Targeting Metabolic Plasticity and Flexibility Dynamics for Cancer Therapy

Since the original work of Otto Warburg in the 1920s that demonstrated cancer cells to prefer glycolysis over mitochondrial respiration even under conditions of sufficient oxygen supply (1, 2), multiple studies have shown that the metabolic reprogramming of cancer is not static, but rather a highly dynamic process. The essentiality and uniqueness of these metabolic alterations enabled new discoveries in cancer diagnosis and therapy and led to the designation of cancer metabolism as one of cancer's hallmarks (3). Indeed, numerous metabolic alterations are co-opted by cancer cells during tumor initiation and progression, to maximize cancer fitness to the ever-changing environmental cues (4). Therefore, continuous metabolic adaptations are key for cancer cell growth and survival. These adaptations are achieved by coordinated intrinsic changes in gene expression leading to suppression and activation of enzymes, as well as by extrinsic fluctuations in the levels of metabolites, which directly induce or repress a specific metabolic pathway.

In this review, we frame the nuances in the metabolic reprogramming during tumor progression as metabolic flexibility (the ability to use different nutrients) and plasticity (the ability to process the same nutrient differently; Fig. 1; ref. 5). In addition, we highlight the dynamic adaptability of metabolism during tumorigenesis as a target for improving response to cancer therapy and for overcoming resistance.

As tumors grow and progress, cancer cells face changing microenvironments, which are composed of different nutrients, metabolites, and cell types. This happens at first because, as the tumor proliferates and grows, differently vascularized areas arise, resulting in gradients of oxygen, available nutrients, and in accumulation of tumor-produced metabolites. Consequently, and depending on the tumor type, metabolic intratumor heterogeneity can arise. In this respect, it has been shown that within the same human lung cancer lesions, cancer cells in low perfused areas of the tumor relied on glucose for energy metabolism, whereas cancer cells in higher perfused areas preferred other nutrients, presumably lactate (6).

A similarly high metabolic flexibility has also been recapitulated when switching cancer cells from an in vitro to an in vivo environment (7), or when comparing growth of the same tumor in two different organs (8). Such metabolic flexibility may be inherent to the heterogeneity of every cancer but also might be restricted to the cancer type and specific oncogenic mutations. Accordingly, it was found that in clear cell renal cell carcinoma (ccRCC), which is in 90% of patients defined by Von Hippel-Lindau (VHL) loss and a consequent pseudohypoxic state, cancer cells relied mostly on glucose and low activity of the tricarboxylic acid (TCA) cycle, as compared with lung and brain tumors (9). It remains to be determined whether the low metabolic flexibility in ccRCC is caused by decreased heterogeneity following the predominance of VHL mutations.

Nutrient availability to cancer cells is also influenced by the stroma. Accordingly, it has been observed that pancreatic ductal adenocarcinoma (PDAC) cells feed on alanine released from stroma-associated pancreatic stellate cells (10). Consequently, targeting the neutral amino acid transporter SLC38A2 impaired PDAC growth (11). Similar stroma-associated pancreatic stellate cell–derived lysophosphatidylcholines supported PDAC membrane synthesis and growth signaling (12). These and similar findings highlight that cancer cells can adapt but also trigger the release of certain nutrients from the stromal compartment.

Interestingly, it was further observed that such conditioning of the nutrient environment by stromal cells also happens during premetastatic niche formation. In particular, primary breast tumor–secreted miR-122 impaired glucose uptake in nontransformed cells, elevating glucose availability in the premetastatic niche of the lung, which ultimately increased the permissiveness of the niche for hosting metastasizing cancer cells (13). In line, there is increasing evidence that cancer cell metabolic flexibility can promote aggressiveness, ultimately leading to metastasis formation. Specifically, it was shown that lactate uptake in primary lesions correlated with aggressive oncologic behavior in patients with non–small cell lung cancer (14). Mechanistically, it was found that lactate consumption supported the antioxidant protection of disseminating cancer cells (15), which is important for increasing survival of metastasizing cancer cells in the circulation (16). Similarly, stimulating palmitate consumption in cancer cells, resulting in the expression of the fatty-acid binding protein CD36, boosted the metastatic potential of human oral cancer orthotopic mouse models, and CD36 expression correlated with poor prognosis in multiple cancers (17). In addition, asparagine availability has been found to boost metastasis formation in experimental breast cancer mouse models (18). Hence, nutrient flexibility may contribute to cancer progression.

