Genome-scale metabolic modeling and machine learning unravel metabolic reprogramming and mast cell role in lung cancer: a multi-level analysis

Introduction

Lung cancer remains one of the most prevalent and lethal cancers worldwide, accounting for a significant proportion of cancer-related deaths (1). The identification of reliable biomarkers for lung cancer is crucial for improving early diagnosis and treatment strategies (2). In particular, the complex interplay between cancer cells and the immune system has become a key focus in understanding lung cancer development and progression. The tumor microenvironment comprises various immune cells that significantly influence cancer progression, with mast cells playing a crucial role in this complex microenvironment (3).

Mast cells, traditionally known for their involvement in allergic reactions and inflammatory responses, have gained increasing attention in cancer research. These multifunctional immune cells are found in various tissues, including the lungs, and have been implicated in both pro-tumorigenic and anti-tumorigenic activities (3). For instance, some studies have reported that a higher number of mast cells in non-small cell lung cancer (NSCLC) tissues are associated with poor patient prognosis (4). Mast cells can contribute to anti-tumor immune responses through various mechanisms, including the promotion of dendritic cell maturation and migration, enhancement of CD8⁺ T cell activation and proliferation, and modulation of regulatory T cells (Tregs) (5). However, mast cells can also exhibit pro-tumorigenic effects by promoting angiogenesis, tissue remodeling, and immunosuppression in certain contexts (6).

Computational methods can complement experimental efforts (7), which are often costly and constrained by limited data. Novel approaches that uncover metabolic signatures associated with lung cancer and mast cell activity could revolutionize diagnostic and prognostic capabilities in clinical settings. Genome-scale metabolic modeling (GSM) has emerged as a powerful tool in cancer research, enabling the comprehensive analysis of cellular metabolism at a systems level. By integrating genomic, transcriptomic, and metabolomic data, GSMs provide a comprehensive view of metabolic processes and their alterations in cancer cells (8). These models have been successfully applied to various cancer types, such as pancreas (9), breast (10), colorectal (11), and brain (12) cancers, offering insights into metabolic vulnerabilities and potential therapeutic targets.

The field of cancer screening has been further enhanced by the integration of machine learning approaches. These computational techniques can identify complex patterns and relationships within large-scale datasets, facilitating the discovery of novel biomarkers and metabolic signatures associated with cancer progression (13). Thus, the combination of GSM and machine learning holds great promise for advancing our understanding of cancer metabolism and improving diagnostic and therapeutic strategies.

This study aims to address a significant gap in the field by developing the first genome-scale metabolic models of lung cancer cells and mast cells in both healthy and cancerous states. By focusing on the metabolic role of mast cells in lung cancer, we seek to uncover novel insights into the complex interactions between these immune cells and the tumor microenvironment. Our objectives include the construction and analysis of GSMs for lung tissue and lung-associated mast cells, as well as the identification of specific metabolic biomarkers that could serve as potential diagnostic or therapeutic targets. Through this comprehensive approach, we aim to elucidate the metabolic reprogramming that occurs in mast cells within the context of lung cancer, potentially uncovering new avenues for targeted therapies and improved diagnostic strategies. By leveraging the power of GSM and machine learning techniques, this study represents a significant step forward in our understanding of the metabolic landscape of lung cancer and the role of mast cells in its development and progression. We present this article in accordance with the TRIPOD reporting checklist (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-555/rc).

Methods

Data acquisition and preprocessing

Gene expression data were obtained from the publicly available Gene Expression Omnibus (GEO) dataset (GSE18842), which initially comprised 91 lung tissue samples: 46 tumor and 45 control samples. Only samples with both tumor and matched adjacent healthy tissue from the same patient were included in the analysis; three samples lacking a corresponding pair were excluded. Thus, the final dataset consisted of 43 paired lung tissue samples (healthy and cancerous) from individual patients. The study setting is retrospective and in silico, based on these publicly available genomic data, with no physical recruitment sites or centers, and represents a general population of lung cancer patients. There were no missing values in the gene expression data for any of the 43 paired samples included in this study; therefore, no special handling of missing data was required. After acquiring the gene expression profiles, we mapped the genes to those present in the Human1 model (14) resulting in a final set of 3,502 genes with metabolic functions and their corresponding expression values. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Cell type deconvolution analysis

To investigate mast cell-specific gene expression within our heterogeneous tissue samples, we employed CIBERSORTx (15), a powerful machine learning tool developed for assessing cellular abundance and cell type-specific gene expression patterns from bulk tissue transcriptome profiles. CIBERSORTx allows for the deconvolution of complex gene expression mixtures, enabling the estimation of cell type proportions and gene expression profiles without the need for physical cell isolation (15).

We utilized CIBERSORTx to impute mast cell-related marker gene expressions from the bulk RNA-seq data for all healthy and cancerous samples in our dataset. This deconvolution method enabled us to estimate the gene expression profiles of mast cells within the heterogeneous tissue samples.

GSM

We developed genome-scale metabolic models using the Human1 model and the iMAT (16) method. First, genes were mapped to their corresponding reactions based on gene-protein-reaction (GPR) associations. Reaction expression levels were then calculated using these gene-to-reaction mappings and the established GPR rules. Reaction expression levels were categorized into three groups: lowly expressed, moderately expressed, and highly expressed. Based on the characteristics of the dataset, two thresholds were established for this categorization: a lower threshold defined as the mean expression value minus half of the standard deviation [mean − 0.5*standard deviation (STD)], and an upper threshold defined as the mean expression value plus half of the standard deviation (mean + 0.5*STD). Reactions with expression levels below the lower threshold were classified as lowly expressed, those above the upper threshold were considered highly expressed, and the reactions with expression levels falling between these two thresholds were categorized as moderately expressed. We chose 0.5*STD based on the characteristics of our dataset, as it provided a suitable distribution of reaction expression across the three expression categories.

