Metabolic Symbiosis Enables Adaptive Resistance to Anti-angiogenic Therapy that Is Dependent on mTOR Signaling

1Judah Folkman’s vision of targeting the tumor neovasculature as a new modality of cancer therapeutics has inspired a series of drugs that either exclusively (e.g., bevacizumab) or primarily (e.g., sunitinib, axitinib, and sorafenib) inhibit VEGF signaling (,; and references therein) with associated beneficial responses, representing proof of principle and new additions to the armamentarium of anti-cancer drugs. However, as with many targeted therapies, clinical responses to angiogenesis inhibitors (AI) are typically limited, manifested as increased, but limited progression-free survival and variable (or no) overall survival (; and references therein). Concurrent with such clinical investigations, a number of preclinical studies of AI in various mouse models of human cancer have revealed multiple forms of adaptive resistance that enable tumors to evade the effects of AI therapy. One such model—RIP1Tag2 transgenic mice, which develop de novo pancreatic neuroendocrine tumors (PanNET) via a multistage pathway—has been particularly instructive: tumorigenesis involves a discrete angiogenic switch, which is necessary for initial tumor formation. AIs, in particular ones targeting the VEGF signaling pathway, show demonstrable efficacy in this model, the results of which motivated clinical trials that led to the approval of one such agent, sunitinib, in human PanNET. Importantly, however, AIs are not curative in this mouse model, much as in human cancer patients. The basis for treatment failure lies in part in the development of multiple forms of adaptive resistance to AIs, including revascularization mediated by alternative pro-angiogenic signaling circuits, recruitment of vascular-protective myeloid cells, and cooption of normal tissue vessels via increased invasion and metastasis.

2The RIP1Tag2 model has also been used to investigate other cancer hallmarks, including resistance to apoptosis and induction of invasiveness, which are variously regulated by signaling from receptor tyrosine kinases, including EGFR, IGFR, IR, and ALK; these signals converge in part on the mTOR kinase, which orchestrates a broad program affecting cell survival and metabolism. Notably, both preclinical and clinical studies have demonstrated the efficacy of targeting mTOR in PanNET.

3We have continued to assess the effects on AI therapy in this PanNET model, including a comparative analysis of two multi-kinase inhibitors, sunitinib and axitinib, both of which target the VEGFR/PDGFR signaling pathways to inhibit new blood vessel growth and disrupt endothelial cells and pericytes of the preexisting tumor vasculature, reducing vascular density and functionality rather than producing the vascularization characteristic of less potent AI such as bevacizumab. Although both compounds have a similar target profile, axitinib exhibits fewer off target effects and toxicity (http://www.accessdata.fda.gov). In the course of further characterizing cellular and histologic responses to the two drugs, in particular their effects on the aforementioned mTOR signaling pathway, we observed an intriguing switch in the pattern of mTOR activity, reflected by phosphorylation of its downstream effector S6. Specifically, the diffuse pattern seen in untreated tumors was replaced by one of focal clusters. This focal reorganization of mTOR signaling motivated the investigation reported below, where we implicate metabolic symbiosis as another mode of adaptive resistance to anti-angiogenic therapy.

4Given that mTOR signaling is prominently involved in the PanNET phenotype (see above), we had asked whether it was affected by disruption of the tumor vasculature with sunitinib. Indeed, mTOR signaling was found to be elevated in a set of sunitinib-treated tumors, as reflected in the heightened levels of phospho-S6 kinase (p-S6). In contrast, sunitinib treatment of cultured mouse PanNET cells did not elicit this upregulation (data not shown). To substantiate this observation, molecular efficacy trials were performed with sunitinib, in which mice with late stage tumors were treated for 7 days followed by analysis of tumor lysates by western blotting. Sunitinib elicited an increase in p-S6 (Figure 1A, left), which was blocked by concomitant administration of the mTOR inhibitor rapamycin (Figure 1A, right), demonstrating that its upregulation is mTOR dependent. Axitinib-treated tumors produced a similar trend (data not shown).

5The elevated mTOR signaling was surprising in that sunitinib- and axitinib-treated tumors become hypoxic in response to collapse of the tumor vasculature, and observations presented below), and mTOR is typically downregulated in hypoxic conditions. Indeed, a derivative PanNET cancer cell line (βTC4) downregulated p-S6 in hypoxic versus normoxic conditions (Figure 1B). Intrigued by this apparent dichotomy, we further investigated mTOR signaling in tumors treated for 4 weeks using immunohistochemistry. As shown in Figure 1C, p-S6 was elevated in sunitinib-treated tumors, but its expression was relocalized from the diffuse pattern seen in control tumors into focal clusters. The specificity of the p-S6 staining in control and sunitinib-treated tumors was confirmed by treatment with rapamycin, which ablated p-S6 expression, even when dosed in combination with sunitinib.

