Candidate Antimetastasis Drugs Suppress the Metastatic Capacity of Breast Cancer Cells by Reducing Membrane Fluidity

To metastasize, a cancer cell must acquire a set of capabilities that are distinct from and potentially incompatible with its ability to flourish in the primary tumor (1). One explanation for how cells can acquire metastatic capacity is that they activate a latent epithelial–mesenchymal transition (EMT) program, which provides cancer cells with increased motility and stem cell properties (2, 3). Regardless of mechanism, the diversity of signals that can activate the metastatic cascade (4–6) poses a challenge in developing therapeutic strategies. A model that explains the acquisition of metastatic traits using a more limited set of targetable cellular events would help in the development of broadly applicable treatments.

Despite the wealth of signals known to promote metastasis, the search for small molecules to target this fatal process has been hampered by a lack of knowledge of a targetable addiction of metastatic cells. Efforts to identify genetic alterations that predispose tumors to metastasis have thus far been unsuccessful (7). To address this deficiency, unbiased in vitro screens for drugs have focused on the stem cell phenotype because metastasis requires the ability to recapitulate a primary tumor in a distant site with the same hierarchy of cell types. Recent screens against stem cells have identified two leading candidates, salinomycin, a potassium channel ionophore; and thioridazine, a dopamine receptor antagonist (8, 9). In both cases, their primary targets have little known bearing on the known biology of metastasis. However, salinomycin has later been shown to inhibit Wnt signaling through an unknown mechanism dependent upon a transmembrane receptor (10). For thioridazine, an association between dopamine receptor activation and stem cells was originally suggested, but an independent screen recovered both dopamine receptor agonists and antagonists, suggesting that the compound likely works through an alternative mechanism (11). Given these results, the only apparent similarity in targets is the dependence upon membrane associated receptors. The lack of a deeper explanation for how these compounds work reveals a limitation in our understanding of the important determinants of metastatic capacity.

To better understand the targets and dependencies of metastasis, our study demonstrates that the fluidity of cell membranes, whether changed by pharmacologic or genetic inducers, controls metastatic capacity. Decreasing fluidity prevents and reverses an EMT and inhibits metastasis in mouse models. In human breast cancers, overexpression of ABCA1, a regulator of membrane cholesterol and fluidity, is associated with increased metastasis. Taken together, these disparate lines of evidence indicate that membrane fluidity is a critical determinant of metastatic capacity and suggest that it may be a viable therapeutic target.

T47D, MCF7, AU565, MDA-MB-231, MDA-MB-361, and ZR-75-1 cell lines were purchased from ATCC in 2011. 4T1 cells were purchased from ATCC in 2013. SUM-159 were obtained from Dr. Jeffrey A. Frost (University of Texas Health Science Center at Houston, Houston, TX); human mammary epithelial (HMLE) and HMLE ER-Twist cells were kindly provided by Dr. Sendurai A. Mani (University of Texas MD Anderson Cancer Center, Houston, TX) in 2013. All cell lines used in this study were used at low passage and shown to be free of mycoplasma (MycoAlert Mycoplasma Detection Kit, Lonza #LT37-618). Except for SUM-159, all cell lines were obtained either from ATCC or the original source (HMLE). We did no further authentication on the SUM-159 cell line.

Additional procedures are described in Supplementary Materials and Methods.

We searched for compounds that can repress a metastatic phenotype (i.e., the set of cellular characteristics associated with metastasis), even if they do not lead to death as in prior screens (8, 9), using an EMT gene expression signature as a model (Fig. 1A). We created a signature based on embryonic development using an approach we previously described (12), although other signatures have been developed (13, 14). Noticing that EMT inducers TWIST1 and ZEB1 (6, 15) are upregulated at day 9.5 (Supplementary Fig. S1A), we selected for our signature the 219 genes that are either induced or repressed at this time (Fig. 1B; Supplementary Table S1). We found that the scores from this signature are significantly higher in cells after undergoing an EMT (across 6-gene expression data sets; Fig. 1C). To further characterize the signature, we performed pathway analysis using GATHER (16) and found 67 significant processes (FDR < 0.05), including many associated with differentiation, development, and motility (Supplementary Fig. S1B). We also generated a network of the altered pathways (17–19) and saw that the signature is associated with upregulation of Wnt signaling, which is an inducer of Snail and EMT (Supplementary Fig. S1C; ref. 20). These results indicate that our EMT signature captures the distinction between the epithelial and mesenchymal phenotypes in cancer.

