Statins as Anticancer Agents in the Era of Precision Medicine

Since their approval by the FDA in the late 1980s, statins have revolutionized the clinical management of high cholesterol. Statins are specific inhibitors of the mevalonate pathway, which is responsible for the de novo synthesis of cholesterol and nonsterol isoprenoids (Fig. 1). Specifically, statins inhibit the conversion of HMG-CoA to mevalonate by inhibiting the rate-limiting enzyme of the mevalonate pathway, HMG-CoA reductase (HMGCR). In addition to its important roles in normal physiology, the mevalonate pathway supports tumorigenesis and is known to be deregulated in human cancers (1–4). As such, there is significant interest in repurposing statins as anticancer agents. Statins have been shown to induce potent tumor-specific apoptosis (5–7). Moreover, many retrospective studies have reported that statin use is associated with reduced cancer risk (8–11), lower cancer grade and stage at diagnosis (12, 13), and reduced recurrence and/or cancer-specific mortality (14–18). Given that statins are FDA-approved, well-tolerated, and are available as generic drugs, they offer an immediate, safe, and inexpensive opportunity to improve cancer patient care and treatment outcomes.

Despite these promising observations, statins have yet to be repurposed and integrated into cancer patient care. Emerging evidence suggests that certain molecular subtypes of cancer are more susceptible to statin therapy than others, highlighting the importance of predictive biomarkers for patient stratification. Moreover, recent clinical trials have provided important insights into how to realistically use these agents in an oncology setting. In this review, we highlight the gaps in knowledge that have precluded the repurposing of statins as anticancer agents, as well as recent advances that will help inform future clinical trial design.

Statins compete with HMG-CoA for binding to the active site of HMGCR, thereby reducing mevalonate synthesis. As a consequence, statins deplete intracellular cholesterol, which triggers a homeostatic feedback mechanism governed by the sterol regulatory element-binding protein (SREBP) family of transcription factors (Fig. 1). Activation of the SREBPs results in the increased expression of mevalonate pathway and sterol metabolism genes, including HMGCR and the low-density lipoprotein (LDL) receptor (LDLR). Increased membrane expression of LDLR leads to enhanced LDL cholesterol (LDL-C) uptake from the bloodstream, thus effectively lowering serum cholesterol levels. As a result, statins are commonly prescribed to reduce the risk of cardiovascular disease or improve survival in patients with cardiovascular disease.

Cholesterol has also been shown to play multifaceted roles in tumorigenesis (reviewed in refs. 1, 19). In specific contexts, statins have been shown to elicit their anticancer effects through the depletion of cholesterol. For example, one study demonstrated that simvastatin decreases the cholesterol content of lipid rafts in prostate cancer cells, which hinders AKT signaling and induces apoptosis (20). Moreover, in a subset of medulloblastoma driven by aberrant Hedgehog signaling, the depletion of cholesterol impairs signal transduction and inhibits cancer cell growth (21). However, in the majority of other reports, exogenous cholesterol is unable to rescue statin-induced apoptosis, highlighting a role for other end products of the mevalonate pathway in cancer cell survival.

In addition to cholesterol, statins also reduce the synthesis of nonsterol isoprenoids, including geranylgeranyl pyrophosphate (GGPP; Fig. 1). Several studies have shown that statin-induced apoptosis can be consistently and fully rescued by exogenous mevalonate or mevalonate-derived GGPP (22–25). These studies not only support that statin-induced apoptosis is an on-target effect, but also reveal that certain cancers rely on GGPP synthesis for survival. GGPP can serve as a substrate for protein prenylation, or as a precursor for the synthesis of other metabolites, such as coenzyme Q (CoQ) and dolichols (1). In recent years, it has become apparent that different cancer cell types have a dependency on distinct fates of GGPP (22–24, 26, 27). In acute myeloid leukemia and multiple myeloma cells, statin-induced apoptosis can be phenocopied by prenylation inhibitors, which suggests that these cancers rely on GGPP synthesis, at least in part, for protein prenylation (22, 23, 28). However, in other cancers where statin-induced cell death can be rescued by exogenous GGPP, statin sensitivity can be uncoupled from effects on protein prenylation (24). Indeed, recent studies have shown that certain tumors rely on the mevalonate pathway for the synthesis GGPP-derived CoQ (26, 27). In these cells, statin treatment leads to oxidative stress and apoptosis (26, 27), which can be rescued by exogenous CoQ (26).