Interestingly, although flexibility seems widely present in the heterogeneous environment of the primary tumor, metastasizing cancer cells may lose some of this flexibility, creating a dependence on a particular nutrient. In line, it was found that inhibiting the lactate transporting protein monocarboxylate transporter 1 (MCT1) and CD36 in patient-derived mouse models of melanoma and oral squamous cell carcinomas, respectively, impaired metastasis formation but did not affect primary tumor growth (15, 17). Similarly, simply restricting dietary asparagine during cancer cell dissemination also reduced metastasis formation (18). This feature of lost metabolic flexibility seems to be continued when cancer cells reach distant organs. Breast cancer cells that seed in the lung environment remodel the extracellular matrix to create a permissive niche (19). For this activity, collagen prolyl-4-hydroxylase (P4HA) is required. Although this enzyme is highly transcriptionally regulated (20), it was recently discovered that pyruvate uptake from the extracellular space via MCT2 is required for P4HA-dependent collagen hydroxylation and for metastasis outgrowth in breast cancer mouse models (21). Notably, the extracellular pyruvate requirement of cancer cells for remodeling the extracellular matrix of the metastatic niche was dependent not on carbon contribution but rather on regulation via metabolite concentrations leading to alanine aminotransferase (ALT2; also known as GPT2)–dependent production of α-ketoglutarate (21). Moreover, it has been found that proline catabolism serves as an energy source in breast cancer cells colonizing the lung, but not in the corresponding primary tumors (22). Interestingly, dependency on proline catabolism was at least in vitro found only during colonization but not once colonies had formed, suggesting that the dependence on proline catabolism may be a transient event during metastasis formation.

The dependencies on certain nutrients or nutrient inflexibility during metastasis may be explained by several mechanisms. First, a reduced heterogeneity within metastasizing cancer cells may explain nutrient inflexibility; that is, the cancer subpopulation most capable of metastasis formation may inherently not be able to use a certain nutrient. Thus, on the macroscopic level, primary tumors show flexibility due to heterogeneity whereas metastases may have lost it due to reduced heterogeneity. Alternatively, the phenotypic changes required for metastasis formation may result in the dependency of cancer cells on certain nutrients. In this case, nutrient flexibility in established secondary tumors should be similar to primary tumors and thus nutrient inflexibility would be transient. Finally, nutrient inflexibility may be simply induced by the environment, and thus the lack of certain nutrients in a given environment may induce nutrient inflexibility. Further studies are needed to address the nature of nutrient inflexibility in (metastatic) cancer cells. These include a further mechanistic understanding as to what extent metabolic heterogeneity and microenvironment-dependent metabolic inflexibility can explain organ-specific metastasis patterns. An important additional and clinically relevant question is whether metastases regain a certain metabolic flexibility and thus to what extent metastasis prevention versus treatment can be achieved with the mechanisms described above.

Besides metabolic flexibility, metabolic plasticity also contributes to effective cancer progression. Interestingly, the metabolic mode cancer cells use to increase their energy production during metastatic outgrowth seems to be dependent on microenvironment-induced plasticity. Specifically, it was found that breast cancer cells metastasizing to the lung show a PPARγ coactivator 1-alpha (PGC1α)–dependent bioenergetic plasticity in glucose metabolism (23), whereas the same breast cancer cells metastasizing to the liver required glycolytic energy production (24). In the latter case, silencing of pyruvate dehydrogenase kinase 1 (PDK1), which shifts cells from glycolytic to mitochondrial energy production, prevented breast cancer–derived liver but not lung metastasis (24). In addition, cancer cell origin seems to alter the preferred energy production of cancer cells colonizing a distant organ. Accordingly, it has been shown that colorectal cancer cells metastasizing to the liver have the plasticity to scavenge extracellular bioenergetics through secretion of creatine kinase, brain-type (CKB) to the extracellular space (25).

The features of the surrounding extracellular matrix also contribute to cancer plasticity. Whereas nontransformed cells die of anoikis once they lose matrix attachment, metastasizing cancer cells can activate antioxidant pathways to evade anoikis (15, 26–28). In addition, it has been found that matrix detached cancer cells of different origin can fuel glutamine carbon into reductive carboxylation (28, 29), and that basal-like breast cancer cells activate the gluconeogenic enzyme fructose bisphosphatase 1 (FBP1; ref. 30) to increase their reactive oxygen species (ROS) scavenging capacity. In line with these findings, metastasizing breast cancer cells can sustain α-ketoglutarate levels through activation of ALT2 (21), rather than the enzyme glutamine dehydrogenase (GDH), which is frequently used by nontransformed cells and primary cancer cells. Targeting ALT2 consequently greatly impaired metastasis formation in experimental breast cancer mouse models (21).

