Understanding the Warburg effect and the prognostic value of stromal caveolin-1 as a marker of a lethal tumor microenvironment

Cancer cells show a broad spectrum of bioenergetic states, with some cells using aerobic glycolysis while others rely on oxidative phosphorylation as their main source of energy. In addition, there is mounting evidence that metabolic coupling occurs in aggressive tumors, between epithelial cancer cells and the stromal compartment, and between well-oxygenated and hypoxic compartments. We recently showed that oxidative stress in the tumor stroma, due to aerobic glycolysis and mitochondrial dysfunction, is important for cancer cell mutagenesis and tumor progression. More specifically , increased autophagy/mitophagy in the tumor stroma drives a form of parasitic epithelial-stromal metabolic coupling. These findings explain why it is effective to treat tumors with either inducers or inhibitors of autophagy, as both would disrupt this energetic coupling. We also discuss evidence that glutamine addiction in cancer cells produces ammonia via oxidative mitochondrial metabolism. Ammonia production in cancer cells, in turn, could then help maintain autophagy in the tumor stromal compartment. In this vicious cycle, the initial glutamine provided to cancer cells would be produced by autophagy in the tumor stroma. Thus, we believe that parasitic epithelial-stromal metabolic coupling has important implications for cancer diagnosis and therapy, for example, in designing novel metabolic imaging techniques and establishing new targeted therapies. In direct support of this notion, we identified a loss of stromal caveolin-1 as a marker of oxidative stress, hypoxia, and autophagy in the tumor microenvironment, explaining its powerful predictive value. Loss of stromal caveolin-1 in breast cancers is associated with early tumor recurrence, metastasis, and drug resistance, leading to poor clinical outcome.

and in fact it is now clear that cancer cells utilize both glycolysis and oxidative phosphorylation to satisfy their metabolic needs. Experimental assessments of ATP produc tion in cancer cells have demonstrated that oxidative pathways play a signifi cant role in energy generation, and may be responsible for about 50 to 80% of the ATP generated [5][6][7][8]. Also, it should be considered that most studies were performed using isolated cancer cells, which may behave very diff erently from cancer cells in vivo, surroun ded by their natural microenvironment [9] (see also Koukourakis and colleagues [10] for another viewpoint).
Second, several studies now clearly indicate that mitochondrial activity and oxidative phosphorylation support tumor growth. Loss-of-function mutations in the TCA cycle gene IDH1 (isocitrate dehydrogenase 1) are found in about 70% of gliomas, but, interestingly, correlate with a better prognosis and improved survival, suggesting that severely decreased activity in one of the TCA cycle enzymes does not favor tumor aggressiveness [11]. Th e mitochondrial protein p32 was shown to maintain high levels of oxidative phosphorylation in human cancer cells and to sustain tumorigenicity in vivo [12]. In addition, STAT3 is known to enhance tumor growth and to predict poor prognosis in human cancers [13]. Interestingly, a pool of STAT3 localizes to the mitochondria, to sustain high levels of mitochondrial respiration [14] and to augment transformation by oncogenic Ras [15,16]. Similarly, the mitochondrial transcription factor A (TFAM), which is required for mitochondrial DNA replication and oxidative phosphorylation, is also required for K-Rasinduced lung tumorigenesis [17]. Finally, when constrained to use glycolysis by depletion of mitochondrial DNA, melanoma B16 cells (B16ρ 0 ) and breast cancer T47D cells (T47Dρ 0 ), show severe impairment of tumorigenicity in vivo [18,19].
