The proper analysis of metabolism in cell-cell interaction is dependent on specific analytic techniques to measure the metabolome, which represents the pool of metabolites from diverse array of metabolic activities [
97,
98]. Here we discuss the different methods that allow us deciphering the metabolic crosstalk between different cells, e.g., of CAFs and tumor cells.
3.1. Mass Spectrometry-Based Metabolomics
Generally, mass spectrometry (MS) is the leading technology used to identify metabolites and quantify their concentration [
99]. It allows obtaining both quantitative and qualitative information about the intracellular or extracellular metabolome composition of a given biological sample [
97]. MS is commonly coupled with gas chromatography (GC) or liquid chromatography (LC), to allow pre-separation of metabolites. GS tandem MS requires the derivatization of metabolites and their conversion to the air phase. This limits the number of molecules that can reliably be measured. GC-MS, however, is an ideal technique for analyzing non-polar, volatile and non-volatile molecules of small molecular weight [
100]. Conversely, LS-MS is well-suited to analyze thermally unstable molecules and requires a less extensive extraction process. These metabolites are then loaded on the MS to identify and decode the metabolic profiles of cells or tissues. This kind of metabolite analysis can be used for a targeted (with pre-defined or selected metabolites targets) and a non-targeted metabolomics approach.
In the studies of CAF and tumor cell interactions, the use of LC-MS is more popular compared with GC-MS (). This is probably due to the broader coverage of metabolite read-outs and a simpler methodical setup in LC-MS compared to GC-MS. A few studies have also used superior versions of the conventional LC, namely ultra-performance liquid chromatography (UPLC) or high-performance liquid chromatography (HPLC) to assess metabolism in tumor–CAF interaction. Both allow the separation of smaller particle sizes (<2 µm for UPLC) and provide a higher speed, resolution, and sensitivity [
101]. However, MS is mostly used to analyze in vitro cultures and ex vivo tumor tissues. More applications of those techniques are needed to further explore CAF and tumor metabolic interactions under in vitro and/or in vivo settings. In the future, they will be particularly useful in delineating the role of small extracellular vesicles, which are derived from CAFs, and are capable of driving tumor metabolic rewiring. Moreover, integrating the MS-measurements with other metabolic approaches such as stable isotope labelling and seahorse assays seems an ideal way to decipher CAF and tumor cell metabolic interactions.
3.2. Metabolic Flux Analysis
While measuring the level of cellular metabolites provides crucial information, it is equally important to determine the pathways that are linked to them and how they interact. On its own, a drop in the levels of a metabolite can either point towards the reduction of the synthesis or the increased consumption of this metabolite via another pathway. One of the metabolic techniques used to determine the pathways that contribute to metabolite secretion is known as ‘metabolic flux analysis’ (MFA). MFA uses nutrients, such as glucose, amino acids, or lipids, which are labelled with the stable isotopes
13C or
14N, as the basis for metabolites’ identification and quantification [
102]. Mass spectrometry is commonly used to analyze the labeled samples. Primarily,
13C-MFA aims to produce a quantitative cellular metabolic map by assigning the values of fluxes to the reactions in the network and by using confidence interval for every predicted flux [
103].
Although MFA has been used to elucidate the metabolic changes that take place in cancer cells and their microenvironment in vitro, it is equally important to perform
13C-MFA under in vivo settings, which has been addressed recently [
101,
104]. In fact, the variations in experimental set-up between different in vitro studies on CAF–tumor interaction has also raised concerns about the most reliable workflow to use. Thus, its usage for understanding CAF–tumor cell interactions is still limited (), particularly under the in vivo or ex vivo settings. However, in vivo flux analysis provides a better comprehensive view of the metabolism at the cellular and whole-organism level which can complement in vitro approaches [
105,
106]. In particular, it will help to understand the nutrient exchange between CAFs and the other components of the TME, such as immune cells or pericytes, which are known to impact tumorigenesis. Furthermore, by using
13C-MFA in vivo, it will be possible to understand the contribution of each metabolic pathway to the generation of CAF-secreted oncogenic metabolites in a physiological relevant context. Accordingly, the integration of MFA with other computational approaches will lead to a better understanding of the CAF–tumor interaction-driven cancer metabolic rewiring in the future.
3.3. Seahorse Extracellular Flux
The Agilent Seahorse XF Analyzer is a powerful tool that can offer a rapid insight into the metabolic state of the cells. Seahorse protocols use a combination of metabolic inhibitors to quantify the rate of glycolysis and/or OXPHOS, via the measurement of extracellular acidification rate (ECAR) and oxygen consumption rate (OCR). By understanding the cellular responses to metabolic drugs, the different respiratory phases can be defined and the cellular bioenergetics status of the cells can be inferred [
107].
In ECAR measurement, cells are stimulated with 2,4-Dinitrophenol (2,4-DNP, an oxidative phosphorylation inhibitor) and 2-Deoxy-D-Glucose (2-DG, a hexokinase inhibitor). 2-4 DNP induces maximal glycolytic capacity, whereas 2-DG is a glucose analog that completely shuts glycolysis. For measuring OCR, cells are exposed to mitochondria perturbing reagents. Oligomycin, carbonyl cyanide-4-(trifluoromethoxy)phenylhydrazone (FCCP) and rotenone coupled with antimycin are added to the cells in a sequential manner to induce changes in mitochondrial activity. Oligomycin blocks the ATP synthase (complex V) and suppresses OCR. FCCP is a mitochondrial uncoupler which disrupts the proton gradient and mitochondrial membrane potential, and drives OCR to a maximal level, which is then reduced to a minimal level by antimycin A and rotenone (complex III and complex I inhibitor), finally disrupting mitochondrial respiration. The administration of these drugs at different time-points causes the fluctuation in the OCR level. These data can then be used to calculate basal respiration, ATP-linked respiration, proton leak, maximal respiration capacity, reserve capacity and non-mitochondrial respiration [
108].