Afterwards, iMAT was used to focus on highly expressed reactions, and to exclude lowly expressed reactions, which often have higher variability (17), from our analysis. This filtering strategy enhances computational efficiency by reducing the number of models’ reactions. In addition, the highly expressed reactions are likely to have a more significant impact on cellular functions and metabolic processes. For each sample, we generated two types of models: lung bulk tissue-specific models and lung-associated mast cells models for all samples. For the lung-associated mast cells models, we used a biomass equation similar to that of macrophages (18) in lung tissue. This approach was necessitated by the lack of a specific biomass equation for mast cells in literature. Additionally, this choice was supported by several shared characteristics between mast cells and macrophages. Both cell types share a myeloid lineage, indicating a common origin (19), and exhibit functional similarities in lung tissue, such as roles in immune response and inflammation (20). Furthermore, both mast cells and macrophages contain granular content (21), which is crucial for their immune defense roles. These similarities justify the use of the macrophage biomass equation as a proxy for mast cells, enabling us to proceed with our analysis despite the lack of mast cell-specific data. Finally, flux balance analysis (FBA) (22) was performed to obtain the reaction flux profiles for each sample. FBA is a widely used computational method in systems biology that calculates the flow of metabolites through a metabolic network, predicting the growth rate of an organism or the rate of production of an important metabolite. It uses linear programming to determine a set of metabolic fluxes that optimize a given objective function, such as biomass production, while satisfying mass balance constraint. In our study, FBA allowed us to predict the metabolic capabilities of our lung tissue and lung-associated mast cell models under steady-state conditions, providing insights into the metabolic differences between normal and cancerous states.

Machine learning analysis

Random forest classifiers for healthy and cancerous case categorizations

We employed a Random Forest classifier to distinguish between healthy and cancerous samples using reaction flux data from FBA as the input. All analyses, including model development and evaluation, were conducted on the same dataset (GSE18842, 43 paired samples). The input consists of 10,812 reaction flux as features that are padded to have the same length for all 86 samples. The dataset was split into training (80%) and test (20%) set, with stratification to ensure balanced class representation. The Random Forest model was implemented using scikit-learn with 1,000 trees, Gini impurity criterion, and a maximum depth of 32. The model’s performance was assessed using 8-fold cross-validation and out-of-bag (OOB) score estimation. Feature importance scores were utilized to identify the most discriminating reactions between healthy and cancerous samples.

Temperature effect analysis: metabolic thermodynamic sensitivity analysis (MTSA)

We developed a novel method for thermodynamics-based calculation of the effect of temperature changes on cancer growth rate and mast cell growth rate. This method, which we call MTSA, integrates various computational approaches to provide a comprehensive understanding of temperature-dependent metabolic changes. Due to the complexity of the models and limitations in kinetic data availability, we made several key assumptions. We assumed that all enzymatic reaction rates follow the Michaelis-Menten equation (23):

ν=Vmax[S]Km+[S][1]

where v is the reaction rate, V_max is the maximum reaction rate, [S] is the substrate concentration, and K_m is the Michaelis constant. Additionally, we assumed that each reaction operates at maximum driving force, as all reactions are in pseudo-steady state and occur rapidly. Furthermore, the reaction rate is calculated based on the rate-limiting substrate. Among all the substrates of a reaction, the one having the minimum value of k_cat/K_m is considered as the rate-limiting substrate. We used the Max/Min driving force (MDF) method (24) to calculate metabolite concentrations that maximize the Gibbs free energy of each reaction. To find the rate of each reaction within the feasible solution space, we employed flux variability analysis (FVA) (25) to determine the maximum flux for each reaction which served as a reference point for generating a range of potential V_max values. The k_cat values for each enzyme were estimated using the DL-Kcat method (26). Through non-linear regression, we determined the Michaelis constant and reaction rate within the feasible space for the rate-limiting substrate of each reaction. Figure 1 shows the MTSA framework. This approach allowed for the robust estimation of kinetic parameters while accounting for the inherent nonlinearity of enzyme-catalyzed reactions (see Appendix 1 for more details). This comprehensive methodology, MTSA, allowed us to investigate the effects of temperature on cancer and mast cell growth rates.

Figure 1 MTSA: an integrated framework for temperature-dependent metabolic modeling. This figure illustrates the comprehensive MTSA framework for analyzing temperature effects on metabolic networks, consisting of four integrated steps: [1] contextualizing tissue specific and mast cell specific genome scale metabolic models using Human1 model and applying iMAT, coupled with DL-Kcat model to predict the Kcat values, FVA for obtaining maximum feasible flux of the reactions, and MDF analysis for thermodynamic optimization; [2] generation of synthetic kinetic data by creating substrate concentration ranges around MDF-determined optimal points and introducing experimental variation with Gaussian noise; [3] estimation of kinetic parameters through Michaelis-Menten equation fitting with multiple initial estimates and quality assessment via R-squared statistics; and [4] analysis of solution spaces at different temperatures (36–40 ℃) to quantify metabolic temperature sensitivity within physiologically relevant ranges, ultimately enabling prediction of temperature-dependent metabolic behavior without extensive experimental measurements. FVA, flux variability analysis; iMAT, Integrative Metabolic Analysis Tool; MDF, Max/Min driving force; MTSA, metabolic thermodynamic sensitivity analysis.

