Three-Dimensional Cell Culture-Based Screening Identifies the Anthelmintic Drug Nitazoxanide as a Candidate for Treatment of Colorectal Cancer

This article is featured in Highlights of This Issue, p. 1271

The cancer drug development process has become increasingly costly and inefficient resulting in few new highly effective drugs reaching the market yearly (1). New strategies for drug discovery are therefore urgently needed. One such strategy is drug repositioning in which a new indication for an existing drug is identified. Using this approach, approved, discontinued, or withdrawn drugs with unrecognized anticancer activity can be rapidly advanced into clinical trials, because much of the required documentation already exists. The availability of compound libraries containing drugs with documented clinical use makes unbiased screening for repositioning candidates an attractive approach.

During the past decades, most screening approaches for identification of new cancer drug candidates have used cell-free assays for detection of specific interactions with known molecular targets (2). However, there has been a renewed interest in drug screening focused on compound-induced phenotypic changes in live cells (3, 4). Monolayer two-dimensional (2D) cultures of human tumor cell lines have been the predominant models in these efforts. However, these models do not mimic the complex pathophysiological conditions present in solid tumors (5, 6). Moreover, most of the standard chemotherapy drugs identified by using 2D models target proliferating cells. In recent years, it has been shown that dormant cells, residing in areas far from blood vessels are, at least partially, responsible for cancer drug treatment failure (7). These cells are highly resistant to standard cytotoxic compounds and, due to poor drug penetration, are also insufficiently exposed to the treatment (7).

Therefore, tumor cells grown in three-dimensions (3D) have been suggested to provide a more clinically relevant model. The multicellular tumor spheroid (MCTS) is one such 3D model. MCTS are known to closely simulate the tumor microenvironment with respect to glucose, oxygen, and lactate distribution, resulting in gene expression and phenotypic changes similar to those observed in vivo (5, 8–10). MCTS not only simulate the harsh conditions present in poorly vascularized tumors but also facilitate simultaneous evaluation of the penetration properties of investigated compounds (11, 12). Moreover, if uniform and equal in size, MTCS can be used for drug activity comparisons (8, 13). We have previously shown that compounds identified in 3D-model screens are distinct from those found using 2D models (14, 15). Thus, tumor cells in 3D models are not necessarily always drug resistant, opening a possibility for identification of more active cancer drugs if large-scale 3D-based drug screening could be pursued.

There are numerous MCTS formation methods (14, 16–20) but most are not suited for high-throughput screening (HTS); they are either costly, difficult to handle in a large scale or the spheroids often vary in size and number. Thus, MCTS-based screening efforts performed so far have used small drug libraries, not exceeding a few hundred compounds (14, 15, 17, 21).

Here, we present a novel MCTS model system and illustrate its suitability for HTS. We applied the model for screening a library of 1,600 clinically tested compounds, making hit compounds suitable for drug repositioning. By comparing the activity in our MCTS model with that in 2D cultures, compounds with 3D-specific activity were identified and further evaluated in a 3D-based clonogenic assay. The five most active hits identified were all compounds reported to target mitochondria, supporting our previous finding that respiration is an attractive target in solid tumors (22).

Among the hit compounds nitazoxanide, an FDA-approved antiprotozoal drug with an excellent pharmacokinetic and safety profile, was selected for further evaluation and demonstrated strong antitumor activity in vivo when combined with a standard chemotherapeutic agent.

Antibodies to 4EBP-1 (#9452), phospho-4EBP-1 (#9459), p70-S6K (#9202), phospho-p70-S6K (#9205), AMP-activated protein kinase (AMPK; #2532), phospho-AMPK (#4188), caspase-3 (#9664S), and Wnt Signaling Antibody Sampler Kit (#2915) were purchased from Cell Signaling Technology. Antibody to c-Myc (#sc-40) was purchased from Santa Cruz Biotechnology. Antibody to β-actin (#A5316) was purchased from Sigma-Aldrich. Antibodies to Ki-67 (#IR626) and CD44 (#M7082) were purchased from Dako Sweden AB. Pimonidazole staining kit was purchased from Hypoxyprobe. JC-1 was purchased from Sigma-Aldrich. Pharmakon 1600 (1,600 compounds) was purchased from MicroSource Discovery Systems Inc. Hit compounds: closantel, niclosamide, nitazoxanide, pyrvinium pamoate, and salinomycin; and standard cytotoxic compounds: doxorubicin and oxaliplatin were purchased from Sigma-Aldrich. Tizoxanide was purchased from Cayman Chemicals. All compounds were dissolved in DMSO. DMSO concentration in cell culture during screening did not exceed 0.32%.

