Quis custodiet ipsos custodies: who watches the watchmen?

Mitochondrial dysfunction in breast cancer cells prevents tumor growth: understanding chemoprevention with metformin

Metformin is a well-established diabetes drug that prevents the onset of most types of human cancers in diabetic patients, especially by targeting cancer stem cells. Metformin exerts its protective effects by functioning as a weak "mitochondrial poison," as it acts as a complex I inhibitor and prevents oxidative mitochondrial metabolism (OXPHOS). Thus, mitochondrial metabolism must play an essential role in promoting tumor growth. To determine the functional role of "mitochondrial health" in breast cancer pathogenesis, here we used mitochondrial uncoupling proteins (UCPs) to genetically induce mitochondrial dysfunction in either human breast cancer cells (MDA-MB-231) or cancer-associated fibroblasts (hTERT-BJ1 cells). Our results directly show that all three UCP family members (UCP-1/2/3) induce autophagy and mitochondrial dysfunction in human breast cancer cells, which results in significant reductions in tumor growth. Conversely, induction of mitochondrial dysfunction in cancer-associated fibroblasts has just the opposite effect. More specifically, overexpression of UCP-1 in stromal fibroblasts increases β-oxidation, ketone body production and the release of ATP-rich vesicles, which "fuels" tumor growth by providing high-energy nutrients in a paracrine fashion to epithelial cancer cells. Hence, the effects of mitochondrial dysfunction are truly compartment-specific. Thus, we conclude that the beneficial anticancer effects of mitochondrial inhibitors (such as metformin) may be attributed to the induction of mitochondrial dysfunction in the epithelial cancer cell compartment. Our studies identify cancer cell mitochondria as a clear target for drug discovery and for novel therapeutic interventions.

Caveolin-1 and accelerated host aging in the breast tumor microenvironment: chemoprevention with rapamycin, an mTOR inhibitor and anti-aging drug

Increasing chronological age is the most significant risk factor for human cancer development. To examine the effects of host aging on mammary tumor growth, we used caveolin (Cav)-1 knockout mice as a bona fide model of accelerated host aging. Mammary tumor cells were orthotopically implanted into these distinct microenvironments (Cav-1(+/+) versus Cav-1(-/-) age-matched young female mice). Mammary tumors grown in a Cav-1-deficient tumor microenvironment have an increased stromal content, with vimentin-positive myofibroblasts (a marker associated with oxidative stress) that are also positive for S6-kinase activation (a marker associated with aging). Mammary tumors grown in a Cav-1-deficient tumor microenvironment were more than fivefold larger than tumors grown in a wild-type microenvironment. Thus, a Cav-1-deficient tumor microenvironment provides a fertile soil for breast cancer tumor growth. Interestingly, the mammary tumor-promoting effects of a Cav-1-deficient microenvironment were estrogen and progesterone independent. In this context, chemoprevention was achieved by using the mammalian target of rapamycin (mTOR) inhibitor and anti-aging drug, rapamycin. Systemic rapamycin treatment of mammary tumors grown in a Cav-1-deficient microenvironment significantly inhibited their tumor growth, decreased their stromal content, and reduced the levels of both vimentin and phospho-S6 in Cav-1-deficient cancer-associated fibroblasts. Since stromal loss of Cav-1 is a marker of a lethal tumor microenvironment in breast tumors, these high-risk patients might benefit from treatment with mTOR inhibitors, such as rapamycin or other rapamycin-related compounds (rapalogues).

Overexpression of nuclear NHERF1 in advanced colorectal cancer: association with hypoxic microenvironment and tumor invasive phenotype