Results

Cell type deconvolution analysis

Understanding the complex cellular composition of the tumor microenvironment is crucial for unraveling the mechanisms of cancer progression. Traditional bulk RNA sequencing data, while informative, presents challenges in distinguishing signals from different cell populations within heterogeneous tissue samples. To address this limitation, we employed CIBERSORTx (15), a computational deconvolution method that enables accurate estimation of cell type proportions from bulk tissue transcriptomics data. This tool is particularly valuable for our study as it allows us to quantify the relative abundance of different immune cell populations, including mast cells, in both healthy and cancerous lung tissues.

CIBERSORTx analysis revealed significant differences in immune cell populations between healthy and cancerous lung tissues, with a particularly notable finding regarding resting mast cells (Figure 2). The analysis demonstrated a highly significant reduction in resting mast cell infiltration in cancerous lung tissues compared to healthy lung tissue (mean fraction ≈0.15 in healthy vs. ≈0.08 in cancer). The strong statistical significance of this finding is supported by both a very low P value (1.315e−4) indicating the extreme unlikelihood that this difference occurred by chance, and a robust t-statistic (4.009) demonstrating that the observed difference is several standard deviations from zero. This substantial decrease suggests a potential shift in mast cell states during cancer development, raising important questions about their role in lung cancer pathogenesis. While our analysis also revealed significant alterations in other immune cell populations, including M2 macrophages (P=4.598e−12, t=−8.060), dendritic cells (resting: P=8.549e−13, t=8.424; activated: P=0.0448, t=−2.037), memory B cells (P=1.926e−6, t=5.118), and regulatory T cells (Tregs) (P=7.526e−4, t=−3.498), this study focuses specifically on understanding the implications of reduced resting mast cells in the tumor microenvironment due to their multifaceted role in lung cancer. The significant decrease in resting mast cells, coupled with the complex remodeling of the immune landscape, provides a strong rationale for investigating their potential role in lung cancer progression or suppression.

Figure 2 Immune cell composition analysis comparing healthy and cancerous lung tissue samples using the CIBERSORTx deconvolution method. The plots display the fractional abundance of 22 distinct immune cell populations, with blue representing healthy samples and red representing cancerous samples. Statistical significance between groups is indicated by asterisks (****, P≤0.0001; ***, P≤0.001; *, P≤0.05; ns: not significant). Notably, resting mast cells (highlighted by a green dashed box) show significant differences (P<0.001) between healthy and cancerous conditions, suggesting their potential role in the lung cancer microenvironment. The y-axis represents the relative fraction of each cell type, and the width of each violin plot indicates the density distribution of the data points.

Based on these significant differences in immune cell populations, particularly the reduction in resting mast cells, we proceeded to develop genome-scale metabolic models to further investigate the metabolic changes associated with lung cancer and the role of mast cells in this disease.

GSM

We successfully developed genome-scale metabolic models for both bulk lung tissue and mast cell-specific samples using the Human1 (14) model and implementing the iMAT (16) method. To validate our genome-scale metabolic models, we conducted in silico nutrient deprivation experiments by systematically analyzing the impact of essential nutrient sources (N, P, C, and S) in the models’ environmental conditions.

Our simulations demonstrated that carbon source elimination resulted in complete cessation of biomass production, as cells cannot synthesize carbon from scratch. Similarly, when all key nutrient sources were eliminated simultaneously, the models predicted complete growth arrest, aligning with the fundamental biological principle that cells cannot sustain growth without essential nutrients (27). These outcomes support the models’ validity and adherence to basic metabolic constraints, confirming their biological accuracy and predictive capabilities. The genome-scale metabolic models generated for bulk lung tissue and mast cells exhibited varying numbers of reactions and metabolites across the 86 samples. Figures S1,S2 illustrate the distribution of these reactions and metabolites across all models.

To further validate the biological relevance of our genome-scale metabolic models, we performed an additional level of model assessment by comparing key metabolic flux patterns characteristic of lung cancer, based on established literature. We leveraged well-recognized metabolic hallmarks of cancer cells to benchmark our models.

First, we evaluated the efficiency of lactate production—a hallmark of metabolic reprogramming in cancer cells known as the Warburg effect. Cancer cells often convert a greater proportion of pyruvate into lactate, even when oxygen is freely available (28,29). To assess this in our models, we quantified both the total lactate output and a “lactate efficiency” metric, defined as the ratio of lactate produced to available pyruvate in each sample (Figure 3A). As shown in Figure 3B, the bar represents the relative total lactate produced (which is slightly reduced in cancer compared to healthy samples, with a fold change of 0.86), while the second bar indicates the lactate-efficiency (which is substantially higher in cancer samples, with a fold change of 5.08). This result reveals that cancer cell models are much more efficient at converting pyruvate to lactate, mirroring the altered glycolytic activity observed in real lung cancer tissues.