Although the examples above imply the existence of metabolic plasticity, they could also be explained by a metabolic switch inherently creating a dependency on the newly activated pathway. To further disentangle metabolic plasticity from metabolic switching during cancer progression, metabolic pathways with dual activity in primary tumors could be studied. A prominent example of metabolic plasticity in primary tumors is the use of fatty acid desaturase 2 (FADS2)–mediated sapienate production in parallel to stearoyl-CoA desaturase-1 (SCD1)–mediated palmitoleate production to generate monounsaturated fatty acids from palmitate (31, 32). Here, neither the inhibition of FADS2 nor SCD1 blocks proliferation of certain cancer cells whereas the dual inhibition of both enzymes leads to impairment of proliferation. However, how this plasticity may change during metastasis formation remains to be determined. Thus, it will be important to define comprehensively to what extent true metabolic plasticity (compared with metabolic switching) contributes to metastasis formation and whether this plasticity is transient or stable and therefore can be exploited to prevent and treat metastases.

A recent single-cell assessment of gene expression of metabolic enzymes from tumor tissue revealed a high degree of heterogeneity, whereby the expression of mitochondrial enzymes exhibited the highest variability within the same tumor (33). In support of this finding, a pioneering effort to generate a new sensor for mitochondrial potential, a readout of mitochondrial health in vivo, showed that in human and mouse lung cancer mitochondrial function is highly heterogeneous (34) and correlates with different glycolytic subtypes. Whether this bioenergetic signature changes over time during tumor progression is currently unclear. Yet, it was shown that during cancer cell extravasation, melanoma cells expressing low levels of PGC1α, a master regulator of mitochondrial function, are selected for survival, and when they colonize the lungs, PGC1α levels are reestablished (35). Of note, the PGC1α-low population showed enhanced migration in vitro and metastasis in vivo, whereas PGC1α-high population drives a proliferative phenotype both in the primary tumor and in the metastatic node (35). A similar connection between PGC1α levels, mitochondrial function, and metastasis was observed in prostate and renal cancers, associating with poor outcome (refs. 36, 37; Table 1). These results are in line with the observation that mitochondrial DNA (mtDNA) depletion in human tumors, which is generally associated with bioenergetic defects, is linked with poor patient prognosis in several human cancers (38). Yet, it should be noted that the connection between mitochondrial defects and metastasis does not apply to all cancer types. For instance, in breast cancer it has been demonstrated that metastatic cells exhibit increased mitochondrial biogenesis and respiration (39) and bioenergetic efficiency (ref. 23; Table 1). These results indicate a requirement for PGC1a and mitochondrial activity for metastasis. Consistent with this view, very recent findings indicate that another component of mitochondrial fitness, mitochondrial morphology, is required for angiogenesis and metastasis, and the genetic ablation of OPA1, a master regulator of mitochondrial fusion, abolished metastasis in mouse models of melanoma (40).

From a mechanistic point of view, the connection between mitochondrial function and cancer progression has been elusive. A seminal paper in 2008 provided the first line of evidence that replacing mitochondria with mitochondria from a metastatic cell line could transfer its aggressiveness (41). Here, they ascribed the increased aggressiveness provided by metastatic mitochondria to increased oxidative stress. However, the putative roles of ROS in metastasis have been challenged and appear more complex and dependent on tumor stage (see below). Another hypothesis is that the metabolic changes that arise from a specific mitochondrial dysfunction could support metastasis. In support of this scenario, we recently showed that the gradual increase in mtDNA heteroplasmy of an mtDNA mutation of ATP6 is directly associated with epithelial–mesenchymal transition (EMT) and increased migration (42). We proposed that the loss of mitochondrial function caused by high levels of heteroplasmy activates glycolysis and the coupling between the glycolytic enzyme GAPDH and the enzyme malate dehydrogenase 1 (MDH1), which we found to colocalize with the cytoskeleton. Although the molecular details have yet to be defined, it is possible that mitochondrial dysfunction could prompt a metabolic rewiring that facilitates cell motility and migration. Nevertheless, it should be underscored that rather than a complete dysfunction, the extent of mitochondrial dysfunction we assessed was limited to 80% of mtDNA heteroplasmy, and cells still exhibited mitochondrial-dependent respiration. Therefore, it is possible that a stronger mitochondrial defect could be counterproductive for cancer cell growth and motility. Indeed, mutations that significantly hamper mitochondrial function could be detrimental to cancer cells, as indicated by the relatively benign nature of oncocytomas, tumors characterized by a significant mitochondrial suppression (43), and by the fact that tumors generated by cells devoid of mtDNA take up healthy mitochondria from neighboring cells to support pyrimidine biosynthesis (44). Consistent with this view, maintaining an appropriate turnover of mitochondria via autophagy is essential for cancer, and when autophagy is inhibited, the accumulation of unhealthy mitochondria leads to cancer cell demise (reviewed in ref. 45).

