Lerner et al., 2020).

Given the strong relationship between the Golgi architecture and a number of higher-order functions triggered by this organelle, such as cell polarization, attachment, migration, proliferation, mitosis, differentiation, etc., and the close physical and functional connection of the Golgi with the ER, it is surprising that the Golgi has remained a largely unexplored target for drug intervention. The finding that EMT cells are sensitive to Golgi-disturbing agents is important not only for the treatment of highly malignant tumors, especially those with a high proportion of EMT cells, but also for uncovering broader implications of the Golgi in EMT plasticity. Further studies on the regulatory mechanisms controlling the function of the Golgi in other epithelial and mesenchymal cells will be important.

HMLE cells expressing Twist or empty vector (HMLE-Twist and HMLE-pBp) were kindly provided by Dr Tamer T. Onder, HMLE cells expressing shRNA targeting E-cadherin or GFP (HMLE-shEcad and HMLE-shGFP) and SUM 159 were kindly provided from Dr Robert A. Weinberg's laboratory. HMLE cells were maintained in a 1:1 mixture of HuMEC Ready Medium (Gibco) and DMEM complemented with 10% FBS, 2 mM L-glutamine, 100 U/ml penicillin, 100 µg/ml streptomycin, 2.5 μg/ml insulin and 0.5 µg/ml hydrocortisone (Sigma-Aldrich). MCF-7 and MDA-MB-231, HepG2 and HCT 116 cell lines were obtained from the ATCC, and maintained in DMEM complemented with 10% FBS, 2 mM L-glutamine, 100 U/ml penicillin and 100 µg/ml streptomycin, whereas SUM 159 cells were grown in Ham's F12 medium with 5% FBS, 1 μg/ml hydrocortisone and 5 μg/ml insulin (Sigma-Aldrich). HeLa cells stably expressing the trans-Golgi marker GalT–GFP and H2B–mCherry (Neumann et al., 2010) were kindly provided by Dr Rainer Pepperkok (EMBL, Heidelberg, Germany) and maintained at the Advanced Light Microscopy Core Facility (ALMF), EMBL, Heidelberg, Germany. All cells were grown in a humidified atmosphere at 37°C with 5% CO₂, and were tested for mycoplasma contamination on a regular basis. EMT and non-EMT status was regularly checked by probing for EMT markers (CD24/CD44 by FACS and E- and N-cadherin by western blot). Salinomycin (S4526), nigericin (N7143), monensin (46468), paclitaxel (T7402), tunicamycin (T7765), brefeldin A (B6542), bafilomycin A (B1793) and GalNAc(α)-O-bn (B4894) were purchased from Sigma-Aldrich.

Cells were seeded (3000 per well) in standard 96-well microtiter plates and left to attach. On the next day, test compounds were added at 10 different concentrations in quadruplicates. The cell growth rate was evaluated after 72 h of incubation by adding MTT reagent (Sigma-Aldrich). Obtained results for each compound were calculated and normalized to mock-treated cells. Dose–response curves were plotted in GraphPad Prism using a non-linear fit.

For mixed culture experiments, cells were seeded in six-well culture plates at 1:1 ratio (75,000 EMT and 75,000 non-EMT cells) and left to attach. Next day, the medium was replaced by medium containing vehicle (DMSO) or test compound. After 72 h of treatment, cells were detached, washed in blocking solution (2% BSA in PBS) and immunolabeled with anti-CD24 PerCP/Cy5.5 and anti-CD44 FITC primary antibodies (BioLegend, 1:50, 311115 and 1:100, 338803, respectively) for 30 min at room temperature (RT). Cells were then washed twice with PBS and transferred to FACS tubes. 10,000 cells were analyzed on a BD FACSCalibur, and the percentages of CD24/CD44-specific populations were calculated in FlowJo (TreeStar, Inc.).