Figure 3 Biological validation of genome-scale metabolic models in lung cancer and healthy samples through common metabolic flux features. (A) Schematic overview of glucose and glutamine metabolism showing entry routes into the TCA cycle, with focus on pyruvate (from glucose) and glutamine utilization in mitochondria. (B) Bar chart showing fold change in total lactate production and lactate-efficiency (lactate produced per available pyruvate) in cancer vs. healthy models. Cancer cells demonstrate markedly higher lactate-efficiency, capturing the well-known behavior of cancer cells in glycolysis pathway. (C) Paired flux analysis showing the distribution of glutamine transport across individual samples, highlighting consistently elevated glutamine flux in cancer compared to healthy models. TCA, tricarboxylic acid.

Next, we examined another characteristic feature of lung cancer metabolism: increased dependence on glutamine, often termed “glutamine addiction”. It is well documented that lung cancer cells display a strong dependence on glutamine metabolism (30,31), frequently exhibiting elevated glutamine uptake and mitochondrial import to fuel the TCA cycle and meet increased biosynthetic and energetic demands (Figure 3A). To determine if our patient-specific cancer models captured this metabolic phenotype, we analyzed the flux of glutamine transport into mitochondria. As illustrated in Figure 3C, each dot represents an individual sample, with cancer samples consistently exhibiting much higher glutamine transport flux values than healthy ones. These results reflects a dramatic increase (15.5-fold) in glutamine transport flux in cancer models relative to healthy controls, indicating markedly higher glutamine uptake.

Together, these results provide strong indirect evidence that our models faithfully recapitulate key metabolic hallmarks of lung cancer. Elevated lactate efficiency and increased glutamine uptake, as predicted by our models, are highly consistent with observations reported in experimental studies. This comprehensive validation process builds on our earlier nutrient deprivation tests and demonstrates that the underlying reaction fluxes in our reconstructed models are robust and biologically meaningful, supporting their application in downstream analyses such as machine learning-based classification and biomarker discovery in lung cancer.

Machine learning analysis

Random forest classification

Leveraging the genome-scale metabolic models developed, we employed a Random Forest classifier to distinguish between healthy and cancerous states using reaction flux features derived from FBA. The classifier achieved robust performance metrics for bulk lung tissue samples, with a mean accuracy of 0.95. The model exhibited balanced precision and recall scores of 0.95 and 0.94, respectively, resulting in an F1-score of 0.94. The confusion matrix reveals effective classification with minimal misclassification between healthy and cancer states (Table S1). For the mast cell models, the classifier achieved perfect classification metrics across all evaluation parameters, with accuracy, precision, recall, and F1-score all reaching 1.0. This exceptional performance can be attributed to the minimal variation in reaction fluxes between samples, as determined through FBA using gene expression data obtained from CIBERSORTx deconvolution. The confusion matrix demonstrates flawless separation between healthy and cancerous states, with zero misclassifications in both categories.

Multi-level analysis reveals complex metabolic reprogramming in lung cancer

Our GSM and random forest classification framework provide a powerful, multi-faceted lens through which to examine metabolic reprogramming in lung cancer. This integrated approach achieved robust differentiation between healthy and cancerous states with a mean accuracy of 0.95, while simultaneously revealing the underlying metabolic features driving this distinction. By combining reaction scores obtained through feature selection and deriving gene importances using GPR relationships; while also calculating pathway importance based on the reactions present in each specific pathway, we gained comprehensive insights spanning from pathway-level activities to individual gene contributions.

Figure 4A highlights the top 40 reactions ranked by their importance in distinguishing between healthy and cancerous states, as determined through feature selection analysis in the first level of our multi-level metabolic alteration analysis in lung cancer. Among these, a reaction involved in the production of benzoate (MAR08643) is identified as the highest-ranked reaction. The large difference in flux values between healthy and cancer samples (Figure S3) is a key reason for this reaction’s top ranking. In healthy samples, this reaction exhibits consistently high flux values, whereas in cancer samples, the flux is predominantly near zero, indicating that this reaction is mostly inactive in the cancerous state.

Figure 4 Multi-level analysis of metabolic alterations in lung cancer: gene, reaction, and pathway level results. (A) This plot displays the top 40 most important reactions in lung cancer metabolism. The reactions are ranked by their importance values (y-axis), with reaction IDs shown on the x-axis. A color gradient visually enhances the importance hierarchy. (B) The bubble plot illustrates the relationship between gene importance and their correlation with cancer state. The x-axis represents Spearman correlation coefficients (ρ), where positive values indicate upregulation and negative values indicate downregulation in cancer. Bubble sizes correspond to the relative importance of each gene, with larger bubbles indicating higher importance. Colors denote different metabolic subsystems, revealing pathway-level patterns. Notable features include CYP27A1, showing high importance (large bubble size) and strong negative correlation in bile acid biosynthesis pathway (purple), and several transport-related genes (pink) with positive correlations. The legend provides a comprehensive mapping of metabolic subsystems to their respective colors, highlighting the diverse metabolic processes involved in lung cancer metabolism. (C) Relative importance distribution of the top 12 metabolic pathways identified in lung cancer. The pie chart shows the percentage contribution of each pathway, with Isolated subsystem (15.1%), Bile acid biosynthesis (12.9%), and Nucleotide metabolism (10.8%) emerging as the most significant pathways.