6Next, we assessed the relationship between p-S6 expression and hypoxia in treated tumors. Co-staining of sunitinib- and axitinib-treated tumors with antibodies to p-S6 and pimonidazole—to reveal regions of hypoxia—revealed an anti-correlation: p-S6 was predominantly expressed in the normoxic, and not hypoxic, regions of the treated tumors (Figure 1D).

7Eventually, as the vascular regression and inhibition of new growth continues, the normoxic regions seen in Figure 1D can coalesce into the focal clusters seen in Figure 1C. This is further illustrated in Figure 1E, which shows a highly necrotic/hypoxic tumor (left); within the regions of fibrosis/necrosis, focal clusters of viable cancer cells can be seen (H&E, middle left). These clusters are highly proliferative (Ki-67, middle right) and are cuffing blood vessels (CD31, right).

8The clustering of viable cancer cells around the few remaining blood vessels inside long-term treated tumors is consistent with their need for oxygen, but also for blood-borne nutrients, in particular glucose. We performed short-term trials using sunitinib alone or in combination with rapamycin and then performed RNA-sequencing (seq) analysis on the treated and control tumors. We conducted enrichment analysis on the top 2,500 unregulated genes in the control versus treated tumors, and the top ten enriched hallmark pathways for sunitinib monotherapy are depicted in Figure 2A, including hypoxia and glycolysis. The enrichment plot for the hallmark glycolysis signature (Figure 2A) includes Slc16a3 (MCT 4) and Ldha (lactate dehydrogenase A; Figure 2B), along with (to a lesser degree) Slc2a1 (GLUT1, data not shown). Both sunitinib and axitinib produced marked reductions in tumor vasculature (Figure S1A) with widespread hypoxia, consistent with the hypoxia signature. Motivated by the glycolysis signature and the known induction of glycolysis by hypoxia, we assessed expression by immunostaining for the lactate transporter MCT4 and the glucose transporter GLUT1, whose elevated co-expression is diagnostic of glycolysis. Concordantly, the striking upregulation of Slc16a3/MCT4 revealed by the RNA profiling of sunitinib-treated tumors was detected by immunohistochemistry in tissue sections from the majority of sunitinib- and axitinib-treated tumors and was very low/absent in control or rapamycin-treated tumors (Figure 2C). In treated tumors, GLUT1 and MCT4 were upregulated in the hypoxic regions, as revealed by co-staining with pimonidazole (Figure 2D). Such upregulation in hypoxic conditions is consistent with previous reports that GLUT1 and MCT4 are regulated by the HIF hypoxia response system. To confirm this interpretation, we evaluated compound mice carrying a tissue-specific gene knock out of HIF1α in the oncogene-expressing cancer cells. Indeed, sunitinib treatment failed to upregulate GLUT1 and MCT4 in the Hif1α-knockout (KO) tumors (Figure 2E), supporting the expectation that HIF1α drives their upregulation.

9The compartmental expression of GLUT1/MCT4 was reminiscent of a phenomenon—observed in certain tumors—known as metabolic symbiosis, wherein hypoxic regions import and metabolize glucose and secrete lactate, while normoxic regions import and metabolize the lactate. We therefore assessed expression of the lactate transporter MCT1 and found that it was upregulated in a distinct compartment from the hypoxic, MCT4 positive region (Figure 2F), although a small minority of cells are MCT1+/MCT4+ double positive (Figure S1B).

10The hypothesis of metabolic symbiosis demands that the secreted lactate is not merely “toxic acidic waste”, but rather a source of fuel that is imported and metabolized by the normoxic cancer cells. We assessed whether cultured βTC3 cells could recapitulate key aspects of the hypoxic response in culture and found that GLUT1 and MCT4 were upregulated under hypoxic conditions (Figure 3A). We further assessed lactate utilization in cultured cancer cells and in tumor-bearing mice, by supplying isotopically labeled lactate or glucose, and assessing their catabolism into metabolites by NMR spectroscopy (Figures 3B and 3C). Thus, βTC3 cancer cells were cultured in hypoxic and normoxic conditions, in which they were supplied with 1–13C-glucose or with 3–13C-lactate in glucose-free media, in both cases similarly supplemented with glutamine and fetal calf serum (FCS); after 20 hr, the cells were harvested for NMR analysis. At this time point, cells cultured in 1–13C-glucose increased the levels of lactate in conditioned media both in hypoxia and normoxia, whereas cells cultured in 3–13C-lactate reduced the levels of lactate in conditioned media under normoxia and slightly increased lactate in hypoxic conditions (Figure 3B, legend). The 13C NMR analysis revealed that in aerobic conditions 3–13C-lactate was catabolized to C4-glutamate, C2- and C3-aspartate (Asp), and C3-alanine (Figure 3B), consistent with previous studies on lactate metabolism by cancer cells and neurons. In contrast, lactate was imported, but not catabolized in hypoxic conditions (Figure 3B). Notably, the lactate catabolites coincide with those produced by 1–13C-glucose, indicating that both sets are likely produced through the same intermediate, pyruvate.