Despite numerous studies implicating the direct, intratumoral inhibition of HMGCR as the mechanism by which statins elicit their anticancer effects, systemic contributions are also likely. It is important to note that, unlike in humans, statin treatment does not reduce serum cholesterol levels in mice (20, 29). While reducing circulating cholesterol levels may add to the benefit of statin therapy in patients with cancer, evidence from preclinical studies support a direct mechanism. Thus, in this review, we focus primarily on the direct effects of statins on cancer cells.

While many epidemiologic studies report positive associations between statin use and cancer patient outcomes, the extent to which statin use confers a benefit is variable between studies (14–18). Several factors might explain this heterogeneity, including interpatient differences in the type of statin, dose, and duration of statin use (discussed further in the next section). Furthermore, it is possible that not all patients with cancer benefit equally from statin therapy. Consistent with this hypothesis, highly heterogeneous responses to statin exposure across panels of cancer cell lines have been reported (23, 24, 30–32), and biomarkers of statin sensitivity have recently been described (Fig. 2). Hence, different tumor subtypes are not equally vulnerable to statin therapy.

If statins are to be repurposed for the precise treatment of cancer, we must first identify which tumor subtypes are vulnerable to statin-mediated HMGCR inhibition. In breast cancer, for example, statin sensitivity has been associated with estrogen receptor (ER) status, where ER-negative breast cancer cells are particularly sensitive to statin exposure (31). These preclinical observations are further supported by clinical data demonstrating greater tumor cell apoptosis after fluvastatin treatment in women with ER-negative breast cancer (33). Independent studies have demonstrated that tumor cells of various origins with higher expression of mesenchymal cell markers (e.g., vimentin) and/or lower expression of epithelial cell markers (e.g., E-cadherin) are highly sensitive to statin treatment (24, 34, 35). Furthermore, statins have been shown to preferentially kill cells induced to undergo epithelial-to-mesenchymal transition (24), suggesting that they may be effective at impairing metastatic disease. Whether ER-negative breast tumors are more sensitive to statins because they are more mesenchymal remains to be determined. Moreover, it remains poorly understood why cancer cells in a mesenchymal state are vulnerable to HMGCR inhibition. Nonetheless, these data further support the concept that statin sensitivity can be stratified by tumor subtype.

In some cancer cells, statin sensitivity is inversely associated with the ability to activate a feedback mechanism in response to mevalonate pathway inhibition. In response to cholesterol depletion, the SREBP family of transcription factors is activated to restore homeostasis (Fig. 1). In certain cancer cells, this feedback mechanism is impaired, which renders them vulnerable to HMGCR inhibition. In multiple myeloma, for example, it was shown that a subset of cell lines and primary patient-derived cells fail to induce the expression of SREBP target genes following statin treatment and readily undergo apoptosis (36). In contrast, cell lines and primary cells with robust statin-induced SREBP activation were resistant to statin exposure (36). Statin sensitivity has subsequently been associated with impaired feedback regulation of the mevalonate pathway in other cancer types, including prostate cancer (32). Further research is required to better understand why some cancer cells have impaired feedback regulation of the mevalonate pathway, and to identify a clinically amenable biomarker that can stratify patients on the basis of this dampened homeostatic response.

In breast cancer, statin sensitivity has been inversely associated with high basal expression of cholesterol biosynthesis genes, including HMGCR (37). This is consistent with a report that acquired resistance to statin exposure in vitro is associated with significantly elevated HMGCR expression (38). However, studies that have evaluated HMGCR expression as a predictive biomarker of statin sensitivity have yielded conflicting results (31, 36–39). This is likely due, in part, to the lack of specificity of many commercially available HMGCR antibodies (2, 37, 40). These observations also suggest that there is a complex relationship between HMGCR expression and statin sensitivity in cancer. On one hand, elevated HMGCR expression and deregulated mevalonate pathway activity can support tumorigenesis and render cancer cells vulnerable to statin treatment (2, 41). In these tumors, elevated HMGCR expression may indicate a tumor dependency, whereby even a slight dampening of mevalonate pathway activity is sufficient to induce tumor-specific cell death. On the other hand, as HMGCR expression continues to increase (e.g., via elevated SREBP activity), higher statin drug concentrations are required to inhibit the mevalonate pathway, thereby decreasing statin sensitivity (32, 36, 38, 42). Hence, careful consideration is required when evaluating the utility of HMGCR expression as a predictive biomarker of statin sensitivity in cancer.