Th ere is also evidence that pro-oncogenic molecules regulate mitochondrial function. Cyclin D1 inhibits mitochondrial function in breast cancer cells [20]. Overexpression of cyclin D1 is observed in about 50% of invasive breast cancers and is associated with a good clinical outcome [21], indicating that inhibition of mitochondrial activity correlates with favorable prognosis. Importantly, it was shown that the oncogene c-Myc stimulates mitochondrial biogenesis, and enhances glutamine metabolism by regulating the expression of mitochondrial glutaminase, the fi rst enzyme in the glutamine utilization pathway [22]. Glutamine is an essential metabolic fuel that is converted to alpha-ketoglutarate and serves as a substrate for the TCA cycle or for glutathione synthesis, to promote energy production and cellular biosynthesis, and to protect against oxidative stress [23]. Interestingly, pharmacological targeting of mito chondrial glutaminase inhibits cancer cell transform ing activity, suggesting that glutamine metabolism and its role in fueling and replenishing the TCA cycle are required for neoplastic transformation [24] (for a recent review on glutamine cancer metabolism, see [2]).

The reverse Warburg eff ect
It is increasingly apparent that the tumor microenvironment regulates neoplastic growth and progression. Activation of the stroma is a critical step required for tumor formation. Among the stromal players, cancer associated fi broblasts (CAFs) have recently taken center stage [25]. CAFs are activated, contractile fi broblasts that display features of myo-fi broblasts, express musclespecifi c actin, and show an increased ability to secrete and remodel the extracellular matrix [26,27]. Th ey are not just neutral spectators, but actively support malignant transformation [28] and metastasis [29], as compared to normal resting fi broblasts. For example, in a humanized mouse model of breast cancer, oncogenically driven human organoids develop tumors only if coinjected with immortalized fi broblasts and not with normal primary fi broblasts, suggesting that stromal activa tion is required for breast cancer formation [30].
Importantly, the tumor stroma dictates clinical outcome and constitutes a source of potential biomarkers [31]. Expression profi ling has identifi ed a cancer-associated stromal signature that predicts good and poor clinical prognosis in breast cancer patients, independently of other factors [32][33][34].
We and others have recently shown that a loss of caveolin-1 (Cav-1) in the stromal compartment is a novel biomarker for predicting poor clinical outcome in all of the most common subtypes of human breast cancer, including the more lethal triple negative subtype [35,36]. A loss of stromal Cav-1 predicts early tumor recurrence, lymphnode metastasis, tamoxifen-resistance, and poor survival. Overall, breast cancer patients with a loss of stromal Cav-1 show a 20% 5-year survival rate, compared to the 80% 5-year survival of patients with high stromal Cav-1 expression. In triple negative patients, the 5-year survival rate is 75.5% for high stromal Cav-1 versus 9.4% for absent stromal Cav-1 [37]. A loss of stromal Cav-1 also predicts progression to invasive disease in ductal carcinoma in situ patients, suggesting that a loss of Cav-1 regulates tumor progression [35]. Similarly, a loss of stromal Cav-1 is associated with advanced disease and metastasis, as well as a high Gleason score, in prostate cancer patients [38].
In order to mechanistically dissect how a loss of stromal Cav-1 induces a lethal microenvironment, we have employed Cav-1(-/-) null mice as a model system. Complementary studies were also performed with a coculture model of normal human fi broblasts and MCF7 breast cancer cells.
Unbiased proteomic and transcriptional analysis has shown that Cav-1(-/-) null bone-marrow-derived stromal cells display the up-regulation of both myo-fi broblast markers and glycolytic enzymes, clearly indicating that a loss of Cav-1 promotes myo-fi broblast conversion and induces aerobic glycolysis in stromal cells [39]. Th us, stromal cells lacking Cav-1 undergo aerobic glycolysis and secrete energy-rich metabolites (such as lactate and pyruvate) that directly feed cancer cells and fuel mitochondrial respiration of adjacent cancer cells. We have termed this novel hypothesis the 'reverse Warburg eff ect' . Th is novel idea has now also been independently confi rmed by Pol and colleagues [40], who demonstrated that genetic ablation of Cav-1 in murine fi broblasts is indeed suffi cient to functionally induce the onset of aerobic glycolysis via mitochondrial dysfunction [40]. Interestingly, immunohistochemistry on human breast cancers lacking stromal Cav-1 has demonstrated the over-expression of key glycolytic enzymes, such as PKM2 and LDH, in the fi broblastic tumor stroma [39].