Over 57% of colorectal cancer (CRC) patients have regional or distant spread of their disease at the time of diagnosis. Despite recent advances, there is a compelling need to better characterize prognostic markers for advanced CRC. The present study investigates protein expression of NHERF1, HIF-1α and TWIST1 and their relationship in distant normal mucosa (DNM), tumor (T) and adjacent normal mucosa (ANM), lymph node metastasis (LNM) and liver metastasis (LM), determining their role as potential markers in advanced stages of human CRC. Overexpression of nuclear NHERF1 was shown in 47% of tumors, which exhibited a significant association with poor histological grade (P=0.0346). Nuclear NHERF1 showed a higher expression in T, LNM and LM than both DNM (P<0.0001) and ANM (P<0.05). Nuclear HIF-1α was significantly higher in T, LNM and LM than DNM and ANM (P<0.05, P<0.001, P<0.0001, respectively). A positive correlation between nuclear NHERF1 and nuclear HIF-1α was found in LNM (r=0.331, P=0.020), where an extended co-localization of the two proteins was demonstrated. TWIST1 was more expressed in T than DNM and ANM (P<0.0001) and was higher in T than LNM and LM (P<0.0001). Moreover, nuclear NHERF1 was directly correlated to TWIST1 (r=0.339, P=0.015) in T samples, where a high co-expression of the two proteins was demonstrated both in no longer polarized epithelial cells and in invasive mesenchymal elements adjacent to hypoxic and perinecrotic colonic areas. Overall, nuclear NHERF1 expression was associated with poorer differentiation grade and with higher expression both of HIF-1α in lymphatic metastasis and TWIST1 in invasive front of tumor. Our results support the oncogenic role of NHERF1 and promote nuclear NHERF1 as a potential new biomarker of advanced CRC.

Overcoming anoikis--pathways to anchorage-independent growth in cancer

Anoikis (or cell-detachment-induced apoptosis) is a self-defense strategy that organisms use to eliminate 'misplaced' cells, i.e. cells that are in an inappropriate location. Occasionally, detached or misplaced cells can overcome anoikis and survive for a certain period of time in the absence of the correct signals from the extracellular matrix (ECM). If cells are able to adapt to their new environment, then they have probably become anchorage-independent, which is one of the hallmarks of cancer cells. Anoikis resistance and anchorage-independency allow tumor cells to expand and invade adjacent tissues, and to disseminate through the body, giving rise to metastasis. Thus, overcoming anoikis is a crucial step in a series of changes that a tumor cell undergoes during malignant transformation. Tumor cells have developed a variety of strategies to bypass or overcome anoikis. Some strategies consist of adaptive cellular changes that allow the cells to behave as they would in the correct environment, so that induction of anoikis is aborted. Other strategies aim to counteract the negative effects of anoikis induction by hyperactivating survival and proliferative cascades. The recently discovered processes of autophagy and entosis also highlight the contribution of these mechanisms to rendering the cells in a dormant state until they receive a signal initiated at the ECM, thereby circumventing anoikis. In all situations, the final outcome is the ability of the tumor to grow and metastasize. A better understanding of the mechanisms underlying anoikis resistance could help to counteract tumor progression and prevent metastasis formation.

Hyperactivation of oxidative mitochondrial metabolism in epithelial cancer cells in situ: visualizing the therapeutic effects of metformin in tumor tissue

We have recently proposed a new mechanism for explaining energy transfer in cancer metabolism. In this scenario, cancer cells behave as metabolic parasites, by extracting nutrients from normal host cells, such as fibroblasts, via the secretion of hydrogen peroxide as the initial trigger. Oxidative stress in the tumor microenvironment then leads to autophagy-driven catabolism, mitochondrial dys-function, and aerobic glycolysis. This, in turn, produces high-energy nutrients (such as L-lactate, ketones, and glutamine) that drive the anabolic growth of tumor cells, via oxidative mitochondrial metabolism. A logical prediction of this new "parasitic" cancer model is that tumor-associated fibroblasts should show evidence of mitochondrial dys-function (mitophagy and aerobic glycolysis). In contrast, epithelial cancer cells should increase their oxidative mitochondrial capacity. To further test this hypothesis, here we subjected frozen sections from human breast tumors to a staining procedure that only detects functional mitochondria. This method detects the in situ enzymatic activity of cytochrome C oxidase (COX), also known as Complex IV. Remarkably, cancer cells show an over-abundance of COX activity, while adjacent stromal cells remain essentially negative. Adjacent normal ductal epithelial cells also show little or no COX activity, relative to epithelial cancer cells. Thus, oxidative mitochondrial activity is selectively amplified in cancer cells. Although COX activity staining has never been applied to cancer tissues, it could now be used routinely to distinguish cancer cells from normal cells, and to establish negative margins during cancer surgery. Similar results were obtained with NADH activity staining, which measures Complex I activity, and succinate dehydrogenase (SDH) activity staining, which measures Complex II activity. COX and NADH activities were blocked by electron transport inhibitors, such as Metformin. This has mechanistic and clinical implications for using Metformin as an anti-cancer drug, both for cancer therapy and chemo-prevention. We also immuno-stained human breast cancers for a series of well-established protein biomarkers of metabolism. More specifically, we now show that cancer-associated fibroblasts over-express markers of autophagy (cathepsin B), mitophagy (BNIP3L), and aerobic glycolysis (MCT4). Conversely, epithelial cancer cells show the over-expression of a mitochondrial membrane marker (TOMM20), as well as key components of Complex IV (MT-CO1) and Complex II (SDH-B). We also validated our observations using a bioinformatics approach with data from > 2,000 breast cancer patients, which showed the transcriptional upregulation of mitochondrial oxidative phosphorylation (OXPHOS) in human breast tumors (p < 10(-20)), and a specific association with metastasis. Therefore, upregulation of OXPHOS in epithelial tumor cells is a common feature of human breast cancers. In summary, our data provide the first functional in vivo evidence that epithelial cancer cells perform enhanced mitochondrial oxidative phosphorylation, allowing them to produce high amounts of ATP. Thus, we believe that mitochondria are both the "powerhouse" and "Achilles' heel" of cancer cells.