There is extensive interplay between the mevalonate pathway and signal transduction in cancer (reviewed in ref. 1), and, therefore, aberrant cell signaling in tumors may confer increased sensitivity to statin therapy. For example, mevalonate-derived farnesyl pyrophosphate (FPP) and GGPP serve as substrates for the posttranslational prenylation of oncoproteins such as RAS and RHO, which is important for their proper localization and function (43). As such, it has long been hypothesized that RAS mutations may be potential biomarkers of statin sensitivity. While activated RAS can sensitize some cells to statins (24, 44), preclinical studies have shown that statin-induced apoptosis is independent of RAS localization and function (23, 24, 44). Moreover, several clinical trials have evaluated statin therapy in patients with RAS-mutant tumors, but the majority of trials failed to demonstrate promising therapeutic responses (45–48). Hence, despite the interplay between RAS and the mevalonate pathway in cancer, RAS status is a poor predictor of statin sensitivity.

A number of recent studies have also implicated TP53 status in modulating cancer cell sensitivity to statins. While wild-type p53 represses the mevalonate pathway (49), loss of p53 and certain gain-of-function p53 mutants have been shown to induce the expression of mevalonate pathway genes (41, 49, 50). Consistently, it has been demonstrated that these p53-null (26) or -mutant (50–52) tumors are dependent on the mevalonate pathway and particularly vulnerable to statin treatment. The latter has been attributed to the roles of the mevalonate pathway in the stability of mutant p53 protein (51–53).

Cancer type–specific biomarkers of statin sensitivity may also exist. For example, in clear-cell renal cell carcinoma, cells driven by loss of the tumor suppressor, von Hippel-Lindau (VHL; ∼90% of tumors), were found to be dependent on the mevalonate pathway for proper RHO and RHO kinase (ROCK) signaling, and were more sensitive to statin treatment compared with VHL wild-type cells (54). Moreover, in multiple myeloma, cancer cells driven by a t(4;14) chromosomal translocation are highly dependent on GGPP synthesis and more sensitive to statin-induced apoptosis compared with other multiple myeloma subtypes (55). While TP53, VHL, and t(4;14) status can potentially predict statin sensitivity, further validation in patients will be required before these biomarkers can be used clinically.

After identifying which patients with cancer might benefit from the addition of a statin to their treatment regimen, the next step is evaluating how best to prescribe these drugs as anticancer agents. Data from epidemiologic, preclinical, and early-phase clinical studies have demonstrated that statin type, dose, and treatment duration are all important variables to consider when evaluating statins as anticancer agents. While all FDA-approved statins are effective in lowering serum cholesterol by inhibiting HMGCR activity in the liver (Table 1), their ability to directly inhibit HMGCR in extrahepatic tumor tissues may be statin type specific. It has been hypothesized that the lipophilic statin drugs are more likely to reach and readily enter extrahepatic cells, whereas hydrophilic statins are more hepatoselective (56). Consistent with this hypothesis, epidemiologic studies have reported that lipophilic, but not hydrophilic, statin use is associated with reduced cancer incidence (10) and recurrence (15) in patients with breast cancer.

Recent clinical studies have reported that the lipophilic statins, atorvastatin (57) and fluvastatin (58), are measurable in prostatic tissue at low nanomolar concentrations after short-term treatment with a typical cholesterol-lowering dose (80 mg/day). While these concentrations are less than those evaluated in most in vitro studies, these lower concentrations, when prescribed in the neoadjuvant setting (discussed further in the next section), were shown to reduce tumor cell proliferation (59) or induce apoptosis (58) in a time-dependent manner. These observations are consistent with epidemiologic (60, 61), preclinical (30, 31, 58, 62), and clinical (33, 58, 59) data, all of which indicate that the anticancer effects of statins are both dose- and time-dependent. This implies that comparable anticancer responses may be achieved using lower statin doses over longer durations versus higher statin doses over a shorter period of time. Phase I dose escalation studies have indeed demonstrated that statins are well-tolerated at doses much higher than typically prescribed for cholesterol management (∼10–30 × higher), at least for defined periods of time (63–66).

Interestingly, while similar concentrations of atorvastatin and fluvastatin were measured in prostatic tissue following acute treatment, only atorvastatin was found to accumulate within the prostate relative to the serum (57, 58). This may have important implications for longer treatment schedules, as the pharmacokinetic properties of specific statins may enable higher achievable drug concentrations within certain tumor tissues over time. The choice of statin and dosing schedule will likely depend on the type of cancer being treated.