In further support of this hypothesis, co-culture with cancer cells promotes Cav-1 down-regulation in adjacent fi broblasts. Importantly, when cultured under homotypic conditions, MCF7 cancer cells have a very low mito chondrial mass (the conventional Warburg eff ect). However, co-culture with fi broblasts, which more closely mirrors the microenvironment of a naturally occurring tumor, promotes a very signifi cant increase in mitochondrial mass in MCF7 cancer cells, suggesting that the Warburg eff ect might be an in vitro artifact [41]. Importantly, lactate administration to homotypic MCF7 cancer cell cultures signifi cantly increases mitochondrial mass, suggesting that lactate administration phenocopies the presence of reactive fi broblasts [41], by promoting mitochondrial biogenesis. Th ese new data indicate that cancer cells and CAFs develop a 'symbiotic' or 'parasitic' relationship, with the vectorial and unilateral transfer of energy from glycolytic stromal cells to oxidative cancer cells (Figures 1 and 2).
Clinically, high glucose uptake has been exploited to monitor tumor growth using position emission tomography (PET) scanning via radiolabeled 2-deoxy-glucose. However, we should acknowledge that PET avidity does not necessarily correlate with high aerobic glycolysis. For example, PET is not useful in clear cell renal carcinomas [42,43], which are the tumor 'prototype' in which the Warburg eff ect should sustain tumor formation. About half of clear cell renal carcinomas are due to mutations in the Von Hippel-Lindau (VHL) gene [44,45], leading to HIF1α constitutive activation and forced induction of aerobic glycolysis [46][47][48]. Conversely, little attention is paid to understanding which cell-type or compartments within a tumor are most PET avid. In fact, the clinical use of PET is well established in Hodgkin's lymphomas [49], which are composed of less than 10% tumor cells, the rest being stromal and infl ammatory cells [50]. Yet, Hodgkin's lymphomas are very PET avid tumors, suggesting that 2-deoxy-glucose uptake may be associated with the tumor stroma. Th at the fi brotic component may be glucose avid is further supported by the notion that PET is clinically used to assess the therapeutic response in gastrointestinal stromal tumors (GIST), which are a subset of tumors of mesenchymal origin. Finally, pilot clinical studies have shown that PET is useful for detection of various forms of fi brosis, including pulmonary fi brosis [51].
Th e reverse Warburg eff ect can also be described as 'metabolic coupling' between supporting glycolytic stromal cells and oxidative tumor cells. Metabolic cooperativity between adjacent cell-compartments is observed in several normal physiological settings. During folliculogenesis and early embryogenesis, oocytes are surrounded by supporting cumulus granulosa (CG) cells. Oocytes lack some crucial metabolic functions, such as the ability to utilize glucose, to produce cholesterol and to transport certain amino acids. Th us, oocytes control CG cell metabolism, promoting their glycolysis, cholesterol synthesis and amino acid uptake [52,53]. Remarkably, both oocytes and zygotes can reach a two-cell stage only in the presence of pyruvate and oxalo acetate, but not using glucose [54,55]. However, upon co-culture with CG cells, oocytes or zygotes can cleave to a two-cell stage also in the presence of glucose, clearly indicating that CG cells metabolize glucose and supply the oocyte with the . This activates reactive oxygen species (ROS) production and autophagy. ROS production in CAFs, via the bystander eff ect, serves to induce random mutagenesis in epithelial cancer cells, leading to double-strand DNA breaks and aneuploidy. Cancer cells mount an anti-oxidant defense and upregulate molecules that protect them against ROS and autophagy, preventing them from undergoing apoptosis. So, stromal fi broblasts conveniently feed and mutagenize cancer cells, while protecting them against death. See the text for more details. A+, autophagy positive; A-, autophagy negative; A R , autophagy resistant.