An indirect in vitro study set-up using the conditioned media, of either the tumor cells or the CAFs, is commonly used to study CAF–tumor cell interaction with Seahorse (). This is obviously due to the simplicity of such a set-up. In addition, most studies have only been performed on a 2D cell culture monolayer so far. However, it is commonly accepted that 3D cell cultures better recapitulate the physiological parameters of a tumor; they have been described to better mimic cell–cell interaction and environmental factors (gradient of oxygen and nutrient supply) compared to their 2D counterparts [
109,
110].
In a more advanced setup, Demircioglu et al., 2020 have used Seahorse to analyze isolated cells from ex vivo tumor samples [
29]. Such an approach may give deeper insights into the physiological and metabolic state of a tumor. Yet, the prior tissue processing steps to isolate the cells need to be considered as these may significantly alter the initial tumor metabolic state.
Thus, further explorations on 3D cultures and direct co-culture settings with Seahorse, in combination with other metabolic measurements (such as stable isotope tracing and non-destructive metabolic imaging approaches), are definitively warranted in order to gain greater mechanistic insights into CAF–tumor cell interaction.
3.4. Computational Approaches to Unravel Tumor-CAF Metabolic Reprogramming
The increasing use of modern analytical techniques has helped characterizing tumor metabolism. However, the heterogeneity of the tumor metabolic rewiring is still not yet fully understood. Studies often only touch upon the phenotypic analysis without providing an in-depth mechanism. This is mostly due to the technical challenges and experimental limitations in the field of metabolomics. While metabolomics studies provide powerful insights into the cellular metabolic status of cells, the information regarding the cause or effect of metabolite changes is still missing, hence limiting our understanding of the underlying mechanisms. The cell’s metabolome constitutes the amplified and integrated signals from various levels, either at the transcriptional level or post-translational level [
111]. Therefore, a multi-omics approach (proteomics, transcriptomics and metabolomics) may provide a bigger picture to identify the key molecular regulators of a complex disease, like cancer. Further integration of ‘omics’ data to build personalized in silico models (i.e., models which represent each patient individually) could shed light on the metabolic heterogeneity and the regulatory interactions between different classes of biomolecules (genes, proteins and metabolites), especially in CAF subpopulations which are continuously being identified.
The principle of network modelling is to present high-throughput data in a robust and predictive way [
112]. Metabolic models are computational tools showing the metabolites as a set of nodes and the enzymatic conversion from one metabolite to another as edges [
113]. The main advantage of computational metabolic networks is the possibility to narrow down the number of potential targets to be validated in vitro, therefore saving time and money. They also allow studying the complex interactions between the tumor and its microenvironment on multiple scales and generating patient-specific models [
114]. As such, they represent important tools to improve drug testing, thereby making treatments more accurate and beneficial to the patients [
115].
Genome-scale metabolic reconstructions, such as Recon2 [
116], iHSA [
117], or Human1 [
118], consist of a library of all the possible reactions that can occur in an organism, as well as a set of genes that control the enzymes and transporters that allow the reactions to take place [
119]. There are two types of metabolic models. Kinetic models rely on ordinary differential equations and kinetic rates. Their applicability is, however, limited, due to a high number of parameters to be determined [
120]. Constraint-based models focus on steady states and do not integrate parameters, which allows reconstructing and analyzing larger models. Over the last few years, these models have gained interest to model the metabolic status of a cell [
121]. One important advantage of constraint-based modelling is the possibility to integrate multi-omics data, such as transcriptomics [
122], proteomics [
123] or metabolomics [
124], to constrain genome-scale metabolic reconstructions and generate context-specific models [
125]. Several algorithms exist that build context-specific models such as iMAT [
126], INIT [
127], mCADRE [
128] and the FASTCORE family [
129,
130,
131]. Finally, in silico gene knock-out can be applied to identify potential drug targets. Gene knock-out consists of blocking all the reactions associated with a specific gene and assesses its effect on the objective function of the model, i.e., biomass production. The genes that significantly reduce growth are considered as essential and, therefore, as potential drug targets [
132].
However, modelling a complex system such as the tumor microenvironment remains challenging, mainly due to the difficulty in modelling the interactions between multiple cell types. Indeed, such models already exist for microbial communities [
133], however, similar approaches applied to multiple human cell types are still widely unexplored. In 2010, Lewis et al. built a constraint-based model for multiple cell interactions in the brain based on Recon1 and successfully simulated Alzheimer’s disease [
134]. In 2015, Capuani et al. reconstructed a small, unconstrained model (75 reactions) for the lactate shuttle between CAFs and tumor cells [
135]. More recently, in 2018 Shan et al. modeled the impact of the tumor microenvironment on the Warburg effect and glutamine addiction in cancer cells and found that the reverse Warburg effect provided growth advantage to the tumors originating from deep tissues [
136]. Finally, in 2019 Damiani et al. developed a new algorithm (single-cell flux balance analysis) that allows integrating single-cell RNA-seq data into models of breast and lung cancer. This method allowed them to model an heterogeneous cancer cell population, to identify cancer cell subpopulations based on their growth rate and enabled to represent metabolite exchange between different cell types [
52,
137].
These first studies strongly demonstrate that there is no doubt that constrain-based modelling combined with single-cell sequencing will become one of the most important tools to study tumor–microenvironment interactions. Additionally, many efforts are now directed towards the integration of signaling data into metabolic models.