Another possible advantage that mitochondrial dysfunction provides to cancer could be induction of an anabolic/antioxidant metabolic rewiring that supports growth and survival under harsh environmental conditions by reducing cancer dependency on oxygen consumption for ATP generation. Indeed, a reduction in mitochondrial respiration together with a parallel increase in glycolysis can enable cancer cells to survive when the tumor is poorly vascularized and the supply of oxygen becomes limited. This hypothesis is supported by the observation that activating oxygen consumption in cancer cells by the expression of uncoupling protein 1 can reduce cancer cell survival (46). A corollary to this metabolic reprogramming is the accumulation of metabolites that have signaling roles and can elicit a phenotype switch that supports survival and metastasis. For instance, it was shown that the accumulation of mitochondrial metabolites such as 2HG, fumarate, and succinate, which are known to increase under conditions of poor oxygenation, could trigger EMT (reviewed in ref. 47).

It is also possible that dysregulation of mitochondrial function during tumor progression has non–cell autonomous functions and can affect the tumor microenvironment. Indeed, tumors that are deficient for Complex I, despite their slow proliferation, induce macrophage infiltration into the tumor and increase tumor malignancy (48).

Finally, ROS have long been considered unwanted by-products of mitochondrial metabolism. More recently, ROS have been suggested to function as important signaling molecules, implicated in many diseases including cancer. However, their role in tumorigenesis is far from clear, due to technical challenges in their detection and modulation, especially in vivo, and due to the fact that their function seems to depend on tumor stage. During the early phases of tumorigenesis, ROS appear to be mutagenic, and therefore they support transformation. Recent evidence indicates that ROS increase upon transformation but their levels are kept in check by antioxidant programs such as that orchestrated by NRF2 (49). Independent experiments showed that mitochondrial ROS are essential for KRAS-mediated tumorgenicity and anchorage-independent growth (50). Of note, cells deficient in mtDNA do not generate ROS and fail to grow in an anchorage-independent manner (50). Therefore, evidence indicates that ROS increase upon transformation and support oncogenic functions, but their lethal effects need to be counteracted by antioxidant programs. Their role in tumor progression is even more debated. The increased ROS generation by dysfunctional mitochondria was initially linked with cancer metastasis in the above-described mitochondria swap experiment between normal and highly aggressive cancer cells (41). Of note, antioxidants appeared to reduce the metastatic potential of these cybrid cancer cells in vivo. Only a year later, it was reported that during cell detachment, one of the key steps of tumor progression that precedes metastasis is that cells experience a burst of oxidative stress, which, if left unchecked, can lead to cell demise (51). Of note, antioxidants appeared to increase the chances of survival of detached cells. To sum up these two lines of evidence, on one hand, ROS increases malignancy; on the other hand, too much ROS is toxic for cancer cells detached from their matrix, with an opposite metastasis-promoting effect of antioxidants. Of note, it is still unclear why cells that detach from the matrix experience this burst in ROS, but it is possible that changes in metabolism elicited by alterations of cellular mechanics are responsible for it.

These two apparently contradictory pieces of evidence revealed that the role of ROS in cancer progression is likely context-dependent. For instance, in the effort to elucidate the role of ROS in tumor progression, Porporato and colleagues found that mitochondrial-derived superoxide, caused by either the suppression of mitochondrial function or mitochondrial overload, increases aggressiveness and metastasis in vivo (52). Of note, the authors found that this increased migration involved the protein tyrosine kinases SRC and PYK2 as downstream effectors and was blunted by mitochondria-specific antioxidants. However, in a model of melanoma, it was found that metastasis activates an antioxidant program to survive and antioxidants increase the efficiency of metastasis (26). Here, the authors concluded that oxidative stress limits the formation of distant metastasis in vivo. Consistent with this view, it was found that melanoma metastases rely on lactate not only as an energy substrate but also as a source of reducing power via the activation of the oxidative arm of the pentose phosphate shunt (15). This work is consistent with the finding that antioxidants increase rate of metastasis in melanoma in preclinical settings (27). Overall, a scenario is emerging whereby ROS affect cancer cells in a dose- and stage-dependent fashion. During the early phases of tumorigenesis, increased ROS, possibly caused by dysregulation of mitochondrial function, could activate signaling cascades that promote transformation. Of note, at this stage, ROS could further shape the fate of early tumors by causing DNA damage and genome instability (53). During this phase, the antioxidant capacity of the cells may be sufficient to avoid cell death. Yet, when the tumor grows and cells start to detach from the matrix, cells experience a burst of oxidative stress. Cells that succeed in responding to this wave of oxidative stress have the ability to extravasate and efficiently metastasize. At both stages, the presence of antioxidant programs is essential to avoid cell death. In support of this view, it was recently shown that in pancreatic cancer the levels of the antioxidant protein TIGAR vary during tumor progression. During the early stages, high TIGAR levels are needed to cope with the oxidative stress caused by transformation. Yet, as the tumor progresses, decreasing levels of TIGAR appear to increase the malignancy of cancer cells, consistent with the selection for cells with higher ROS and higher invasive capacity. At a later stage, though, TIGAR levels go up again to buffer the oxidative stress experienced by metastatic cells (54).