With the validated EMT signature, we identified drugs that can perturb it using the Connectivity Map, a database of gene expression changes triggered by pharmacologic compounds (Fig. 1D; ref. 21). Analysis of 1,309 compounds, including FDA-approved drugs and other bioactive molecules, yielded predictions of drugs that repress an EMT gene expression signature (antimetastasis drugs) (Fig. 1E; Supplementary Table S2). The top scoring antimetastasis drugs are inhibitors of known EMT pathways, such as vimentin expression (withaferin A; ref. 22), HDAC dependency (vorinostat, trichostatin A; ref. 23), HSP90 response (tanespimycin, geldanamycin; ref. 24), or PI3K activation (LY294002; ref. 25). Furthermore, one of the compounds identified in a prior screen, thioridazine, is also predicted to be antimetastasis. The other, salinomycin, was not profiled in this database. To validate the robustness of the results, we found a statistically significant overlap (P = 0.0004) with the connectivity scores obtained using a previously published EMT signature (13). All but thioridazine were confirmed by the core EMT signature to be antimetastatic (P = 0.007; Supplementary Fig. S1D).

To test whether the drugs predicted in silico can alter metastatic phenotypes in vitro, we focused on four with a range of mechanisms: alprostadil (prostaglandin agonist), amitriptyline (tricyclic antidepressant), haloperidol (dopamine receptor antagonist), and maprotiline (tetracyclic antidepressant); as well as two previously identified drugs, salinomycin and thioridazine.

To validate whether the predicted drugs can affect metastatic capacity, we tested whether they impact cell motility or mammosphere formation, cellular characteristics that are necessary for metastasis. To measure motility, we performed transwell migration and wound healing assays and found that the drugs significantly decreased the motility of cells (Fig. 2A–C and Supplementary Fig. S2A–S2C). In addition, the drugs could also repress the ability of the cells to form mammospheres, an in vitro assay for early progenitor stem cell characteristics in breast cancer cells (Fig. 2D and Supplementary Fig. S2D; ref. 26). The drug concentrations used varied from 1 to 20 μmol/L (see Supplementary Materials and Methods).

To evaluate whether the assays were confounded by an impact on cell proliferation or death, we measured the impact of the drugs on cell number. At seven days, the maximum duration of the prior assays, drug treatment minimally impacted the cells, leading to a 17% to 18% decrease that cannot explain the magnitude of the changes in the metastatic phenotypes (Supplementary Fig. S2E). One exception is alprostadil, which showed a maximum of 50% decrease (one cell division) at seven days. However, even this drug had no impact after one day, indicating that the changes in the short term motility assays also cannot be explained by a change in cell number.

The prior experiments demonstrated that antimetastasis drugs could decrease metastatic phenotypes. However, in tumors, a metastatic phenotype may be maintained by signaling sustained via genetic or epigenetic changes or from extracellular sources. To see whether the target of the drugs is dominant over the presence of a continuous metastasis-inducing signal, we next tested whether antimetastasis drugs can reverse an EMT in cells that have undergone the transition or prevent cells from entering one. For the first experiment, we used HMLE cells (nontumorigenic immortalized human mammary epithelial cells; ref. 27) that stably express Twist, which forced the cells to undergo an EMT as seen by loss of the E-cadherin epithelial marker and gain of mesenchymal ones (Fig. 2E and F). In these cells, five days of antimetastasis drug treatment resulted in reassertion of E-cadherin and loss of mesenchymal markers, even under continued Twist expression, showing that the cells reverted to an epithelial state despite continued expression of a potent EMT inducer.

