Approaches to Study Tumor Metabolism in Mice

Abstract

Altered metabolism is a hallmark of cancer that reflects the increased energetic and biosynthetic demands of proliferating cancer cells. Accumulating evidence suggests that many aspects of tumor metabolism in vivo differ from those of cancer cells in culture, and that the tumor microenvironment plays a major role in driving these differences. Apart from cancer cells, tumors are composed of diverse cell types including vascular cells, fibroblasts, and immune cells. Nutrient sharing and competition between different cell types and access to nutrients in tumor tissue are not well modeled in standard monoculture systems. Tumors can also interact with whole body metabolism and share nutrients with peripheral tissues such as fat and muscle. Hence, studying some aspects of cancer metabolism in mouse models is useful for examining the complex metabolic interactions between cells and tissues. In this article, we discuss how advances in mouse modeling have enabled studies of tumor metabolism in vivo, providing insight into metabolic factors that influence cancer cell behavior in the physiological context of a whole organism. These studies are providing insight into how to leverage altered cancer metabolism for improved therapy.

Metabolic alterations in tumors relative to healthy tissues support tumor growth and cancer progression. These metabolic changes are driven both by dysregulated growth signaling and by how cells utilize available nutrients to adapt and proliferate in the conditions found in specific tissues. Early studies by Warburg in the 1920s using rat tumor slices demonstrated that many tumors increase glucose uptake and ferment much of that glucose into lactate even in the presence of oxygen, a phenomenon known as the Warburg effect or aerobic glycolysis (Warburg et al. 1927; Warburg 1956). The increased conversion of glucose to lactate by tumors in vivo was confirmed in live chickens bearing Rous sarcoma virus-driven cancer (Cori and Cori 1925), and the increased glucose uptake associated with many cancers has subsequently been leveraged via FDG-PET imaging as both a staging tool and a marker of therapy response in patients (Engelman et al. 2008; Ben-Haim and Ell 2009; Almuhaideb et al. 2011). This cancer-specific metabolic rewiring suggests that cancer metabolism can be targeted for therapy, and indeed antimetabolite chemotherapies are a mainstay of treatment for many malignancies (Wang et al. 2013; Luengo et al. 2017). Nevertheless, few therapies have been intentionally developed to take advantage of altered tumor metabolism to treat cancer.

Over the past decade, a resurgence of interest in cancer metabolism has led to additional insights into how glucose metabolism is reprogrammed in cancer cells. It is now appreciated that aerobic glycolysis supports the utilization of glucose carbon for production of glycolytic and tricarboxylic acid (TCA) cycle intermediates needed to synthesize the lipids, amino acids, and nucleotides cells need to proliferate (DeBerardinis et al. 2007; Jones and Thompson 2009; DeBerardinis and Chandel 2016; Vander Heiden and DeBerardinis 2017). Studying how this macromolecule synthesis is regulated has further expanded our understanding of cancer metabolism beyond glucose metabolism. For example, cancer cells adopt different metabolic strategies for using amino acids. Many cancer cells consume significant amounts of the amino acid glutamine in culture (Nicklin et al. 2009), which serves as a nitrogen donor for the synthesis of amino acids and nucleotides and supplies anaplerotic carbon to the TCA cycle (DeBerardinis et al. 2007; Jones and Thompson 2009; Wise and Thompson 2010). Some cancer cells also depend on exogenous serine to proliferate (Locasale et al. 2011; Possemato et al. 2011). Serine is a nonessential amino acid (NEAA) that can be synthesized from glycolytic intermediates through the serine synthesis pathway or be taken up from the environment (Snell 1984; Snell and Weber 1986; DeBerardinis 2011; Amelio et al. 2014). Beyond its requirement for protein, serine is needed for many biosynthetic pathways including phospholipid synthesis (Spector and Yorek 1985; Hickman et al. 2011; Amelio et al. 2014), sphingolipids (Muthusamy et al. 2020), nucleotide synthesis, and methylation reactions (Tibbetts and Appling 2010). As a final example, increased fatty acid synthesis is important for the growth of some tumors (Medes et al. 1953; Swinnen et al. 2006; DeBerardinis and Thompson 2012; Santos and Schulze 2012), while other cancers are dependent on uptake of fatty acids from their environment (Currie et al. 2013). Cytoplasmic acetyl-CoA is used to synthesize palmitate, the initial product of de novo fatty acid synthesis, which can then be further elongated and desaturated to produce various saturated and monounsaturated fatty acids (Currie et al. 2013). These fatty acids are then used for the synthesis of complex lipids, which supports the membrane biogenesis needed for cell proliferation.

The mechanisms that drive metabolic rewiring in cancer cells are being actively investigated, but there is extensive evidence that cancer-promoting genetic alterations in oncogenes and tumor suppressor genes involved in growth signaling regulate metabolic pathways in cancer cells. For instance, the MYC proto-oncogene regulates the expression of many enzymes involved in glycolysis, the TCA cycle, the mitochondrial electron transport chain, nucleotide synthesis, fatty acid synthesis, and the uptake and metabolism of amino acids (Stine et al. 2015; DeBerardinis and Chandel 2016). Activation of receptor tyrosine kinases also influences carbohydrate and fatty acid metabolism via activation of the PI3K/Akt, RAS/MAPK, and Wnt/β-catenin signaling pathways through a variety of mechanisms (Sever and Brugge 2015). Loss of p53 affects the metabolism of both glucose and amino acids, including serine (Puzio-Kuter 2011; Labuschagne et al. 2018; Tajan et al. 2018; Humpton and Vousden 2019). Furthermore, some cancer cells possess genetic alterations in genes encoding metabolic enzymes. For instance, amplification of the genomic locus on human chromosome 1p12 encoding the serine synthesis pathway gene phosphoglycerate dehydrogenase (PHGDH) is found in some breast cancers and melanomas (Beroukhim et al. 2010), and high PHGDH expression is important for the growth of these tumors (Locasale et al. 2011; Possemato et al. 2011). Studies in mouse models argue increased PHGDH expression is particularly important for tumors growing in tissues where serine levels are limiting (Maddocks et al. 2017; Sullivan et al. 2019b). Another prominent example is mutations in isocitrate dehydrogenase (IDH), which generate the oncometabolite 2-HG and contribute to the pathogenesis of some cancers (Parsons et al. 2008; Dang et al. 2010; Andronesi et al. 2012). Engineering IDH mutations into mouse models can lead to premalignant lesions, and these findings helped motivate efforts to develop IDH inhibitors to treat IDH-mutant malignancies (Sasaki et al. 2012; Wang et al. 2013; Hirata et al. 2015; Fortin et al. 2023).