A+
A-or A R

Cancer Cells
glycolytic product pyruvate to permit maturation [54]. In this regard, transcriptional profi ling of CG cells and oocytes has revealed that key enzymes of the glycolytic pathway (aldolase A, enolase 1, LDHA, and PKM2) are very highly expressed in CG cells, but are undetectable in oocytes [56]. Th ese results indicate that granulosa cells support the development and maturation of oocytes and early embryos by providing them with essential nutrients, such as oxidative phosphorylation substrates (pyruvate and lactate [57]), amino acids [58], and cholesterol [59]. Metabolic coupling is also normally observed in the brain between astrocytes and neurons, and this is known as 'neuron-glia metabolic coupling' [60,61]. In this regard, glycolytic astrocytes generate high levels of lactate to support mitochondrial oxidative phosphorylation in adjacent neurons. Consistent with this idea, LDH is preferentially expressed in astrocytes, and the mitochondrial enzyme pyruvate dehydrogenase is pre sent selectively within the neurons [62,63]. Also, in skeletal muscle it is well established that glycolytic fast-twitch muscle fi bers produce lactate, which is then secreted via monocarboxylate transporter (MCT) 4. Lactate is then taken up by the slow-twitch fi bers expressing MCT1 and utilized as a substrate for oxidative phosphorylation [64,65]. Th is form of metabolic coupling is known as the 'skeletal muscle lactate shuttle' [66].
As such, it is perhaps not surprising that tumors may have also developed a form of metabolic coupling, specifi cally involving tumor-stromal interactions. In further support of the existence of a 'lactate shuttle' in human tumors, we have now shown that CAFs express MCT4 (for lactate extrusion), while breast cancer cells express MCT1 (for lactate uptake) (Figure 3) [67]. Interestingly, MCT4 expression in CAFs is induced by oxidative stress, and MCT4 is a known HIF1α target gene (see Discussion below) [67]. . This leads to the onset of both autophagy and mitophagy, as well as aerobic glycolysis, which then produces recycled nutrients (such as lactate, ketones, and glutamine). These high-energy chemical building blocks can then be transferred and used as fuel in the tricarboxylic acid cycle (TCA) in adjacent cancer cells. The outcome is high ATP production in cancer cells, and protection against cell death. ROS, reactive oxygen species. (b) Homotypic cultures (upper panels) of MCF7 cells (right) and hTERT-fi broblasts (left) were immunostained with a mitochondrial membrane antibody (red). Note that mitochondrial mass is lower in mono-cultures of MCF7 cells compared to fi broblasts. However, co-culture of MCF7 cells with fi broblasts (lower panel) induces a dramatic increase in mitochondrial mass in the 'central MCF7 cell colony' , outlined by the dotted white oval. In contrast, mitochondrial mass is decreased in co-cultured fi broblasts. Panel (b) was modifi ed and reproduced with permission from [41,78].

Oxidative stress: random mutagenesis and protection against cell death
Recent studies have dissected the mechanism(s) by which a loss of stromal Cav-1 leads to an aggressive breast cancer phenotype, and have shown that oxidative stress plays a central role. Th e role of oxidative stress in sustaining tumor growth is underscored by the observation that reactive oxygen species (ROS)-mediated myo-fi broblast conversion is suffi cient to reduce tumor-free survival, and increases metastatic potential in a mammary tumor mouse model [68].
Gene expression profi ling of Cav-1(-/-) null bonemarrow-derived stromal cells has shown the upregulation of gene transcripts associated with ROS produc tion, and over-expression of the transcriptional targets of HIF1α and NFκB, suggesting that a loss of stromal Cav-1 induces oxidative stress, mimics hypoxia, and stimulates infl ammation [41].