Overall, these accumulating data suggest that mitochondrial function plays a key role during tumor progression and that an important feature lies in the ability of the mitochondrial function level to be context-specific. Therefore, an attractive therapeutic strategy could be to target mitochondrial potential for metabolic adaptability.

Importantly, the host metabolism affects tumor metabolic adaptability potential, primarily by one's genetics. In congenital cancer predisposition syndromes, the kinetics of cancer development is promoted by an inherent metabolic rewiring caused by constitutive activation of signaling pathways and transcriptional programs that regulate anabolic metabolism, that is, RASopathies, Li–Fraumani syndrome, and Cowden syndrome. In some syndromes, such as Proteus and Beckwith–Wiedemann syndromes, the uncontrolled cellular growth manifests as overgrowth of a specific tissue or of the whole body, as well as in tumor predisposition (55). Several cancer predisposition syndromes involve mutations in metabolic genes that cause cell toxicity leading to cancer, that is, the development of hepatocellular carcinoma in patients diagnosed with tyrosinaemia type I (56, 57), or the accumulation of metabolites with oncogenic activity. For instance, in the hereditary cancer syndromes caused by germline mutations in SDH or FH, accumulation of the oncometabolites fumarate and succinate, respectively, have been proposed to cause cancerous transformation (58, 59). To what extent the host metabolism predisposes to cancer in these tumor predisposition syndromes is currently unknown. In the case of SDH/FH-deficient tumors, patients are heterozygous for SDH/FH mutations and it will be important to determine whether the remaining wild-type allele is sufficient to maintain host physiology. It is possible that in some tissues the heterozygous loss of SDH/FH might give rise to mitochondrial dysfunction and/or mild accumulation of fumarate/succinate, affecting host metabolism and, possibly, impinging on the ability of the immune system to eradicate SDH/FH-deficient clones. Interestingly, mitochondrial dysfunction syndromes pose an increased cancer risk, as can be seen in mitochondrial depletion syndromes (60).

Some congenital metabolic syndromes involve chronic changes in metabolic flux that recapitulate similar changes to those observed in cancer cells, including enhanced pentose phosphate pathway (PPP) activity, lactate production, and high synthesis of lipids and nucleotides, predisposing to tumorigenesis. For example, in glycogen storage disease type 1a, the deficiency in the glycolytic glucose-6-phosphatase complex leads to hypoglycemia, lactic acidosis, hyperlipidemia, and increased shunting via the PPP (61).

In addition, changes in host metabolic capacities due to aging, diet, health status, and physical activity can affect cancer progression either directly or indirectly via the microbiome, which can generate metabolites affecting tumorigenesis (62). With age, degenerative changes in the host metabolism create an inflexible and less fertile environment, which may drive tumors to metastasize (63). Importantly, the surrounding fibroblasts in aged patients with melanoma secrete high amounts of lipids that enable drug resistance; the rich lipid tumor microenvironment promotes the upregulation of the tumor fatty acid transport protein 2 (FATP2), which supports mitochondrial metabolism and cancer cell survival under therapy-induced stress (64).

The host health status is a crucial determinant of its cancer risk (65). In patients with obesity, cancer development is facilitated by metabolic reprogramming caused by the prevalence of diabetes and insulin resistance, leading to IGF1 overproduction and to enhanced shunting of glycolytic intermediates via the PPP (66, 67). Conversely, a calorie-restriction diet, which likely constrains metabolic adaptability, affects the signaling of IGF1, PI3K, PTEN, and mTOR, resulting in inhibition of cancer growth. Interestingly, alterations in the circadian clock, which centrally regulates daily rhythms of cellular metabolism, have been shown to predict poor survival in patients with cancer and to associate with increased incidence of several cancers in shift workers (68). Furthermore, perturbations of circadian clock components led to increased c-MYC expression and to metabolic dysregulations, consequently promoting lung tumor progression and decreasing survival (69).

Hence, host fitness determines one's capacity to accommodate cancer demands for nutrients, thus affecting cancer metabolic adaptability.