HCT116 GFP and HT-29 GFP (human epithelial colon carcinoma cell lines constitutively expressing green fluorescent protein) were purchased from Anticancer Inc. in 2009 and 2014, respectively. HCT116 and HT-29 cell lines were obtained from the ATCC in 2009. The cell banks performed authentications by short tandem repeat analysis. No further authentication was performed in our laboratory. All experiments with purchased cell lines were performed within 6 month after resuscitation. Cells were cultured in McCoy's 5A Modified Medium (Sigma-Aldrich) + v/v 10% inactivated fetal calf serum, antibiotics (streptomycin 50 μg/mL and penicillin 60 μg/mL) and 2 mmol/L l-glutamine at 37°C in 5% CO₂.

Spheroids were formed from HCT116 GFP or HT-29 GFP cells for 7 days without medium change. In 50 μL of fresh medium, 10,000 cells per well were plated into 384-well F-bottom Ultra Low Attachment plates (Corning) using Biomek 2000 (Beckman Coulter). To decrease liquid evaporation, plates were covered with humidified MicroClime Environmental Microplate Lids (Labcyte). To induce cell aggregation, plates were placed in a position, in which corner P1 (bottom-left) was located lower than other corners of the plate. Plates were incubated in this “tilted” position in 37°C, humidified atmosphere containing 5% CO₂ for 3 hours. Subsequently, plates were placed on the laboratory rocker (Vari-Mix Platform Rocker; Thermo Scientific), laying on the P1-P24 edge in a way that angle between rocker's shelf and plate's bottom was around 85°. The rocker was set to the following settings: 3 hours stationary, 15-minute rocking at a speed 5 rpm and maximum rocking angle 48°. The plates were incubated on the rocker in 37°C for 4 days. After this, they were incubated in 37°C for 3 days in a “tilted” position.

Following pimonidazole (200 μmol/L for 1 hour, for pimonidazole adducts stainings) treatment spheroids were washed with PBS, fixed with 4% formalin in PBS, dehydrated with 70% ethanol, embedded in paraffin, and sectioned. REAL EnVision Detection System (Dako, K5007) was used to visualize the target antigen. The sections were deparaffinized and microwaved in Tris–EDTA buffer (pH 9.0) or Citrate buffer (pH 6.0) to unmask the epitopes. Sections were incubated for 5 minutes in peroxidaze blocking solution (Dako, S2023). Following antibody dilutions were used: pimonidazole (1:50), Ki-67 (RTU), caspase-3 (1:100), and CD44 (1:100). After 30-minute incubation with the primary antibody at room temperature, the sections were washed and incubated with Dako REAL EnVision/HRP for 20 minutes, washed, and incubated with DAB for 10 minutes. The sections were counterstained with hematoxylin for 5 minutes.

Pharmakon 1600 drugs were transferred to spheroid plates using Echo Liquid Handler 550 (Labcyte), resulting in final drug concentration of 20 μmol/L (for plates' layouts, see Supplementary Materials and Methods). On days 2 to 7 of drug incubation, mean spheroid GFP fluorescence intensity was measured every day using ArrayScan VTI Reader (Cellomics Inc.). Images were acquired for green fluorescence and bright-field channels using 5× objective and suitable filters. The average pixel intensity of each spheroid was quantified using the BioApplication Morphology Explorer (Cellomics Inc.). One image per well was acquired. As a secondary spheroid viability assessment method, resazurin-based TOX8 assay was performed on day 7. TOX8 solution (5 μL) was added to each well. The plate was incubated in 37°C for 4 hours and then resorufin fluorescence was measured with FLUOstar OPTIMA plate reader (BMG Labtech), with excitation/emission filter settings 544 nm/590 nm. A compound was defined as a hit in GFP-based viability assay when mean spheroid GFP fluorescence intensity from the well containing this compound was lower than mean GFP fluorescence intensity of negative controls by at least 3 standard deviations (SD) of negative controls' mean fluorescence intensity. A compound was defined as a hit in TOX8 assay when the resorufin fluorescence intensity from the well containing this compound was lower than 50% of mean resorufin fluorescence intensity of negative controls.

For monolayer experiments, 2,500 HCT116 cells per well were plated in 50 μL of fresh medium into 384-well cell-culture treated plates (Nunc) and cultivated for 24 hours in 37°C before drug addition. Screening plates' layouts were identical as in spheroid plates. Final drug concentration used was 10 μmol/L. In monolayer experiments, cell viability was assessed after 72 hours of incubation with drugs using fluorescence microculture cytotoxicity assay (FMCA), as previously described (23).

For the evaluation of the assay, one spheroid plate was prepared. One hundred and fifty-two wells were treated with 20 μmol/L clofazimine for 72 hours. Viability measurements were performed at identical time points and using the same methods as in screening. Z-factor was calculated as recommended previously (24).