11Next, tumor-bearing mice treated for 10 days with sunitinib, rapamycin+sunitinib, or vehicle control were infused with 3–13C-lactate and the tumors excised and evaluated by NMR. Both treated and control tumors metabolized lactate (Figure 3C). Additionally, the branched metabolic pathways involved in lactate catabolism may be altered in activity. The prominent 3–13C-alanine band produced from 3–13C-lactate → 3–13C-pyruvate + glutamate → 3–13C-Ala + α-ketoglutarate is relatively undiminished versus the glutamate and Asp species in the sunitinib-treated tumors; notably, the pathway producing 3–13C-Ala + α-ketoglutarate is involved in sustaining tricarboxylic acid (TCA) cycle intermediates important both for energy production and biosynthesis of cellular building blocks, a process referred to as anaplerosis. Intriguingly, RNaseq and RT-quantitative (q)PCR analysis (Table S1) of sunitinib versus control tumors revealed upregulation of Gpt2, an aminotransferase catalyzing the transamination that produces alanine + α-ketoglutarate from pyruvate and glutamate. In addition, Gls2, which converts glutamine to glutamate in the first step of its metabolism, is also upregulated (Table S1). In contrast, the rapamycin+sunitinib-treated tumors did not produce 3–13C-Ala. Rather, lactate-derived pyruvate evidently was converted to acetyl-CoA to enter the TCA cycle, producing C2- and C4-glutamate and C2- and C3-Asp (Figure 3C). Thus, the data suggest that glutamine-associated catabolism of lactate to alanine becomes favored in the context of sunitinib therapy, which in the combination-treated tumors is repressed—directly or indirectly—by mTOR inhibition. Collectively, these observations, along with previous studies indicating the link between glutamine metabolism and mTOR signaling, led us to consider the possible involvement of glutamine and the observed metabolism of lactate in the induction of mTOR signaling.

12Motivated by the observation that expression of Gls2 was increased in sunitinib-treated tumors (Table S1), we asked whether it was expressed in the hypoxic or normoxic compartment. Analysis of sunitinib-treated and control untreated tumors by fluorescent staining with anti-glutaminase 2 (GLS2) and pimonidazole demonstrated elevated expression of glutaminase 2 in both cases, and it could be found in both normoxic and hypoxic compartments in sunitinib-treated tumors (Figure 4A), a result that was substantiated via RT-qPCR analysis of βTC3s cultured under normoxic and hypoxic growth conditions (Table S2); we also analyzed additional metabolic genes for differential expression and found significant upregulation of Slc2a1, Slc16a3, and Ldha in hypoxic conditions, whereas Slc2a2, Ldhb, and Gpt2 were elevated in normoxic growth conditions (Table S2).

