Background
From antiquity
1 [
1] to the present day cancer has plagued humanity; in 2018 cancer claimed an estimated 9.6 million lives, one in six deaths worldwide [
2]. Despite the many consequential improvements in cancer treatment, there remains a clinical imperative to identify novel therapies with improved efficacy and diminished toxicity. In this context, statins [3-hydroxy-3-methylglutaryl-CoA (HMG-CoA) reductase inhibitors], the most commonly prescribed class of pharmaceuticals worldwide, have engendered promise and drawn scrutiny.
In 1971 Akira Endo isolated the first statin progenitor, citrinin [
3]. Soon it was shown to be an inhibitor of HMG-CoA reductase [
4], the rate limiting step in the mevalonate pathway. Although citrinin proved to be nephrotoxic, in 1976 the Endo laboratory [
5,
6] and a British group [
7] independently isolated ML-236B, also called compactin. Eleven years later, the FDA approved the first commercial statin: naturally derived lovastatin. Subsequently six statins, including two semi-synthetic and four synthetic formulations, have entered the marketplace and represent primary therapy for the prevention of cardiovascular disease. Today, an estimated 30 million people worldwide take statins.
Within a few years of the introduction of statins, however, concerns regarding their safety emerged, notably an associated increase in non-cardiac mortality [
8]. Animal studies suggested that they might be carcinogenic: when given statins at doses equivalent to those commonly prescribed in humans, rats developed lymphomas and carcinomas of the liver, stomach, lung, and thyroid [
9]. Consequently, large randomized clinical trials were undertaken to evaluate not only the efficacy of statins but also any associated risk of cancer. Ironically, by demonstrating reduced incidences of colorectal carcinoma, prostate cancer and melanoma [
10], these studies were the first to indicate that statins might prevent cancer. To date, preclinical and clinical data suggest chemopreventive effects of statins against a variety of cancers including those of the breast [
11], colon [
12], lung [
13], liver [
14], pancreas [
15], and prostate [
16].
Several lines of evidence have suggested that statins might also have value in the treatment of cancer: statins modulate the mevalonate pathway [
17], which ultimately modifies the posttranslational processing of proteins involved in cell cycle control; cancer cells exhibit increased synthesis, receptor mediated uptake, and degradation of cholesterol (reviewed in [
18]); and, disrupted cholesterol homeostasis has been demonstrated in various tumor models (see, for example, [
19]). In 1998, Matar, et al. [
20], published a landmark study: a short course of lovastatin in rats inhibited primary fibrosarcoma growth and diminished the size and number of experimentally induced lung metastases. Subsequently, numerous publications have supported the notion that statins exert anticancer activity through mevalonate-dependent and -independent mechanisms, as recently reviewed [
17,
21].
Disappointingly, statins
alone have not proven effective as anticancer therapy; however, there is evidence that statins might potentiate the effects of anti-cancer drugs [
22]. A recent systematic review and meta-analysis by Mei et al. [
23] (which included 95 studies, 1,111,407 patients and more than 18 cancer types) compared statin users to individuals not taking statins. The patients receiving statins in conjunction with chemotherapy experienced a 30% reduction in all-cause mortality, a 40% reduction in cancer-specific mortality, and prolonged progression-free, recurrence-free, and disease-free survivals. Yet soon after the publication of Mei’s analysis, Farooqui et al. [
24] published a systematic review and meta-analysis of 10 randomized controlled studies involving 1881 individuals with stage 3 or 4 cancers, in which statin use did not improve progression-free or overall survival.
Reaching a sound assessment from clinical trials of the value of statins as adjuncts to conventional chemotherapy is confounded by the numbers of different drugs – both statins and chemotherapeutic agents – featured in the various investigations, and the multiplicity of cancer types treated. We reasoned that the possible therapeutic benefits of statins in the context of chemotherapy are unlikely the global consequence of statin administration but instead are specific to the interacting drug combinations. Therefore, we chose to investigate a subset of statin – chemotherapeutic drug interactions by rigorously assaying a single statin, lovastatin (which has similar dissociation constants with HMG Co-A reductase from budding yeast and mammalian sources [
25]), in combination with each of ten FDA-approved chemotherapeutic agents having a variety of mechanisms of action.
