The immunological Warburg effect: Can a metabolic‐tumor‐stroma score (MeTS) guide cancer immunotherapy?

The “glycolytic switch” also known as the “Warburg effect” is a key feature of tumor cells and leads to the accumulation of lactate and protons in the tumor environment. Intriguingly, non‐malignant lymphocytes or stromal cells such as tumor‐associated macrophages and cancer‐associated fibroblasts contribute to the lactate accumulation in the tumor environment, a phenomenon described as the “Reverse Warburg effect.” Localized lactic acidosis has a strong immunosuppressive effect and mediates an immune escape of tumors. However, some tumors do not display the Warburg phenotype and either rely on respiration or appear as a mosaic of cells with different metabolic properties. Based on these findings and on the knowledge that T cell infiltration is predictive for patient outcome, we suggest a metabolic‐tumor‐stroma score to determine the likelihood of a successful anti‐tumor immune response: (a) a respiring tumor with high T cell infiltration (“hot”); (b) a reverse Warburg type with respiring tumor cells but glycolytic stromal cells; (c) a mixed type with glycolytic and respiring compartments; and (d) a glycolytic (Warburg) tumor with low T cell infiltration (“cold”). Here, we provide evidence that these types can be independent of the organ of origin, prognostically relevant and might help select the appropriate immunotherapy approach.