13Next, we investigated possible links between glutamine, lactate metabolism, and mTOR signaling. First, we assessed the importance of glutamine for cell proliferation of βTC3 cells by EdU incorporation, which revealed minimal difference between the various culture conditions containing glutamine, with or without glucose and/or lactate; there was, however, a block in cell-cycle progression under glucose without glutamine conditions (w/wo lactate) (Figure S2A), highlighting the dependence of these cells on glutamine. Then, we evaluated mTOR signaling in βTC3 cells (Figure 4B) cultured in different combinations of lactate and glutamine. Cells were cultured in FCS and glutamine-supplemented, glucose-free media, with or without lactate. Net changes in lactate levels in conditioned media (CM) were assessed. Lactate levels in glutamine-supplemented, glucose-free cell media decreased reproducibly under aerobic conditions, consistent with its uptake by βTC3 cells (Figures 4B and S2B). Moreover, there was significant upregulation of p-S6 levels in cells cultured in lactate + glutamine versus glutamine alone (Figure 4B, top and bottom), correlating with a net reduction of lactate in CM. As expected, upregulation of p-S6 could be reversed by rapamycin. Notably, p-S6 upregulation could also be reversed by inhibiting lactate uptake with CHC or 7ACC2, two distinctive inhibitors of the MCT1 lactate transporter (Figure 4B, top and bottom). In addition, a glutaminase inhibitor, DON (Figure 4B, top and bottom), and an alanine aminotransferase inhibitor, AOA (Figure 4B, top), also reversed p-S6 upregulation, whereas a glutaminase 1-specific inhibitor, BPTES, did not (Figure 4B, bottom). We observed similar effects of lactate and glutamine on mTOR signaling in lactate-avid SiHa ovarian cancer cells (Figure S2C), which have been extensively used to study metabolic symbiosis. Collectively, the data support a mechanism whereby cancer cells take up and metabolize lactate in the context of bioavailable glutamine in normoxic, but not hypoxic conditions (Figures 3B and 3C), thereby upregulating mTOR signaling (Figure 4B). A schematic depicting the branched metabolic pathways involved in lactate catabolism in control and treated Rip1Tag2 PanNET tumors is depicted in Figure S2D. In addition, we found that βTC3 cells cultured in lactate + glutamine upregulated α-ketoglutarate compared to cells cultured in glutamine only (Figure 4C). Concordantly, Sonveaux and colleagues recently reported a link between lactate and glutamine metabolism in cultured SiHa and HeLa cancer cells under oxidative conditions.

14In light of the proposition that potent AIs were inducing tumors to adapt a strategy of metabolic symbiosis, and the observed relocalization of mTOR signaling into the normoxic compartment of putative symbiotic clusters, we investigated the impact of blocking mTOR signaling on the AI-induced symbiosis and consequent tumor phenotypes. Regression trials (depicted in Figure 5A) were initiated at 13 weeks, when the mice already had appreciable tumor burden, and continued to a defined endpoint 2 weeks later (15 weeks), when most untreated animals have succumbed to hypoglycemia attributable to increased insulin secretion from the multiple pancreatic tumors that develop. Reasoning that vascular collapse might affect the bioavailability of rapamycin in the tumor microenvironment, we tested a trial regimen in which R+S and R+A arms were dosed for an initial 2D with rapamycin alone, followed by continuous dosing in combination with the AI (called “stagger” or “St”). Another cohort of R+S was dosed simultaneously with both drugs from the beginning (“simultaneous” or “Sim”). All three monotherapies (Mono-S, red triangles; Mono-R, yellow triangles; and Mono-A, green circles) showed anti-tumor activity, blocking further growth to produce tumor stasis compared to 13 week controls (Figure 5B, blue circles). In marked contrast, each of the combinations of rapamycin plus an AI (R+S-St, purple diamonds; R+S-Sim, purple triangles; and R+A-St, rose squares) produced significant tumor regression versus the 13 week starting time point (Figure 5B, blue circles). An additional 4-week-long intervention trial was performed with sunitinib, starting at 11 weeks, when the tumors were smaller, until the same defined 15 week endpoint (Figure 5C). Again, while Mono-S (red triangles) and Mono-R (yellow triangles) produced significantly smaller tumors than vehicle-treated controls (blue circles), their combination (R+S-St, purple diamonds and R+S-Sim, purple triangles) produced significantly lower tumor burden than either efficacious monotherapy. Other dosing strategies were also assessed, including a shift from Mono-R to Mono-S (Figure 5C, R->S, green circles), which appeared indistinguishable from each monotherapy in regard to tumor burden (TB). There was also an efficacious response to the combination R+S when sunitinib was reduced by half (Figure 5C, R+1/2S, blue-gray circles). Thus, both experimental therapeutic trials demonstrated anti-tumoral efficacy for the combination of the mTOR inhibitor rapamycin and a potent AI.

15In order to further characterize the therapeutic response, open endpoint survival trials were performed with the combinations, in comparison to monotherapy- and vehicle-treated controls. Cohorts of mice were treated starting at 13 weeks until they became moribund and were sacrificed (Figure 5D). All treatments, including the monotherapy arms, led to a significant survival advantage versus vehicle-treated mice. End stage tumors had progressed to a considerable size on rapamycin monotherapy, larger on average than untreated controls, indicating an adaptive resistance involving enhanced growth rates (Figure S3A), as reported previously. Overall, TB at end stage was higher in regimens involving axitinib versus sunitinib (Figure S3A). Interestingly, both AIs produced tumors that never revascularized (data not shown). Additionally, the combination therapies reduced the frequency of metastasis compared to both AI monotherapies, despite the significantly longer survival of mice in the dual therapy arms (Figure S3B). It is unclear why the mice with only residual TB were dying, which is a topic worthy of future investigation.