To test whether the drugs can prevent an EMT, we used an inducible EMT system comprised of HMLE cells carrying tamoxifen-inducible Twist (2). We treated cells with antimetastasis drugs for seven days and then activated Twist with tamoxifen. While untreated cells underwent an EMT, cells treated with antimetastasis drugs did not undergo the same change (Fig. 2G). This shows that antimetastasis drugs can prevent the EMT despite the presence of an EMT-inducing signal. We confirmed that a stem cell phenotype was also reverted in a mammosphere assay (Fig. 2H and I). Similar results could also be seen in T47D cells using the EMT inducer Snail (Supplementary Fig. S2F and S2G).

Taken together, these experiments demonstrate that drugs with a broad range of established targets can inhibit both metastatic phenotypes and EMT across a range of breast cancer cell lines (T47D, nonmetastatic, luminal; MCF7, nonmetastatic, luminal; MDA-MB-231, metastatic, claudin low; and SUM-159, metastatic, claudin low; ref. 28). Combined with the observation that they can inhibit an EMT, it suggests that they affect a target that is necessary for an EMT but also has a general capacity to tune metastatic phenotypes across cell states.

Due to the large number of pathways and the breadth of drugs that can reduce metastatic phenotypes, we surmised that they may be controlled by a broad cellular characteristic. To identify it, we examined the drugs found in our in silico screen and noticed a strong correlation between the induction of an EMT signature and a reported ability to alter plasma membrane fluidity (defined as the rate that a molecule can diffuse laterally across the membrane; Fig. 3A). This mechanism is supported by prior studies showing that cell lines from lung (29), sarcoma (30), and breast tumors (31) that metastasize have higher fluidity. On the basis of these results, we asked whether the observed fluidity change is a correlated byproduct of, or a requirement for, EMT and metastasis.

To test whether our predicted antimetastasis drugs can alter plasma membrane fluidity, we used a FRAP assay (Fig. 3B). After treating cells, we found that haloperidol decreases the fluidity of the plasma membrane (Fig. 3C), showing a lower number of mobile molecules as well as a decreased rate of diffusion (Fig. 3D).

To test a broader range of molecules, we switched to a more tractable pyrene fluidity assay. As a control, we included oleic acid, an 18-chain monounsaturated fatty acid that fluidizes membranes by disrupting lipid packing (32). Using this approach, we confirmed that EMT inducers and oleic acid increase fluidity, whereas antimetastasis drugs decrease it across a range of cell lines (Fig. 3E; Supplementary Fig. S3A).

The prior experiments were performed in a monolayer culture that lacks the 3D organization of the in vivo state, which has been seen to alter the response of cells to drugs such as chemotherapies (33). Therefore, to determine whether 3D organization impacts the ability of drugs to alter fluidity, we cultured MDA-MB-231 and T47D cells in 3D, where they assumed stellate (characterized by invasive projections) and mass (with disorganized nuclei and robust cell–cell adhesions) morphologies, respectively (Supplementary Fig. S3B; ref. 34). Although cells in 3D culture exhibit higher fluidity than in 2D, the drugs could nevertheless reduce fluidity similar to that seen in 2D (Fig. 3F and G and Supplementary Fig. S3C and S3D).

Next, to distinguish whether fluidity is correlated with, or causally represses, metastatic phenotypes, we isolated the molecular targets of the drugs from fluidity by treating cells concurrently with both an antimetastasis drug (that decreases fluidity) and oleic acid (that increases fluidity), resulting in no net change (Fig. 3H and Supplementary Fig. S3E). When fluidity was unchanged, antimetastasis drugs could no longer decrease either cell migration or mammosphere formation in MDA-MB-231 or T47D cells (Fig. 3I and J and Supplementary Fig. S3F and S3G). This demonstrates that antimetastasis drugs decrease metastatic phenotypes through a fluidity-dependent mechanism.