One hundred and sixty two spheroid-active compounds were screened against spheroids in duplicates at five different concentrations from range 2 to 32 μmol/L. Then, 41 most active compounds were tested in spheroid or monolayer cultures in triplicates at 10 different concentrations from range 0.5 to 32 μmol/L. GFP-based and TOX8 readout methods were used for viability evaluation in spheroids and FMCA was used for monolayer cultures. Twelve 3D-selective compounds (with 3D-based IC₅₀s lower than 2D-based IC₅₀s after 3 days of drug treatment) were chosen for further evaluation. For 3D-selective activity visualization, spheroids or monolayer HCT116 GFP cell cultures were treated with 10 μmol/L nitazoxanide or 15 μmol/L oxaliplatin for 72 hours. Every 24-hour images for 3D and 2D cultures were acquired using ArrayScan VTI Reader (Cellomics) or IncuCyte FLR (Essen BioScience), respectively.

Spheroids were treated with drugs for 24 to 72 hours, depending on a particular experiment. Following the treatment, spheroids were washed with PBS and centrifuged. Supernatant was aspirated and 50 μL of AccuMax (PAA Laboratories GmbH) was added. After 30-minute incubation in 37°C, spheroids were pipetted vigorously 30 times to obtain single-cell suspensions. Ten microliters or all (for Supplementary Fig. S5) of each suspension was added to 3 mL of fresh medium, mixed and plated into 6-well (35 mm diameter) Nunclon Surface plates (Nunc). Following seeding, plates were incubated in 37°C for 10 days. Then, colonies were washed with PBS, preserved with methanol, stained with 5% Giemsa dye in PBS and counted. Wells too dense to count were assumed to contain over 200 colonies. Each drug incubation variant was performed in triplicate.

The Seahorse XF analyzer was used as indicated by the manufacturer (Seahorse Bioscience). Seventy thousand HCT116 cells per well in 100 μL culture medium were plated in XF24-plate containing blank controls. Before the measurements, medium was replaced with 500 μL of Seahorse assay media (1 mmol/L pyruvate, 25 mmol/L glucose, pH 7.4) at 37°C without CO₂ for 1 hour. Oxygen consumption rate (OCR) values were measured by XF24 Extracellular Flux Analyzer. Oligomycin and FCCP were used at 1 or 0.5 μmol/L, respectively.

Twenty-five hundred HCT116 cells per well were plated in 50 μL of fresh medium into optical bottom, black 384-well cell-culture–treated plates (Nunc) and cultivated for 48 hours in 37°C before drug addition. Following 2-hour drug treatment, 35 μL of medium was aspirated and warm staining solution added (containing Hoechst 33342 for staining nuclei and JC-1 in PBS). Final concentration of JC-1 was 2.5 μg/mL. The plate was incubated in 37°C for 20 minutes, washed three times with PBS, and read immediately in Cellomics ArrayScan VTI Reader. Images were acquired for blue (Hoechst 33342) and red (JC-1) fluorescence channels using appropriate filters and 20× objective. The average pixel intensities from detected cytoplasmic spots (mitochondria) were quantified using the BioApplication Spot Detector (Cellomics Inc.). At least 1,000 cells per well were analyzed and at least 7 wells per condition were measured.

HCT116 cells were treated with 17 μmol/L nitazoxanide. Following treatment, basal OCR was measured with Seahorse XF analyzer after 24, 72, 120, or 168 hours of incubation in drug-free medium (cell number was assessed before each measurement and cells were split when necessary).

Twenty-five hundred HCT116 cells per well were plated in 50 μL of fresh media (DMEM containing 10% inactivated fetal calf serum, streptomycin 50 μg/mL, penicillin 60 μg/mL, and 2 mmol/L l-glutamine) with or without glucose into 384-well cell-culture–treated plates (Nunc) and cultivated for 24 hours in 37°C before drug addition. Final hit compounds were added in triplicates. Cell viability was assessed after 72-hour drug incubation using FMCA.

Spheroids were treated with nitazoxanide or tizoxanide at five concentrations from range 0.1 to 10 μmol/L for 24 hours. After treatment, spheroids were washed with PBS and kept in −80°C until further handling. For one experiment, at least 16 spheroids for each drug concentration were used. Western blotting was then performed as described previously (22). Primary antibodies were used at the following dilutions: β-actin (1:10,000), C-myc (1:1,000), AMPK (1:1,000), phospho-AMPK (1:1,000), p70 (1:1,000), phospho-p70 (1:1,000), 4EBP1 (1:1,000), phospho-4EBP1 (1:1,000), and Wnt kit antibodies (1:1,000). Horseradish peroxidase (HRP)–conjugated anti-rabbit and anti-mouse antibodies were used at 1:5,000.