Co-cu ltures of normal human fi broblasts and MCF7 cells indicate that cancer cells use oxidative stress as a weapon to trigger the conversion of adjacent fi broblasts into myo-fi broblasts [69]. Cancer cell-induced oxidative

CAFs Cancer Cells
Lactate Ketones stress potently perturbs the behavior of adjacent fi broblasts, induces the lysosomal-mediated degradation of Cav-1, and promotes mitochondrial dysfunction, resulting in increased aerobic glycolysis [41]. In turn, these glycolytic fi broblasts support tumor cell mitochondrial respiration and growth by actively transferring highenergy nutrients (such as lactate and pyruvate) to cancer cells.
In support of the 'reverse Warburg eff ect' , comparison of the tumor-promoting properties of two fi broblast lines (named CL3 and CL4) with either mitochondrial or glycolytic metabolism has revealed that aerobic glycolysis in CAFs greatly promotes tumor formation. CL3 fi broblasts show oxidative metabolism and increased mitochon drial mass, whereas CL4 fi broblasts display a shift towards aerobic glycolysis and increased lactate production [70]. Interestingly, in a xenograft model, CL4 fi broblasts enhance the growth of mammary tumors by approxi mately eight-fold compared to CL3 cells, without a detectable increase in angiogenesis. Consistent with this growth-promoting eff ect, CL4 fi broblasts also increase the mitochondrial mass of co-cultured breast cancer cells [71].
An oxidative-stress-rich micro-environment generates DNA damage in both cancer and stromal cells (Figure 1). We have shown that MCF7 cancer cells induce oxidative stress and promote DNA double-strand breaks in associated stromal cells, which are blocked by antioxidant treatments [41]. Similarly, after three-dimensional co-culture with prostate cancer cells, bone-derived stromal cells undergo stable cytogenetic modifi cations by a ROS-mediated mechanism [72].
Conversely, in an MCF7-fi broblast co-culture model, MCF7 cancer cells undergo aneuploidy and random muta genesis [41], suggesting that CAFs facilitate the dynamic search for a more aggressive 'mutator phenotype' in cancer cells ( Figure 1). Functionally, fi broblasts provide cancer cells with a six-fold protection against cell death [41] (Figure 2), via the up-regulation of the antiapoptotic protein TIGAR [41]. Cav-1 knockdown fi broblasts provide even greater protection for cancer cells against apoptosis, clearly indicating that a loss of stromal Cav-1 in humans may greatly facilitate tumor growth by suppression of cancer cell death [41].
Th us, data supporting the 'reverse Warburg eff ect' indicate that cancer cells and fi broblasts are metabolically coupled and mutagenically co-evolving [41,73]. Cancer cells use oxidative stress to corrupt adjacent fi broblasts and to induce their metabolic re-programming [73]. In this way, fi broblasts secrete energy rich metabolites that facilitate cancer cell survival. In addition, fi broblasts promote mutagenesis of cancer cells, leading to a more aggressive 'aneuploid' phenotype [41,73]. Aneuploidy in cancer cells is known to be associated with poor clinical outcome.

Hypoxia, autophagy, and mitophagy in the tumor stroma
Metabolomic profi ling reveals that Cav-1(-/-) null mammary fat pads display a highly catabolic metabolism, with the increased release of several metabolites, such as amino acids, ribose and nucleotides, and a shift towards gluconeogenesis, as well as mitochondrial dysfunction [74]. Th ese changes are consistent with increased autophagy, mitophagy and aerobic glycolysis, all processes that are induced by oxidative stress [74]. Autophagy or 'self-eating' is the process by which cells degrade their own cellular components to survive during starvation or to eliminate damaged organelles after oxidative stress. Mitophagy, or mitochondrial-autophagy, is particularly important to remove damaged ROS-generating mitochondria.
