A drug screening assay on cancer cells chronically adapted to acidosis

RPMI-1640 (SH30255.01), trypsin (SH40003.12), phosphate-buffered saline (PBS, SH40003.12) and Fetal Bovine Serum (FBS) (SV30160.03) were purchased from HyClone. Sodium Bicarbonate (25080) and RPMI-1640 without NaHCO₃ (51800) were purchased from Gibco. Bafilomycin A1 (BafA1, B1793), protease cocktail tablets EDTA-free, phosphatase inhibitors (P5726, P0044) and Verteporfin were purchased from Sigma. Protein assay dye reagent concentrate (500-0006), protein assay standard (5000-0007) and dry milk (170-6404) were purchased from Bio-Rad. The following antibodies were used: LC3B (Cell Signaling Technology, 2775), SQTM1/p62 (BD Biosciences, 610833) and β-actin (Sigma, A5441). HRP-conjugated anti-rabbit (NA934V) and anti-mouse (NXA931) antibodies, ECL system (RPN2106) and PVDF membranes (RPN303F) were purchased from GE Healthcare. The fluorogenic substrate Suc-LLVY-AMC (S-280) was purchased from Boston Biochem (USA).

The colon carcinoma cell line HCT116 was cultured in RPMI-1640 medium supplemented with 10% FBS, antibiotics and 2 g/l NaHCO₃. The low pH adapted HCT-116 cell line (AA-HCT116) was obtained by growing the parental cells in RPMI-1640 medium buffered at pH 6.8 (0.15 g/l NaHCO₃) for 3 months as previously described [13, 14]. In the settings used, media pH was stable over 48 h and dropped by about 0.1–0.2 units after 3 days in culture for cells kept at pH 6.8 and by 0.3–0.4 units for parental cells. Since parental cells have a higher proliferation rate, so was medium pH dropping faster and cells were passaged every 2–3 days. AA-HCT116 cells were passaged every 4–5 days. The h-TERT RPE1 cell line was cultured in DMEM F12 medium supplemented with 10% FBS and antibiotics. The MelJuSo UbG76V-YFP cell line was cultured in DMEM medium completed with 10% FBS and antibiotics.

HCT116 cells were plated at 60,000 cells/well in 100 µl medium in XF24-well cell plates. Subsequently medium was replaced with 500 µl Seahorse assay medium (1 mM pyruvate, 25 mM glucose and 2 mM glutamine). Oligomycin, FCCP, rotenone and antimycin A were injected according to the Seahorse XF cell mito stress kit. The mitochondrial function was assessed detecting the ECAR and OCR values by using the Seahorse XF analyzer.

AA-HCT116 cells were plated at 15,000 cells/well in flat-bottom 96-wells plate, left under the laminar hood for 1 h and then moved to the incubator. On the next day, all plates were treated simultaneously with 10 µM of each of the compounds from the Prestwick chemical library (one compound per well). For each plate, quadruplicate wells were used as untreated controls and quadruplicate wells used for the positive control (VLX600 10 µM) [15]. Plates were then moved back to the incubator and 48 h later cell viability was assessed by the acid phosphatase assay [16]. Hits were identified using a threshold of 70% inhibition of cell viability. Hit confirmation was performed testing each compound in a dose range 0–15 µM in duplicate wells.

Cells were plated in 96-well plates and the next day treated with different concentrations of the specific compounds. Forty-eight hours after treatment cell viability was evaluated by using the acid phosphatase assay.

For analysis of clonogenic cell survival, HCT116 and AA-HCT116 cells were plated at 300 cells/well in 6-well plates and treated with VP for 8 or 48 h, following culture for 6–8 days in drug-free media. Plates were then washed in PBS and colonies were stained with Giemsa.

Clonogenic cell survival was also assessed on HCT116 MCS treated with different compounds. Briefly, MCS were obtained by a modification of the hanging-drop method applied to 96-well plates as previously described [17]. After drug treatment for 3 days, the MCS were collected, washed in PBS, and a single cell suspension was obtained after incubation with Accutase (Biolegend, 423201). Equal volumes of cells for each treatment condition were plated in triplicate in 6-well plates for 6–8 days, followed by staining the colonies with Giemsa. The software ImageJ was used to perform semi-automated colony counting.

Cells were plated in black 96-wells plates (Costar) and allowed to adhere overnight.