Emerging evidence suggests that genetic alterations, while important, are not sufficient to define all cancer-associated metabolic changes. For example, KRAS mutations are present in ∼25% of tumors, making them one of the most common genetic mutations linked to lung, colorectal, and pancreatic cancers. Oncogenic KRAS activation is a crucial event that acts as a molecular switch to activate various intracellular signaling pathways related to nutrient sensing and metabolism as well as transcription factors regulating cell proliferation, migration, transformation, and survival. Kras activation and p53 deletion in the mouse pancreas or lung initiates pancreatic ductal adenocarcinoma (PDAC) or non-small-cell lung carcinoma (NSCLC), respectively (Tuveson and Jacks 2002; Hingorani et al. 2005; Walrath et al. 2010; Young et al. 2011; Kersten et al. 2017). Therefore, modeling PDAC and NSCLC in the mouse is valuable for learning more about nutrient usage in the tumor. Two examples of Cre recombinase-dependent genetically engineered mouse models (GEMMs) of PDAC include LSL-Kras^G12D/+Trp53^−/− (KP^−/−C) and LSL-Kras^G12D/+Trp53^R172H/+ (KPC) (Hingorani et al. 2003, 2005; Bardeesy et al. 2006). Those Cre-dependent alleles can be activated in the pancreas by Pdx-1-Cre, Ptfa-1-Cre (p48), or their tamoxifen-inducible alleles (Kawaguchi et al. 2002; Hingorani et al. 2003; Nakhai et al. 2007; Kopinke et al. 2012). For NSCLC GEMMs, Cre recombinase adenovirus (Ad-Cre) can be injected into the lung to activate the same Kras and Trp53 alleles to initiate lung tumors (DuPage et al. 2009). Interestingly, when comparing PDAC versus NSCLC tumors with the same Kras and Trp53 mutations, NSCLC tumors are dependent on metabolism of branched chain amino acids (BCAAs) for growth, whereas BCAA metabolism is dispensable for PDAC tumor growth (Mayers et al. 2016). Similarly, NSCLC cancer cells cultured in standard media are dependent on glutamine to proliferate, while the same cells growing as tumors in vivo are less dependent on glutamine metabolism (Davidson et al. 2016). In fact, most cells in culture exhibit some glutamine dependence, while other mouse cancer models are also less dependent on glutamine (Yuneva et al. 2012; Son et al. 2013; Sellers et al. 2015; Davidson et al. 2016). Cancer cells also require access to specific nutrients to proliferate in some tissues as metastases even though they are less dependent on those nutrients in the primary tumor (Tasdogan et al. 2020; Ferraro et al. 2021; Rinaldi et al. 2021; Rossi et al. 2022; Sivanand et al. 2022; Altea-Manzano et al. 2023). Together, these data argue that the in vivo tumor microenvironment and the cancer tissue of origin are both determinants of how metabolism is constrained in proliferating cancer cells. Critically, results like these illustrate the importance of studying tumor metabolism in the appropriate physiological context that is provided by mouse models.

In this article, we discuss how mouse models have enabled studies of tumor metabolism. Various approaches for examining tumor metabolism in mice are providing insight into how the tumor microenvironment plays a critical role in defining cancer cell metabolic dependencies. These include using in vivo stable isotope nutrient labeling to monitor metabolic pathway activity, developing methodologies to study metabolic interactions between cancer and noncancer cells within tumors, probing interactions between tumors and normal tissues, and developing cell culture systems that model physiological nutrient levels. Applying these approaches is providing greater insight into the metabolic factors that influence cell behavior in the physiological context of the whole organism and could lead to new strategies to improve cancer therapy.

STABLE ISOTOPE METABOLITE LABELING CAN BE USED TO PROBE METABOLIC ACTIVITY IN VIVO

One tool to assay metabolism in vivo is tracing the fate of stable isotope-labeled metabolites. This approach involves administering ¹³C, ²H, and/or ¹⁵N-labeled nutrients to animals followed by tissue harvest and mass spectrometry to measure label uptake and incorporation into various metabolic pathways. This technique has been used extensively to study both normal tissue and tumor metabolism (Davidson et al. 2016; Hui et al. 2017; Spinelli et al. 2017; Jang et al. 2019; Lau et al. 2020; Tasdogan et al. 2020; Bartman et al. 2021, 2023; Faubert et al. 2021; Dong et al. 2022; Nascentes Melo et al. 2022; Rossi et al. 2022).