16The combination of rapamycin plus sunitinib treatment for 1–2 weeks reproducibly elicited tumor regression, characterized by widespread necrosis and fibrosis, and markedly reduced proliferation, particularly in large tumors (Figures S4A, S4B, and S6). In contrast, while 8 weeks of Mono-S also produced tumors with pronounced necrosis and fibrosis, the remaining tumor cells were highly proliferative (Figure 1E). When 13- to 15-week-treated tumors were analyzed comparatively for proliferation rates via Ki-67 staining, there was a significant reduction in proliferation in R+S-treated tumors versus controls and both monotherapies (Figure S4B). Tumors treated with Mono-R had limited necrosis (data not shown), consistent with this and our previous study indicating it elicited heighted rates of apoptosis without appreciable necrosis (Figure S4C).

17Seeking to understand the basis for this heightened necrosis, and having identified putative symbiotic clusters of cancer cells associated with remaining blood vessels in the Mono-S-treated tumors, we asked whether rapamycin was affecting symbiosis in the R+S-treated tumors. The clusters could still be detected, as revealed by co-staining with MCT1 and MCT4, but their morphology was altered. Surprisingly, we observed that cancer cells in the hypoxic GLUT1high, MCT4high, and mTORlow compartment were selectively depleted compared to those cells in the normoxic GLUT1low, MCT1high, and mTORhigh regions where rapamycin was acting to suppress mTOR signaling. Only a rim of MCT4 cells could be detected, flanking regions of necrosis, in contrast to the multiple layers of such cells seen in the Mono-S-treated tumors (Figure 6A). In contrast, rapamycin did not obviously disrupt the (previously p-S6high) MCT1high compartment (Figure 6A). Thus, inhibition of mTOR signaling in the normoxic compartment was disrupting the symbiosis, but via an unexpected paracrine mechanism.

18A tenant of the symbiosis hypothesis is that the normoxic cancer cells have not upregulated glucose transporters, in contrast to the hypoxic cells (Figure 2D), thereby sparing limited bioavailable glucose to diffuse from the nearest blood vessel to distal, but still viable hypoxic cancer cells. We wondered, therefore, if mTOR inhibition affected expression of glucose transporters? We first surveyed untreated tumors and observed heterogeneous expression of GLUT1 and a related transporter GLUT2, which is involved in glucose homeostasis in pancreatic islets (Figure S5A). A third transporter, GLUT3, was not consistently expressed (data not shown). Mono-S- and R+S-treated tumors were stained with antibodies that recognized pimonidazole adducts, indicative of hypoxia, and anti-GLUT1 and anti-GLUT2 antibodies (Figure 6B). The analysis revealed appreciable upregulation of GLUT2 expression (red staining) in the normoxic regions of R+S, but not Mono-S-treated tumors (Figure 6B, bottom); the Mono-S tumors express very little GLUT2 (or GLUT1) in normoxic regions proximal to the vessels. In contrast to GLUT2, GLUT1 is not widely upregulated in the normoxic compartment of R+S-treated tumors. We also performed comparative IHC using automated staining with GLUT1 and GLUT2 and saw a similar upregulation of GLUT2 in the normoxic regions of R+S- versus Mono-S-treated tumors (Figure S5B). Western blot analysis on tumors also confirmed GLUT2 upregulation in R+S-treated tumors versus Mono-S (Figure S5C), as did qRT-PCR (Table S1). There was also a modest upregulation of GLUT2 in βTC3 cells cultured in glutamine+lactate+rapamycin versus glutamine ± lactate (Figure S5D), conditions we believe may be a surrogate for those found in the oxidative cancer cells inside AI-treated tumors. Although consistent with the in vivo results, further work is required to substantiate these findings and determine the molecular pathway that governs this regulation. Thus, we envision that the normoxic compartment of the putative symbiotic clusters switches, in the context of mTOR inhibition, to importing and metabolizing blood-borne glucose instead of sparing it to favor lactate. Consequently, the vascular non-proximal hypoxic cells would starve from glucose deprivation and die; likely exacerbated by lactic acidosis, since the normoxic cells are now net secretors of lactate as a consequence of aerobic glycolysis. Transmission electron microscopy (Figure S6) of cancer cells in the peri-necrotic region of a R+S-treated tumor reveals cells with granulated cytoplasm and grossly distended mitochondria, and abundant insulin granules, potentially indicative of the extreme conditions of insufficient oxygen and glucose, and an acidic microenvironment. To further substantiate these associations, Figure S7 shows images of additional Control, Mono-S-, and R+S-treated tumors on serial sections stained with anti-MCT1, anti-MCT4, and pimonidazole, matched with anti-phospho-S6 ribosomal protein (pS6), -GLUT1, and -GLUT2. These images further document the upregulation of MCT1/4 in Mono-S- and R+S-treated tumors versus untreated controls and abundant MCT1 staining in the R+S tumors. The pimonidazole/pS6 panels show strong and relatively homogenous staining of pS6 in non-hypoxic control tumors, and compartmentalized pS6 staining in the normoxic regions of Mono-S tumors, but no reactivity in the R+S-treated tumors. The TEM and immunostaining analysis both support the interpretation that the hypoxic compartment is being eliminated faster than the normoxic one in the R+S-treated tumors.