On the next day, VP was added in a range 0–15 μM for 4 h. After removal of drug containing medium, cells were extensively washed with PBS. VP fluorescence was recorded with excitation wavelength of 450 nm and an emission wavelength of 695 nm [18] using a fluorescence microplate reader Tecan Infinite M1000 (Tecan, Männedorf, Switzerland). For each condition quadruplicate wells were used and VP fluorescence was normalized to protein content for each well.

Qualitative analysis of intracellular VP was performed by laser-scanning fluorescence microscopy with a Leica confocal microscope.

Analysis of the autophagic flux by Western blot was done on different cell lines. Experiments were performed by plating cells in their appropriate medium and allowing them to adhere overnight. On the next day, VP was added at specified concentrations for 4, 8 or 24 h and BafA1 (100 nM) was added during the last 2 h of incubation. Cells were then collected for Western blotting. Analysis of the autophagic flux was performed using the ratio of normalised levels of LC3-II in the presence and absence of BafA1. All experiments were performed in darkness conditions.

Cells were washed with PBS on ice and collected by scraping in cold PBS. The cell pellet was lysed in RIPA buffer (150 mM NaCl, 50 mM Tris pH 7.4, 1% Nonidet P-40, 0.1% SDS and 0.5% sodium deoxycholate) in presence of protease and phosphatase inhibitors. The protein concentration was determined using the Biorad Protein Assay (Biorad Laboratories) and equal amount of proteins (20 µg) was loaded on pre-casted acrylamide gels (4–12% SDS-PAGE, NuPage). The proteins were transferred from the gel to PVDF membrane for 2 h at 4 °C. Red Ponceau staining of the membranes verified the proper loading and transfer. Membranes were blocked in 5% blotting grade dry milk in TBS with 0.1% Tween (TBS-T) for 1 h at room temperature and then incubated with primary antibodies diluted in 5% BSA in TBS-T overnight at 4 °C. The next day membranes were washed and incubated for 1 h at room temperature with the appropriate HRP-conjugated secondary antibody and the binding was revealed by the ECL system.

The fluorogenic substrate Suc-LLVY-AMC was used to measure the chymotrypsin-like activity of the proteasome on cellular lysates. Briefly, cells were treated with VP or BZ for 4 h, lysed in presence of 0.05% NP40, 1 µM DTT and 2 µM ATP, followed by the addition of Suc-LLVY-AMC (10 μM). Fluorescence intensity was recorded using a Tecan Infinite M1000 microplate reader with excitation wavelength 380 nm and emission wavelength 460 nm.

The cell line MelJuSo UbG76 V-YFP was used as reporter of proteasome inhibition [19]. Cells were plated overnight and then exposed to VP or BZ. Detection of accumulated Ub-YFP was performed by Western blot analysis using an anti-GFP antibody.

RNA sequencing was performed in biological triplicate using Strand-specific TruSeq library preparation and Ribo-Zero ribosomal depletion. Tophat2, HTSeq, and DESeq 2 were utilized for alignment, quantification, and differential expression analysis, respectively, with the hg19 genome and Ensembl v73 [20‐22]. All bioinformatics were preformed in the R programming language [23] and scripts related to the differential expression analysis, gene ontology analysis, and community detection analysis are publically available (https://​github.​com/​GranderLab/​acidAdaptedRNAse​q) as an R package facilitating reproduction of the analysis and associated figures. Principal component analysis and hierarchical clustering was performed using the prcomp and heatmap.2 functions, respectively, within the R statistical environment. Data was centered and scaled (mean = 0, sd = 1) for each gene previous to clustering and regularize log transformed previous to PCA.

Significant (alpha < 0.05) genes in the differential expression analysis were used for biological processes GO term enrichment analysis with the topGO software [24]. All quantified genes, defined as counts per million > 1 in at least 3 samples, were utilized as the gene universe. Terms with less than 5 annotated genes were not included in the significance testing procedure. Significance testing was performed using the classic Fisher method and the top 1000 terms with the lowest p-value were included in downstream analysis unless otherwise specified.