Because our objective was to compare the ten drug pairings with each other, it was important to use a neutral cellular substrate, rather than a particular human cancer cell line which because of its inherent cellular origin and genetic mutations might favor or disadvantage a particular drug pair. We therefore chose as the cellular substrate the exceptionally well-studied model organism,
Saccharomyces cerevisiae [
26]. There is ample precedent for utilizing budding yeast in research related to cancer and its therapies [
26‐
28]. Many of its genes have human orthologs [
29] and many cell signaling pathways now recognized as critical to oncogenesis were first identified and/or extensively studied in yeast. Importantly,
S. cerevisiae has been utilized in numerous pharmacologic studies and high throughput screens [
30‐
40], including ones specifically focused on anticancer drug research (reviewed in [
41]). In anticipation of future studies of the genetic basis of interactions of statin-drug combinations, we created a balanced pool of heterozygous deletion strains (marked by DNA bar codes) of
S. cerevisiae essential genes. Barcoded pools [
42] were employed in the three largest yeast chemogenomic screens [
32,
34,
36] which are collated in the NetwoRx data base [
33]. (The pools also provide a resource for investigating genetically driven resistance to drug treatments [
43,
44].) We configured a 96-well microplate assay compatible with the
Combenefit dual-drug interaction software [
45]; cell concentration data, read spectrophotometrically, were submitted to rigorous statistical analysis for synergistic or antagonistic interactions, calculated according to the Loewe additivity model [
46] which is part of the
Combenefit package.
Our data demonstrate that combining lovastatin with conventional chemotherapeutic agents results in drastically different interactions, ranging from strong synergism to profound antagonism, sometimes within the same concentration space. Of the ten chemotherapeutic drugs, four (tamoxifen, doxorubicin, methotrexate, and rapamycin) exhibited net synergism with lovastatin; two drugs (gemcitabine and 5-fluorouracil) had neutral scores; and four (epothilone, cisplatin, cyclophosphamide, and etoposide) displayed net antagonism. As proof of principle, two of the drug combinations, tamoxifen/lovastatin and cisplatin/lovastatin, were further evaluated in human cancer cell lines. The results in cell lines were generally accordant with the data obtained with S. cerevisiae but with variations in the patterns of synergism and antagonism between individual cell lines, even of the same cancer type.
Discussion
The fundamental question we sought to answer in this study is whether statin-chemotherapy interactions are a manifestation of a global effect of statins or are, as we hypothesized, specific to individual chemotherapeutic drugs. To that end, we assayed a single statin, lovastatin, paired with each of ten commonly prescribed chemotherapeutic agents, with various mechanisms of action (Table
1). The cellular substrate was the model organism,
Saccharomyces cerevisiae; a balanced pool of heterozygous deletions of all essential genes was created for this study. Highly reproducible dilution assays were performed in microplates; cell densities were compiled with a plate reader. The data from the assays were submitted to rigorous statistical analysis to identify synergistic and antagonistic interactions, according to the Loewe additivity model, implemented with the Combenefit software package [
45].
The Combenefit program compiles the data from the plate replicates for a given drug pair and computes various global metrics according to assumptions dictated by the chosen model. The metric “SUM_SYN_ANT”, which is as summation of all interactions within the concentration space, proved most useful. In accordance with our hypothesis, the ten chemotherapeutic agents exhibited a spectrum of global interactions, ranging from synergism to neutrality to antagonism (Fig.
1a). Four chemotherapeutic drugs paired with lovastatin exhibited strong, statistically significant, synergistic interactions: tamoxifen, doxorubicin, methotrexate and rapamycin. Two of the ten chemotherapeutic drugs - gemcitabine and 5-fluorouracil - had neutral interactions with lovastatin; and four – epothilone, cisplatin, cyclophosphamide and etoposide - exhibited net antagonism. Global metrics however provide only a snapshot of drug-drug interactions. The synergism score matrices were more informative; they revealed that synergism and antagonism often resided in the same concentration space, sometimes with potential clinical implications (see later).