In the previous sections, we have highlighted some of the most investigated metabolic pathways and how they contribute to tumor initiation and progression toward metastasis. However, the determinants of these metabolic changes are only poorly understood. Are they intrinsically present in the original tumor, or are they acquired during tumor evolution? We propose the following two scenarios to explain the adaptability of cancer (Fig. 2). In the first scenario, an initial tumor mass is composed of phenotypically identical clones. Depending on nutrient and oxygen availability, and likely external forces, some cells will undergo a phenotypic switch that prepares them to face the new environment. Only cells that have sufficient metabolic flexibility and plasticity will survive, extravasate, and contribute to metastasis. In this scenario, the adaptability can be driven by (epi)genetic changes that modulate the expression of metabolic genes. In addition, external cues, such as nutrients or growth factors, can further regulate these metabolic enzymes. Interestingly, chromatin structure and function depend on the availability of ATP, methyl donors, and other metabolites (aKG, succinate, fumarate, etc.) that regulate chromatin modifiers. Therefore, it is possible that when a group of cancer cells experience changes in nutrient availability, the resulting epigenetic and transcriptional response is what drives their adaptation phase. In this scenario, the resulting metabolic phenotype is plastic and can be reverted or further changed depending on the new metabolic niche the cancer cell will experience. Such a scenario is supported by experiments showing that consecutive reimplantations of secondary lesions into the primary site reestablish the original metastatic phenotype (26).

The second scenario relies on the intrinsic phenotypical heterogeneity of the tumor mass. Here, we postulate that within a tumor mass, genetically identical cancer cells exhibit an intrinsic variability in their metabolic phenotype. Although this intrinsic metabolic heterogeneity can be initially “neutral,” that is, it does not enhance growth or survival in the primary tumor, it may give rise to clones that can adapt to a new, harsher metabolic environment. It is possible that only a few cells within the tumor tissue will have the appropriate metabolic configuration that enables survival, but these will be sufficient to either invade the tissue or extravasate. This scenario is supported by some evidence. For instance, it has been demonstrated in silico that metabolic changes during evolution may be an indirect consequence of the inherent structure of a given metabolic network, which provides flexibility to use other energy sources. For instance, the metabolic network used for growth on glucose could enable the use of many more carbon sources when glucose becomes scarce (70). In addition, almost a decade ago it was demonstrated that mutant RAS could generate subclones that express high levels of the glucose transporter GLUT1. This condition, which does not provide a growth advantage when glucose is available, provides unique advantages to the mutant cells when glucose becomes scarce (71). Although it is unclear how this heterogeneity in the expression of GLUT1 is maintained, it is possible that epigenetic mechanisms govern it. We speculate that additional mutations that “fix” this metabolic configuration might arise in the clonal population, allowing for the emergence of a stable clone. It is possible that the genetic heterogeneity observed in tumors arises as a combination of nongenetic preadaptations and subsequent fixation of the phenotype by mutations in the genome.

In addition, recent articles suggest that it is the existence of cancer stem cells with inherently high metabolic plasticity that generates resistance to therapy by allowing dynamic transitioning between different metabolic phenotypes, enabling tumors to regain growth following therapy (72). Indeed, it has been reported that cancer stem cells are able to switch to a glycolytic metabolism when oxidative phosphorylation (OXPHOS) is blocked (73). Interestingly, a combined analysis of biological, biochemical, pharmacologic, and genetic studies revealed that cancer stem cells' stemness may arise from metabolic events occurring in noncancer stem cells. Specific metabolic hits are thought to affect chromatin organization and activate epigenetic programs involved in the metabolic-driven reprogramming of cancer stem cells (74). According to this, the identification of key metabolic processes involved in this reprogramming might be useful to identify and target cancer stem cell survival.

Tumor metabolic plasticity and flexibility contribute to resistance in most types of anticancer therapy. Undoubtedly, one of the main contributors to this metabolic adaptability potential is genetic heterogeneity. Here, resistance evolves by clonal selection of a specific signaling pathway that promotes the required metabolic rewiring that can meet the stress imposed by the drug. Because most anticancer therapies target the uncontrolled proliferation of cancer cells, the purpose of the compensatory metabolic reprograming is to restore cancer cell survival and growth. This is best exemplified in chemoresistance caused by plasticity in glucose metabolism. For example, following cisplatin chemotherapy, several key glycolytic enzymes as HK2, PFK, and PKM2, and glucose transporters as GLUT1, are induced in cervical cancer by activating adenosine monophosphate–activated protein kinase (AMPK) signaling (75, 76). This augmentation in glycolysis results in high levels of glycolytic intermediates for branching pathways such as the PPP that support nucleotide synthesis and redox homeostasis. Importantly, the resultant high lactate secretion generates a hypoxic microenvironment that limits drug entry into the cells. In addition, increased glucose consumption by upregulation of glucose transporters cues the cell to glucose deprivation and activates the stress machinery to induce autophagy and escape apoptosis (77). In parallel, AMPK also promotes glutamine metabolism, which by itself contributes to chemoresistance by supplying substrates to the TCA cycle to preserve mitochondrial function and support cancer cell survival (78). Indeed, combining the chemotherapy cisplatin with metabolic inhibitors of glycolysis, such as 3-BrPA (3-bromopyruvate), a specific inhibitor of HK2 kinase, increases chemotherapy efficacy (79, 80). Increasing mitochondrial metabolism can also provide an escape route for cancer cell survival from the effects of therapy; more than half of patients with melanoma with BRAF mutations and a consequent MAPK activation develop resistance to MAPK inhibitors that is dependent on mitochondrial OXPHOS (81). Indeed, therapy combining a MAPK inhibitor with gamitrinib, a small molecule targeting the mitochondria, augmented the efficacy of MAPK inhibitor treatment in melanoma cells by inducing mitochondrial dysfunction and inhibiting tumor bioenergetics (81).