A promising therapeutic space for the use of statins is soon after diagnosis to delay the need for more aggressive treatment and/or improve the outcome of first-line therapy. In a series of window-of-opportunity trials in breast and prostate cancer, lipophilic statin treatment showed evidence of reduced tumor cell proliferation and increased apoptosis in a subset of patients. In these studies, short-term neoadjuvant treatment (between 1.5 and 12 weeks) with a cholesterol-lowering dose of either fluvastatin (33, 58) or atorvastatin (39, 59) was evaluated. In all four studies, pretreatment biopsy samples were compared with surgical material obtained after statin treatment. Immunohistochemistry was then performed to evaluate markers of tumor cell proliferation (Ki67) and/or apoptosis (cleaved caspase-3). Fluvastatin treatment was reported to increase tumor cell apoptosis in patients with high-grade breast cancer (33) and localized prostate cancer (58), where greater increases were observed in patients on a higher dose (33) or treated for longer durations (58). Similarly, neoadjuvant atorvastatin therapy was shown to reduce tumor cell proliferation in patients with primary invasive breast cancer (39). Subsequent microarray analysis in these same paired clinical samples revealed atorvastatin-induced effects on genes associated with apoptosis and reduced MAPK signaling (67). While neoadjuvant atorvastatin therapy was not found to reduce intratumoral Ki67 staining in patients with prostate cancer overall, a significant decrease in Ki67 was observed in patients on atorvastatin for greater than 28 days (59); however, a similar response in Ki67 was not observed following fluvastatin therapy (58).

Not only do these studies reinforce that the anticancer effects of statins are both dose- and time-dependent, but they further highlight that certain subgroups of patients may benefit more than others. For example, neoadjuvant fluvastatin treatment in breast cancer was found to decrease Ki67 and increase cleaved caspase-3 expression in patients with ER-negative, high-grade tumors (33), which is consistent with ER-negative breast cancer cells being particularly vulnerable to statin exposure (31). Future studies are needed to evaluate the potential long-term benefits that these effects may have on disease progression.

In phase I/II studies that have evaluated statins in combination with various standard-of-care therapies, promising responses have been reported in some patients, ranging from stable disease to complete responses (64, 66, 68, 69). While it is premature to draw conclusions about statin efficacy from these studies, these data provide important information that should be considered when designing future randomized controlled trials (RCTs). For example, the variable responses observed when considering mixed patient populations suggest that there are likely specific subsets of patients with cancer who might benefit from statin therapy, highlighting the need for predictive biomarkers to inform patient stratification.

Statin therapy has been evaluated in a number of RCTs (Table 2), including phase III trials in patients with small-cell lung cancer (SCLC; ref. 70), metastatic colorectal cancer (71), advanced hepatocellular carcinoma (72), or advanced gastric cancer (73). In these studies, the addition of 40 mg/day pravastatin (70, 72) or simvastatin (71, 73) to standard chemotherapy offered no additional benefit compared with chemotherapy alone. While disappointing, these studies were designed and initiated prior to evidence demonstrating that specific tumor subtypes are more vulnerable to statins than others. No phase III study to date has stratified patients on the basis of molecular markers predictive of statin sensitivity; however, post hoc analyses may uncover that a particular subgroup of patients benefited from statin therapy in these phase III trials. Moreover, given our increasing understanding of the differences between statin drugs and their differential ability to accumulate in extrahepatic tissues (57, 58, 74), choice of statin drug is an important factor. Both pravastatin and simvastatin at 40 mg/day are moderate-intensity prescriptions (Table 1), and, therefore, higher doses or prescription of a higher intensity statin might have yielded greater responses in these studies. Drug combination strategies to potentiate the anticancer activity of statin drugs might also be considered for future RCTs.

Statins have been evaluated in combination with various classes of other anticancer agents, including targeted therapeutics against different oncogenic signaling pathways and epigenetic modifiers (Table 3). Combining a statin with other targeted therapies can enhance their anticancer activity and overcome potential drug resistance mechanisms.

As with any combination therapy approach, not only is there the potential for synergistic anticancer activity, but there is also the possibility of drug–drug interactions that lead to increased toxicity. Hence, careful consideration must be given to drug selection. In addition to differences in solubility, statins also differ from one another in how they are metabolized (Table 1). For example, atorvastatin is highly lipophilic, but is primarily metabolized by cytochrome P450 (CYP450) 3A4 (CYP3A4). CYP3A4 function is modulated by certain foods and several commonly prescribed medications (including many chemotherapeutics), and, therefore, lipophilic statins metabolized by other enzymes, such as fluvastatin or pitavastatin, may offer a lower potential for unwanted drug–drug interactions (75). Moreover, some statins, such as lovastatin, have been shown to interact with and modulate P-glycoprotein activity (a major drug efflux pump; refs. 30, 76). These factors must be considered when evaluating statins in combination with other targeted therapies.