Seven- to 9-week-old female NMRI nu/nu mice (Crl:NMRI-Foxn1nu; Charles River) were housed under standard conditions. One hundred microliters of suspension containing 5 × 10⁶ HCT116-GFP cells in culturing medium was injected into the right rear flank of each animal. When tumors reached a volume of >100 mm³, animals were randomized into control (treated with vehicle) or treatment (nitazoxanide, irinotecan, or combination) groups. Nitazoxanide (200 mg/kg; OnTarget Chemistry) was administered by gavage in 5 mL of 1% CMC in PBS/kg twice daily (once daily during weekends) for 28 to 30 days. Irinotecan (40 mg/kg; Actavis) was injected intraperitoneally in 10 mL/kg of 0.9% NaCl once weekly, starting on day 3. Caliper measurements of tumor volume were performed every third day. Animals sacrificed before day 28 were not included in the results analysis. Nonrepresentative animals were excluded on the basis of tumor observations (for details see Results) or outlier test (tumors, which volume was below Q1-1.5IQR or above Q3+1.5IQR; one animal removed from combination group). On days 28 to 30, the GFP fluorescence of the xenograft tumors was measured with a CCD camera (IVIS, Spectrum, Caliper Life Sciences) and the images were analyzed for radiant efficiency in Living Image 4.2. After this, animals were terminated, tumors dissected out and weighted. The experiments were performed with approval of local ethical committee Stockholm North (N447/12).

A one-sided unpaired t test for in vivo data analysis and Pearson correlation were performed using Prism software.

The aim of this study was to evaluate whether clinically used substances show unforeseen anticancer activity in an in vitro 3D solid tumor model. The 3D model was developed to mimic the microenvironment in the deep parenchyma of solid tumors with respect to hypoxia, nutrient deprivation, and low pH. MCTS were formed in 384-well plates using GFP-labeled HCT116 colon cancer cells. They were cultivated without medium change for 7 days prior to drug exposure. This is a significant variation from conventional spheroid formation protocols, in which fresh medium is added during the incubation period. During the 7-day-long culture period, glucose concentration in the culture medium dropped from 15.5 to 4.66 mmol/L and further down to 2.90 mmol/L after 3 additional days (drug exposure). Corresponding medium pH decreased from 7.47 to 7.04 at day 7 and to 6.78 after 10 days (Supplementary Fig. S1). These values closely correspond to those reported for hypoxic regions in solid tumors (25).

The procedure resulted in the formation of approximately 500 μm Ø MCTS with hypoxic cores (pimonidazole staining), low proliferation rates (Ki-67 staining), and central regions with increased apoptosis (active caspase-3 staining), as shown in Fig. 1A. GFP fluorescence is, as previously reported (19), a simple, reliable, and noninvasive surrogate marker of spheroid viability, well suited for read-out in drug screening (Fig. 1B–D) as well as dose-response and time-course experiments (Supplementary Fig. S2). As a secondary read-out for MCTS viability, we used a resazurin-based assay, TOX8, previously shown suitable for such measurements (20). Results obtained with these two assays showed a high degree of concordance (Supplementary Fig. S3). The TOX8 assay made it possible to include hits that would have been missed if only GFP-signal was used for hit selection, such as pyrvinium pamoate (due to their auto-fluorescence in GFP emission spectrum; Supplementary Fig. S3). Both assays performed well with Z′-factors > 0.5, as calculated according to Zhang and colleagues (24).

We screened the collection of 1,600 clinically active compounds on MCTS, followed by hit-validation (See Fig. 1F for a schematic overview and Materials and Methods for details). In parallel, we screened this library using HCT116 cells [wild-type, previously shown to respond to treatment identically as GFP-labeled HCT116 cells (19)] grown as 2D monolayer cultures. Comparison of the results from the 3D- and 2D-based screens showed that most clinically used cytotoxic drugs have only modest efficacy in spheroids (Fig. 1E), in agreement with previous studies (14, 19). Clinically used cytotoxic drugs target proliferating cells, which can explain their lack of efficacy in MCTS, in which cell proliferation is low (Fig. 1A).

A comparison of the results of the 2D and 3D drug screens (Fig. 1E) indicated the existence of compounds with selective activity toward spheroids. Forty-one compounds were selected on the basis of activity in the 3D-model. To validate the results, we performed extensive dose–response experiments using both 3D and 2D cultures. These experiments resulted in the identification of 12 compounds with preferential activity toward the 3D model (Supplementary Table S1). The 3D-selectivity of one of these, nitazoxanide, is presented in Fig. 2A and C. The 3D-selective activity of nitazoxanide could be observed as spheroids lost viability (decreased GFP signal), at a concentration at which 2D-cultured cells were actively proliferating. This pattern was in sharp contrast with that of mitomycin C (Fig. 2B) and oxaliplatin (Fig. 2C).