To understand the relationship between the significant GO terms, the GO graph was retrieved for all significant (alpha < 0.05) GO terms using the GOSim Bioconductor package [25, 26]. The GO graph was then utilized for community detection via the spin-glass algorithm [27] from the igraph package [28] using a maximum of 200 possible spin states. The authority score was subsequently calculated for all nodes in the GO graph to determine the node with the highest in-degree for each community [29]. The GO graph was then collapsed on the nodes with the highest authority scores (community node) by merging all nodes into said node and simplifying the graphs edges. In cases when several nodes had identical authority scores equaling the max authority score for that community’s nodes, one was chosen at random to represent the community. All nodes not presently contained in the collapsed graph, were merged into the graph and edges were re-drawn between the merged nodes and the community node. Results were visualized with the graph package [30].

All data were obtained from at least three experimental replicates and presented as mean ± SEM, if not otherwise indicated. Data analysis was performed using Graphpad Prism 6.0 (GraphPad Software, CA). Differences between groups were analysed with parametric or non-parametric tests according to the distribution of the values. The specific test used is indicated in figure legends. The significance level was set as P < 0.05.

It has been previously shown that cancer cells are less sensitive and/or completely resistant to several clinically used drugs in conditions of acidosis, especially when the compounds are basic [8, 31‐33]. To allow studies of such resistance, the colon carcinoma cell line HCT116 was adapted to grow in conditions of chronic acidosis at pH 6.8 starting from the bulk cell population. We refer to this acid-adapted sub-line as AA-HCT116 [13, 14]. We then performed a small validation screen to test the respective sensitivities of HCT116 and AA-HCT116 cells towards a small panel of clinically used drugs. These span multiple oncology related drug classes, including DNA damaging agents (doxorubicin, cisplatin, methotrexate, 5FU), a microtubule stabilizer (Paclitaxel), tyrosine kinase inhibitors (Sorafenib and Erlotinib), a mTOR inhibitor (Everolimus), HDAC inhibitors (Vorinostat and Panobinostat), a proteasome inhibitor (Bortezomib) and an autophagy inhibitor (Chloroquine). As shown in Fig. 1, cells chronically adapted to acidosis are less sensitive to most of the drugs under these conditions, while showing similar level of sensitivity towards Everolimus and Sorafenib. None of the tested drugs were more cytotoxic in conditions of acidosis. In this validation run, we used salinomycin as a positive control since we recently reported that this compound has a preferential cytotoxicity under acidic conditions [14]. These results confirm that acidosis negatively affects the efficacy of anticancer drugs. As reported below, these findings prompted us to use this model system in a broader search for compounds with preferential activity towards acid-adapted cells, but prior to this we initiated a more thorough characterization of the AA-HCT116 cells.

The adapted AA-HCT116 cells showed dramatic phenotypic changes in comparison to the parental cell line, showing a more mesenchymal-like morphology, a lower proliferation rate (doubling time 17 ± 1.5 vs 23 ± 1.6 h) and increased forward (size) and side scatter (granularity) when analyzed by flow cytometry (Fig. 2a, b). RNA sequencing (RNA-seq) performed on both cell lines showed that AA-HCT116 cells have a unique transcription profile, further indicating that specific molecular changes had taken place during adaptation to low pH conditions (Fig. 2c). Changes to cellular transcriptomes in response to such conditioning have also been reported by others [34]. Analysis of the metabolic profile by using an XF Analyser indicated that AA-HCT116 cells are characterized by a lower basal and oligomycin-induced ECAR, suggesting a decreased glycolytic rate compared to parental cells (Fig. 2d). AA-HCT116 cells also tend to have a lower oxygen consumption rate, although this difference was not statistically significant (Fig. 2e).

We utilized RNA-seq data to perform differential expression analysis and discovered 4796 genes to be significantly (alpha < 0.05) altered in AA-HCT116 cells vs. their parental cell line, with 1283 of these genes exhibiting a fold change > 2 (Additional file 1: Figure S1A).

Gene ontology (GO) term enrichment analysis resulted in 579 significantly (alpha < 0.05) enriched terms. Since many of these terms are related to similar biological processes, we desired to summarize these results by identifying groups of similar terms. To achieve this we utilized a community detection algorithm [27] to deduce similar terms within the GO graph using the terms found to be significant, as well as, their ancestors. This resulted in the detection of 39 communities (Fig. 3a). In the communities detected, we identified many terms known to be associated with the biology of acid adapted cells such as cell differentiation, cell death, and cell adhesion, as well as, novel associations such as autophagy, cellular metabolism, and Wnt signaling. Analysis of the gene expression patterns in each community revealed strong contrasts in gene expression between parental and acid adapted cells (Fig. 3b). Finally, analysis of individual expression patterns within each community term revealed both known and novel players in the biology of acid adapted cells (Fig. 3c).