Cells were collected by trypsinization, pelleted (250 g for 5 min) and washed with PBS on ice. Pellets were lysed with RIPA buffer supplemented with complete protease inhibitor (Roche) supplemented with phosphatase inhibitors (NaF and Na₃VO₄, Sigma-Aldrich) and subsequently sonicated using Bioruptor (Diagenode). The total concentration of proteins in cell lysates was measured by using the Pierce BCA Protein Assay Kit (Thermo Fisher Scientific) according to the recommendations from the manufacturer. 50 µg of protein per sample were loaded on a gel, separated using SDS-PAGE and transferred to PVDF membrane (Carl Roth). After protein transfer, membranes were stained using Naphthol Blue solution (0.1% Naphthol Blue in 10% methanol, 2% acetic acid) to visualize transferred proteins that was used to evaluate equal sample loading. Next, membranes were incubated in blocking solution [5% non-fat milk in Tris-buffer saline with 0.1% Tween 20 (TBST)] for 20 min at room temperature. After blocking, membranes were incubated with primary antibodies at 4°C overnight. Primary antibodies used were against: E-cadherin (1:500, Becton-Dickinson, 610181), N-cadherin (1:500, Becton-Dickinson, 610920), proN-cadherin (1:500, R&D Systems, AF1388), EGFR (1:1000, Cell Signaling Technology, 4267), TFRC (1:500, Thermo Fisher Scientific, 13-6800), eIF2α (1:1000, Cell Signaling Technology, 9722), p-eIF2α (1:1000, Cell Signaling Technologies, 3597) and CREB-2 (1:100, SCBT, sc-200). After washing the membranes with TBST, membranes were incubated with complementary HRP-conjugated secondary antibody (sheep anti-mouse IgG–HRP, 1:10,000, GE Healthcare, NA931; goat anti-rabbit IgG–HRP, 1:5000, Bio-Rad, 1706515; rabbit anti-sheep IgG–HRP, 1:3000, Invitrogen, 61-8620) for 2 h at room temperature and subsequently washed again with TBST before detection of signal by Western Lightning Plus-ECL reagent (Perkin-Elmer). Emitted signal from membranes was collected and visualized with UVITEC imaging system (Cleaver Scientific) and final images were prepared in Photoshop CS2 and Illustrator CS2 (Adobe). Uncropped images of the blots shown in Figs 3, 4 and Fig. S4 are shown in Fig. S6. To remove N-glycans from proteins, cell lysates were additionally treated with 16 units of PNGase F (New England Biolabs) per µl of lysate complemented with 1% NP-40. De-glycosylation reaction was performed at 37°C for 60 min.

Cells were seeded onto poly-L-lysine-treated glass cover slides, left to attach and the next day treated with compounds as indicated. Cellular compartments were stained according to the following protocol. After 24 h of incubation slides were fixated with 4% PFA in PBS for 10 min and then permeabilized with blocking solution (2% BSA, 0.05% Tween-20 in PBS) for another 10 min. Slides were then incubated with primary antibody in blocking solution for 30 min at RT. Primary antibodies used were against: GM130 (1:500, BD Biosciences, 610822), GIANTIN (1:2000, Abcam, ab24586), ERGIC53 (1:1000, Enzo Life Sciences, ALX-804-602) and β′COP (1:500, rabbit polyclonal anti-β′COP antibody was a kind gift from Dr Rainer Pepperkok, EMBL Heidelberg, Germany). Slides were then washed and incubated with complementary goat anti-rabbit IgG conjugated to Alexa Fluor 488, goat anti-mouse IgG conjugated to Alexa Fluor 568 or goat anti-mouse IgG conjugated to Alexa Fluor 488 secondary antibody (Thermo Fisher Scientific, A-11034, A-11031 and A-11001, respectively) diluted 1:300 in blocking solution at RT. Fianlly, slides were extensively washed with blocking solution, and in the last washing step 100 ng/ml of DAPI was used as a counter stain for DNA. Slides were then mounted onto glass slides using Fluoromount antifade reagent (Sigma-Aldrich). Images were collected on a Leica SP8 confocal microscope, processed in ImageJ and prepared for publication in Illustrator CS2 (Adobe). Live-cell imaging of HeLa cells stably expressing the fluorescent markers H2B–mCherry and GalT–GFP for 24 h at a time-lapse of 15 min was performed with automated epifluorescence microscope (IX-81; Olympus Europe) using an in-house modified version of ScanR and an image-based autofocus routine, as described previously (Neumann et al., 2010).