Stable isotope metabolite labeling has predominantly been used to consider what fuels are consumed by tumors. These experiments typically involve infusion of a labeled metabolite to isotopic steady state within tissues of interest. For example, the decreased glutamine dependency of some NSCLC tumors in mice, described above, was revealed in part by in vivo ¹³C-glutamine tracing. While ¹³C-glutamine highly labels TCA cycle metabolites in cultured NSCLC cells, it contributes minimally to the TCA cycle in tumors derived from Kras-driven autochthonous mouse models of NSCLC (Davidson et al. 2016). Subsequent studies have highlighted other metabolites as potential novel fuel sources for tumors. For instance, infusion of ¹³C-lactate into tumor-bearing animals or human patients resulted in substantial labeling of TCA cycle intermediates, suggesting that some tumors may consume lactate as a carbon source (Hui et al. 2017; Faubert et al. 2020, 2021; Bartman et al. 2023). Similarly, injecting mice with ¹⁵N-ammonia led to labeling of amino acids in breast cancer xenografts, leading to the conclusion that these tumors can utilize ammonia (Spinelli et al. 2017). However, because many metabolic reactions are rapidly reversible, including transport of metabolites in and out of cells, it is possible for cells to incorporate the isotopic label from a given metabolite without net consuming that metabolite, a phenomenon termed exchange flux (Locasale et al. 2011; Shestov et al. 2014; DeBerardinis and Chandel 2016; Muir et al. 2018; Ying et al. 2019; Liu et al. 2020). Thus, label incorporation into downstream metabolites alone may not provide conclusive evidence of net metabolite consumption or production, and additional experiments are needed to assess net consumption or production of metabolites and understand fuel choice across different cancers. Measurement of arteriovenous metabolite differences across a tumor is a classic approach that could be informative in this case but is technically challenging in mice (Jang et al. 2019). Adding further complexity, it is possible that different tumor cell subsets, such as cancer cells versus noncancer cells, may have distinct fuel preferences that will not be captured in bulk tumor measurements. This was proposed at least in the case of lactate (DeBerardinis and Chandel 2016; Hui et al. 2017; Bartman et al. 2023). Hence, methods to dissect heterogenous tumors into different cell populations to study cell–cell metabolic interactions could be important to understand some phenotypes (Ma et al. 2019; Lau et al. 2020; DeVilbiss et al. 2021; Sheldon et al. 2021; Datta et al. 2022; Kerk et al. 2022).

Data derived from in vivo stable isotope metabolite-labeling experiments have also been used to model flux through metabolic pathways in tissues (Buescher et al. 2015; Hui et al. 2017; Bartman et al. 2021, 2023). To do this, the kinetics of label incorporation into metabolites of interest within tissues are measured over time, combined with mathematical modeling to estimate metabolic pathway fluxes (Fig. 1). For instance, TCA cycle fluxes in healthy mouse tissues and mouse tumors were recently calculated based on labeling from infused lactate and glutamine (Bartman et al. 2023). Here, different GEMMs and xenograft tumors that model multiple types of cancer, including PDAC, NSCLC, colorectal cancer, and leukemia, were used to calculate TCA cycle flux. Combined with glucose uptake rates quantified by using isotope-labeled 2-deoxyglucose infusions, it was estimated that healthy mouse tissues generate most of their ATP (at least 90%) from the TCA cycle and oxidative phosphorylation. Surprisingly, primary solid tumors displayed suppressed TCA cycling relative to normal tissues, even though they were also estimated to produce the majority of their ATP oxidatively. As a result, this study suggested that solid tumors generally produce ATP at a slower rate than healthy tissues. One proposed explanation for this conclusion is that tumors reduce the synthesis of proteins required for tissue-specific functions (Bartman et al. 2023). One caveat of these experiments is that bulk tissues were analyzed, and other work has shown that the distribution of glucose uptake varies substantially across cell types in tissues, with macrophages being a major consumer of glucose in some cancers (Reinfeld et al. 2021). Therefore, further refinement of stable isotope metabolite-labeling approaches in heterogenous cell types is needed to fully understand nutrient use by tumors.

Figure 1.

Approaches to study tumor metabolism in vivo. (1) Stable isotope labeling is an approach to examine metabolite usage in vivo. Stable isotope-labeled metabolites such as U-¹³C-glucose can be infused into mice over time to trace label incorporation into downstream metabolic pathways. Both steady-state labeling and dynamic labeling, together with metabolic modeling, can provide information about nutrient preferences and metabolic fluxes in tumors and tissues. This approach can also be applied to study metabolic interactions between cells in tissues, and between tumors and other organs. (2) Tumors have different cell populations, including cancer-associated fibroblasts (CAFs), macrophages, lymphocytes, and cancer cells. Extracellular matrix (ECM) is also an important stromal feature of the tumor microenvironment. Interactions between these distinct cell types and each other, as well as the ECM, can be studied using cell-sorting strategies and mass spectrometry imaging. (3) Metabolic interactions between tumors and normal tissues can be explored using dual recombinase genetic mouse models to enable separate genetic perturbations in different tissues. For example, pancreatic tumors can be generated in the pancreas by using Flp-FRT to activate the Kras oncogene and inactivate p53 in the pancreas, and another gene of interest, such as MuRF-1, can be deleted in the muscle by using the Cre-loxP system. (Biorender was used to generate the figure.)

From a technical perspective, the most common way to administer stable isotope-labeled nutrients to animals is through an intravenous route. One protocol involves surgical implantation of a catheter into the jugular vein of a mouse. After postoperative treatment, recovery, and catheter maintenance, a labeled metabolite solution can be intravenously infused into conscious mice by using a microdialysis pump (Davidson et al. 2016; Hui et al. 2017; Park et al. 2018; Lau et al. 2020). An alternative intravenous approach is to use a tail vein catheter for the infusion; however, this protocol requires animals to be anesthetized during the infusion (Buescher et al. 2015; Faubert et al. 2021; Sheldon et al. 2021). A direct comparison of these two approaches to assess whether the intravenous infusion method impacts tissue and tumor metabolite labeling is needed, particularly since anesthesia can lead to metabolic alterations (Kharasch and Thummel 1993; Makaryus et al. 2011; Rao et al. 2016; Guo et al. 2017; Sheldon et al. 2021). In addition, because both of these approaches are technically challenging, some groups have used repeated intravenous or intraperitoneal bolus injections of labeled nutrients into conscious mice (Grima-Reyes et al. 2021; Sheldon et al. 2021; Lee et al. 2023). While this method has provided some insights, interpretation is confounded by the kinetics of label accumulation and decay, which could affect labeling patterns in tissues.