Metastasis is comprised of complex physiologic events that are not present in vitro, including influences from the microenvironment and immune system. To model these factors, we tested our antimetastasis drugs in a syngeneic immunocompetent mouse model of metastasis. We injected 4T1 cells orthotopically into the mammary fat pads, a model of the stroma of normal mammary glands, of BALB/c mice. Once a day, we injected mice intraperitoneally with saline, alprostadil, or haloperidol at doses analogous to those given to humans clinically. To isolate fluidity from other targets of the drugs, we also treated with a combination of the drugs and oleic acid. During the course of the experiment, all mice developed primary tumors in the mammary gland and were sacrificed after 31 days. The haloperidol-treated mice exhibited lethargy, a well-known side effect of the drug, immediately after injections, which diminished over the next hours, while the alprostadil-treated mice showed no obvious changes in behavior or weight. The sizes of the primary tumors were similar across all groups, indicating that the drugs had little or no impact on tumor growth (Fig. 4A). However, the fluidity of the tumors treated with drugs decreased, showing that the drugs affected the tumors, except when balanced by oleic acid injections (Fig. 4B). Furthermore, drug treatment led to significantly fewer metastatic nodules on the surface of the lung, which was reversed when fluidity was restored with oleic acid (Fig. 4C and D). The tumors treated with oleic acid showed an increased number of nodules that did not achieve statistical significance. However, half these lungs were permeated with metastatic growth such that nodules could not be counted accurately, exceeding the limits of the assay. To more sensitively measure tumor burden, we sectioned the lungs, stained with hematoxylin and eosin, and calculated the metastatic burden as the fraction of cells that were cancerous. This revealed that oleic acid does significantly increase metastatic burden (Fig. 4E and F). Finally, to check if the antimetastasis drugs can affect the mesenchymal status of the tumors, we stained for the expression of vimentin, revealing that the antimetastasis drugs resulted in tumors with decreased vimentin expression, while oleic acid had the opposite effect (Fig. 4G and H). Taken together, these results demonstrate that antimetastasis drugs significantly reduced breast cancer metastasis, dependent upon a decrease in the membrane fluidity of the primary tumor.

As a more fluid membrane is required for metastasis, we reasoned that metastatic tumor cells must somehow increase their fluidity. One possible mechanism is an alteration in cellular cholesterol, a determinant of membrane fluidity (35), which we confirmed by titrating cholesterol in MDA-MB-231 cells (Fig. 5A and B). We treated MDA-MB-231 cells with our antimetastasis drugs and found that they increased cholesterol levels (Fig. 5C), consistent with the decrease in fluidity. Furthermore, cholesterol also leads to a loss of metastatic phenotypes, as measured by migration and mammosphere formation (Fig. 5D–G). Indeed, the total cellular cholesterol levels are inversely correlated with membrane fluidity and metastatic phenotypes.

As cholesterol is a potent inhibitor of metastatic phenotypes, we predicted that the EMT program may increase metastatic capacity by reducing cholesterol. To identify this mechanism, we searched for changes in the expression of genes that control cellular cholesterol levels across an EMT (36). This bioinformatic analysis revealed that an ATP-binding cassette transporter ABCA1, which functions as a cholesterol reverse transporter (37), is significantly upregulated in an EMT (Fig. 6A). To confirm this, we expressed Snail or Twist in HMLE cells and a range of luminal breast cancer cell lines and measured ABCA1 gene expression. In these conditions, the expression of ABCA1 was increased 4- to 12-fold (Fig. 6B).

As a potential mechanism to explain their activity, we tested whether the antimetastasis drugs reduce ABCA1 expression. We treated T47D cells induced to undergo an EMT and MDA-MB-231 cells with a range of antimetastasis drugs and found that ABCA1 gene and protein expression was significantly decreased (Fig. 6C and D). Further, in sections from lung metastases, the tumors from mice treated with antimetastasis drugs had significantly fewer cells with high ABCA1 expression (Fig. 6E and F). Thus, ABCA1 is an EMT-responsive gene that can be inhibited by the treatment of antimetastasis drugs.