The 12 3D-selective compounds were subsequently challenged in a spheroid-based clonogenic assay in order to select compounds with specificity toward the dormant hypoxic and nutrient-deprived cancer cells in the MCTS. Thus, to focus on 3D-selectivity rather than absolute potency, spheroids were exposed to the compounds at their respective 2D IC₅₀ concentrations. Following washout of the drugs and spheroid dispersal, single-cell suspensions were seeded in fresh medium into 6-well culturing plates and left for 10 days to regrow. Five of the compounds, closantel, nitazoxanide, niclosamide, pyrvinium pamoate, and salinomycin abolished colony formation (Fig. 2D). Moreover, the preferential activity toward the 3D model was not limited to the HCT116 cell line. Similar results were observed when the compounds were tested in the colon carcinoma cell line HT-29 (Supplementary Fig. S4). Interestingly, the five final hit compounds were not as active in HCT116 and HT-29 spheroids that were formed with addition of fresh medium during the culture period (Supplementary Fig. S5). Therefore, if a standard spheroid model rather than our medium-stressed version had been used, these compounds would most likely not have been identified as 3D-selective.

Because the five final 3D-selective molecules have known mechanisms of action, we had a possibility to identify cellular targets that confer this context-dependent vulnerability. A literature review (Supplementary Table S2) revealed that all final hit compounds target mitochondria and oxidative phosphorylation (OXPHOS), albeit by different mechanisms. Niclosamide, closantel, and nitazoxanide [which is rapidly converted to its active metabolite tizoxanide (26)] are all anthelmintic compounds that cause uncoupling of mitochondrial membrane potential (26–31). In fact, these compounds share an identical pharmacophore (highlighted in red; Supplementary Table S2). Pyrvinium pamoate, also an anthelmintic drug, targets the fumarate-reductase system and has been previously recognized for its high activity against spheroids and glucose-deprived cells (32, 33). The ionophore salinomycin, which has recently gained attention for its activity against cancer stem cells (CSC), has also been identified as an OXPHOS inhibitor (34).

Because inhibition of mitochondrial respiration was seemingly a common denominator among the final hit compounds, we characterized their effects on oxygen consumption and mitochondrial function. The compounds were assessed at concentrations up to their 2D IC₅₀s for their effects on OCR in monolayer HCT116 cells, using a known uncoupler of mitochondrial membrane potential (FCCP) as a positive control. Hit compounds reported to be uncouplers, that is, nitazoxanide, niclosamide, and closantel as well as FCCP increased OCR at low concentrations (Fig. 3A–D). However, as the concentration increased, the period of elevated OCR became shorter and was followed by shutdown of mitochondrial respiration, indicated by a rapid decrease of OCR. Importantly, these effects were observed for the uncouplers below their 2D IC₅₀ concentrations. When even higher concentrations of the uncoupling hit compounds or FCCP were used, increase in OCR could not be detected. At drug concentrations sufficient to cause a complete shutdown of mitochondrial respiration, late addition of FCCP caused no increase in OCR (Fig. 3B–D, highlighted with circle).

The effect of the remaining two hit compounds, pyrvinium pamoate and salinomycin, on OCR was different from what was observed for the uncouplers. Pyrvinium pamoate at the 2D IC₅₀ concentration (1.5 μmol/L) resulted in an immediate decrease of OCR (Fig. 3E). This effect was weaker at lower concentrations and absent at 0.1 μmol/L. An increase in OCR, as caused by uncouplers, was not observed. Salinomycin at 1 μmol/L caused a slight increase in OCR (Fig. 3F). However, inhibition of OCR by salinomycin used at its 2D IC₅₀ concentration (5 μmol/L) was weaker than the inhibition caused by corresponding concentrations of uncouplers. Moreover, late addition of FCCP induced an increase in OCR (Fig. 3F), opposite to the effects observed after exposure to high concentrations of the uncouplers.

We confirmed these results with JC-1 staining, a probe detecting polarization state of mitochondrial membrane. At 2D IC₅₀ concentrations 2-hour exposure to uncouplers or pyrvinium pamoate resulted in the complete depolarization of mitochondrial membrane (Fig. 3G). In contrast, exposure to salinomycin increased the amount of red fluorescent aggregates within mitochondria, indicating their extreme hyperpolarization (Fig. 3H). This effect of salinomycin has been previously reported (35).

Monolayer-based characterization of mitochondrial dysfunction was in agreement with time-dependent changes in pimonidazole staining of MCTS. After 4 or 24 hours of exposure to the hit compounds or the uncoupler CCCP (positive control), spheroids were exposed to pimonidazole and stained for pimonidazole adducts. Four-hour exposure to the compounds that stimulated OCR in 2D culture (nitazoxanide, niclosamide, closantel, and salinomycin) increased the hypoxic area within spheroids (Fig. 3H), which is consistent with elevated consumption of oxygen in 2D cultures (Fig. 3A–D, and F). A similar effect was observed in spheroids treated with CCCP. In contrast, exposure to pyrvinium pamoate decreased the hypoxic area in spheroids (Fig. 3H), which is in accordance with an immediate decrease of OCR in 2D cultures (Fig. 3E). After 24-hour exposure to uncouplers, hypoxic areas in spheroids disappeared or decreased substantially, indicating shutdown of mitochondrial respiration (Fig. 3H).