Mitochondrial oxidative stress in cancer-associated fibroblasts drives lactate production, promoting breast cancer tumor growth: understanding the aging and cancer connection

Increasing chronological age is the most significant risk factor for cancer. Recently, we proposed a new paradigm for understanding the role of the aging and the tumor microenvironment in cancer onset. In this model, cancer cells induce oxidative stress in adjacent stromal fibroblasts. This, in turn, causes several changes in the phenotype of the fibroblast including mitochondrial dysfunction, hydrogen peroxide production, and aerobic glycolysis, resulting in high levels of L-lactate production. L-lactate is then transferred from these glycolytic fibroblasts to adjacent epithelial cancer cells and used as "fuel" for oxidative mitochondrial metabolism.  Here, we created a new pre-clinical model system to directly test this hypothesis experimentally. To synthetically generate glycolytic fibroblasts, we genetically-induced mitochondrial dysfunction by knocking down TFAM using an sh-RNA approach.  TFAM is mitochondrial transcription factor A, which is important in functionally maintaining the mitochondrial respiratory chain. Interestingly, TFAM-deficient fibroblasts showed evidence of mitochondrial dysfunction and oxidative stress, with the loss of certain mitochondrial respiratory chain components, and the over-production of hydrogen peroxide and L-lactate. Thus, TFAM-deficient fibroblasts underwent metabolic reprogramming towards aerobic glycolysis.  Most importantly, TFAM-deficient fibroblasts significantly promoted tumor growth, as assayed using a human breast cancer (MDA-MB-231) xenograft model. These increases in glycolytic fibroblast driven tumor growth were independent of tumor angiogenesis. Mechanistically, TFAM-deficient fibroblasts increased the mitochondrial activity of adjacent epithelial cancer cells in a co-culture system, as seen using MitoTracker. Finally, TFAM-deficient fibroblasts also showed a loss of caveolin-1 (Cav-1), a known breast cancer stromal biomarker. Loss of stromal fibroblast Cav-1 is associated with early tumor recurrence, metastasis, and treatment failure, resulting in poor clinical outcome in breast cancer patients. Thus, this new experimental model system, employing glycolytic fibroblasts, may be highly clinically relevant. These studies also have implications for understanding the role of hydrogen peroxide production in oxidative damage and "host cell aging," in providing a permissive metabolic microenvironment for promoting and sustaining tumor growth.

Hydrogen peroxide fuels aging, inflammation, cancer metabolism and metastasis: the seed and soil also needs "fertilizer"

In 1889, Dr. Stephen Paget proposed the "seed and soil" hypothesis, which states that cancer cells (the seeds) need the proper microenvironment (the soil) for them to grow, spread and metastasize systemically. In this hypothesis, Dr. Paget rightfully recognized that the tumor microenvironment has an important role to play in cancer progression and metastasis. In this regard, a series of recent studies have elegantly shown that the production of hydrogen peroxide, by both cancer cells and cancer-associated fibroblasts, may provide the necessary "fertilizer," by driving accelerated aging, DNA damage, inflammation and cancer metabolism, in the tumor microenvironment. By secreting hydrogen peroxide, cancer cells and fibroblasts are mimicking the behavior of immune cells (macrophages/neutrophils), driving local and systemic inflammation, via the innate immune response (NFκB). Thus, we should consider using various therapeutic strategies (such as catalase and/or other anti-oxidants) to neutralize the production of cancer-associated hydrogen peroxide, thereby preventing tumor-stroma co-evolution and metastasis. The implications of these findings for overcoming chemo-resistance in cancer cells are also discussed in the context of hydrogen peroxide production and cancer metabolism.