Although ABCA1 was previously found to decrease cell proliferation, its relationship with metastasis is still unknown (38). To determine whether ABCA1-regulated cholesterol levels control metastatic capacity, we used an RNAi strategy and knocked down ABCA1 in MDA-MB-231 cells (Supplementary Fig. S4A). This led to an increase in cellular free cholesterol, consistent with a diminished ability of the cells to efflux cholesterol (Fig. 7A). The cholesterol levels were also inversely correlated with changes in membrane fluidity (Fig. 7B). We measured cell proliferation for one week and observed no change (Supplementary Fig. S4B). As our model would predict, we also observed in these cells a decreased cell motility (Fig. 7C and D and Supplementary Fig. S4C). In each of these experiments, we could rescue the phenotype by restoring cellular cholesterol with MβCD, a compound that extracts cholesterol from cell membranes (39). This demonstrates that the changes in phenotype were dependent upon its impact on cholesterol and not another function of ABCA1. Taken together, these results show that ABCA1 determines cell migration through the regulation of cholesterol levels.

Seeing that knockdown of ABCA1 can alter metastatic phenotypes, we next tested whether cyclosporin A, an ABCA1 inhibitor, has similar effects (40). Treatment of MDA-MB-231 cells with this compound led to a significant increase in cellular cholesterol concomitant with a reduction in fluidity (Supplementary Fig. S4D and S4E). Consistent with this result, we also observed a decrease in migration and mammosphere formation (Supplementary Fig. S4F–S4H). Finally, we tested whether cyclosporin A can inhibit an EMT in a T47D model. We induced an EMT by ectopic expression of Snail, and found that it can restore epithelial transcriptional markers, along with a reduction in mammosphere-forming ability (Supplementary Fig. S4I and S4J). These results confirm that the inhibition of ABCA1 function decreases metastatic phenotypes.

Having seen the importance of ABCA1 in vitro, we tested whether it drives metastasis in vivo. We injected the control and MDA-MB-231/shABCA1 cells (with two different shRNAs) into the tail veins of immunocompromised NOD/SCID mice and sacrificed them after 49 days. We stained the lungs for expression of ABCA1 and confirmed that the tumors from ABCA1 knockdown had lower ABCA1 expression (Supplementary Fig. S5A and S5B). At that time, they had significantly more metastatic nodules both on the surface (Fig. 7E and F) and in the interior of the lungs (Fig. 7G). To check the growth rate of the cells in vivo, we injected MDA-MB-231/control and MDA-MB-231/shABCA1-3 cells orthotopically into the mammary fat pads of NOD/SCID mice and saw no difference in the size of the resultant tumors, in contrast to a prior report, potentially because in the prior study, ABCA1 was ectopically overexpressed in a genetic background, where it promotes apoptosis (Supplementary Fig. S5C; ref. 38). In any case, our results demonstrate that ABCA1 significantly enhances metastatic colonization in vivo.

Given that ABCA1 is expressed during EMT and can drive metastatic properties in vitro and metastasis in vivo, we predicted that it may be a mechanism used by human tumors to increase metastatic capacity. To test this, we examined ABCA1 expression across four publicly available datasets of breast cancer patients that were monitored for distant metastases (Supplementary Fig. S5D). Among tumors that metastasized within five years, 41%±28% had high ABCA1 expression (Fig. 7H). Those patients acquired metastases significantly earlier than those with normal ABCA1 expression. The time required for 25% of the tumors to metastasize was 9±4 years shorter when ABCA1 was high. Hundred percent of patients with high ABCA1 had metastases by 12.5 years, compared with only 28% of patients with normal ABCA1. The strong association between ABCA1 and distant metastases in human breast cancers, coupled with the preclinical evidence revealing the mechanism, shows that ABCA1 is a potent inducer of metastatic capacity in human breast cancers. This demonstrates a novel strategy for human tumors to acquire metastatic capacity, namely the regulation of fluidity through alteration in cellular cholesterol homeostasis.