The near-complete suppression of this reaction in cancer samples suggests that it is downregulated as part of metabolic reprogramming in lung cancer. This downregulation may reflect a strategy by lung cancer cells to suppress reactions involved in benzoate production. Benzoate has been implicated in modulating inflammatory responses through NF-κB activation, a pathway that plays a dual role in cancer biology. While NF-κB activation can promote tumor progression by supporting cell survival, angiogenesis, and metastasis, excessive inflammation can attract immune cells capable of attacking tumor cells (32). By reducing flux through this reaction and limiting benzoate production, lung cancer cells may avoid triggering pro-inflammatory responses that could disrupt their immunosuppressive microenvironment or harm their survival. Other notable reactions include those involved in bile acid biosynthesis, such as MAR03214 and MAR01685, which will be discussed in detail in subsequent sections.

In the next level of our analysis, we examined gene-level correlations with lung cancer metabolic reprogramming. Our correlation analysis, visualized in Figure 4B, revealed distinct patterns of both positive and negative associations, with Spearman correlations ranging from −0.8 to 0.8. The relative importance of each gene, represented by bubble size, highlighted key players in metabolic reprogramming. Most notably, sterol 27-hydroxylase, CYP27A1, emerged as the most significantly downregulated gene in lung cancer, aligning with previous studies showing reduced CYP27A1 expression in lung adenocarcinoma (33). This finding points to substantial perturbation of cholesterol metabolism in the alternative bile acid biosynthesis pathway in lung cancer cells, which we explore in detail in subsequent sections.

In addition, several NUP genes, colored green in Figure 4B, are associated with the DNA transport reaction, which facilitates the movement of DNA between the cytosol and nucleus. These genes demonstrate moderate importance but exhibit high correlations and a prominent distribution among the top-ranked genes. This confirms the involvement of previously reported NUP genes in lung cancer (34,35) while also uncovering novel associations with NUP42, NUP133 and NUP188. These genes, not previously well-established as being linked to lung cancer, were found to be upregulated in cancer samples alongside other known cancer-related nucleoporins shown in Figure 4B.

At the pathway level, our analysis revealed widespread metabolic alterations across multiple subsystems (Figure 4C). The Isolated subsystem, which includes reactions that are not easily categorized into specific metabolic pathways, emerged as the most significant contributor, accounting for 15.1% of the total importance. This was followed by the bile acid biosynthesis pathway, contributing 12.9%, and nucleotide metabolism, which ranked third with 10.8%.

The strength of our integrated approach lies in its ability to connect pathway-level changes with specific gene activities. For instance, the high importance of the bile acid biosynthesis pathway in our analysis was traced back to specific enzymatic reactions and gene expression changes, particularly the downregulation of CYP27A1. Similarly, we identified how alterations in individual genes contributed to broader pathway dysregulation, providing a mechanistic understanding of metabolic reprogramming in lung cancer. These interconnected findings demonstrate the complex nature of metabolic adaptations in lung cancer cells and highlight the importance of analyzing these changes at multiple biological levels simultaneously.

Our multi-level analysis also revealed the significant upregulation of inosine triphosphate pyrophosphatase (ITPA) in lung cancer compared to healthy tissue, aligning with previous study on NSCLC (36). ITPA’s dual involvement in nucleotide and purine metabolism pathways presented an interesting case study for our analytical approach. By simultaneously examining gene expression, reaction fluxes, and pathway relationships, we identified a specific reaction {H2O [c] + XTP [c] ⇒ H+ [c] + PPi [c] + xanthosine-5-phosphate [c]} as particularly crucial in the nucleotide metabolism pathway. While ITPA has traditionally been viewed as a housekeeping gene responsible for nucleotide pool maintenance (37), our findings suggest a more specialized role in lung cancer metabolism. Beyond its known function in hydrolyzing non-canonical purine nucleotides, the marked upregulation of ITPA appears to serve a critical function in lung cancer cell survival. This adaptation may be particularly important given the elevated demand for purines in rapidly proliferating cancer cells, where ITPA’s enhanced activity helps maintain genome stability by efficiently removing potentially mutagenic non-canonical purines like XTP from the nucleotide pool. In the following sections, we will delve into the biological relevance of the most important findings presented here, providing a detailed explanation.

Aminoacyl-tRNA biosynthesis and alternative bile acid biosynthesis pathway in lung cancer

Building on the pathway-level insights, we now turn our focus to specific metabolic pathways that exhibit notable differences between lung cancer and healthy lung tissue. In particular, we will examine the aminoacyl-tRNA biosynthesis and bile acid biosynthesis pathways to better understand their distinct roles in lung cancer metabolism. We begin with the less prominent but functionally significant aminoacyl-tRNA biosynthesis pathway, followed by the more extensively altered bile acid biosynthesis pathway.

Aminoacyl-tRNA biosynthesis in lung cancer

Our multi-level analysis revealed that while all amino acid utilization reactions in the aminoacyl-tRNA biosynthesis pathway are important, four specific amino acids—valine, isoleucine, histidine, and lysine—demonstrated higher importance scores and showed selective upregulation in lung cancer cells.

In particular, the reaction corresponding to valine synthesis emerged as highly significant in our RF classifier analysis (Figure 5A). The synthesis of valine aminoacyl-tRNA (tRNA-Val) exhibited consistently elevated activity in cancerous tissues compared to healthy samples (Figure 5B). These scatter plots comparing reaction fluxes between healthy and cancer samples demonstrated a clear upward shift, with a higher density of points appearing above the y=x line, indicating systematically higher flux values in cancer samples. The density distribution plots further supported this pattern, revealing a distinct rightward shift in cancer samples that suggests increased aminoacyl-tRNA synthesis activity. Quantitative analysis of metabolic fluxes (Figure 5C) showed that median flux values for valine aminoacylation reactions were elevated by approximately 15% in cancer cells relative to healthy tissue, representing a 1.15-fold increase.