The anticancer mechanism of antiangiogenic therapy is to reduce tumor vascularity and cause tissue hypoxia. Yet, inducing hypoxia triggers upregulation of HIF1 that can lead to worse outcomes in terms of resistance (82). HIF1 induces the expression of glycolysis-related genes such as GLUT1, GLUT3, PDK1, PKM2, PFKFB3, GYS1, ENO1, LDHA, HK2, and GAPDH, again enhancing glycolysis and its branching pathways (83). In addition, hypoxia in the tumor microenvironment leads to lipolysis and the release of free fatty acids, while, in parallel, hypoxia increases fatty-acid uptake by the tumor via upregulation of the fatty-acid importer CD36 expression. Within tumor cells, hypoxia increases glutamine uptake, providing substrates for the TCA cycle for the synthesis of citrate and for ATP synthesis by OXPHOS. The resultant increase in ATP and metabolite levels supports tumor proliferation and contributes to cancer resistance (84). Here again, combined therapy of nintedanib and 3PO, a selective glycolytic inhibitor of 6-Phosphofructo-2-kinase/fructose-2,6-bisphosphatase (PFKFB3), was synergistic in inhibiting tumor growth in a breast cancer model (85). Although the increase in glycolysis may be a general metabolic strategy for chemoresistance, other tumors including melanoma and hematologic malignancies appear to increase OXPHOS in resistant clones (reviewed in ref. 86). For instance, in melanoma, the increase in OXPHOS in resistant clones is supported by PGC1α and is needed to buffer oxidative stress (87). In line with this finding, in chronic myeloid leukemia, targeting mitochondrial function can eradicate therapy-resistant cells (88).

In some cancer therapies that target a selected metabolic vulnerability, resistance may occur when the nutrient circumstances change following therapy. Under the new conditions, cancer cells benefit from selecting for clones with reprogramming of the targeted metabolic pathway. For example, in multiple human malignancies including melanoma, hepatocellular carcinoma, prostate cancers, osteosarcoma, and mesothelioma, argininosuccinate synthetase 1 (ASS1) is silenced by epigenetic methylation. ASS1 is a urea cycle enzyme that outside the liver participates in the citrulline–arginine cycle for the generation of arginine and its downstream metabolites (89). Silencing of ASS1 increases the availability of its substrate, aspartate, for the synthesis of pyrimidines that are utilized for DNA and RNA synthesis that supports cell proliferation (90). Tumors with ASS1 silencing are hence proliferative but in parallel become arginine auxotrophic, meaning they cannot synthesize arginine and require extracellular arginine supplementation for survival. This metabolic vulnerability is taken advantage of for therapy; arginine starvation agents such as PEGylated arginine deiminase (ADI-PEG20) and human arginase 1, which degrade extracellular arginine, are in various stages of clinical trials (91). Yet, resistance to arginine-degrading treatment develops and involves reexpression of ASS1 caused by binding of MYC to its promoter (92). Such evolving resistance exemplifies the advantage of epigenetic flexibility over genetic rigidity determined by mutations.

Importantly, the tumor microenvironment can provide tumors with metabolites that enable metabolic plasticity and flexibility, leading to cancer resistance. Increasing autophagy and exosome secretion in the microenvironment can provide essential metabolites such as amino acids, fatty acids, and nucleic acids to support the metabolic flexibility required for cancer cell survival and growth under nutrient deprivation (93). The exosome cargo can also provide metabolic plasticity that affects tumor progression by carrying exosome-associated miRNAs to the tumor cell. For example, in malignant mesothelioma cells, miR-126, an angiogenesis inducer, can regulate cancer metabolism by decreasing the levels of its downstream target insulin receptor substrate-1, causing upregulation in the expression of oxidative stress defense and gluconeogenesis genes (94). The resultant increase in glucose yields a glycolytic shift, which supports a less malignant mesothelioma phenotype (95).

Thus, metabolic adaptability is an important component in cancer resistance to therapy, and targeting it directly will likely be therapeutically beneficial. An attractive therapeutic approach could be to exploit these metabolic features to enhance therapy by targeting cancer adaptive mechanisms, in particular cancer heterogeneity and the specific signaling molecules that enable metabolic rewiring.