Cells were seeded onto 10 cm plates (1.3×10⁶ cells per plate) and left to attach overnight. Next day, the medium was replaced by medium containing vehicle (DMSO) or 0.2 µM Sal. Ater 24 h, total RNA from biological triplicates of HMLE-Twist and HMLE-pBp cells was extracted (12 biological samples in total) with Quick-RNA Miniprep kit (Zymo Research) according to the manufacturer's protocol. DNase I was used to remove possible DNA contamination. RNA concentration was measured on Qubit 3.0 Fluorometer (Thermo Fisher Scientific) and the quality of RNA was additionally checked with Bioanalyzer 2100 (lowest RIN=9.7, Agilent). cDNA library was prepared with TrueSeq Stranded mRNA kit (Illumina) from 750 ng of RNA (average cDNA size was ∼345 bp). The library was sequenced on a NextSeq500 (Illumina) using high output flow-cell in paired-end mode (75+75 cycles). Raw data yielded over 690 million high-quality reads (Q30>75%). Reads were uploaded to BaseSpace Sequence Hub (Illumina) and aligned to human reference genome (hg19) via RNA-Seq Alingment app (STAR alignment, ver 1.1.1). Differential expression of genes after treatment for each cell line and between EMT and non-EMT cell lines was executed via DESeq2 app (ver 1.1.0) in paired mode. The data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus (Edgar et al., 2002) and are accessible through GEO Series accession number (GSE124975).

For Gene Ontology analysis we used Gene Ontology Enrichment Analysis and Visualization tool (GOrilla, http://cbl-gorilla.cs.technion.ac.il/) by loading single ranked genes from each differential gene expression experiment (genes were ranked according to differential expression from highest to lowest). Additionally, we performed GO analysis of overlapping genes from different experimental settings with Panther GO algorithm (http://www.geneontology.org/). Venn diagrams were plotted via the web tool of Ghent University (http://bioinformatics.psb.ugent.be/webtools/Venn/).

For gene set enrichment analysis (GSEA), expression data obtained after treatment with Sal in both EMT and non-EMT cells was analyzed for enrichment of top 100 genes induced by monensin (GSE5258; Lamb et al., 2006; Hieronymus et al., 2006), nigericin and HyT36 Golgi UPR (GSE99490; Serebrenik et al., 2018), doxorubicin (GSE39042; Flamant et al., 2012), 4MU-xyloside (GSE117938; Sasaki et al., 2019), brefeldin A (GSE94861; Yamaguchi et al., 2017), tunicamycin and thapsigargin (GSE24500). In addition, GSEA was performed for genes in gene ontology category ‘Glycosylation’ downloaded from the Molecular signature database (MsigDB). All GSEA analyses were executed with the use of GSEA software downloaded from Broad Institute (MIT) (Subramanian et al., 2005).

Raw data from initial compound screen by Gupta et al. (2009) was downloaded and analyzed in R software. To generate additional compounds with selective properties, we performed three algorithm searches, which yielded 32, 54 and 183 compounds combining to total of 227 unique compounds. Algorithms were: (1) shGFP BsubvalueA (or B) > of mean of BsubvalueA (or B) AND shEcad ZscoreA (or B) is < mean – 1 s.d. of ZscoreA (or B); (2) shGFP BsubvalueA (or B) > of mean of average (BsubvalueA, BsubvalueB) AND shEcad ZscoreA (or B) is < mean – 1 s.d. of ZscoreA (or B); (3) shGFP BsubvalueA (or B) > of mean – 1 s.d. of BsubvalueA (or B) AND shEcad ZscoreA (or B) is < mean – 2 s.d. of ZscoreA (or B).