Figure 5 Comparative analysis of aminoacyl-tRNA biosynthesis pathway reactions in lung cancer versus healthy tissue. (A) Schematic representation of the valine aminoacylation reaction [tRNA(Val)], showing the synthesis of L-valyl-tRNA(Val) from valine, ATP, and tRNA. (B) Flux comparison between healthy and cancer samples for the valine aminoacylation reaction, including scatter plot (bottom left) with y=x reference line and density distribution (upper left and bottom right). (C) Box plot showing the flux distribution comparison between healthy and cancer samples for tRNA(Val) synthesis. (D) Schematic representation of the isoleucine aminoacylation reaction [tRNA(Ile)], showing the synthesis of L-isoleucyl-tRNA(Ile) from isoleucine, ATP, and tRNA. (E) Flux comparison between healthy and cancer samples for the isoleucine aminoacylation reaction, including scatter plot (bottom left) with y=x reference line and density distribution (upper left and bottom right). (F) Box plot showing the flux distribution comparison between healthy and cancer samples for tRNA(Ile) synthesis. Both reactions demonstrate consistently higher flux values in cancer samples compared to healthy tissue, with approximately 15% elevation in median flux values, indicating increased aminoacyl-tRNA synthesis activity in lung cancer metabolism.

Similarly, the isoleucine aminoacylation reaction also emerged as highly significant in our analysis (Figure 5D). The synthesis of isoleucine aminoacyl-tRNA (tRNA-Ile) showed comparable patterns of elevated activity in cancerous tissues (Figure 5E), with quantitative analysis (Figure 5F) demonstrating similar increases in median flux values for these aminoacylation reactions in cancer cells compared to healthy tissue.

Similar upregulation patterns were observed for histidine and lysine aminoacyl-tRNA synthetase reactions (Figure S4). While our analysis underscores the importance of all amino acids in the aminoacyl-tRNA pathway, the preferential utilization of these four specific amino acids—valine, isoleucine, histidine, and lysine—suggests a specialized metabolic adaptation in lung cancer cells. This selective upregulation aligns with previous studies that have identified altered branched-chain amino acid metabolism in lung cancer (38,39), but our multi-level approach uniquely reveals the specificity of this adaptation to these four amino acids. The enhanced activity of these particular aminoacyl-tRNA synthesis reactions across the majority of cancer samples points to a precise metabolic reprogramming that may be crucial for supporting the increased energy demands and protein synthesis requirements of rapidly proliferating cancer cells, while also highlighting the overall importance of the entire aminoacyl-tRNA biosynthesis pathway in cancer metabolism.

Alternative bile acid biosynthesis in lung cancer

Bile acid biosynthesis proceeds through two distinct pathways: the classical (neutral) pathway and the alternative (acidic) pathway. While the classical pathway accounts for approximately 90% of bile acid synthesis and primarily occurs in the liver (40), our analysis highlights the significance of the alternative pathway in lung cancer. A segment of the bile acid biosynthesis pathway is illustrated in Figure 6A. It demonstrates a specific series of enzymatic reactions that begin in the cytosol and continue into the mitochondria, focusing on the conversion of cholesterol to 3alpha,7alpha,12alpha-trihydroxy-5beta-cholestanate through multiple intermediates. The process initiates with cholesterol in the cytosol, where it undergoes hydroxylation by CH25H (MAR01786) to form 26-hydroxycholesterol. This intermediate is further modified through several enzymatic steps to produce 3alpha,7alpha,12alpha-trihydroxy-5beta-cholestan-26-al. The pathway then proceeds with the action of CYP27A1 (MAR01611) to generate 3alpha,7alpha,12alpha-trihydroxy-5beta-cholestanate.

Figure 6 Analysis of CYP27A1-catalyzed reaction (MAR03214) in bile acid biosynthesis pathway. (A) Schematic representation of the bile acid synthesis pathway showing the conversion of cholesterol to bile acids through multiple steps, including the mitochondrial CYP27A1-catalyzed reaction (MAR03214) that interconverts 3alpha,7alpha,12alpha-trihydroxy-5beta-cholestan-27-al and 3alpha,7alpha,12alpha-trihydroxy-5beta-cholestanate. A detailed legend shows the steroid nucleus structure and functional groups. (B) Flux comparison between healthy and cancer samples for reaction MAR03214, including scatter plot with y=x reference line (left) and density distribution (right), showing reduced flux values in cancer samples. (C) Box plot demonstrating the flux distribution comparison between healthy and cancer samples, highlighting significantly lower flux values in cancer tissues.

A critical aspect of this pathway is the transport of metabolites across the mitochondrial membrane, facilitated by specific transporters (SLC27A5 or SLC27A2, MAR01620). Once inside the mitochondria, the pathway culminates in a reversible reaction (MAR03214) catalyzed by CYP27A1, which mediates the interconversion between 3alpha,7alpha,12 alpha-trihydroxy-5beta-cholestan-27-al and 3alpha,7alpha,12 alpha-trihydroxy-5beta-cholestanate. This mitochondrial reaction represents a key step in the alternative (acidic) pathway of bile acid biosynthesis, which, although accounting for only about 10% of total bile acid synthesis, plays a particularly significant role in lung cancer metabolism. This important reaction (MAR03214) was also identified by our RF classifier as a discriminating reaction between cancer and healthy states. The flux comparison and median flux values for this reaction are shown in Figure 6B,6C, respectively, which demonstrate significantly reduced activity in cancer cells compared to healthy tissues.