Having demonstrated net synergism of four drug combinations in the yeast model, we selected one of them – tamoxifen and lovastatin - for additional study in human cancer cell lines. This combination, which exhibited the strongest synergy on day two, was intriguing because of its substantial efficacy in
S. cerevisiae, which is devoid of estrogen receptors. (Although an estrogen binding protein has been identified in budding yeast [
51,
52], the protein demonstrated negligible binding of tamoxifen [
51].) The efficacy of tamoxifen alone (Fig.
2b) or paired with lovastatin (Fig.
2c, d) therefore likely results from activation of one or more of tamoxifen’s several known [
47,
53‐
58], or as yet undiscovered, off-target pathways (reviewed in [
53]).
Of the three human breast cancer cell lines assayed, MCF-7 – which possesses estrogen receptors – demonstrated strong synergy with lovastatin (Fig.
7c), in a pattern resembling that seen with yeast (Fig.
2d). In the other two breast cancer cell lines tested (see Fig.
7f, i), tamoxifen had only scattered, weak synergy or was strongly synergistic with only the highest concentration of lovastatin. Together, these observations frame a paradox: the combination of tamoxifen plus lovastatin was strongly synergistic in an organism,
S. cerevisiae, devoid of estrogen receptors (ERs), yet displayed a similar pattern of strong synergism only in a breast cancer cell line, MCF7, which possesses ERs. (It is important when interpreting our experiments with tamoxifen in either yeast or the MCF7 cell line to remain cognizant of two well-documented observations: first, the parent drug tamoxifen – sometimes misleadingly labeled a “prodrug” [
59,
60] – does
not require its metabolism in order to be pharmacologically active [
59‐
61]; and, second, tamoxifen has well-documented pharmacologic effects independent of the estrogen receptor [
47,
53‐
58,
62].)
Curiously, the EC
50 of lovastatin showed substantially greater variability between breast cancer cell lines than did the EC
50 of tamoxifen. A possible resolution of these seeming contradictions is suggested by the reports of Radin and Patel [
53] and Tan et al. [
54]; higher concentrations (in the micromolar range) of tamoxifen are required to engage ER-independent pathways than for ER-dependent mechanisms. It therefore seems plausible that a statin might lend efficacy to tamoxifen in ER-negative breast cancer in which tamoxifen is otherwise impotent; however, targeted delivery of the two drugs would perhaps be required to achieve the requisite concentration ratios of the two drugs [
63].
We sought further proof of principle in support of our
S. cerevisiae-based experimental design by testing the combination of cisplatin and lovastatin in three human cell lines: A549 (lung adenocarcinoma); HT29 (colon adenocarcinoma); and MCF7 (breast carcinoma). This combination was chosen as a surrogate for the six neutral or antagonistic pairings because it so commonly prescribed for a variety of malignancies. In the yeast model, the combination exhibited strong antagonism (Table
2; Fig.
1a, d; Fig.
6d; Additional file
2: Fig. S6d). Congruent with our
S. cerevisiae data, the metrics for the three cell lines were antagonistic, even more so than those calculated from the yeast assays (Table
2), lending further support to the validity of our model.
Two features of the experimental design merit further discussion. Because our intent was to compare the ten statin-chemotherapeutic drug pairs with each other, an important consideration was that the cellular target be a neutral one, free of the biologic and genetic biases inherent in every cancer cell line; choosing any one human cancer cell line for the assays risked biasing the assays for or against one or another drug pair. In the Background we set forth our rationale for choosing the model organism
Saccharomyces cerevisiae. Although the utility of budding yeast in drug studies may be unfamiliar to some, there is ample precent for the choice. For example, three large chemogenomic studies [
32,
34,
36] have been performed; these are consolidated in the searchable data base NetwoRx [
33], which includes 466 drugs and compounds. (This data base, as well as other published literature, guided our selection of drug concentrations for the
S. cerevisiae studies.) The balanced pool of heterozygous deletion strains of all essential genes, which we created in order to have a cellular substrate with consistent genetic diversity, proved to be only modestly more sensitive to the various drugs than the cognate wildtype strain. However, because each deletion is bar-coded, the pool provides a useful resource for the analysis of genetic targets and resistance mechanisms of drug treatments, studies beyond the scope of this report.