One approach to potentiate statin-induced apoptosis is via combination treatment with SREBP inhibitors. Similar to normal cells, statin treatment triggers the activation of the SREBPs in most cancer cells (Fig. 1). Statin-mediated activation of the SREBPs, particularly SREBP2, results in the induction of mevalonate pathway gene expression, including the upregulation of HMGCR. Knockdown of SREBF2 (the gene that encodes SREBP2) via RNAi suppresses this feedback loop and sensitizes cancer cells to statin-induced death (32, 77). Consistent with this result, our group identified that the drug dipyridamole, an agent approved for the secondary prevention of cerebral ischemia, could synergize with statins to induce apoptosis in hematologic cancer (42) and prostate cancer (32) cells. Mechanistically, we and others have shown that dipyridamole inhibits statin-induced SREBP activation, thereby preventing the upregulation of mevalonate pathway genes in response to statin exposure (32, 38, 42). By impairing this feedback mechanism, dipyridamole significantly reduces the concentration of statin drug needed to inhibit the mevalonate pathway and induce apoptosis. Because both statins and dipyridamole are FDA-approved drugs, there is interest in advancing this drug combination to clinical trials in oncology.

Epidemiologic evidence supports a positive association between statin use and response to antiandrogen therapy in patients with prostate cancer (78–81). These data are further supported by preclinical studies showing that the combination of a statin with either abiraterone acetate or enzalutamide enhances cytotoxicity in prostate cancer cell lines (82, 83). Moreover, enzalutamide-resistant prostate cancer cells upregulate HMGCR expression, and treatment with simvastatin resensitizes these cells to enzalutamide (84).

A number of mechanisms have been proposed for the interaction between statins and antiandrogen therapy. First, cholesterol is a precursor for androgen biosynthesis, and, therefore, statin-mediated cholesterol depletion may also reduce intratumoral androgen levels. Consistent with this hypothesis, statins have been shown to inhibit androgen receptor (AR) activity in prostate cancer cell lines (82–85). In these same studies, statins were also found to reduce AR expression (82–85), possibly via the inhibition of AKT/mTOR signaling (84). Moreover, statin use has been associated with reduced serum PSA levels in patients with prostate cancer (86), which is regulated by AR. Finally, certain statin drugs can compete with dehydroepiandrosterone sulfate, a testosterone precursor, for binding to a transporter at the surface of prostate cancer cells, and, therefore, block androgen uptake (78). Taken together, statins can enhance antiandrogen therapy through multiple mechanisms. Prospective clinical trials are warranted to evaluate the combination of a statin and antiandrogen therapy in patients with advanced prostate cancer and other steroid hormone–driven malignancies.

Given the interplay between the mevalonate pathway and oncogenic signal transduction (1), numerous studies have evaluated the combination of a statin with various agents that target cell signaling (Table 3). For example, statins synergize with EGFR inhibitors, including erlotinib and gefitinib, to induce cell death in a number of different cancer cell types in vitro (87). Statins have been shown to inhibit ligand-induced EGFR activation and downstream AKT signaling, which can be reversed by exogenous GGPP (88, 89). Statin-mediated inhibition of AKT has further been implicated as a mechanism for overcoming resistance to EGFR inhibitors in non–small cell lung cancer (NSCLC) cells (90, 91). Combined treatment with a statin and EGFR inhibitor has been evaluated in phase II RCTs in patients with NSCLC (Table 2). Simvastatin (40 mg/day) in combination with gefitinib resulted in higher tumor response rates and longer progression-free survival (PFS) compared with gefitinib alone in patients with EGFR wild-type nonadenocarcinomas (92); however, similar responses were not observed when simvastatin was combined with afatinib, a second-generation EGFR inhibitor (93).

An emerging area of investigation is the combination of statins and epigenetic inhibitors, including histone deacetylase (HDAC) and bromodomain inhibitors (refs. 3, 94–96; Table 3). A series of dual-action compounds has also been developed, where the hydroxamate group of vorinostat, an HDAC inhibitor, was fused to lovastatin (97). The resulting HMGCR-HDAC dual inhibitors have been shown to possess potent and selective anticancer activity (98, 99). In mouse models of colorectal cancer, treatment with a dual HMGCR-HDAC inhibitor significantly reduced intestinal inflammation, decreased tumor burden, and impaired metastasis (98).