Figure 7A illustrates the other specific segment of the bile acid biosynthesis pathway, focusing on the conversion of cholesterol to 3alpha,7alpha-dihydroxy-5beta-cholestan-26-al through multiple enzymatic steps. The pathway begins with cholesterol, which contains a characteristic steroid nucleus structure comprising three hexagonal rings and one pentagonal ring, along with a variable-length carbon side chain. The initial transformation is catalyzed by CH25H through reaction MAR01784, which introduces a hydroxyl group to form 26-hydroxycholesterol.

Figure 7 Analysis of the initial CYP27A1-mediated reaction (MAR01685) in the alternative bile acid synthesis pathway. (A) Pathway diagram shows how cholesterol is transformed through multiple enzymatic steps to generate the metabolites involved in reaction MAR01685, which was identified as a key discriminating reaction showing significant downregulation in cancer compared to healthy states. The initial conversion is catalyzed by CH25H (MAR01784), followed by additional enzymatic modifications leading to the metabolites required for the CYP27A1-catalyzed reaction (MAR01685). The legend depicts the fundamental steroid nucleus structure of cholesterol and the functional groups that are attached to this structure during the sequential conversion of cholesterol to pathway intermediates. (B) Flux comparison between healthy and cancer samples for reaction MAR01685, including scatter plot with y = x reference line (left) and density distribution (right), showing reduced flux values in cancer samples. (C) Box plot demonstrating the flux distribution comparison between healthy and cancer samples, highlighting significantly lower flux values in cancer tissues.

Following this initial hydroxylation, the pathway proceeds through an intermediate step to generate 5β-cholestane-3α,7α,26-triol. The final step in this sequence is catalyzed by CYP27A1 through reaction MAR01685, which oxidizes the hydroxyl group at position 26 to form an aldehyde group, resulting in the production of 3alpha,7alpha-dihydroxy-5beta-cholestan-26-al. Throughout these transformations, the steroid nucleus structure remains intact while modifications occur primarily on the side chain and at specific positions of the ring system, as indicated by the presence of hydroxyl (-OH) groups (shown as blue circles) and the final aldehyde (-CHO) group (represented by an orange triangle) in Figure 7A.

Recent studies have demonstrated that alterations in this pulmonary bile acid synthesis pathway, particularly changes in CYP27A1 expression and activity, may be linked to various lung pathologies, including certain types of lung cancer (33). Our RF classifier identified two key reactions as important discriminating factors in the bile acid biosynthesis pathway, both catalyzed by the CYP27A1 gene (Figures 6A,7A). The altered activity of CYP27A1 affects the mitochondrial metabolism of cholesterol derivatives, potentially influencing cellular homeostasis and signaling pathways in lung cancer cells.

The flux through reactions catalyzed by CYP27A1 was significantly reduced in lung cancer samples compared to healthy tissues (Figures 6B,7B). The plots comparing reaction fluxes between healthy and cancer samples demonstrate a clear downward shift, with a higher density of points appearing below the y=x line, indicating systematically lower flux values in cancer samples compared to healthy tissues. This pattern shows two distinct characteristics in the density distribution plots: for positive fluxes, there is a leftward shift in cancer samples indicating reduced activity, while for negative fluxes, cancer samples show a marked shift toward zero flux values compared to the substantial negative fluxes observed in healthy tissues. For reaction MAR03214 (Figure 6B), the flux comparison reveals a consistent pattern of decreased activity in cancer samples, with most data points clustering below the reference line. Similarly, for reaction MAR01685 (Figure 7B), the scatter plot shows a pronounced downward deviation from the y=x line, indicating substantially reduced reaction flux in cancerous tissues. These systematic reductions are quantitatively reflected in the median flux values for these reactions, which were markedly lower in cancer cells compared to healthy tissues (Figures 6C,7C), suggesting a significant dysregulation of the CYP27A1-mediated bile acid synthesis pathway in lung cancer metabolism.

This decrease in CYP27A1 activity suggests a potential downregulation of bile acid biosynthesis in lung cancer. These reaction flux alterations show how lung cancer cells change their metabolism, particularly in amino acid processing and bile acid production. While the bulk tissue analysis provided valuable insights, we also applied our multi-level approach to the mast cell-specific models to further investigate the role of this kind of immune cell in lung cancer metabolism.

The role of mast cells in lung cancer

Given the metabolic shifts observed in lung cancer, we next sought to explore how these alterations manifest in specific immune cell populations. Since mast cells play a crucial role in the tumor microenvironment, as evidenced by Figure 2 which demonstrates their high proportion in lung cancer-healthy tissue samples and a notable difference in their distribution between healthy and cancerous tissues, we applied our framework to mast cell-specific metabolic models to further investigate their involvement in lung cancer metabolism. Using reaction fluxes obtained from FBA of the mast cell metabolic models, we implemented a Random Forest classifier to differentiate between healthy and cancerous states. Our analysis revealed significant alterations in mast cell metabolism and function in lung cancer microenvironment. Our RF classifier identified key important reactions in the metabolic network associated with mast cell function in lung cancer. Subsequently, we utilized GPR associations, as discussed earlier, related to these reactions to determine the most important genes. This analysis revealed several genes crucial to mast cell function that are significantly altered in the lung cancer state. Figure 8A displays gene importance distribution, where genes are arranged with color intensities ranging from low importance (light yellow) to high importance (dark red), effectively illustrating the varying significance of different genes in mast cell function.