Cancer cells metabolically "fertilize" the tumor microenvironment with hydrogen peroxide, driving the Warburg effect: implications for PET imaging of human tumors

Previously, we proposed that cancer cells behave as metabolic parasites, as they use targeted oxidative stress as a "weapon" to extract recycled nutrients from adjacent stromal cells. Oxidative stress in cancer-associated fibroblasts triggers autophagy and  mitophagy, resulting in compartmentalized cellular catabolism, loss of mitochondrial function, and the onset of aerobic glycolysis, in the tumor stroma. As such, cancer-associated fibroblasts produce high-energy nutrients (such as lactate and ketones) that fuel mitochondrial biogenesis, and oxidative metabolism in cancer cells. We have termed this new energy-transfer mechanism the "reverse Warburg effect." To further test the validity of this hypothesis, here we used an in vitro MCF7-fibroblast co-culture system, and quantitatively measured a variety of metabolic parameters by FACS analysis (analogous to laser-capture micro-dissection).  Mitochondrial activity, glucose uptake, and ROS production were measured with highly-sensitive fluorescent probes (MitoTracker, NBD-2-deoxy-glucose, and DCF-DA). Interestingly, using this approach, we directly show that cancer cells initially secrete hydrogen peroxide that then triggers oxidative stress in neighboring fibroblasts. Thus, oxidative stress is contagious (spreads like a virus) and is propagated laterally and vectorially from cancer cells to adjacent fibroblasts. Experimentally, we show that oxidative stress in cancer-associated fibroblasts quantitatively reduces mitochondrial activity, and increases glucose uptake, as the fibroblasts become more dependent on aerobic glycolysis.  Conversely, co-cultured cancer cells show significant increases in mitochondrial activity, and corresponding reductions in both glucose uptake and GLUT1 expression. Pre-treatment of co-cultures with extracellular catalase (an anti-oxidant enzyme that detoxifies hydrogen peroxide) blocks the onset of oxidative stress, and potently induces the death of cancer cells, likely via starvation.  Given that cancer-associated fibroblasts show the largest increases in glucose uptake, we suggest that PET imaging of human tumors, with Fluoro-2-deoxy-D-glucose (F-2-DG), may be specifically detecting the tumor stroma, rather than epithelial cancer cells.

The autophagic tumor stroma model of cancer: Role of oxidative stress and ketone production in fueling tumor cell metabolism

A loss of stromal Cav-1 in the tumor fibroblast compartment is associated with early tumor recurrence, lymph-node metastasis, and tamoxifen-resistance, resulting in poor clinical outcome in breast cancer patients. Here, we have used Cav-1 (-/-) null mice as a pre-clinical model for this "lethal tumor micro-environment." Metabolic profiling of Cav-1 (-/-) mammary fat pads revealed the upregulation of numerous metabolites (nearly 100), indicative of a major catabolic phenotype. Our results are consistent with the induction of oxidative stress, mitochondrial dysfunction, and autophagy/mitophagy. The two most prominent metabolites that emerged from this analysis were ADMA (asymmetric dimethyl arginine) and BHB (beta-hydroxybutyrate; a ketone body), which are markers of oxidative stress and mitochondrial dysfunction, respectively. Transcriptional profiling of Cav-1 (-/-) stromal cells and human tumor stroma from breast cancer patients directly supported an association with oxidative stress, mitochondrial dysfunction, and autophagy/mitophagy, as well as ADMA and ketone production. MircoRNA profiling of Cav-1 (-/-) stromal cells revealed the upregulation of two key cancer-related miR's, namely miR-31 and miR-34c. Consistent with our metabolic findings, these miR's are associated with oxidative stress (miR-34c) or activation of the hypoxic response/HIF1a (miR-31), which is sufficient to drive authophagy/mitophagy. Thus, via an unbiased comprehensive analysis of a lethal tumor micro-environment, we have identified a number of candidate biomarkers (ADMA, ketones, and miR-31/34c) that could be used to identify high-risk cancer patients at diagnosis, for treatment stratification and/or for evaluating therapeutic efficacy during anti-cancer therapy. We propose that the levels of these key biomarkers (ADMA, ketones/BHB, miR-31, and miR-34c) could be (1) assayed using serum or plasma from cancer patients, or (2) performed directly on excised tumor tissue. Importantly, induction of oxidative stress and autophagy/mitophagy in the tumor stromal compartment provides a means by which epithelial cancer cells can directly "feed off" of stromal-derived essential nutrients, chemical building blocks (amino acids, nucleotides), and energy-rich metabolites (glutamine, pyruvate, ketones/BHB), driving tumor progression and metastasis. Essentially, aggressive cancer cells are "eating" the cancer-associated fibroblasts via autophagy/mitophagy in the tumor micro-environment. Lastly, we discuss that this "Autophagic Tumor Stroma Model of Cancer Metabolism" provides a viable solution to the "Autophagy Paradox" in cancer etiology and chemo-therapy.