| HIS TORIC AL TIMELINE-THE WARBURG EFFEC T THEN AND NOW
In the early 20s of the last century, Warburg, Posener, and Negelein 1 discovered a unique behavior of tumor tissue in vitro.
They examined respiration and glycolysis of different tissue sections and found that tumors exhibit an unusually high glycolytic activity and production of lactic acid from glucose when compared with normal tissues. Surprisingly, glycolysis was "aerobic" and was not inhibited by oxygen in malignant cells meaning that cancer cells lack the "Pasteur effect." Carl F. Cori and Gerty Cori confirmed this finding in living animals. Interestingly, the lactic acid content of tumors was low in starving animals with low-glucose levels and the authors concluded "up to a certain limit an excess of lactic acid can be completely eliminated into the blood stream." In another set of experiments, the Cori laboratory compared blood that had passed through a tumor with normal venous blood in chicken and observed decreased sugar and increased lactic acid in blood from the tumor-bearing wing. 2,3 In 1926, Otto Warburg further analyzed the energy metabolism of tumors and discussed possible ways to kill tumor cells through a "lack of energy". 4 Warburg measured glucose and lactic acid in tumor arteries and veins and found greater differences than the Coris. His results showed that the concentration of glucose falls by about 57% after passing through a tumor, whereas normal tissues consume about 2%-18% of the arterial glucose. Tumor veins contained clearly more lactic acid than arteries, but this was not the case for normal tissues. He stated that tumor cells obtain energy in two ways, by respiration and fermentation and calculated that about 66% of the glucose is used in fermentation and the rest for respiration, but this may vary in different parts of the tumor based on heterogeneity in glucose and oxygen availability. He concluded that "it is necessary to stop both respiration and fermentation, if cells are to be killed for want of energy". 5 Later in his life, he changed his side of view and claimed that tumor cells are characterized by a damaged respiration. 6,7 In contrast to these observations and numerous studies since, two recent publications stated that the tumor environment contains no elevated lactate levels in the interstitial fluid, 8,9 and lactate may accumulate preferentially in tumor cells. In this case, the tumor stroma, for example, immune cells, would not face high-lactate concentrations. However, Sullivan and colleagues examined pancreatic ductal adenocarcinoma (PDAC), a tumor with high stromal/fibroblast content. A possible explanation for low lactate levels in this tumor model is a metabolic symbiosis in heterogeneous tumors consisting of fibroblasts and highly glycolytic tumor cells.
Here, tumor cells produce and secrete lactate which is consumed and used as a fuel for respiration by adjacent stromal cells. 10,11 Alternatively, it has been shown by Sonveaux and colleagues that respiring tumor cells in oxygenated regions take up lactate and metabolize it. 12 Low lactate levels may also occur in tumors where lactate is eliminated by the blood stream in well-perfused tumor regions. But compromised blood perfusion is a typical feature of tumors and may help to build up extracellular levels of lactate and protons in the tumor environment especially in tumors with highly accelerated glycolysis. In tumors with extremely high glycolytic activity, elevated lactate levels can also be detected in sera of tumor patients. 13,14 Accordingly, lactate levels have been suggested as a prognostic biomarker in high-grade primary brain tumors and metastatic lung cancer where elevated pretreatment serum lactate levels were associated with worse progression free survival. 15,16 Bringing an additional layer of complexity, in the "Reverse Warburg Effect" stromal cells such as fibroblasts or macrophages undergo "aerobic glycolysis" and produce lactic acid which can either accumulate or be utilized by cancer cells for mitochondrial oxidative phosphorylation (OXPHOS). [17][18][19] Thus, not only malignant cells but also stromal cells contribute to tumor lactic acidosis. Furthermore, acidosis is not only associated with lactic acid secretion, as another major source of protons is CO 2 , produced in more oxygenated tumor areas which is hydrated into HCO3 − and H + ions by carbonic anhydrases leading to acidification. 20 Today, we know that tumors often represent a mosaic of tumor cells with different metabolic properties. While some tumors rely more on oxygen, others can be more glycolytic. Metabolic heterogeneity regarding glucose metabolism within and between human lung tumors was nicely demonstrated by the group of DeBerardinis using intraoperative 13C-glucose infusions in patients. 21 The authors concluded that enhanced glucose uptake by lung tumors supplies glucose oxidation rather than enhanced lactate fermentation. On the other hand, in vivo isotope tracing in human clear cell renal cell carcinomas (ccRCC) revealed enhanced glycolysis and minimal glucose oxidation compared to adjacent kidney consistent with the classical Warburg phenotype. 22 A metabolic heterogeneity among ccRCC tumors was demonstrated by Brooks et al with low or no 18F fluoro-deoxy-glucose (FDG) uptake in some tumors but uniformly high uptake in other tumors. 23 We studied T cell infiltration in relation to tumor glucose transporter 1 (GLUT1) expression in ccRCC and identified two major tumor types. Classical Warburg tumors with high GLUT1 expression and low T cell infiltration, and tumors with low tumor GLUT1 expression and high T cell infiltration in the tumor (Figures 1 and 2). 24 In line a recent publication described a negative correlation between increased glycolysis and CD8 T cell infiltration in colon cancer. 25 What is more, GLUT1 expression inversely correlated with numbers of CD3 + T cells in HNSCC and lung SCC. 26 Two molecular subgroups with low-and high OXPHOS were also identified in high-grade serous ovarian cancer. While low-OX-PHOS tumors mainly exhibit a glycolytic metabolism, high-OX-PHOS tumors rely on oxidative phosphorylation supported by glutamine and fatty acid oxidation. Furthermore, distinct metabolic subtypes were described in PDAC cell line models. Here, lipogenic tumor cell lines showed higher oxygen consumption and greater mitochondrial content compared with glycolytic tumor lines. In primary pancreatic tumor samples, the lipid subtype was strongly associated with an epithelial phenotype, whereas the glycolytic subtype was associated with a mesenchymal phenotype. 27 Based on the negative correlation between glycolytic activity, GLUT expression and T cell infiltration, these data suggest a stronger immune cell infiltration in high-OXPHOS tumors compared to tumors with glycolytic activity in ovarian and pancreatic cancer ( Figure 1).
These data also implicate that the "Warburg phenotype" is not characteristic for all tumor entities and not even for all tumor cells within one given tumor. Nevertheless, the work of Otto Warburg and colleagues has an enormous impact on current cancer research and diagnostics. Uptake of the positron-labeled glucose analogue 18F-fluoro-deoxy-glucose determined with positron emission tomography (18F-FDG-PET) is a well-established method for tumor diagnosis and staging. Even disseminated and hematologic malignancies, such as lymphoma can be imaged by FDG-PET, 28,29 demonstrating that not only solid tumors but also leukemia and lymphoma cells of different origin display an accelerated glucose metabolism, which was confirmed by gene expression and functional analysis. [30][31][32] Moreover, the discovery that oncogenes and tumor suppressor genes are closely linked to and regulate the Warburg effect has led to a renewed interest in tumor metabolism which is now regarded as a hallmark of cancer. 33