An autophagy/mitophagy program is also triggered by hypoxia [75,76]. Hypoxia is a common feature of solid tumors, and promotes cancer progression, invasion and metastasis [77]. Interestingly, via induction of autophagy, hypoxia is suffi cient to induce a dramatic loss of Cav-1 in fi broblasts. Th e hypoxia-induced loss of Cav-1 can be inhibited by the autophagy inhibitor chloroquine, or by pharmacological inhibition of HIF1α [69,78]. Conversely, small interfering RNA-mediated Cav-1 knock-down is suffi cient to induce pseudo-hypoxia, with HIF1α and NFκB activation, and to promote autophagy/mitophagy, as well as a loss of mitochondrial membrane potential in stromal cells [78]. Th ese results indicate that a loss of stromal Cav-1 is a marker of hypoxia and oxidative stress. In a co-culture model, autophagy in cancer-associated fi broblasts was shown to promote tumor cell survival via the induction of the pro-autophagic HIF1α and NFκB pathways in the tumor stromal microenvironment [78]. Finally, the mitophagy marker Bnip3L is selectively upregulated in the stroma of human breast cancers lacking Cav-1, but is notably absent from the adjacent breast cancer epithelial cells [78].
Another study has shown that cell-specifi c induction of autophagy by HIF1α activation in fi broblasts or MDA-MB-231 cells diff erentially aff ects tumor growth. In a xenograft model, HIF1α activation in fi broblasts greatly enhances the tumorigenicity of co-injected MDA-MB-231 cells, whereas HIF1α activation in MDA-MB-231 cancer cells suppresses tumor growth [79]. Importantly, in this experimental setting, the levels of tumor angiogenesis were unchanged. As HIF1α triggers autophagy in both fi broblasts and cancer cells, these data demonstrate that the role of autophagy in driving tumor formation is celltype specifi c, and that stromal autophagy, and not cancer cell autophagy, favors tumor growth.
Several studies have demonstrated that the overexpression of autophagic markers, such as ATG16L and cathepsin K and D, in the stroma and not in tumor cells predicts poor prognosis [80][81][82]. Similarly, loss of autophagic markers, such as Beclin 1, in tumor cells correlates with poor clinical outcome, suggesting that activation of an autophagic program in tumor cells reduces tumor aggressiveness [83].
Metabolome profi ling of several types of human cancer tissues versus corresponding normal tissues have consistently shown that cancer tissues are highly catabolic, with the signifi cant accumulation of many amino acids and TCA cycle metabolites [84,85]. Th e levels of reduced glutathione were decreased in primary and metastatic prostate cancers compared to benign adjacent prostate tissue, suggesting that aggressive disease is associated with increased oxidative stress [86]. Also, these data show that the tumor microenvironment has increased oxidative-stress-induced autophagy and increased catabolism.
Taken together, all these fi ndings suggest an integrated model whereby a loss of stromal Cav-1 induces autophagy/ mitophagy in the tumor stroma, via oxidative stress. Th is creates a catabolic micro-environment with the local accumulation of chemical building blocks and recycled nutrients (such as amino acids and nucleotides), directly feeding cancer cells to sustain their survival and growth. We have termed this novel idea the 'autophagic tumor stroma model of cancer' [74]. Th is new paradigm may explain the 'autophagy paradox' , which is based on the fact that both the systemic inhibition and systemic stimulation of autophagy prevent tumor formation [87]. We propose that vectorial energy transfer from the tumor stroma to cancer cells directly sustains tumor growth, and that interruption of such metabolic coupling will block tumor growth. Autophagy inhibitors (such as chloroquine) functionally block the catabolic transfer of metabolites from the stroma to the tumor, inducing cancer cell starvation and death [88]. Conversely, autophagy inducers (such as rapamycin) promote autophagy in tumor cells and induce cell death [89]. Th us, both inhibitors and inducers of autophagy will have a similar eff ect by severing the metabolic coupling of the stroma and tumor cells, resulting in tumor growth inhibition (cutting 'off ' the fuel supply).