The autophagic tumor stroma model of cancer or "battery-operated tumor growth": A simple solution to the autophagy paradox

The role of autophagy in tumorigenesis is controversial. Both autophagy inhibitors (chloroquine) and autophagy promoters (rapamycin) block tumorigenesis by unknown mechanism(s). This is called the "Autophagy Paradox". We have recently reported a simple solution to this paradox. We demonstrated that epithelial cancer cells use oxidative stress to induce autophagy in the tumor microenvironment. As a consequence, the autophagic tumor stroma generates recycled nutrients that can then be used as chemical building blocks by anabolic epithelial cancer cells. This model results in a net energy transfer from the tumor stroma to epithelial cancer cells (an energy imbalance), thereby promoting tumor growth. This net energy transfer is both unilateral and vectorial, from the tumor stroma to the epithelial cancer cells, representing a true host-parasite relationship. We have termed this new paradigm "The Autophagic Tumor Stroma Model of Cancer Cell Metabolism" or "Battery-Operated Tumor Growth". In this sense, autophagy in the tumor stroma serves as a "battery" to fuel tumor growth, progression and metastasis, independently of angiogenesis. Using this model, the systemic induction of autophagy will prevent epithelial cancer cells from using recycled nutrients, while the systemic inhibiton of autophagy will prevent stromal cells from producing recycled nutrients-both effectively "starving" cancer cells. We discuss the idea that tumor cells could become resistant to the systemic induction of autophagy, by the upregulation of natural endogenous autophagy inhibitors in cancer cells. Alternatively, tumor cells could also become resistant to the systemic induction of autophagy, by the genetic silencing/deletion of pro-autophagic molecules, such as Beclin1. If autophagy resistance develops in cancer cells, then the systemic inhibition of autophagy would provide a therapeutic solution to this type of drug resistance, as it would still target autophagy in the tumor stroma. As such, an anti-cancer therapy that combines the alternating use of both autophagy promoters and autophagy inhibitors would be expected to prevent the onset of drug resistance. We also discuss why anti-angiogenic therapy has been found to promote tumor recurrence, progression and metastasis. More specifically, anti-angiogenic therapy would induce autophagy in the tumor stroma via the induction of stromal hypoxia, thereby converting a non-aggressive tumor type to a "lethal" aggressive tumor phenotype. Thus, uncoupling the metabolic parasitic relationship between cancer cells and an autophagic tumor stroma may hold great promise for anti-cancer therapy. Finally, we believe that autophagy in the tumor stroma is the local microscopic counterpart of systemic wasting (cancer-associated cachexia), which is associated with advanced and metastatic cancers. Cachexia in cancer patients is not due to decreased energy intake, but instead involves an increased basal metabolic rate and increased energy expenditures, resulting in a negative energy balance. Importantly, when tumors were surgically excised, this increased metabolic rate returned to normal levels. This view of cachexia, resulting in energy transfer to the tumor, is consistent with our hypothesis. So, cancer-associated cachexia may start locally as stromal autophagy, and then spread systemically. As such, stromal autophagy may be the requisite precursor of systemic cancer-associated cachexia.