Our investigations revealed that membrane fluidity is causally associated with metastasis, provides a mechanism that explains the efficacy of previously and newly identified antimetastasis drugs, and shows that breast tumors have altered cholesterol efflux associated with metastasis (Fig. 7I). The fact that the efficacy of antimetastasis drugs depends on their ability to reduce fluidity enables the search for more potent therapeutics.

The unexpected alterations of cholesterol efflux in breast cancer raise the possibility that systemic cholesterol levels may affect metastasis. This question has been addressed with epidemiologic studies starting 85 years ago with the observation that cancer patients have less heart diseases, initiating investigations that continue almost a century later (41). A range of studies found a correlation between low levels of plasma cholesterol and increased mortality from cancer, although a minority saw no significant effect (Supplementary Table S3). One explanation for the correlation is that some patients may harbor undiagnosed tumors that are metabolizing cholesterol, leading to lower plasma cholesterol levels (42). This is supported by data showing that cholesterol and its metabolites can promote tumor initiation (43). However, even after controlling for this by lag-time analyses, the association remains.

The conflicting interpretations of the association between cholesterol and cancer suggest a complex relationship between cholesterol and disease that is not yet fully understood. As cholesterol affects tumor initiation, proliferation, and metastasis through a range of mechanisms, it is likely that cholesterol plays both promoting and inhibiting roles throughout tumor development. To better understand it, studies must be carried out that carefully isolate the contribution of cholesterol to different hallmarks of the cancer phenotype. This study reveals the impact of fluidity and cholesterol specifically on metastasis. While cells may utilize ABCA1 to promote metastasis, the same gene may also inhibit the development of tumors by inhibiting cellular proliferation (38). Similarly, as inhibitors of cholesterol synthesis, one would predict that statins would promote metastasis. However, the literature on statins is ambiguous, showing both tumorigenic and antitumorigenic effects, potentially due to pleiotropic effects of the drug.

Finally, the ability of fluidity to inhibit metastasis has implications for therapy. Nearly all the antimetastasis drugs we identified are FDA approved. These drugs were able to reduce the fluidity of primary tumors and decrease metastatic capacity in vivo at dosages analogous to those given to humans with no obvious additional systemic toxicity. However, this raises questions regarding the impact of long-term nonspecific decrease of fluidity across an organism. Suggesting that the treatment is not evidently toxic is the fact that one of the drugs, haloperidol, is used to treat chronic psychiatric disorders. Thus, the apparent wealth in available drugs that can affect membrane fluidity, observed by others and ourselves, suggests that a viable antimetastatic therapy may already exist.

No potential conflicts of interest were disclosed.

Conception and design: W. Zhao, J.T. Chang

Development of methodology: B.C. Urban, Y. Zuo, Z. Tan, J.T. Chang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): W. Zhao, S. Prijic, B.C. Urban, M.J. Tisza, Y. Zuo, L. Li, S.A. Mani, J.T. Chang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): W. Zhao, S. Prijic, B.C. Urban, M.J. Tisza, X. Chen, J.T. Chang

Writing, review, and/or revision of the manuscript: W. Zhao, S. Prijic, S.A. Mani, J.T. Chang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Zuo, J.T. Chang

Study supervision: J.T. Chang

The authors thank Drs. Jeff Frost, Guangwei Du, Ilya Levental, Jaime Rodriguez Canales, and Jeff Rosen for reagents, technical assistance, and insights.

This work was supported by grant R1006 from the Cancer Prevention and Research Institute of Texas (J.T. Chang), NIH/NCI CA155243 (S.A. Mani), and UTHealth Innovation in Cancer Prevention Research Training Program Postdoctoral Fellowship from CPRIT grant RP140103 (S. Prijic).

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.