The mechanism by which statins and different epigenetic inhibitors interact remains poorly characterized. However, given that acetyl-CoA is required for both protein acetylation and mevalonate metabolism, it is possible that simultaneously inhibiting multiple acetyl-CoA–dependent processes is detrimental to tumor cells. Indeed, both histone acetylation and mevalonate pathway gene expression are upregulated in pancreatic adenocarcinoma (PDAC) tumors, and cotreatment with atorvastatin and a bromodomain inhibitor, JQ1, significantly impairs PDAC cell growth (3).

Statins have also been shown to elicit immunomodulatory effects (reviewed in ref. 100). High cholesterol in the tumor microenvironment and in tumor-infiltrating CD8⁺ T cells is associated with elevated expression of immune checkpoint proteins and enhanced T-cell exhaustion, which allows tumor cells to escape immune surveillance (101). Importantly, reducing cholesterol levels in the tumor microenvironment or in CD8⁺ T cells restores T-cell antitumor activity (101). These critical observations highlight the potential for statins to be combined with immunotherapy for the treatment of cancer. An independent study that evaluated the vaccine adjuvant activity of statins revealed that lipophilic statins, such as simvastatin, induce a strong Th1 and cytotoxic T-cell response in mice and enhance the therapeutic response to cancer vaccination (102). In particular, the inhibition of protein prenylation in antigen-presenting cells enhanced antigen presentation and T-cell activation (102). This favorable antitumor response was further potentiated by programmed cell death protein-1 (PD-1) blockade, which resulted in prolonged survival in mice inoculated with melanoma or human papillomavirus–associated tumors (102). As most preclinical studies to date have evaluated the anticancer activity of statins in vitro or in immunocompromised animal models, future investigation into the immunomodulatory properties of statins will undoubtably open exciting avenues of research with important clinical implications.

If statins are to be integrated into cancer patient care, a precision medicine approach is necessary. In this review, we highlighted recent advances and outlined important considerations for advancing statins to clinical trials in oncology. We also proposed key questions that should be the focus of future research (Table 4). As not all tumors are vulnerable to statin-mediated mevalonate pathway inhibition, the development of predictive biomarkers of statin sensitivity is crucial for patient stratification. We have highlighted some promising preclinical biomarkers of statin sensitivity, which can be validated in future clinical trials by enriching for patients with these tumor features. In addition, post hoc analyses of completed, unbiased RCTs may similarly reveal novel biomarkers of statin response. However, few statin RCTs in oncology have been performed to date, and those that have been performed evaluated moderate-intensity statin regimens. Given the increasing evidence that certain statins may be better suited as anticancer agents than others, coupled with data indicating that statin-induced apoptosis is both dose- and time-dependent, careful consideration is required when deciding which statin(s) and dosing schedules to evaluate clinically. It is also unlikely that statins will be prescribed as a monotherapy, and, therefore, further investigation into drug combination strategies will remain an important area of research. As a number of preclinical potentiators of statin-induced cancer cell death have already been described, many of which are FDA-approved, immediate phase I/II studies are possible. The outcome of these studies will provide important insights into how to realistically use these immediately available, well-tolerated, and inexpensive agents as precision anticancer therapeutics.

No potential conflicts of interest were disclosed.

Opinions, interpretations, conclusions, and recommendations are those of the authors and are not necessarily endorsed by the Department of Defense.

We thank all members of the Penn laboratory, Drs. Rebecca Gladdy, Benjamin Haibe-Kains, Rob Hamilton, and David Hedley for helpful discussions and critical review of this article. L.Z. Penn holds a Tier 1 Canada Research Chair in Molecular Oncology. The Penn laboratory was supported by funding from the Canadian Institutes of Health Research (CIHR; FRN: 142263), Terry Fox Research Institute (New Frontiers PPG-1064), Canadian Cancer Society (grant #706394), and the Office of the Assistant Secretary of Defense for Health Affairs, through the Breast Cancer Research Program (award no., W81XWH-16-1-0068). J. Longo was supported by a CIHR Doctoral Research award, J.E. van Leeuwen was supported by a Princess Margaret Hospital Foundation Graduate Fellowship in Cancer Research, and M. Elbaz was supported by a George Knudson Memorial Fellowship.

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.