HIF1-alpha functions as a tumor promoter in cancer associated fibroblasts, and as a tumor suppressor in breast cancer cells: Autophagy drives compartment-specific oncogenesis

Our recent studies have mechanistically implicated a loss of stromal Cav-1 expression and HIF1-alpha-activation in driving the cancer-associated fibroblast phenotype, through the paracrine production of nutrients via autophagy and aerobic glycolysis. However, it remains unknown if HIF1a-activation is sufficient to confer the cancer-associated fibroblast phenotype. To test this hypothesis directly, we stably-expressed activated HIF1a in fibroblasts and then examined their ability to promote tumor growth using a xenograft model employing human breast cancer cells (MDA-MB-231). Fibroblasts harboring activated HIF1a showed a dramatic reduction in Cav-1 levels and a shift towards aerobic glycolysis, as evidenced by a loss of mitochondrial activity, and an increase in lactate production. Activated HIF1a also induced BNIP3 and BNIP3L expression, markers for the autophagic destruction of mitochondria. Most importantly, fibroblasts expressing activated HIF1a increased tumor mass by ∼2-fold and tumor volume by ∼3-fold, without a significant increase in tumor angiogenesis. In this context, HIF1a also induced an increase in the lymph node metastasis of cancer cells. Similar results were obtained by driving NFκB activation in fibroblasts, another inducer of autophagy. Thus, activated HIF1a is sufficient to functionally confer the cancer-associated fibroblast phenotype. It is also known that HIF1a expression is required for the induction of autophagy in cancer cells. As such, we next directly expressed activated HIF1a in MDA-MB-231 cells and assessed its effect on tumor growth via xenograft analysis. Surprisingly, activated HIF1a in cancer cells dramatically suppressed tumor growth, resulting in a 2-fold reduction in tumor mass and a 3-fold reduction in tumor volume. We conclude that HIF1a activation in different cell types can either promote or repress tumorigenesis. Based on these studies, we suggest that autophagy in cancer-associated fibroblasts promotes tumor growth via the paracrine production of recycled nutrients, which can directly "feed" cancer cells. Conversely, autophagy in cancer cells represses tumor growth via their "self-digestion". Thus, we should consider that the activities of various known oncogenes and tumor-suppressors may be compartment and cell-type specific, and are not necessarily an intrinsic property of the molecule itself. As such, other "classic" oncogenes and tumor suppressors will have to be re-evaluated to determine their compartment specific effects on tumor growth and metastasis. Lastly, our results provide direct experimental support for the recently proposed "Autophagic Tumor Stroma Model of Cancer".

Understanding the causes of multidrug resistance in cancer: a comparison of doxorubicin and sunitinib

Multiple molecular, cellular, micro-environmental and systemic causes of anticancer drug resistance have been identified during the last 25 years. At the same time, genome-wide analysis of human tumor tissues has made it possible in principle to assess the expression of critical genes or mutations that determine the response of an individual patient's tumor to drug treatment. Why then do we, with a few exceptions, such as mutation analysis of the EGFR to guide the use of EGFR inhibitors, have no predictive tests to assess a patient's drug sensitivity profile. The problem urges the more with the expanding choice of drugs, which may be beneficial for a fraction of patients only. In this review we discuss recent studies and insights on mechanisms of anticancer drug resistance and try to answer the question: do we understand why a patient responds or fails to respond to therapy? We focus on doxorubicin as example of a classical cytotoxic, DNA damaging agent and on sunitinib, as example of the new generation of (receptor) tyrosine kinase-targeted agents. For both drugs, classical tumor cell autonomous resistance mechanisms, such as drug efflux transporters and mutations in the tumor cell's survival signaling pathways, as well as micro-environment-related resistance mechanisms, such as changes in tumor stromal cell composition, matrix proteins, vascularity, oxygenation and energy metabolism may play a role. Novel agents that target specific mutations in the tumor cell's damage repair (e.g. PARP inhibitors) or that target tumor survival pathways, such as Akt inhibitors, glycolysis inhibitors or mTOR inhibitors, are of high interest. In order to increase the therapeutic index of treatments, fine-tuned synergistic combinations of new and/or classical cytotoxic agents will be designed. More quantitative assessment of potential resistance mechanisms in real tumors and in real time, such as by kinase profiling methodology, will be developed to allow more precise prediction of the optimal drug combination to treat each patient.