Controlling for certain variables can help with interpretation of nutrient-labeling experiments in animals. For example, during blood and tissue extraction, it is critical to minimize the processing time and to keep samples cold to ensure that metabolite levels remain as stable as possible. For tissue samples, snap-freezing (or quenching) of the samples immediately after harvesting is recommended where possible (Smith et al. 2020). A delay in the snap-freeze step affected metabolite levels in mouse heart and skeletal muscle, how they differ between wild-type and genetically modified tissues, and shifts in both ¹³C-labeled metabolite abundances and enrichment (Rauckhorst et al. 2022). By and large, studies characterizing label distribution in mouse cancer models report tissue dissection and snap-freezing time to be less than 5 min (Lau et al. 2020; Zeng et al. 2022; Bartman et al. 2023). Once frozen, tissues can be stored at −80°C for prolonged periods with minimal impact on many, but not all, metabolites prior to extraction for mass spectrometry analysis (Lau et al. 2020; Smith et al. 2020; DeVilbiss et al. 2021; Zeng et al. 2022).

Most studies using stable isotope tracing in mice involve infusion of labeled nutrients to isotopic steady state, where the amount of isotopically labeled metabolites is constant over time. These conditions enable comparisons of labeling patterns across different tissues, as non-steady-state measurements might otherwise be confounded by differences in tissue uptake rates. However, one unavoidable caveat of these experiments is that despite infusing a single labeled nutrient, tissue metabolite labeling is not necessarily directly derived from that nutrient. For example, infused ¹³C-glucose might be converted to amino acids in the liver, which are then released into circulation to contribute to labeling in other tissues. As a result, tissues are exposed to a mixture of labeled metabolites that is dependent on whole animal physiology and metabolism, and tissue metabolite labeling from the infused nutrient is always a combination of both direct and indirect contributions.

Another consideration for studying a system at metabolic steady state is that tissue metabolic activity ideally should change minimally during the infusion. Of note, the infusion itself can affect circulating levels of nutrients over time. for example, infusing excessive levels of ¹³C-glucose to increase blood enrichment will impact whole-body metabolism by raising glucose and insulin levels, which can subsequently influence labeling patterns in tissue metabolites (Lau et al. 2020). Efforts have been made to determine a ¹³C-glucose infusion rate that achieves sufficient labeling of tissue metabolites without raising blood glucose and insulin (Lau et al. 2020; Sheldon et al. 2021). The timing of feeding and fasting should also be considered, as multiple studies have demonstrated that metabolite levels are influenced by food intake (Dong et al. 2022) and the circadian clock (Smith et al. 2020; Dong et al. 2022). It is also important to consider that stress conditions can alter metabolism, and minimizing manipulation of the experimental models where possible before and during blood sampling can aid interpretation of results (Dong et al. 2022).

Another limitation of intravenous labeled nutrient infusion is that infusion times are restricted, with 24 h being a practical upper limit (Lau et al. 2020). Most stable isotope infusions in mice are conducted over 2.5–6 h (Lau et al. 2020; Sivanand et al. 2022; Bartman et al. 2023). With this labeling duration, some metabolite classes, such as lipids, may not be substantially labeled and will not have reached steady state (Lau et al. 2020). Because of this, tracer delivery through the diet or drinking water is another approach that has been used for longer-term labeling experiments (Hill et al. 2004; Miller et al. 2020). For instance, deuterated water (D₂O) administration in drinking water leads to deuterium (²H) incorporation into NEAAs, which are then used to synthesize proteins, such that protein synthesis rates in tissues can be assessed by hydrolyzing tissue proteins and measuring label incorporation into amino acids (Miller et al. 2020). D₂O can also be used to monitor fatty acid synthesis because each fatty acid elongation step incorporates three deuterons into the fatty acid (two from NADP[²H] and one from D₂O) (Diraison et al. 1996).

Administering labeled nutrients through the diet has also been used for long-term labeling experiments. For instance, mice with Kras-driven lung and pancreatic tumors were exposed to an amino acid–defined diet in which 20% of leucine and valine were ¹³C-labeled. After a week of exposure to labeled diets, all groups of mice exhibited ∼10%¹³C-BCAA enrichment in plasma, 5% in pancreatic tumors, and 10% in lung tumors. These differences suggested that Kras-driven lung tumors use circulating BCAAs to a greater extent than PDAC tumors (Mayers et al. 2016). Amino acid–labeled diets have also been used to assess protein turnover in muscle in a mouse model of pancreatic cancer (Mayers et al. 2014). Another example is use of ¹⁵N-labeled diets made from Spirulina, a nitrogen fixing blue-green algae that can be grown in ¹⁵N-labeled N₂ to label most amino acids in the organism. This approach has been used in rats and mice to assess newly synthesized proteins in vivo (Tessier et al. 2021). Recently, ¹³C- and ¹⁵N-labeled algal protein-based diets were used to determine the metabolic inputs to specific bacteria based on labeling of bacteria-specific peptide sequences. Lactate, 3-hydroxybutyrate, and urea were noted to pass from the host organism to the gut microbiome, providing insight into tumor–microbiome interactions in cancer (Zeng et al. 2022).