| WARBURG EFFEC T IN TUMOR S TROMAL CELL S
The observed composition of tumors regarding both a cellular and a metabolic heterogeneity lead us to propose the metabolic-tumor- In a small cohort of patients, we analyzed stroma-rich tumors for GLUT1 and CD3 expression by immunohistochemistry and detected GLUT1 positive CAFs in a pancreatic carcinoma patient, however, only in a subset of stromal cells (MeTS2a, Figure 3). In colon carcinoma and mammary carcinoma tissues, CAFs were mainly GLUT1 negative and these areas were often highly infiltrated by CD3positive T cells (MeTS3a, Figure 3).
In light of these observations, high 18F-FDG uptake can also reflect an intense ongoing host reaction, such as after radiation therapy, where 18F-FDG uptake by tumor tissues is often increased despite the decreased viability of tumor cells. In an interesting study, Mamede and colleagues analyzed the distribution of 18F-FDG in immunocompetent and immunodeficient mice bearing the same murine ovarian squamous cell carcinoma. Despite comparable tumor growth and histology in both mice strains, the 18F-FDG signals were significantly higher in tumors in immunocompetent than in immunodeficient mice. 46 Thus, it was unlikely the tumor metabolic activity, but rather an immune activation that led to the increased 18F-FDG uptake. 46,52,53 In a murine mammary carcinoma model, 18F-FDG accumulation was even higher in macrophages in the outer zones of necrosis than in tumor cells, 54  leading to HGF production, thus closing the vicious circle. 58 In a clinical study, breast cancer patients with stage 0/1 tumors were treated

| MOLECUL AR BACKG ROUND OF THE WARBURG EFFEC T IN TUMOR AND S TROMAL CELL S
Warburg proposed that cancer is caused by a metabolic "switch" from mitochondrial respiration to aerobic glycolysis. However, in following years numerous groups provided evidence of the important role of respiration for tumor growth in several preclinical models and human tumor entities. We know that cancer development is caused by genetic alterations and signaling networks downstream

| E VOLUTI ON OF THE WARBURG EFFEC T-B ENEFITS FOR THE TUMOR
In 2004, Gatenby and Gillies asked "why do cancers have high aerobic glycolysis?" 100 and proposed that "cell populations with upregulated glycolysis and acid resistance have a powerful growth advantage, which promotes proliferation and invasion." However, the metabolism of glucose to lactate is less efficient compared to oxidative phosphorylation, at least in terms of ATP production per mol of glucose. Why would a proliferating cell use a less efficient metabolism? A possible explanation is that proliferating cells have requirements that extend beyond ATP as they need to replicate all of its cellular contents such as nucleotides, amino acids, and lipids.
Vander Heiden and colleagues proposed that "the metabolism of cancer cells, and indeed all proliferating cells, is adapted to facilitate the uptake and incorporation of nutrients into the biomass needed to produce a new cell". 101 Would this explain why the Warburg effect provides an evolutionary benefit for tumor cells? The excess lactate secretion that accompanies the Warburg effect leads to a loss of three carbons that could be utilized for building blocks or ATP production-an inefficient use of resources. In a more recent publication, the Vander Heiden group investigated the fraction of carbon mass in cells derived from different nutrients and found that the majority of carbon mass is not derived from glucose but rather from glutamine indicating that high glycolysis supports cell proliferation through mechanisms beyond providing carbon for biosynthesis. 102 Alongside its role in providing carbon for building blocks and energy generation, aerobic glycolysis results in a high rate of lactate production. The maintenance of the glycolytic flux requires a continuous export of lactate and protons from the cancer cell, which is Another integral factor of sustained tumor growth and metastasis is angiogenesis. Tumor-derived lactate can induce vessel formation through stimulation of VEGF production by endothelial cells. 105 Vegran et al 106  In addition to the generally accepted mechanism of tumor vascularization through sprouting of endothelial cells from pre-existing vessels, some studies suggest a contribution of stem cell-derived endothelial progenitors as well as cells from the myeloid lineage. We found that incubation of tumor-associated DCs with pro-angiogenic factors, such as vascular endothelial growth factor and oncostatin M, led to trans-differentiation of DCs into endothelial-like cells. 107

| IMMUNOLOG IC AL CONS EQUEN CE S OF THE WARBURG EFFEC T-A ME TABOLI C IMMUNE CHECKP OINT
Several publications underline the importance of immune cell infiltration for patient outcome. Galon and colleagues suggested in 2006 that the type, density, and location of immune cells within colorectal tumor samples are a better predictor of patient survival than classical histopathological methods. 108 We and others have shown that tumor-derived lactate strongly inhibits both T cell and NK cell function 14,84,109 and the differentiation and activation of myeloid cells. 110,111 This indicates that the tumor-promoting effect of lactate and acidification may in part be related to its immunosuppressive function and the metabolic phenotype of tumors may be decisive for T cell activity and thereby for patient prognosis.