For the analysis of UPR-related splicing of XBP1, 150,000 cells were seeded in 6-well plates and left to attach. The next day, the medium was replaced with medium containing vehicle (DMSO) or tested compounds. After 24 h of treatment, cells were washed with ice-cold PBS, and RNA was isolated from cells using TRIzol reagent (Sigma). Reverse transcription and subsequent PCR amplification was performed with OneTaq RT-PCR Kit (New England Biolabs). Primers were: GAPDH-For (5′-GAGTCAACGGATTTGGTCGT-3′), GAPDH-Rev (5′-TTGATTTTGGAGGGATCTCG-3′), XBP1-For (5′-AAACAGAGTAGCAGCTCAGACTGC-3′) and XBP1-Rev (5′-TCCTTCTGGGTAGACCTCTGGGAG-3′).

Cells were grown in 24-well plates (75,000 cells/well) in 1 ml medium containing vehicle (DMSO) or compounds. After 24 h, the medium above cells was collected and analyzed for the amount of IL-8 secreted. ELISA was performed with Human IL-8 ELISA Ready-SET-Go kit (Affymetrix eBioscience) according to the manufacturer's protocol.

Cells were seeded onto poly-L-Lysine-treated glass cover slides, left to attach and the next day treated with compounds as indicated. Lipid droplets were stained according to the following protocol. After 24 h of incubation, slides were washed with PBS twice and then fixated with 4% PFA in PBS for 15 min at RT. After fixation, cells were again washed twice with PBS and then stained with 150 nM Nile Red (Sigma-Aldrich, 72485) in PBS for 15 min at RT. Cells were then washed with PBS twice and mounted onto microscopy slides using Fluoromount antifade reagent containing 100 ng/ml DAPI (Sigma-Aldrich). Images were collected on a Leica SP8 confocal microscope, processed in ImageJ and prepared for publication in Illustrator CS2 (Adobe).

500,000 cells in suspension were loaded with 3 µM Fluo-4 AM (Thermo Fisher Scientific, F14201) in DMEM without serum supplemented with 0.02% Pluronic-F127 (Sigma, P2443) for 60 min at 37°C. The cells were then washed in HBSS and left in DMEM without serum for additional 20 min at 37°C to allow complete de-esterification of intracellular Fluo-4 AM. Cells were then washed twice with HBSS and, after the last wash a 400 µl of compound in HBSS was added to cells (final concentration as indicated). Green fluorescence of 10,000 cells was measured 10 and 30 min after the addition of compound on a BD FACSCalibur. The collected data was further analyzed in FlowJo (TreeStar).

25,000 cells were seeded on a 3.5 cm dishes with glass bottom (ibidi, 81156) and left to attach overnight. On the day of imaging, cells were loaded with Fluo-4 AM as for flow cytometry above. For fluorescence imaging, cells were washed twice with HBSS with 1.2 ml was left over cells for imaging. After the desired area with cells was found, cells were imaged every 15 s for 10 min. 400 µl of 4× concentration of compound in HBSS was added directly to the cells (final concentration indicated) under the microscope before the second image (after ∼10 s). Images were collected on a Leica SP8 confocal microscope with a 40× immersion objective. The Fluo-4 fluorescence intensity of multiple cells was measured in ImageJ and normalized to the same cell from the first image.

500,000 cells in DMEM were treated with Sal at indicated concentrations. After 30 min of incubation 2.5 µg/ml of JC-1 mitochondrial potential fluorescent probe (eBioscience, EBI-65-0851-38) was added to the cells for additional 30 min of incubation (1 h total of Sal treatment). Cells were then placed on ice and analyzed on a BD FACSCalibur flow cytometer, and the percentages of red and green fluorescent specific populations were calculated in FlowJo (TreeStar).