APPROACHES FOR STUDYING CELL–CELL INTERACTIONS WITHIN TISSUES

Studying metabolic interactions between different cell types within tissues remains a major challenge for the field. Tumors are composed of many cell types, including tumor-associated fibroblasts, neurons, T cells, macrophages, and other immune cells (Fig. 1; Valkenburg et al. 2018). Cancer-associated fibroblasts are a subset of stromal cells that can promote tumor progression (Valkenburg et al. 2018; Barrett and Puré 2020), and macrophages can constitute a majority of the cell population in some tumors (Poh and Ernst 2018; Mantovani et al. 2022). PDAC tumors are notable in that cancer cells are the minority of the cell population (Öhlund et al. 2014, 2017; Lau et al. 2020; Sahai et al. 2020; Geeraerts et al. 2021). Studies on interactions between PDAC cells and stromal cells in the tumor microenvironment of KPC or KP^−/−C PDAC GEMMs, PDAC GEMM-derived syngeneic allografts, and human cell line xenografts have shown that nutrient exchange can affect disease phenotypes such as metastasis and drug sensitivity (Dalin et al. 2019; Banh et al. 2020; Lau et al. 2020; Francescone et al. 2021; Tian et al. 2021; Datta et al. 2022; Kerk et al. 2022). For example, it was demonstrated by using both human PDAC xenografts and PDAC GEMM-derived syngeneic allografts that pancreatic stellate cells support PDAC tumor metabolism through autophagic alanine secretion (Sousa et al. 2016) and nucleoside secretion (Yuan et al. 2022). Furthermore, by using optical imaging in PDAC GEMMs, it was found that pancreatic stellate cells and cancer cells have direct physical interactions that promote a more oxidized redox state in cancer cells within tumors (Datta et al. 2022; Kerk et al. 2022). As a final example, it was also demonstrated that neuron-derived serine can support protein synthesis in human pancreatic cancer cell line–derived orthotopic xenografts (Banh et al. 2020). These examples show how metabolic interactions between distinct cells within tumors can influence tumor metabolism and growth.

To study metabolic interactions between cells in tumors, it is critical to assess the metabolism of distinct cell types within mixed populations. How to do this is challenging because most metabolites turn over rapidly compared to the time needed for cell isolation. The most common method to isolate cell types from a complex cell population is flow cytometry. However, flow cytometry has several disadvantages including (1) loss of cell types during tissue dissociation, (2) metabolite exchange with the dissociation medium and buffers needed to sort cells, (3) turnover of metabolites during the time of sorting, and (4) limits on detection of less abundant metabolites from rare cell types in a population that makes sorting impractical for obtaining enough cells to assess levels of those metabolites. Nevertheless, several groups have attempted to optimize flow cytometry to conduct metabolomic profiling of distinct cell types such as hematopoietic stem cells (HSCs) and various cell subsets from tumors (Ma et al. 2019; DeVilbiss et al. 2021; Qi et al. 2021; Sheldon et al. 2021; Meacham et al. 2022). In one study, hundreds of metabolites were quantified from hematopoietic stem and progenitor cells either by isolating large numbers of cells or by pooling HSCs from 120 mice to perform a single experiment (Takubo et al. 2013; Naka et al. 2015). In a more recent example, rapid cell isolation by flow cytometry was coupled with an LC-MS approach to quantify metabolites in rare hematopoietic stem cell populations in mice (Qi et al. 2021; Meacham et al. 2022). The same research group further refined this method by using hydrophilic liquid interaction chromatography and high-sensitivity orbitrap mass spectrometry with improved chromatographic separation, increased mass resolution, minimized ion suppression, and elimination of the sample drying step in their extraction. This approach was applied to characterize the metabolism of circulating human melanoma cells from the blood of xenografted mice (DeVilbiss et al. 2021). Nevertheless, not all metabolites will be stable under these conditions, and appropriate controls are needed to use mass spectrometry to study metabolites in rare populations.

Similar challenges apply to using sorting approaches to study different cell populations in tumors. Recently, a rapid magnetic immunoprecipitation-based cell isolation approach was used to study glucose metabolism in purified T-cell subpopulations undergoing immune responses in vivo (Sheldon et al. 2021; DeVilbiss et al. 2021; Hartmann et al. 2021; Kilgour et al. 2022). Uniformly labeled ¹³C-glucose was infused into pathogen-infected mice for 2 to 6 h, and then T-cell populations were isolated using antibody-based magnetic bead separation in a process that took 35 min to obtain material to be extracted for mass spectrometry (Sheldon et al. 2021). As expected, liver and spleen-labeling patterns were strongly correlated with blood-labeled fractions. However, unlike T-cell responses in vitro, which involve increased aerobic glycolysis (Pearce et al. 2013; MacIver et al. 2013), this study found that highly proliferative effector T cells displayed lower glucose flux to lactate, showing how metabolism in subpopulations of cells can also be impacted by the in vivo tissue environment. With a similar approach, human ovarian tumors were mechanically disaggregated and filtered into a single-cell suspension prior to bead-based cell enrichment of immune cell populations. Adenosine, kynurenine, and 1-methylnicotinamide (MNA) metabolites were noted to differ (Kilgour et al. 2022), but it remains possible that some differences resulted from continued metabolism that occurred during cell isolation. Nevertheless, the presence of differences argues that the metabolomes of these different cells were likely different from each other in tumors.

Another approach to study distinct cell populations is the analysis of isotope-labeled metabolite incorporation into stable macromolecules, as this can overcome many of the challenges associated with metabolism continuing during the time needed for sorting (Lau et al. 2020). Unlike free polar metabolites, proteins, DNA, and lipids remain stable during sorting, and labeling of amino acids, nucleotides, and fatty acids within these macromolecules can reflect how cells metabolized the provided stable isotope-labeled nutrient to synthesize these metabolites. This approach was applied to understand differences between how PDAC cells versus pancreatic stellate cells metabolize glucose (Lau et al. 2020). In this study, both KP^−/−C and KPC PDAC mouse models were examined. By hydrolyzing proteins in these sorted cell populations to examine glucose labeling of amino acids, it was found that pyruvate carboxylase (PC) and malic enzyme 1 (ME1) were more active in PDAC cells compared to pancreatic stellate cells. These results were corroborated by the finding that loss of PC or ME1 impairs PDAC tumor growth (Lau et al. 2020). One caveat of this approach is that labeling will not reach steady state, so differences in labeled nutrient uptake between cells must be considered for data interpretation.