After 24-hour exposure to the uncouplers, we were still able to retrieve intact spheroids, whereas exposure to salinomycin or pyrvinium pamoate for this period resulted in the loss of spheroid integrity, suggesting earlier cell death (Fig. 3H). Therefore, we examined the clonogenicity of dissociated spheroids in relation to time of exposure to the five final hit compounds. Continuous drug exposure for 48 hours was required to achieve effective loss of clonogenicity using mitochondrial uncouplers, whereas pyrvinium pamoate and salinomycin induced an earlier loss of clonogenicity (Fig. 4A and B). This demonstrates that cells in compromised microenvironments can survive with impaired mitochondrial respiration for a limited period and that mitochondria are able to recover from exposure to the tested OXPHOS inhibitors. The latter was evident from assessment of OCR in drug-free medium following exposure of HCT116 to nitazoxanide for 24 hours. OCR gradually increased after removal of the drug and returned to normal levels after 7 days (Fig. 4C). Thus, continuous drug exposure is required to eliminate tumor regrowth potential. This is an important finding considering an optimal treatment schedule for compounds targeting mitochondria.

In addition to oxygen availability, glucose is an established limiting factor for dormant cells in MCTS (21, 22). Cells experiencing low nutrient concentrations would be particularly dependent on mitochondrial respiration in order to meet their energy demands. We therefore tested the influence of glucose starvation on the response to the final hit compounds. Monolayer HCT116 cells cultured in medium without glucose were more sensitive to treatment with all five hit compounds than when cultured under standard glucose conditions (Fig. 5A–E). Conversely, higher survival rates were observed for cells treated with doxorubicin in no-glucose conditions (Fig. 5F), presumably because of their lower proliferation rate during glucose starvation, rendering the cells less sensitive to the DNA-damaging agent. Thus, maintaining functional mitochondrial respiration seems critical for survival in glucose-deprived conditions.

According to our findings, efficient eradication of cancer cells in glucose-deprived conditions through inhibition of mitochondrial respiration relies on continuous drug exposure for a sustained period of time. Of the final five hit compounds identified in the screen, nitazoxanide stands out as the only one that, without major side effects, reaches high systemic concentrations after oral administration (Supplementary Table S2; refs. 36, 37). Therefore, nitazoxanide was selected as the drug with the highest repositioning potential. Importantly, we verified that its active metabolite tizoxanide (which is the only species detectable in plasma after oral administration of nitazoxanide) was as active as the parental drug in vitro (Supplementary Fig. S6).

Our results suggested that impairment of mitochondrial function under conditions of nutrient starvation leads to an energy catastrophe and cell death. Energy stress is a trigger for activation of AMPK. Therefore, we examined whether treatment with nitazoxanide would result in the phosphorylation of AMPK in spheroids. Indeed, after 24-hour exposure of HCT116 spheroids to nitazoxanide, increased levels of phosphorylated AMPK could be observed (Fig. 5G), suggesting an increase in AMP:ATP ratio. This increase was observed at 0.5 μmol/L, and was slightly weaker at higher concentrations.

AMPK activation is associated with inhibition of mammalian target of rapamycin (mTOR), which is one of the downstream targets of AMPK (38). Twenty-four-hour exposure to nitazoxanide resulted in dose-dependent downregulation of phosphorylated 4EBP1 and p70 in HCT116 spheroids (Fig. 5H), indicating inhibition of mTOR pathway. Because treatment with nitazoxanide causes energetic stress in cancer cells, we predicted it to also indirectly downregulate other oncogenic pathways. Indeed, treatment with nitazoxanide caused a decrease in c-Myc levels and inhibition of Wnt signaling (Supplementary Fig. S6G). Notably, nitazoxanide has been previously reported to strongly reduce levels of c-Myc in cancer cell lines (39). Its inhibitory effects on Wnt pathway have not been reported so far.

Because solid tumors in vivo harbor both quiescent and proliferating cells, we reasoned that successful treatment would likely rely on a combination of compounds targeting these distinct cell populations. To test this hypothesis, we treated spheroids formed with medium change (i.e., not medium-stressed), which comprise both quiescent and proliferating cell populations (Supplementary Fig. S7A), with nitazoxanide. Seventy-two hours of continuous treatment did not completely inhibit colony formation (Supplementary Fig. S7B and S7C). Exposure to the standard cytotoxic drug irinotecan, which targets proliferating cells, did not result in a strong inhibition of colony regrowth as well (Supplementary Fig. S7B and S7C). However, when both drugs were used in combination, colony formation was strongly abrogated. These results give a rationale for using nitazoxanide in combination with irinotecan rather than as a single agent for in vivo experiments.