A reasonable strategy to target tumor heterogeneity is based on identifying and targeting driver mutations (e.g., in BRAF) or a signaling pathway shared by multiple mutations in the same cancer (e.g., mTOR). Here, although a single drug can target multiple clones, the majority of patients eventually develop resistance, and hence combining these signaling inhibitor drugs with metabolic inhibitors will likely be more efficient (96, 97). Indeed, concurrent inhibition of BRAF and glycolysis, or a combination of MAPK and mitochondrial inhibitors, induces cell death in BRAF inhibitor–resistant melanoma cells (81, 98).

Another approach to diminish tumor heterogeneity is to stress tumors to develop a metabolic dependency that sensitizes tumor cells to specific therapies. For example, methionine is essential for protein synthesis, one carbon metabolism and nucleotide synthesis, gene regulation by DNA methylation, and redox metabolism, which are all essential for carcinogenesis (99). Restricting methionine by diet has been shown to sensitize PDX models of colorectal cancer to chemotherapy with 5-fluorouracil (5-FU; ref. 100). To identify drugs that are relevant for implementing this approach, a metabolic sensitivity assay can be added to high-throughput drug screens.

Another strategy for targeting cancer adaptability potential is to target cancer stem cells by, for example, using antibodies against cancer stem cell–specific cell surface markers such as CD20 and CD52 (101). Similarly, attempts at targeting specific proteins that enable the switch between different metabolic states may also prove to be an attractive strategy. For example, the HSP90 molecular chaperone controls metabolic rewiring either through direct binding to chromatin or via control of transcription factors and epigenetic effectors (102). Indeed, HSP90 interacts with and modulates several signaling pathways involved in metabolic plasticity including MYC, HIF1α, and AKT/PKB. HSP90 can also influence cancer metabolism by directly binding glycolytic enzymes such as GAPDH and PKM2. Specifically, the HSP90 mitochondrial isoform TRAP1 stabilizes the binding of HK2 to the mitochondrial voltage-dependent anion channel (VDAC), maximizing its activity, and also binds and inhibits the activity of the respiratory chain complex II [succinate dehydrogenase (SDH); ref. 103]. Likewise, reintroducing wild-type p53 or inhibiting c-MET might decrease metabolic flexibility potential (104).

There is no doubt that tumor evolution is determined mostly by the pressure to supply its metabolic needs under changing environmental metabolic stresses. This ability for metabolic adaptation to fluctuating stresses should hence be regarded as tumors' metabolic Achilles' heel, and therapies targeting this should be included in the future arsenal against cancer (Fig. 3).

Notably, cancer cells' survival mechanisms such as metabolic adaptability are hijacked from normal cells. The signaling and metabolic pathways providing this metabolic adaptability are all used by normal cells, and especially by proliferating cells, under physiologic conditions. Hence, in targeting cancer's metabolic plasticity and flexibility features, it may be challenging to balance between causing side effects and developing therapy resistance. Along these lines, accumulating articles suggest changing the goal of cancer management from cure to chronic disease. Here, the idea of treatment is to avoid inducing extreme stress on cancer cells that presumably select for more aggressive clones, and rather aim to restrict and control cancer growth. For this, one would need to continuously identify targetable molecular changes and tailor treatments that are predicted to be most effective against resistance and relapse. This may be achieved by monitoring cancer burden with imaging together with following the precise molecular signature of the evolving cancer using liquid biopsies and repeated sequencing (105).

Undoubtedly, more preclinical research and clinical trials must be completed before such interventions become common practice in cancer therapy. In the optimal scenario, combinatory drugs will abolish cancer's adaptability potential, whereas in the more realistic scenario, we should aim to restrict the ability of cancer to adapt as much as possible.

S.-M. Fendt reports funding from Bayer (fee for service), Merck (fee for service), Black Belt Therapeutics (fee for service), and from Fund + (consulting) outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

We acknowledge and thank the Weizmann Institute for providing financial and infrastructural support. A. Erez is incumbent of the Leah Omenn Career Development Chair and is supported by research grants from the European research program (ERC818943) and from the Israel Science Foundation (860/18). A. Erez received additional support from The Moross Integrated Cancer Center, Sagol Institute for Longevity Research, Adelis Foundation, Rising Tide Foundation, and from Manya and Adolph Zarovinsky. C. Frezza's work is funded by the MRC Core award grant MRC_MC_UU_12022/6, the ERC (ERC819920), and CRUK Programme Foundation award C51061/A27453. S.-M. Fendt acknowledges funding from the ERC Consolidator Grant Agreement no. ERC771486, FWO – Research Projects (G098120N and G088318N), KU Leuven – Methusalem Co-Funding, and Fonds Baillet Latour.