Thus, AA-HCT116 cells maintained under acidic conditions showed characteristics resembling less differentiated cells and with a decreased proliferation rate, further supporting the use of these cells as a model system for identification of compounds targeting cancer cells under these clinically relevant conditions.

The drug screening assay was applied in a 96-well plate format following the optimization of multiple parameters including FBS concentration and cell densities. In the screening assay the compound VLX600 (10 µM) was used as positive quality control (QC), resulting in about 20% cell viability (Fig. 4a). The Z-factor of the assay was calculated based on controls in each plate and averaged 0.6.

The Prestwick chemical library was applied for our search of cytotoxic compounds with efficacy under acidic conditions. Cells were exposed to 10 µM of each of the 1280 compounds diluted in DMSO for 48 h and cell viability was assessed by acid phosphatase assay. Figure 4b displays the percent inhibition of cell viability observed in the presence of library compounds (blue marker) in comparison to the QC (green marker) and DMSO (dark red marker), which were used to define 100% and 0% inhibition, respectively in data normalization. Hits inducing a loss in cell viability > 70% were selected for confirmation studies in full concentration response assays. Out of the initial 55 hits, 11 compounds were confirmed as having a cytotoxic activity in the range 1–10 µM in both culture conditions (Table 1). To further filter for compounds with what we regarded as a desirable in vitro-based efficacy versus safety profile, we compared the concentration-dependent effects of the 11 hits on cell viability in parental HCT116 cells, AA-HCT116 cells, and in the immortalized epithelial cells, h-TERT RPE1. Among the clinically used compounds, only Verteporfin (Fig. 4b, c) and Auranofin (Additional file 2: Figure S2) showed a small window of decreased toxicity towards normal cells and a tendency for preferential activity towards AA-HCT116 cells. Of the remaining hits from the Prestwick library, some turned out to have detergent-like structures and were thus excluded (Additional file 2: Figure S2). Amongst hits that were not pursued at this stage there are also agents with bactericidal activity, although these could be of interest in follow-up studies.

VP (trade name Visudyne^®) is a benzoporphyrine derivative used as a photodynamic agent for the treatment of macular degeneration, with limited toxicity. Recent studies have proposed that VP has anticancer activity both in vitro and in preclinical animal models [35, 36]. Moreover, VP has been suggested as a specific inhibitor of the YAP-1 protein whose role is important in the biology of colon carcinoma [37].

Firstly, we confirmed that VP was equally effective in reducing viability of AA-HCT116 cells with respect to their parental cell line by viability assay (Fig. 5A) and clonogenic cell survival (Fig. 5b) using monolayer cultures. Data from Fig. 5a indicates that the IC₅₀ of VP for HCT116 and AA-HCT116 cells are 3.4 ±0 .77 μM and 1.72 ± 0.43 μM, respectively (P≤0.05). Taken together these results suggest that VP can be effective in killing cancer cells in metabolically stressed environments. To confirm this, we tested the activity of VP in the 3D model of multicellular spheroids (MCS). MCS more accurately reproduce the biochemical and metabolic heterogeneity of the tumor environment by containing hypoxic and acidic regions when compared to monolayer cultures [5]. Interestingly, many of the hit compounds identified in our screen have been reported to inhibit mitochondrial function in 3D drug screening models using HCT116 cells suggesting some selectivity for metabolically stressed cores [38]. We observed that VP was effective in reducing cell viability of MCS (Fig. 5c), similar to compounds like salinomycin (SAL) and VLX600 which have been recently reported to have a strong killing activity in MCS [6, 38]. Given the light-sensitive nature of VP we conducted experiments under conditions where the cultures were either expose or not to ambient light. When the activity of VP on MCS was analysed in terms of clonogenic cell survival we observed a dramatic, light-dependent inhibitory effect even at very low VP doses (Fig. 5d). In fact, despite being a photosensitizing agent, VP has been reported to inhibit cell growth of different cancer cell lines even without light activation [39, 40]. In line with these findings, VP exposure to ambient light resulted in a dramatic enhancement in cytotoxic activity in both HCT116 and AA-HCT116 cells (Fig. 5e, f). Importantly, in our experimental settings we did not employ the specific 690 nm wavelength laser light that is routinely used for photodynamic activation of VP, but instead we simply exposed the treated cell cultures to ambient light for 30 min before moving the plates into the incubator.