Finally, spatial metabolomics using mass spectrometry imaging (MSI) is an emerging technology that may overcome many of the limitations of current methods to study cell populations in tissues by visualizing the distribution of biomolecules in tissue sections. This approach was initially applied to image the distribution of large proteins (Luxembourg et al. 2004), and has been increasingly applied to image metabolites in tissues (Shimma et al. 2007; Hayasaka et al. 2008; Harada et al. 2009). There are several advantages of MSI. While labels can be incorporated, MSI does not require any labels or specific probes (Shimma et al. 2007). MSI is also a nontargeted imaging method that can be used to image many types of metabolites simultaneously (Hayasaka et al. 2008; Patterson et al. 2018). Matrix-assisted laser desorption/ionization-MSI (MALDI-MSI) can measure metabolites with a spatial resolution that can resolve single cells (∼5–50 µM), thereby providing metabolic information on both different cell populations and their relative positions to each other in the tissue. For instance, MALDI-MSI revealed abnormal distribution of phospholipids in colon cancer liver metastasis (Shimma et al. 2007), highlighting how this approach can uncover aspects of tumor metabolism that are not evident using other techniques. Importantly, MALDI-MSI has been combined with stable isotope labeling in mice to study how nutrients are consumed spatially within a tissue (Wang et al. 2022; Bartman et al. 2023; Planque et al. 2023). MALDI-MSI was combined with ¹³C lactate infusions to visualize relative TCA flux in healthy and metastatic regions of the lung and used to argue that metastatic nodules in the lung from primary breast cancer xenografts have higher TCA flux compared with the surrounding healthy lung regions (Bartman et al. 2023). MALDI-MSI can also be coupled with histology (Aichler and Walch 2015; Patterson et al. 2018). For example, HER2 immunochemistry was combined with MSI to spatially distinguish whether ¹³C incorporation from glucose into lipid species in mouse brain tissues with human breast cancer xenograft metastases reflected the labeling of cancer cells or normal brain tissue (Ferraro et al. 2021).

However, there are several challenges associated with MALDI-MSI, such as high cost and complex data analysis. Additionally, MALDI-MSI can have low sensitivity for some metabolites, and ion suppression from variable levels of different metabolites at each point in space can affect detection of the same metabolite across different regions of the tumor. MALDI-MSI can also cause delocalization of metabolites during the matrix tissue-mounting step, which reduces the specificity of metabolites detected at individual spatial points (Zemaitis et al. 2021). To address this, several MSI platforms have been developed without MALDI, including desorption electrospray ionization–mass spectrometry imaging (DESI-MSI) and secondary ion mass spectrometry (SIMS) (Soudah et al. 2023). While SIMS and MALDI-MSI have single-cell/subcellular spatial resolution, DESI-MSI has a lower spatial resolution of ∼200 μm. As MSI techniques continue to be refined, these approaches will be a powerful tool for studying metabolic heterogeneity among distinct cell types in tumors.

APPROACHES FOR STUDYING METABOLIC RELATIONSHIPS BETWEEN DIFFERENT TISSUES

Interactions between tumor cells, the immune system, and peripheral tissues is evident from epidemiological studies that have shown that obesity is linked to a higher risk of 13 different kinds of cancer, including cancers in the breast, colon and rectum, uterus, stomach, kidneys, liver, and pancreas (Yuan et al. 2020). Indeed, studies in model organisms including mice and Drosophila have demonstrated that extensive metabolic cross talk occurs between distant organs during development and in diseases such as cancer (Alvarez-Ochoa et al. 2021). These interactions between organs help maintain whole-body energy homeostasis and coordinate tissue growth with nutrient availability (Droujinine and Perrimon 2016; Alvarez-Ochoa et al. 2021). Most work in this area has focused on how hormonal cues mediate this intertissue cross talk (Owusu-Ansah and Perrimon 2015), and emerging work is beginning to focus more on how tumor metabolism is modulated in the context of whole-body metabolism.

One major whole-body alteration associated with the progression of many cancers is cancer-associated cachexia, which is defined as unintended body weight loss with specific losses of muscle and adipose tissue (Kir et al. 2014; Danai et al. 2018; Martin et al. 2021; Yuan et al. 2022; Bilgic et al. 2023). Studies in KP^−/−C and KPC mouse models of pancreatic cancer showed that breakdown of muscle and adipose tissue occur early in the progression of this disease (Mayers et al. 2014; Danai et al. 2018), with a similar phenomenon observed in patients (Babic et al. 2019). The mechanisms driving tumor-induced tissue wasting are poorly understood and can be disease-specific. Experiments in cultured cells have focused mainly on how tumor-derived ligands can trigger both tumor growth and changes in host tissues (Herrington et al. 2001; Kir et al. 2014, 2016; Mueller et al. 2016; Danai et al. 2018; Weber et al. 2022). Notably, it is likely that processes that degrade proteins in muscle tissue, such as autophagy and proteasomal degradation, contribute to the elevated levels of circulating BCAAs, which is associated with PDAC progression and cachexia (Mayers et al. 2014; Danai et al. 2018; Neyroud et al. 2023). Indeed, disrupting autophagy in the host can affect tumor growth arguing that tissue breakdown or turnover can provide some nutrients for the tumor (Poillet-Perez et al. 2018; Khezri et al. 2021). It has also been shown that the muscle-specific ubiquitin E3 ligase MuRF1 is required for PDAC-associated muscle wasting, and genetic inactivation of MuRF1 alters both systemic and tumor metabolism and delays tumor growth (Neyroud et al. 2023). Pancreatic enzyme insufficiency and reduced food intake can also contribute to tissue wasting in mouse PDAC models (Michaelis et al. 2017; Danai et al. 2018; Martin et al. 2021; Olson et al. 2021). Interestingly, while pancreatic enzyme replacement can reverse some tissue loss in these PDAC models, it also led to decreased survival, further suggesting an interplay between systemic nutrients released from peripheral tissues and the regulation of PDAC metabolism and tumor growth.