19In this and the two companion reports in this issue of Cell Reports from and, we describe an unanticipated new mode of adaptive resistance—metabolic symbiosis—that is induced in response to potent anti-angiogenic therapies that cause vascular collapse and consequent hypoxia. We show that cancer cells compartmentalize themselves in the acute condition of vascular insufficiency into comparatively hypoxic and normoxic compartments, based on their relative proximity to the few remaining functional blood vessels. The hypoxic cancer cells induce expression of the glucose importer GLUT1 and the lactate exporter MCT4, which is dependent on HIF1α. The normoxic cells express the lactate transporter MCT1, and we show that both normoxic cancer cells in culture and tumors in vivo import and catabolize lactate; thus lactate is not just toxic waste from glycolysis, but rather is being used for energy metabolism. Moreover, our data reveal that the metabolic branch of lactate catabolism involving glutamine metabolism is favored, and this branch is blocked in tumors treated with rapamycin and anti-angiogenic therapy. Moreover, PanNET cancer cells cultured in the presence of lactate and glutamine upregulate phospho-S6 ribosomal protein (p-S6); the proposition that glutamine metabolism upregulates mTOR activity is consistent with previous studies in other cell culture systems. The association of lactate and glutamine catabolism in the regulation of mTOR in these conditions of stressful vascular insufficiency is further substantiated by our results showing that mTOR upregulation can be suppressed by impairing lactate uptake with known inhibitors of MCT1, or by inhibiting enzymes involved in glutaminolysis, namely glutaminase 2 that converts glutamine to glutamate in the first step of its metabolism and an alanine aminotransferase that in turn catalyzes the production of alanine + α-ketoglutarate from (lactate-derived) pyruvate and glutamate. Other studies have reported that mTOR regulates glutamine flux and glutaminase levels.

20The observed upregulation/segregation of mTOR activity into the normoxic cell compartment induced by two potent AIs, and the resultant putative metabolic symbiosis, can be disrupted by concomitant inhibition of mTOR signaling with rapamycin. Remarkably, the initial impact of inhibiting mTOR in normoxic cancer cells is the necrotic death of the hypoxic cells, which lack elevated levels of mTOR signaling. An explanation for this intercellular crosstalk is offered by the observation that another glucose importer, GLUT2, is upregulated in normoxic peri-vascular cancer cells in response to rapamycin. The data suggest that the normoxic cancer cells proximal to the few remaining vessels consume the available blood-borne glucose when mTOR is inhibited, thereby starving distal hypoxic cancer cells, leading to their demise. Additionally, our NMR analysis of tumors suggests that mTOR inhibition blocks lactate catabolism via the anaplerotic pathway, and this abrogation of lactate consumption may lead to the intra-cellular and peri-cellular accumulation of toxic levels of lactate. The regulatory pathways involved in GLUT2 induction and metabolic reprogramming when the normoxic compartment is subjected to mTOR inhibition warrant future investigation.

21Remarkably, as summarized in the introduction, multiple modes of adaptive or evasive resistance to therapies targeting tumor angiogenesis have been reported in experimental therapeutic trials performed using conventional transplant and genetically engineered de novo mouse models of cancer. Of these, only heighted invasiveness has been clearly implicated as an adaptive resistance mechanism in a human cancer, namely glioblastoma (; and references therein), although similar mechanisms can be envisioned to underlay the transitory clinical responses to AI therapy in virtually all tested forms of human cancer. Notably, the accompanying report by, this issue of Cell Reports) documents the induction of MCT1/MCT4 lactate transporters in a pattern consistent with metabolic symbiosis in renal cell carcinoma (RCC) patient-derived xenograph tumor (PDX) models treated with AI therapy. Moreover, analysis of biopsies of tumors from RCC patients treated with sunitinib produced histologic signatures of metabolic symbiosis that were exacerbated in relapsing/progressing (resistant) tumors from patients on anti-angiogenic therapy. Their data further show that this metabolic patterning is associated with TOR signaling, since its inhibition can evidently block this characteristic patterning. This study in human RCC, along with the accompanying article by, this issue of Cell Reports) in a mouse model of breast cancer and the results presented herein collectively and compellingly add metabolic symbiosis to the roster of adaptive/evasive resistance mechanisms to anti-angiogenic therapy.