Consequently, nitazoxanide and irinotecan were tested in a mouse xenograft model. HCT116 GFP cells were implanted into NMRI nu/nu mice. After tumor establishment (volume >100 mm³), mice were exposed to either vehicle control, nitazoxanide (gavage, 200 mg/kg twice daily), irinotecan (intraperitoneal injection, 40 mg/kg once weekly), or a combination of the two compounds. Tumor volumes were monitored for 28 days followed by tumor dissection for assessment of tumor weight. Nitazoxanide alone did not cause inhibition of tumor growth (Fig. 6A–D), reminiscent of the poor effect of the drug in proliferating cells in 2D culture (Fig. 2A). Consequently, the lack of effect of nitazoxanide as a single agent may be due to the rapid cell proliferation in xenograft models, with volume doubling times of just a few days. This could indicate that a substantial fraction of cancer cells in these models has a good access to glucose. This situation may not apply to slow-growing tumors in humans, in which doubling times can be often measured in months rather than days.

The topoisomerase 1 inhibitor irinotecan, expected to target proliferating cells, produced strong tumor growth inhibition (Fig. 6A–D). Nitazoxanide potentiated the effect of irinotecan causing a significant reduction in tumor growth compared with irinotecan alone (P = 0.030; Fig. 6A). Similar results were observed after analysis of tumor weight after dissection (P = 0.068; Fig. 6B) and GFP fluorescence of tumors in vivo (P = 0.091; Fig. 6C and D). Interestingly, in the combination group, we observed late-onset, sudden apparent volume increase of tumors in two individuals, which both showed very low GFP fluorescence intensity, indicating low viable cell number (Fig. 6D, framed). When dissected, one of the tumors was filled with viscous fluid. The other, after rapid volume increase, was ruptured and decreased from 0.73 to 0.49 mL in 2 days. Thus, the animal had to be sacrificed at that point due to ethical reasons and the full treatment schedule could not be completed. This phenomenon was exclusively observed in the combination group. Because of the tumors being not representative, these animals were excluded from data analysis. However, these observations could indicate even more pronounced anticancer activity of the drug combination in the two individuals. Importantly, nitazoxanide caused no toxicity, such as diarrhea or skin rash, neither did it cause any change in body weight (Fig. 6E). Irinotecan, on the other hand, produced a significant decrease in body weight indicating systemic adverse effects.

In this work, we present a novel approach for preclinical anticancer drug identification, which combines two concepts that have been recently gaining attention in cancer research. First, high-throughput drug screening was performed using a library consisting of drugs that have been tested clinically and in many cases FDA-approved. This presents an opportunity to find molecules suitable for drug repositioning and rapid advancement into clinical trials. Second, to identify drugs targeting quiescent cancer cells and mimic in vivo tumor microenvironment, we used a new MCTS model well suited for HTS. However, in contrast to most other spheroid-based approaches, the present method did not involve any medium change throughout the whole spheroid culture period. This enabled us to better mimic conditions observed in dormant tumor regions in vivo, in terms of glucose concentration and pH. Our screening system demonstrated excellent reproducibility and capacity to test thousands of compounds. For identification of drugs selectively active against medium-stressed spheroids, we challenged screening hits in parallel in spheroid- and monolayer-based dose–response experiments. Interestingly, standard cytotoxic agents were preferentially active against 2D cultures with only minor effects in the 3D counterpart while for several other compounds, we observed opposite effects. These results support the notion that cytotoxic compounds have limited activity in 3D models, but challenge the common opinion that spheroid models are more resistant to chemicals in general.

Low oxygen and glucose levels in dormant tumor regions in vivo are believed to result from both poor blood supply and high glucose consumption by cancer cells. It has been established that rapidly proliferating cancer cells shift from oxidative phosphorylation to aerobic glycolysis, using it as a prime pathway to fulfill their energy requirements, a phenomenon known as the Warburg effect (40). However, oxidative phosphorylation has recently been found to be required for cancer cell proliferation under low glucose conditions (41). Moreover, OXPHOS was reported to be hyperactive in epithelial cancer cells in situ (42). This growing body of evidence indicates that cells in microenvironments where glucose is not abundant would depend on OXPHOS rather than solely glycolysis to meet their energetic requirements. This hypothesis is in a good agreement with the presented results. We found that compounds most active in 3D cultures targeted mitochondrial respiration by different mechanisms. This confirmed our and others' recent observations that oxidative phosphorylation is an attractive target for anticancer therapy (22, 43, 44).