An approach for examining these intertissue metabolic interactions is to use dual recombinase genetic mouse models to separate genetic perturbations across tissues. Classical Cre-loxP-based GEMMs rely on a single Cre-mediated recombination step to activate mutant oncogene expression and genetic inactivation of tumor suppressors. Similarly, the flippase-FRT (Flp-FRT) uses a Flp-recombinase instead of Cre-recombinase. An inducible dual-recombinase system was developed by combining Flp-FRT and Cre-loxP recombination technologies to improve GEMMs of pancreatic cancer and enable sequential genetic manipulation by two independent recombination systems in distinct tissues (Schönhuber et al. 2014). In this model, deletion of p53 and knockin of the Kras^G12D mutation is achieved with an FRT-flanked Trp53 allele (Trp53 FRT/FRT) and an FRT-stop-FRT (FSF) Kras^G12D knockin allele (FSF-Kras^G12D/+) that are activated via pancreas-specific expression of Flp (Pdx1-Flp) (Young et al. 2011; Lee et al. 2012; Schönhuber et al. 2014). An additional floxed allele of a gene of interest is then expressed under the control of tissue-specific Cre recombinase to achieve deletion of a gene in another tissue (Fig. 1). This model was recently used to delete arginase 1 (Arg1) in myeloid cell lineages with the Cre-loxP system concomitantly with the generation of PDAC tumors with the Flp-FRT system (Menjivar et al. 2023). Genetic inactivation of Arg1 in macrophages delayed formation of invasive disease, while increasing cytotoxic T-cell infiltration. This result suggested that macrophage-derived arginase in the tumor stroma mainly regulates immune suppression in the pancreatic tumor. Therefore, a dual recombinase GEMM approach allows for (1) independent targeting of tumor cell–autonomous and non-tumor cell–autonomous pathways/processes, (2) inactivation of one or more genes in tissues, (3) sequential induction of genetic alterations to model human multistep carcinogenesis and metastasis, and (4) genetic analysis of tumor cell subpopulations such as cancer stem cells, tumor-associated fibroblasts, and immune cell subpopulations (Schönhuber et al. 2014; Menjivar et al. 2023). A major challenge with this dual recombinase system is the amount of time needed to generate mice with both the Flp-FRT and Cre-loxP alleles needed to target the multiple genes of interest. An alternative quicker approach is to transplant cancer cells to form tumors in syngeneic mice that already harbor a disruption of genes of interest within a specific tissue (Poillet-Perez et al. 2018; Apiz Saab et al. 2023; Bilgic et al. 2023; Neyroud et al. 2023). For example, host-specific whole-body knockout of either autophagy-related gene 5 (Atg5) or autophagy-related gene 7 (Atg7) under a ubiquitously expressed tamoxifen-inducible Cre impairs the growth of multiple different allografted tumors (Poillet-Perez et al. 2018). Coupling these genetic models with metabolomics and stable isotope metabolite-labeling approaches described above can be a powerful tool for characterizing how gene function in normal tissues impacts tumor metabolism in mouse models.

PHYSIOLOGICAL MEDIA CAN MIMIC NUTRIENT LEVELS IN THE TUMOR MICROENVIRONMENT

Finally, while the approaches described above are powerful for studying cancer metabolism in vivo, the complexity of animal physiology does not allow the range and flexibility of experimentation that is possible in cultured cells. Therefore, many laboratories are making efforts to refine tissue culture models by improving culture media to better model the metabolism that occurs within tumors in vivo. One strategy is to culture cells directly in biological fluids or tissue extracts. An advantage of this approach is that it, by definition, represents nutrient conditions that cells could experience in vivo (Fig. 2). For example, a study demonstrated that cancer cells dependent on glutamine when cultured in RPMI are less dependent on glutamine when cultured in 100% adult bovine serum (ABS), which more closely resembles nutrient concentrations in blood (Muir et al. 2017). This phenotype also tracked with sensitivity to glutaminase inhibitors and was used to understand a phenotype first noted via studies in mouse models (Davidson et al. 2016). By supplementing ABS with nutrients at levels found in RPMI, it was found that high concentrations of the amino acid cystine in RPMI drive glutamine dependency. Mechanistically, high environmental cystine levels stimulate cystine uptake through the cystine/glutamate antiporter xCT/SLC7A11, which drives glutamate export from the cell. As a result, cells become dependent on extracellular glutamine for TCA cycle anaplerosis (Muir et al. 2017). This observation provides clear evidence that environmental nutrient levels influence how cancer cells use metabolites to support cell proliferation and highlights the utility of culturing cells in different media to better understand metabolic differences between mouse models and cell culture.

Figure 2.

Advantages and disadvantages of altering nutrients in culture media to study tumor metabolism. Cancer cells can be grown in different media nutrient conditions. Standard culture media formulations have several advantages such as low cost and application to multiple cancer cell lines. On the other hand, it fails to replicate the nutrient levels found in tumors and tissues. Culturing cells in a media with physiological nutrient levels can mimic some environments, including what is measured in blood and in tumor or tissue interstitial fluid (TIF). However, preparing these medias can be challenging because of the need to quantify absolute metabolite levels in biological fluids and the time and effort needed to formulate the media.