IFNγ expression in tumor-infiltrating T cells and NK cells, thereby in-
hibiting tumor immunosurveillance and promoting tumor growth. 84 A relation of the tumor metabolic status with an anti-tumor T cell response has been observed in several entities and prominent examples are summarized in Table 1.
But what is the underlying mechanism for the profound impair- Renal cell carcinoma Lower T cell infiltration, inferior response to checkpoint blockade 24,172 Colon cancer Lower CD8 T cell infiltration 25 Head and neck squamous cell carcinoma Lower CD8 T cell infiltration 26 Lung squamous cell carcinoma Lower CD8 T cell infiltration 26 Lung adenocarcinoma Lower T cell infiltration 55 Lung carcinoma Impaired anti-tumor T cell response 17 Prostate cancer Reduced anti-tumoral Th1 cells  132,133 Moreover, high LDH activity has been described in the synovial fluid of RA patients with a shift from LDH1 and LDH2 isoforms to the glycolysis-associated isoforms LDH4 and LDH5. 134 Lactate is then taken up via the sodium-lactate transporters in T cells thereby leading to entrapment and functional changes that drive chronic inflammation. 135 Lactate uptake also results in increased IL-17 production via PKM2/STAT3 signaling and enhanced fatty acid synthesis. 136 Interestingly, the presence of synovial lactate has also been proposed as a fast clinical diagnostic tool to identify patients with septic arthritis. 137,138 Long known and widely accepted, the acidification of the skin represents a pillar of its barrier function. It has been observed that low pH of the skin regulates its permeability, improves the integrity and cohesion of stratum corneum (SC), and increases anti-microbial defenses. 139   and reduce lactate secretion. In addition, immune-modulatory drugs such as lenalidomide act on these transporters by disrupting the MCT-CD147 axis, which is essential for their membrane expression. 162 A highly glycolytic tumor metabolism is also associated with resistance to conventional therapies. Such Warburg effect-mediated therapy resistance were observed with the proteasome inhibitor bortezomib in multiple myeloma cells, carboplatin in non-small-cell lung 163 and paclitaxel in lung cancer cells. 164 Furthermore, a study by Zhao et al revealed that heat shock factor 1 and LDH-A drive glycolysis and induce resistance to trastuzumab, an anti-HER2 receptor antibody in breast cancer cells. 165 Mechanisms of how tumor glycolysis mediates therapy resistance are not completely elucidated, but this phenomenon has repeatedly been linked to the activity of P-glycoprotein (P-gp), which actively pumps cytotoxic drugs such as doxorubicin and paclitaxel out of the cell and the activity of P-gp increases in hypoxia and acidosis. 166 Another mechanism how tumors benefit from acidosis, and which is frequently neglected, is the "ion trapping"-a process where charged compounds such as chemotherapeutics cannot pass a cellular membrane due to decreased permeability. 167 168,169 Other strategies for stromal cell targeting were summarized by Dykes et al. 170 Given that the accumulation of lactate and acidification block the anti-tumor function of T and NK cells and foster the differentiation and activity of immune cell populations supporting tumor growth such as Tregs or MDSCs, the Warburg effect limits the success of immunotherapeutic approaches. In line, the efficacy of adoptive T cell transfer can be limited due to increased tumor glycolytic activity. 171 Furthermore, first studies in humans show a correlation between a high glycolytic activity in tumors and a low response rate to checkpoint blockade. 119,172 Accordingly, good response to immunotherapy was associated with enriched mitochondrial metabolism in melanoma patients. 173 Surprisingly, the opposite finding was re- As the Warburg phenotype is not specific for tumor cells or tumor-promoting stromal cells, but a common feature of proliferating and activated immune cells, such as effector T cells and NK cells, 34,35,37,38,178,179 the application of glycolytic inhibitors could exert side effects on the immune system. Buffering, therefore, might be an alternative strategy to reduce the negative effects of tumor acidification not interfering with immune cell activation. 104,180 Accordingly, different buffering approaches such as the administration of bicarbonate or proton pump inhibitors (PPIs) promoted immunotherapy. [181][182][183] Moreover, Vishvakarma and collegues showed that the application of PPIs resulted in an enhanced recruitment of M1 macrophages and shifted the cytokine profile toward tumor cytotoxic cytokines. Finally, this study showed that macrophages isolated from PPI treated tumors, which were adoptively transferred in tumor-bearing mice, showed superior capacity to control tumor growth. 184 Glycolytic restriction might not affect all immune cell populations to the same extent; it has been shown that under low-glucose conditions, T cells show a remarkable flexibility and while proliferation decreases, effector functions are preserved. [127][128][129]185 Therefore, glycolytic inhibitors might be beneficial even in a tumor setting. In line, lenalidomide has been shown to promote IL-2 expression in T cells. 186 Notably, the response to adoptive T cell increasing levels of lactate and acidification, targeting glycolysis and/or stromal cells might be essential to allow an effective immune response, which is of special importance in the context of cancer immunotherapy.