Pharmacological properties of compounds such as solubility, logD and pKa are determinants of drug sensitivity [8, 32]. VP is an acidic compound with a predicted pKa of 4.12 and in line with this we observed increased intracellular VP levels by approximately threefold in AA-HCT116 cells as compared to parental HCT116 cells (Fig. 6a). This increased accumulation in acidic conditions was confirmed by confocal fluorescence microscopy (Fig. 6b) on AA-HCT116 and by fluorimetric analysis on parental HCT116 cells transiently exposed to pH 6.8 (Fig. 6b). These data suggest that acidic conditions enhance uptake and intracellular VP availability.

Due to the reported activity of VP as an autophagy inhibitor, we next investigated whether it differentially affected autophagic flux (AF) in parental and AA-HCT116 cells. Surprisingly, we observed that AF measured by WB analysis of LC3-II was not significantly altered under VP treatment in either cell line (Fig. 7a). In line with RNAseq data in Fig. 3c, we observe that basal expression of SQSTM1 is increased in AA-HCT116 cells. SQSTM1 is a protein that functions as an autophagy receptor shuttling ubiquitinated proteins for autophagosomal mediated degradation. VP treatment caused the disappearance of the autophagy cargo receptor SQSTM1 at the expected molecular weight. VP was recently reported to cause cross-linking of SQSTM1 [41] and in fact, we observed that VP treatment caused a concentration-dependent accumulation of high-MW (> 62 kDa) SQSTM1 that was concomitant with loss of SQSTM1 at the expected MW (Additional file 3: Figure S3A). Such an effect of VP on protein oligomerization was recently reported for STAT3 and mTOR [36, 42]. Treatment of HCT116 and AA-HCT116 with VP induced protein oligomerization of mTOR and GRP78 also in our hands (Additional file 3: Figure S3B, C).

We reasoned that disturbance of SQSTM1 function might alter the dynamics of the UPS in AA-HCT116 cells. Utilizing the RNA-seq data, we could identify a clear differential expression of UPS-related genes in AA-HCT116 cells, suggesting that changes in the UPS could be responsible for the increased sensitivity of AA-HCT116 cells to VP treatment (Fig. 7b). Gene ontology (GO) analysis further supported this observation, revealing enrichment of several UPS related gene ontology terms (Additional file 3: Figure S3D). As observed for Bortezomib (BZ), a known proteasome inhibitor, VP treatment also induced a concentration-dependent accumulation of poly-ubiquitinated proteins and concurrent decrease in the amount of free ubiquitin (Fig. 7c), which normally indicates defective proteasomal degradation and/or deubiquitination activity. Interestingly, VP seems to induce the appearance of polyubiquitin complexes with MW higher than those induced by Bortezomib, suggesting some differences in the mechanism of action of these two drugs. To investigate this possibility, we used Suc LLVY-AMC, a fluorogenic substrate for measuring the chymotrypsin-like peptidase activity of the 20S proteasome. HCT116 and AA-HCT116 cells were treated with 1 and 5 µM VP, with BZ used as positive control. As shown in Fig. 7d–e, while the proteolytic activity of the proteasome was strongly reduced by treatment with BZ, no major effects were observed on either of the cell lines treated with VP. Consistent with this observation, VP did not affect degradation of the proteasome substrate UbG76 V-YFP in the MelJuSo reporter cell line, indicating that proteasomal degradation is fully functional in VP-treated cells (Fig. 7f). Moreover, unlike BZ, VP does not affect the expression of cyclin-dependent kinase inhibitor CDKN1A, a protein whose turnover is regulated by proteasomal degradation (Fig. 7g). Due to the light-dependent effects of VP on cell viability, we assessed the effects of light exposure during the treatment and protein extraction steps. As shown in Additional file 3: Figure S3E, the presence of high MW polyubiquitinated proteins was observed when all experimental procedures were conducted in presence of light. In contrast no polyubiquitinated proteins were observed when all procedures were conducted in complete darkness. Taken together, these experiments show that although VP induced the accumulation of high MW ubiquitinated proteins, this effect is light dependent and not due to inhibition of proteasomal activity.