Th is model may also explain why enthusiasm for antiangiogenic therapy has been dampened. In most cases, the clinical benefi ts are short term, and more importantly, new data suggest an unexpected link between anti-angiogenic treatments and metastasis. In pre-clinical models, anti-vascular endothelial growth factor (anti-VEGF) drugs (sunitinib and anti-VEGFR2 blocking antibodies) were shown to inhibit localized tumor formation, but potently induced relapse and metastasis [90][91][92]. Th us, by inducing hypoxia in the tumor microenvironment, antiangiogenic drugs may create a more favorable metastatic niche [93]. Hypoxia-induced autophagy may play a role by generat ing a catabolic micro-environment rich in chemical building blocks that can be directly used by cancer cells to sustain malignant transformation and metastatic progression.
Finally, the autophagic tumor stroma model can also provide an explanation for systemic cachexia, which is progressive skeletal muscle and adipose tissue wasting, aff ecting up to 50% of all cancer patients and resulting in severe weight loss and shortened survival [94]. Cachexia is the result of increased energy consumption and higher metabolic rates [95]. Based on our data, we envision that cancer leads to a generalized catabolic state via an autophagic-mechanism that generates building blocks and starves the rest of the body. While the exact signaling pathways governing this phenomenon are not yet fully elucidated, it is clear that oxidative stress-induced autophagy functions as a driver of muscle wasting [96]. For example, skeletal muscles from tumor-bearing mice showed impaired Akt activation and a more than 50-fold induction of Bnip3, a well recognized marker of autophagy/mitophagy [97]. Th ese fi ndings also help explain why patients with metabolic syndrome and diabetes have an increased risk for the development of multiple epithelial cancers, due to their constitutive and systemic activation of the autophagic program, and the over-production of high-energy nutrients, such as lactate and ketones [74].
We have recently used laser-capture micro-dissection of the tumor stroma from human breast cancers to directly validate that a loss of stromal Cav-1 is transcriptionally associated with oxidative stress, hypoxia, autophagy, and mitochondrial dysfunction, via gene-set enrichment analysis (Figure 4) [98]. In addition, we see that oxidative stress in CAFs induces cytokine production via NFκB activation, directly linking infl ammation with autophagy/catabolism in the tumor stroma [99]. So, cachexia may start locally as stromal autophagy, and then spread systemically via cytokine production and infl ammation, which also drive autophagy [99].

Glutamine utilization, ammonia production, and autophagy in the tumor stroma
In direct support that cancer cells use mitochondrial oxidative metabolism, many investigators have shown that cancer cells are 'addicted' to glutamine [24,100]. Gluta mine is a non-essential amino acid that is metabolized to glutamate and enters the TCA cycle as alphaketoglutarate, resulting in high ATP generation via oxidative phosphorylation [2,22,[101][102][103].
Recent studies also show that ammonia is a by-product of glutaminolysis [104][105][106]. In addition, ammonia can act as a diff usible inducer of autophagy [104][105][106]. Given these observations, glutamine addiction in cancer cells provides another mechanism for driving and/or main tain ing autophagy in the tumor micro-environment ( Figure 5). In support of this idea, we have previously shown that a loss of Cav-1 in the stroma is suffi cient to drive autophagy, resulting in increased glutamine produc tion in the tumor micro-environment [74]. Th us, this concept defi nes a new vicious cycle in which autophagy in the tumor stroma transfers glutamine to cancer cells, and the by-product of this metabolism, ammonia, maintains autophagic glutamine production ( Figure 5). Th is model fi ts well with the 'autophagic tumor stroma model of cancer metabolism' , in which energyrich recycled nutrients (lactate, ketones, and glutamine) fuel oxidative mitochondrial metabolism in cancer cells.