22In a broader context, the existence of metabolic symbiosis in tumors has been previously reported by Dewhirst, Sonveaux, Feron, and colleagues to arise spontaneously in certain transplant tumor models. They have described in a series of publications a metabolic symbiosis that conveyed decreased reliance on bioavailable glucose that limited the collateral damage from secreted lactate and consequent lactic acidosis, achieved by symbiotic cells importing and utilizing the lactate produced by glycolysis. The concept of metabolic symbiosis in neoplasia has been extended to interactions between cancer cells and cancer-associated fibroblasts endothelial cells, macrophages (; and references therein), and also documented in pancreatic adenocarcinoma. Its physiological basis lays in analogous symbiotic relationships operative in certain normal tissues, in particular brain and muscle (; and references therein). The concept can now be extended further, with the demonstrations that metabolic symbiosis can be induced in response to a serious environmental stress—angiogenesis inhibition, with consequent dropout and insufficiency of the tumor vasculature, and widespread hypoxia—in models of pancreatic neuroendocrine, breast, and renal cancer and in human RCC (this report, and the accompanying reports by and ).

23The question arises as to whether disruption of this mode of resistance might have value in extending the duration of efficacious responses by anti-angiogenic therapies. We show that metabolic symbiosis induced by sunitinib/axitinib can be disrupted by concomitant inhibition of mTOR signaling, resulting in significant reductions in TB and viability, but with only modest (albeit significant) extension in survival. The limited survival benefit suggests that “rapalogs” may not prove to be ideal drugs to disrupt this form of adaptive resistance; moreover, combinations of sunitinib/axitinib with the rapalog everolimus have been tested clinically, with limited benefit and significant toxicity. An alternative and potentially more attractive strategy may be to inhibit the MCT1 and/or MCT4 lactate transporters with highly selective drugs. Indeed, the preliminary studies reported herein and in the companion paper by Christofori et al. (2016), in which expression of MCT4 was suppressed genetically, encourages further preclinical and then potentially clinical evaluation, given that a number of MCT1 and MCT4 inhibitors are in pharmacological development.

24An important further consideration is that tumors faced with vascular dropout and insufficiency evoked by potent AIs may activate multiple modes of adaptive/evasive resistance, in different regions of the tumor microenvironment. For example, in the RIP1Tag2 mouse model of PanNET, sunitinib is apparently inducing two distinctive modes of adaptive resistance, namely, symbiosis in hypoxic regions of regressing tumors, and, as reported previously, increased invasion and metastasis, in part by upregulating cMET signaling. It is presently unclear whether these invading cancer cells also induce symbiosis upon reaching a size/location where diffusion of oxygen and glucose from co-opted normal vessels is insufficient, evoking hypoxia and, in turn, symbiosis. In fact, others have postulated that tumor acidity may promote local invasion. Interestingly, we observed that co-inhibiting angiogenesis and mTOR reduces the incidence of metastasis, but it is unclear whether this effect relates to disruption of symbiosis or some other physiological parameter. Future studies in preclinical and clinical trials may shed light on the interplay between adaptive resistance mechanisms and means to concomitantly disrupt them.

25An interesting question, not fully answered by the current studies, is how these metabolic clusters are induced to form, and in particular, why the peri-vascular normoxic cancer cells choose to take up and metabolize lactate and spare the limited glucose for their hypoxic brethren. A plausible explanation for the formation and disruption of metabolic symbiosis is illustrated schematically in Figure 7, and might transpire as follows: extensive vascular collapse elicited by the AIs results in hypoxia to trigger the HIF1α transcription factor; HIF1α induces GLUT1 and MCT4 and activates glycolysis in the hypoxic cancer cells, leading to high levels of lactate secretion; accumulating extracellular lactate induces expression of the MCT1 lactate transporter and, consequently, import of lactate into the normoxic cancer cells; in concert with serum-derived glutamine, lactate is catabolized in the normoxic cancer cells with consequent induction of mTOR signaling to promote tumor metabolism.