An alternative approach for culturing cancer cells in physiological nutrient conditions that avoids the use of biological fluids with undefined nutrient levels is to formulate synthetic media that match the levels of metabolites found in blood or tissues (Cantor et al. 2017; Ackermann and Tardito 2019; Vande Voorde et al. 2019). Serum-like modified eagle's medium (SMEM) was first formulated in 2015 with glucose, pyruvate, and amino acid concentrations similar to those found in human blood (Schug et al. 2015; Tardito et al. 2015; McKee and Komarova 2017). In 2017, human plasma-like medium (HPLM) was formulated to contain a more comprehensive set of metabolites and minerals at concentrations measured from adult human plasma, including amino acid derivatives, ketone bodies, and products of nucleic acid catabolism (Cantor et al. 2017). Plasmax is also another cell culture medium, which mimics the metabolite profile of human plasma (Vande Voorde et al. 2019). Arginine and pyruvate are ∼10-fold less abundant in Plasmax than in DMEM, which subsequently impacts cancer cell metabolism (Vande Voorde et al. 2019). These observations highlight the utility of culturing cells in physiological media to study some aspects of metabolism that may not be possible to study in animal models.

One limitation of using media like SMEM, ABS, HPLM, and Plasmax is that they assume metabolite levels in blood represent the nutrients that are directly available to cancer cells within a tumor. This is not always the case, because in solid tumors and other tissues, cells obtain nutrients from the interstitial fluid that surrounds them. The composition of tumor interstitial fluid (TIF) does not match that of blood (Sullivan et al. 2019a) but rather depends on factors such as tissue type, resident cell populations, vascularization, and efficiency of lymphatic drainage (Wagner and Wiig 2015; Kaymak et al. 2021). Therefore, constructing synthetic tissue culture media that mimics nutrient levels in TIF is an alternative approach to model the effects of in vivo tumor nutrient levels on cancer cells using an in vitro culture system (Muir and Vander Heiden 2018; Apiz Saab et al. 2023).

Many efforts have been made to quantitatively measure metabolite levels in biological fluids. For example, TIF was isolated from KRAS-driven pancreatic and lung tumors by placing tumors on a fine mesh and subjecting the samples to low-speed centrifugation, which enabled TIF isolation without causing extensive cell lysis (Sullivan et al. 2019a). It was found that the metabolite composition of TIF is most heavily influenced by the tumor tissue of origin, tumor location, and diet, while tumor genetics had less of an effect (Sullivan et al. 2019a). These observations suggest that media formulated to contain TIF nutrient levels could be used to study the metabolism of different tumor types. However, a caveat with this approach is that it assumes that TIF metabolite levels are uniformly distributed across a tumor, which is unlikely to be the case. Nevertheless, compared to blood metabolite levels, TIF may better represent the nutrients that cancer cells are exposed to in a tumor, and it remains to be seen whether media that reflects TIF nutrient levels will better recapitulate in vivo tumor metabolism compared to classic cell culture approaches.

Several considerations are needed when using media with physiological nutrient levels to culture cells (Fig. 2). First, because metabolite levels in physiological media are typically much lower than those in classic media such as DMEM and RPMI, frequent media changes are needed to ensure that nutrients are not depleted below critical levels during cell culture. An approach that can address this limitation is to culture cells in a continuous-flow system that continually refreshes the nutrients in media over time, such as a chemostat or Nutrostat (Birsoy et al. 2014). Another consideration when using physiological media such as HPLM and Plasmax is that while these media better reflect the concentrations of polar metabolites in the blood, levels of nonpolar metabolites (such as lipids) and proteins (including growth factors) are more difficult to comprehensively quantify in blood and to reconstitute in culture media. The extracellular matrix in tumors is yet another variable that can affect cancer cell metabolism that is difficult to model in vitro. Culturing cells as organoids or in other 3D culture systems may be a better model for studying some cancer phenotypes compared to 2D monoculture on plastic; however, while some studies have found that 3D culture impacts metabolism and can be used to probe metabolic interactions between different cell types (Coloff et al. 2016; van Gorsel et al. 2019; Lau et al. 2020; Hu and Lazar 2022), 3D culture does not recapitulate all aspects of tumor metabolism. Nevertheless, transplantation of organoids in vivo can lead to tumor formation with a stromal architecture that betters recapitulates that found in autochthonous tumors (Luo et al. 2021). Therefore, combining 3D culture systems with physiological media may be one approach to further improve our ability to model tumor metabolism in cultured cells and provide a way to study aspects of metabolism that are less feasible to do in animal models. For example, physiological media has been used in several recent studies to perform high-throughput functional genomic or chemical compound screening to identify metabolic dependencies in cancer cells (Biancur et al. 2021; Rossiter et al. 2021; Zhu et al. 2021; Abbott et al. 2023). Determining whether these metabolic dependencies are mirrored in tumors growing in vivo will be a critical test of whether these physiological media systems will be more predictive for identifying effective therapeutics to use in the clinic.

CLOSING REMARKS

Tumors have high metabolic plasticity to respond and adapt to nutritional changes in their environment. While initial studies established a critical role for glucose and glutamine metabolism in cultured cancer cells, recent studies suggest that the metabolism of tumors in mice and humans can differ from that of cancer cells in culture. The tumor microenvironment contributes to many of these differences, as cancer cells in vivo interact with different nutrient levels than they do in culture, as well as with noncancer stromal and immune cells, and with normal tissue and whole-body metabolism. These interactions can impact tumor responses to cancer therapy, highlighting the importance of understanding the metabolic factors that influence tumor progression in the context of a whole organism. Advances in mouse modeling have enhanced our ability to study tumor metabolism, with tools such as physiological culture media, in vivo stable isotope metabolite labeling, improved mass spectrometry methods to monitor metabolism in distinct tumor cell populations, and dual recombinase systems to probe tumor interactions with normal tissues. Application of these methodologies to study cancer will continue to enable new technical innovations and generate insights into how altered cancer metabolism can be leveraged to improve cancer therapy.