Cells were seeded onto 10 cm plates (10⁶ cells per plate) and left to attach overnight. Next day, the medium was discarded, cells were washed with PBS four times and 9 ml of medium (without FBS or supplements) containing vehicle (DMSO) or compound was added to each plate. After 48 h, plates were placed on ice, medium was collected into tubes, and residual cells and large debris was pelleted at 2000 g for 5 min. Supernatant was collected and concentrated with Amicon Ultra centrifugal filter devices with 10 kDa cut-off membrane (Merck). Protein concentration was measured by Pierce BCA Protein Assay Kit (Thermo Fisher Scientific) according to the recommendations from the manufacturer. 15 µg of proteins from each sample were denatured with the addition of SDS and by incubation at 65°C. The excess of SDS was neutralized with Igepal-CA630 (Sigma-Aldrich) and N-glycans were released following the addition of PNGase F (Promega) in PBS. The released N-glycans were labeled with 4-amino-N-(2-diethylaminoethyl) benzamide (Procainamide; ProA). Free label and reducing agent were removed from the samples using hydrophilic interaction liquid chromatography solid-phase extraction (HILIC-SPE). Fluorescently labeled N-glycans were analyzed by HILIC-UPLC-FLR-MS/MS on an Acquity UPLC instrument (Waters) consisting of a quaternary solvent manager, sample manager, and an FLR fluorescence detector set with excitation and emission wavelengths of 310 and 370 nm, respectively. N-glycans were separated on a Waters BEH Glycan chromatography column, 150×2.1 mm i.d., 1.7 μm BEH particles, with 100 mM ammonium formate, pH 4.4, as solvent A and ACN as solvent B. The separation method used a gradient of 32–38% solvent A for 16 min and 38–55% solvent A for next 24 min. A flow rate of 0.56 ml/min was constant during the whole analytical run. Samples were maintained at 10°C before injection, and the separation temperature was 25°C. The UPLC was hyphenated to the Bruker compact Q-Tof mass spectrometer (MS), both controlled using HyStar software (Bruker) version 4.1. UPLC was coupled to MS via Ion Booster (Bruker) ion source. Capillary voltage was set to 2250 V with nebulizing gas at pressure of 5.5 Bar. Drying gas was applied to source at a flow rate of 4 l/min and temperature of 300°C, while vaporizer temperature was set to 300°C and flow rate of 5 l/min. Nitrogen was used as a source gas, whereas argon was used as collision gas. Ion energy was set to 5 eV, transfer time was 100 μs. Spectra were recorded in mass range from 50 m/z to 4000 m/z at a rate of 0.5 Hz. For MS acquisition collision energy was set to 4 eV. Fragment spectra were acquired using auto MS/MS mode. HILIC-UPLC-FLR chromatograms were used for quantification, and abundance of each glycan was expressed as percentage of total integrated area. N-glycan compositions and structural features were assigned using GlycoMod (ExPASy) (Cooper et al., 2001) (http://web.expasy.org/glycomod/) and GlycoWorkbench (Ceroni et al., 2008).

HMLE-pBp and HMLE-Twist cell lines were seeded at a concentration of 200,000 cells per well in six-well plates. The next day, cells were treated in triplicate with Sal (0.2 μM or 5 µM) and incubated for 24 h. To prepare calibrators, an equivalent number of cells were seeded in five wells without treatment. For cell harvesting, the cell culture medium was removed, and the cells were washed with 2 ml PBS and incubated with 0.33 ml of 0.05% trypsin containing 1 mM ethylenediamine tetraacetate (EDTA, Sigma). After incubation at 37°C for 7 min, cells were collected in 1 ml of complete culture medium in 1.5-ml centrifuge tubes, counted centrifuged at 250 g for 5 min, washed with PBS, and centrifuged again (also 250 g for 5 min) to collect the cell pellet. Equal numbers of HMLE-pBp and HMLE-Twist cells were collected for each treatment (0.2 µM and 5 µM Sal) to avoid bias due to lower cell number caused by treatment. Calibration standards were freshly prepared in biological matrix by spiking Sal into the cell pellet samples at final concentrations of 5, 25, 50, 250 and 500 nM for both cell lines individually. 250 nM Sal solution was used as the QC sample. The linearity of the assay was determined from calibration curves obtained by linear regression; peak area was plotted against analyte concentration.