Lessons from other paradigms: an infectious parasitic cancer cell that metastasizes and captures mitochondrial DNA from host cells
We have recently proposed that cancer cells behave like 'parasites' , by inducing oxidative stress in normal host fi broblasts, resulting in the production of recycled nutrients via autophagy [41,69,78]. Th is is exactly the same mechanism by which infectious parasites (such as malaria) obtain nutrients and are propagated by inducing oxidative stress and autophagy in host cells [107][108][109]. In this regard, malaria is an 'intracellular' parasite, while cancer cells may be thought of as 'extracellular' parasites. Th is explains why chloroquine is both an eff ective antimalarial drug and an eff ective anti-tumor agent, as it functions as an autophagy inhibitor [110], cutting off the 'fuel supply' in both disease states.
Are there any examples of cancer cells that act as infectious parasites? Surprisingly, the answer is yes [111]. Th ere are four known types: canine transmissible venereal tumor (CTVT), which is a sexually transmitted disease that occurs in feral dogs [112] -it is transmitted through coitus, licking/biting, or sniffi ng; devil facial tumor disease, spread among Tasmanian devils via facial biting [113,114]; contagious reticulum cell sarcoma, which is transmitted between Syrian hamsters via mosquito bites [115,116]; and malignant fi brous histiocytoma, which was transmitted from a cancer patient to his surgeon [117] during an injury that occurred when the surgeon was operating.
Th ese parasitic cancers are all transmitted by allografting, so the cancer cells literally 'metastasize' from one aff ected host to another 'naïve' host, in an infectious manner that does not involve a virus.
Experimentally, the most is known about CTVT. Th is disease is thought to have originated in ancient wolves or coyotes, and the tumor cells themselves act as the infectious agent. As such, these cancer cells are genetically distinct from their hosts, as determined by genomic sequence analysis of their nuclear DNA [118]. CTVT represents the oldest known cancer cell line that has been continuously propagated, most likely for more than 10,000 years [118,119].
Interestingly, it has recently been shown that CTVT tumor cells survive by periodically 'capturing' mito chondrial DNA from their hosts [119,120]. Th us, it has been suggested that these CVCT tumor cells have survived for more than 10,000 years by maintaining and renewing their capacity for oxidative mitochondrial metabolism by 'stealing' host cell mitochondrial DNA [119,120]. In accordance with this idea, CTVT is highly sensitive to adriamycin/doxorubicin therapy [121], a chemo-thera peu tic agent that also functions as a mitochondrial poison.
Similarly, it has been independently shown that human cancer cells can 'steal' live mitochondria or mitochondrial DNA [122] from adjacent mesenchymal stem cells in culture, which then rescues aerobic glycolysis in these cancer cells [122]. Th is is known as mitochondrial transfer [122]. Interestingly, others have shown that metastatic breast cancer cells show the up-regulation of numerous mitochondrial proteins [123], specifi cally associated with oxidative phosphorylation, as seen by unbiased proteomic analysis [123].
Th us, increased mitochondrial oxidative metabolism may be a key driver of tumor cell metastasis. In further support of this argument, treatment of MCF7 cancer cells with lactate is indeed suffi cient to induce mitochondrial biogenesis in these cells [41,78] (Figure 6). To determine if these fi ndings may be clinically relevant, a lactate-induced gene signature was recently generated using MCF7 cells [124]. Th is gene signature shows that lactate induces 'stemness' in cancer cells, and this lactateinduced gene signature predicts poor clinical outcome (including tumor recurrence and metastasis) in breast cancer patients [124] (Figure 6). Th ese fi ndings are consistent with experiments showing that intraperitoneal injection of lactate in an MDA-MB-231 xenograft model results in an approximately ten-fold increase in lung metastasis [125].

Competing interests
The authors declare that they have no competing interests. Lactate-treated (10 mM) MCF7 cells were subjected to transcriptional profi ling (exon-array) to generate a gene signature. Interestingly, this lactate-induced gene signature predicts poor clinical outcome in estrogen receptorpositive/luminal A breast cancer patients, which represent nearly 60 to 70% of cases, and is the most common type of breast cancer. Modifi ed and reproduced with permission from [124].