26Notably, mTOR expression must be both cause and consequence, since its pharmacological inhibition results in the normoxic cancer cells upregulating a glucose transporter (GLUT2), and apparently switching to secrete rather than import lactate, resulting both in insufficient glucose for the hypoxic cells and likely toxic acidosis in the extracellular microenvironment. In addition, mTOR has been shown by others to be upregulated by and also to control glutamine metabolism. We extend these results by showing that its inhibition blocks the conversion of lactate-derived pyruvate/glutamate to alanine/α-ketoglutarate in R+S-treated tumors, disrupting the production of TCA cycle intermediates, potentially contributing to increased levels of (toxic and non-catabolized) lactate. While plausible, this rationale for the assembly and disruption of symbiosis will require future experimental validation.

27In conclusion, the induction of metabolic symbiosis in response to vascular insufficiency elicited by potent AIs further illustrates the remarkable propensity of cancer cells to circumvent barriers that arise during tumorigenesis, tumor progression, and in response to treatment. This symbiotic adaptation evidently constitutes another mechanism that enables tumors to overcome the damaging effects of targeted therapies, by co-opting cellular processes intended for important homeostatic purposes that instead serve the evolving cancer.

28Sunitinib and axitinib were purchased from LC laboratories. Rapamune (rapamycin) was purchased from Galexis. Please see Supplemental Information for details on dosing regimens, formulations, etc.

29All mice used in this study were maintained in a pathogen-free barrier animal facility of the Swiss Federal Institute of Technology Lausanne (EFPL) in accord with Swiss regulations for the care and use of mice in experimental research.

30βTC3 cells were acclimated to “low” glucose (6.25 mM glucose + 4 mM Gln +10% FCS). They were subsequently plated at 7 × 106 cells/10 cm tissue culture dish in low glucose culture media for 2 days, switched to 0% glucose/0% Gln + 10% FCS ON, then cultured in 20 mM 3–13C-lactate or 20 mM 1–13C-glucose + 4 mM Gln (no glucose) for 16 hr in hypoxic (1% O2/5% CO2) or normoxic (21% O2/5% CO2) conditions, and harvested for NMR. Plates were harvested individually on ice, rinsed twice with PBS, and collected in 0.9 M perchloric acid, which was subsequently titrated to approximately neutral pH with 0.9 M potassium hydroxide (KOH); extracts were lyophilized for NMR.

31Mice were treated for 10 days with sunitinib or vehicle control and were infused with 3–13-C-lactate in PBS, by intraperitoneal (i.p.) injection, over a period of 90 min. Pentobarbital was administered, and the mice were perfused intravenously (i.v.) with a fluoresceinated lectin. Tumors were excised and immediately quick frozen in liquid nitrogen. For NMR, the frozen tumors were pulverized in liquid nitrogen, 0.9 M perchloric acid was added to the pulverized tumor tissue, rotated at 4°C for 2 hr, and titrated with 0.9 M KOH to a final of approximately pH 6–7.

32NMR analysis was performed using a Bruker AVII 800 MHz spectrometer (18.8 T) equipped with a triple channel 5 mm cryoprobe (CPTCIz). 13C and 1H spectra were acquired using 30° angle pulses on 13C with decoupling on proton (zgpg). A repetition delay (d1) of 3 s and 2,048 scans (ns) was used. Spectra were processed with MestreNova software 10 using 5 Hz of line broadening. The chemical shift scale was calibrated with 3–13C lactate (21 ppm) as a reference, according to and. To prepare samples for introduction into the NMR, 300 μl of pulverized cell solution was mixed with 100 μl of D2O (99.9% deuterated), shaken for 2 min, and directly poured into a 5 mm NMR tube.

33Statistical analysis was performed using GraphPad Prism (GraphPad Software) software, and a log rank test was performed using software from the Broad Institute (http://bioinf.wehi.edu.au/software/russell/logrank/index.html). Due to the small sample size (as few as eight animals per treatment group) and the fact that not all the data are normally distributed, a non-parametric, suitable Mann-Whitney test was used (Figures 1A, 5B, 5C, S1A, S3A, S4B, and S4C; Tables S1 and S2), the log rank test was used for survival studies (Figure 5D), and the Student’s t test was used for Figures 4C, S5C, and S5D. For the incidence of metastasis during survival trials (Figure S3B), data were analyzed using R, where the event is “metastasis” and the horizontal axis is “age of sacrifice due to declining health status”, we tested whether the metastases are more frequent in one group versus the other (function “survdiff”).

34E.A. and D.H. designed the study and wrote the manuscript. E.A. performed the experiments and analyzed the data. P.M. produced and analyzed the NMR data; C.M.W. analyzed/quantitated mTOR signaling in tumors; L.L. and M.-W.P. produced the RNA-seq data; and M.-W.P. contributed to all aspects of the preclinical studies and performed RT-qPCR experiments and ImageJ analysis. S.S. performed bioinformatics analysis.