Finally, we identified nitazoxanide as an OXPHOS inhibitor suitable for direct drug repositioning. It has been approved for human use, with a favorable pharmacokinetic profile and very limited side effects (37).

Our screen was designed to find drugs with selective activity toward quiescent tumor regions. Therefore, nitazoxanide was not expected to show anticancer activity as a single drug in high-proliferative tumor models. This prediction was experimentally verified, because nitazoxanide alone did not impair xenograft tumor growth. However, nitazoxanide potentiated the effect of irinotecan in vivo. This is an important finding considering repositioning nitazoxanide as an anticancer agent. Our results pave the way for further testing in syngeneic (45) or hetero-/orthotopic patient-derived xenograft models (46) prior to clinical advancement. Consequently, putative clinical trials should focus on using nitazoxanide in combination with standard chemotherapy. One example for similar approach was the clinical trial, which resulted in approval of bevacizumab (Avastin), the anti-VEGF antibody, which in combination with chemotherapy prolonged median survival of colorectal cancer patients (47). Notably, when used as a single agent, progression-free survival of patients treated with bevacizumab alone was shorter than in patients treated with chemotherapy alone.

Our results showed that quiescent cancer cells are able to survive nitazoxanide-induced inhibition of mitochondrial phosphorylation and recover from the treatment lasting up to 24 hours. Therefore, nitazoxanide treatment regimens would have to be designed in a way that ensures continuous high plasma concentration of the drug. Clinical studies indicate that nitazoxanide is generally well tolerated in single doses up to 4 g (48) and its maximal plasma concentration is reached 2 to 6 hours after administration (36). Average maximal plasma concentrations after single dose of only 500 mg exceed 6 μmol/L (36), that is, twice the nitazoxanide IC₅₀ concentration in HCT116 spheroids (3 μmol/L). It may seem surprising that a mitochondrial function inhibitor with good systemic distribution does not induce severe adverse effects, in particular cardiac or neurological. However, another uncoupler of OXPHOS, dinitrophenol (DNP), has been used with good tolerability for treatment of nutritional disorders in thousands of patients before its withdrawal from the market (49). Acute toxicity of DNP has been reported only when used at high doses. Conversely, irreversible inhibitors of OXPHOS such as cyanide or antimycin A cause death within a few minutes. Therefore, reversibility of OXPHOS inhibition seems important for drug general toxicity.

Treatment of HCT116 spheroids with nitazoxanide resulted in activation of AMPK pathway and inhibition of mTOR pathway, indicating energetic deficit. Importantly, similar molecular effects of mitochondrial respiration inhibition were recently observed in drug-resistant ovarian CSC populations (44). One of the final screening hits, salinomycin, has also been found to selectively target breast cancer stem cells (50). This body of evidence, together with the fact that HCT116 cell line consists mainly of CSCs [(51); interestingly, culturing HCT116 cells as spheroids increased the expression of stem cell marker CD44, as previously reported (19) and shown in Supplementary Fig. S8] could indicate a novel strategy for targeting cancer stem cells, which is in agreement with recent reports (52–54). Taken together, our findings are in accordance with the hypothesis on dormant cells being CSCs, an issue that is currently intensively investigated.

No potential conflicts of interest were disclosed.

Conception and design: W. Senkowski, S. Linder, R. Larsson, M. Fryknäs

Development of methodology: W. Senkowski, X. Zhang, M. Fryknäs

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): W. Senkowski, X. Zhang, M.H. Olofsson, R. Isacson, U. Höglund, P. Nygren, R. Larsson, M. Fryknäs

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): W. Senkowski, X. Zhang, R. Isacson, M. Gustafsson, P. Nygren, S. Linder, R. Larsson, M. Fryknäs

Writing, review, and/or revision of the manuscript: W. Senkowski, X. Zhang, R. Isacson, P. Nygren, S. Linder, R. Larsson, M. Fryknäs

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): X. Zhang, R. Isacson, M. Gustafsson

Study supervision: S. Linder, M. Fryknäs

The authors thank Paola Pellegrini and Angelo De Milito for help with pH measurements and to Jan Siljason for expert immunohistochemistry stainings. Skillful technical assistance of Lena Lenhammar, Christina Leek, and Nasrin Najafi is gratefully acknowledged.

This study was supported by the Swedish Cancer Society, Swedish Foundation for Strategic Research, Swedish Research Council and the Lions Cancer Research Fund. P. Nygren was supported by Strategiska Forskningsstiftelsen (SSF) and Lions Cancerforskningsfond; S. Linder by Cancerfonden, Barncancerfonden, Radiumhemmets Forskningsfonder, Vetenskapsrådet, SSF; R. Larsson by Cancerfonden, SSF, Lions Cancerforskningsfond; and M. Fryknäs by Lions Cancerforskningsfond.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.