Sample preparation for mass spectrometry (MS) was performed by a liquid-liquid extraction method modified according to Li et al. (2010). For extraction from the cell pellet, all samples and calibrants were homogenized for 15 min by vortexing and sonication in 200 µl of PBS. Then, 600 µl of methanol:water (90:10) mixture was added to precipitate the proteins. The mixture was again shaken and sonicated for 15 min and then shaken at room temperature for 2 h to extract the compound of interest (Sal). The samples were centrifuged at maximum speed (21,130 g) for 10 min, and the supernatant containing the organic phase was transferred to an autosampler vial for analysis MS. The clear solution of the sample or calibrant was injected in a volume of 5 µl onto the LC column.

LC-MS/MS analysis was carried out using an Agilent Technologies 1200 series HPLC system equipped with an binary pump, a vacuum membrane degasser, an automated autosampler and injector interfaced with 6420 triple quadrupole mass spectrometer with electrospray ionization source (ESI) (Agilent Technologies Inc., Palo Alto, CA, USA). The separation was performed on Kinetex C18 column (75×4.6 mm, 2.6 μm paricle size) (Phenomenex, Torrance, USA). Solvents for the analysis were 0.1% formic acid (FA) in water (solvent A) and methanol (solvent B). The gradient was applied as follows: 0 min 30% A, 0–12 min 30–0% A, 12–15 min 0% A, 15.1–20 min 30% A. The flow rate was 0.3 ml/min, and the retention time of Sal was 15.7 min. The electrospray ionization source was operated in a positive mode and samples were detected in the multiple reaction monitoring (MRM) mode with a dwell time of 200 ms per MRM transition. The desolvation gas temperature was 300°C with flow rate of 8.0 l/min. The capillary voltage was 3.5 kV. The collision gas was nitrogen. The MRM transitions of precursor to product ion pairs were m/z 773–531 (qualification transition) and 773–431 (quantification transition). Fragmentor voltage for salinomycin was 135 V and collision energy was set at 40 V. All data acquisition and processing was performed using Agilent MassHunter software.

All graphics with error bars are presented as mean±s.d. except Fig. S3D where mean±s.e.m. was plotted. To determine statistical significance between samples, one-way ANOVA with Dunnett's (Fig. 1A, Fig. S1D, Fig. 3C,D, Fig. S3B) or Tukey's (Fig. 6 and Tables S2 and S3) multiple comparison post-hoc test and two-way ANOVA with Bonferroni's multiple comparisons test was used (Fig. 5B). Statistical calculations were performed in GraphPad Prism and generation of graphics in Excel, R Studio, and Adobe Illustrator CS2 (NS, non-significant; *P<0.05, **P<0.01 and ***P<0.001).

The authors thank Dr Robert A. Weinberg and Dr Tamer T. Onder for providing us with HMLE and SUM 159 cell lines. We thank Dr Rainer Pepperkok and his laboratory for help, expertise and shared reagents regarding microscopy of cellular compartments. Ana-Matea Mikecin Dražić was seconded to the Advanced Light Microscopy Core Facility (ALMF), EMBL, Heidelberg, under WP2 of the FP7-REGPOT-2012-2013-1 InnoMol project. Support from the ALMF Core Facility at EMBL-Heidelberg is acknowledged. We are also thankful to Dr Siniša Volarević and his laboratory for UPR reagents, Dr Oliver Vugrek and Filip Rokić for help with RNA-Seq experiment, Lucija Horvat for assistance with confocal microscopy, Dr Lidija Brkljačić and Dr Ivanka Jerić for performing the LC-MS/MS analysis of salinomycin and Dr Fran Supek for critical reading of the manuscript.

The peer review history is available online at https://journals.biologists.com/jcs/lookup/doi/10.1242/jcs.260934.reviewer-comments.pdf