A second important consideration in experimental design was the choice of synergy model. The Combenefit software package [
45] renders three models (Bliss, HSA and Loewe). The Bliss and Loewe model are arguably the most popular synergy models, but all synergy models have inherent flaws [
49,
64,
65]. The probabilistic Bliss model assumes independent but competing drug actions whereas the Loewe additivity model assumes nonindependence; that is, the two drugs may interact with the same targets or pathways [
49]. Because of the remarkable pleiotropy of statins (reviewed in detail in [
66]), including interactions with a variety of signaling pathways, we posited nonindependence of lovastatin and the individual chemotherapeutic drugs and therefore chose the Loewe model as more appropriate.
Undue reliance upon global metrics, as for example in high throughput studies, risks overlooking potentially useful synergistic interactions which are confined to portions of the concentration space, often with coexisting antagonism (see Figs.
4d,e and
5c, f). Lehar et al. [
65] posit that such patterns of dose-response surfaces of drug combinations are the consequence of drug-induced interacting pathways, both “on-target” and “off-target” [
67,
68]. Lovastatin in combination with rapamycin proffers a case in point: lower concentrations of rapamycin interact synergistically with higher concentrations of lovastatin yet the converse yields significant antagonism (Fig.
5c, f). A comparison of the interactions of doxorubicin versus methotrexate with lovastatin further illustrate pitfalls of global metrics. The SUM_SYN_ANT score for doxorubicin is higher (+ 5.43) than for methotrexate (+ 2.93). However, the synergistic interactions for doxorubicin occur only with the highest two concentrations tested (Fig.
3d), a clinically problematic pattern. By contrast, the synergistic interactions of methotrexate are found with lower concentrations of both drugs (Fig.
4d, e). This suggests that adding a statin might increase the efficacy of methotrexate while allowing a reduction in its dosage, with attendant mitigation of methotrexate’s substantial toxicity. These observations lead to a sobering conclusion: the therapeutic consequences – be they advantageous or detrimental – of a given statin/chemotherapeutic drug combination may hinge upon the concentrations of each achieved at the tumor site. Consequently, targeted delivery strategies [
63] with precise control of concentration ratios of the two drugs [
69] may merit consideration.
In considering the Loewe synergy scores for the various drug interactions, we concur with the view “that any synergy model should be treated as an exploratory ranking statistic for prioritization of the most potent combinations for further evaluation…” [
49]. To that end, we identified four chemotherapeutic drugs – tamoxifen, doxorubicin, methotrexate and rapamycin – which show strong synergistic interactions with lovastatin and merit further investigation. However, we stress that the data presented in this report should not be taken as - nor was our experimental approach intended to generate - prescriptive guidance for the clinical application of statins as adjuvants to conventional chemotherapeutic agents. That said, our results do identify statin/chemotherapeutic drug combinations warranting further study in cell lines, co-cultures, organoids and animal experiments.
In the Background we stated that an objective of our study was to illuminate the confounding literature bearing on the adjunctive role of statins in chemotherapy and cited the systemic reviews and meta-analyses by Farooqui et al. [
24] and Mei et al. [
23]. Farooqui et al. demonstrated that the addition of a statin to conventional therapy failed to improve progression-free or overall survival. Of the ten studies included in their analysis, six incorporated - either as the sole chemotherapeutic drug or as a component of a multi-drug regimen - agents which we found to be either neutral or antagonistic: etoposide, cisplatin, gemcitabine, and 5-FU (one protocol included both cisplatin and epirubicin, which is related to doxorubicin). Of the remaining four studies, one specified whole brain irradiation, and three incorporated drugs (afatinib, thalidomide and gefitinib) which we have not assayed. Thus, our data and the Farooqi meta-analysis provide mutually supportive, albeit circumstantial, evidence affirming the lack of efficacy of at least four specific statin plus chemotherapeutic drug combinations. Unfortunately, the larger systemic review by Mei et al. [
23], which demonstrated beneficial effects of statins upon survival in cancer patients, did not specify the individual chemotherapeutic drugs used in the 95 studies included in their analysis.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.