Figure 8 Analysis of mast cell metabolism and metabolic characteristics in lung cancer. (A) Showing the distribution of gene importance in mast cell function, with color intensity indicating significance level (dark red), high (red), medium (orange), and low (light yellow). (B) Ranking the top 30 metabolic pathways by their importance in distinguishing mast cell behavior between normal and lung cancer conditions. (C) Schematic representation of histamine transport dynamics in mast cells, Increased histamine degranulation patterns for mast cells in lung cancer compared to healthy controls, indicating enhanced histamine release into the extracellular space. The plot compares histamine flux pattern between healthy and cancer states (Min, Max, mean, standard deviation of reaction fluxes). (D) Glutamine transport in mast cells with corresponding plot demonstrating metabolic flux differences between normal (blue) and cancer (red) conditions, suggesting increased energy demands during the lung cancer state. The plots in panels (C) and (D) quantitatively represent the differential metabolic patterns observed in normal versus cancerous states, highlighting significant alterations in both histamine and glutamine metabolism.

Metabolic crosstalk and nutrient competition

GPR-based analysis of the top random forest-ranked reactions identified several transport-related genes as central to mast cell adaptation (Figure 8A). Notably, the SLCO gene family (SLCO1B3, SLCO1B1, SLCO1A2) emerged as highly important, encoding organic anion transporting polypeptides responsible for bile acid import (41). Given the prominent alteration of bile acid biosynthesis in lung cancer, our findings suggest that cancer-associated mast cells engage in cross-compartmental metabolic interactions which may modulate the local availability and signaling of these metabolites. Furthermore, SLC7A6 (y+LAT2 transporter) was also ranked as highly important, mediating uptake of both essential cationic and neutral amino acids including glutamine and arginine—key substrates for anabolic growth and mediator production (42). SLC25A20, a mitochondrial carnitine/acylcarnitine translocase, supports increased fatty acid import for beta-oxidation (43), signifying a shift in energy homeostasis. Collectively, these changes suggest mast cells in lung cancer adapt to the nutrient landscape by enhancing uptake and catabolism of critical substrates, a likely mechanism for functional activation and survival within a metabolically competitive tumor microenvironment. These results show that fatty acid oxidation is one of the main energy supplies for mast cell activation and function in the context of lung cancer metabolism.

The identification of these transport-related genes as highly important through the GPR analysis emphasizes how mast cell function in lung cancer is heavily dependent on their ability to maintain proper nutrient exchange and energy metabolism. Their collective high importance suggests that metabolic reprogramming, particularly in terms of amino acid transport and energy generation, is a crucial aspect of mast cell adaptation to the tumor microenvironment.

Cytokine and mediator signaling

Figure 8B quantifies the relative importance of different metabolic pathways, with tyrosine metabolism, transport reactions, exchange and demand reactions, and fatty acid oxidation as the most important pathways. The consistent height of several top pathways indicates multiple metabolic processes are simultaneously altered in cancer-associated mast cells. In addition to transport reactions and fatty acid oxidation (with their role in mast cell activation was already explained in the previous section), pathway-level analysis (Figure 8B) highlighted tyrosine metabolism as one of the most significantly altered pathways in cancer-associated mast cells. This metabolic shift is consistent with enhanced activity of the KIT receptor pathway, a canonical regulator of mast cell activation (44). Upon engagement by stem cell factor (SCF), KIT signaling triggers downstream phosphorylation events that activate key effectors (e.g., ERK, AKT) (45), promoting mast cell survival, proliferation, and increased biosynthesis of inflammatory and bioactive mediators. The upregulation of tyrosine metabolism thus provides a mechanistic link to the observed increase in secretion of proteases (such as tryptase and chymase) and other cytokines, which can remodel the extracellular matrix (46), facilitate tumor cell invasion, and reshape the immunological domain. This metabolic alteration, alongside other pathways shown in Figure 8B, creates an interconnected network that supports mast cell functions and tumor growth and progression in lung cancer, suggesting potential therapeutic targets through tyrosine kinase inhibition.

Immune suppression and tumor promotion

The most striking functional changes are captured in Figure 8C,8D, which reveal two key adaptations in the reaction-level analysis. Figure 8C reveals an amplified histamine flux pattern in cancer conditions (red area) compared to normal conditions, with notably higher values across statistical parameters. Specifically, cancer conditions show elevated maximum flux rates and a higher mean value, while maintaining a broader range between minimum and maximum values. The increased standard deviation in cancer conditions suggests more variable histamine flux dynamics, while the median values indicate that this enhanced histamine transport is sustained rather than driven by occasional spikes. This enhanced histamine transport suggests mast cells become more actively degranulating in the lung cancer environment. This Functional flux analysis provided further mechanistic resolution. Specifically, the enhanced histamine secretion is a hallmark of mast cell degranulation and has immune-suppressive consequences, including inhibition of anti-tumor T cell responses (47).

The metabolic shift is further highlighted in Figure 8D, where the radar plot compares glutamine transport parameters between healthy and cancer conditions. The healthy cases exhibit relatively balanced values, with the minimum and maximum being roughly similar and a standard deviation close to zero. In contrast, the cancer cases display a negative minimum, a positive maximum, and a significantly larger standard deviation. The shift of glutamine transport reaction fluxes towards negative in cancer conditions likely reflects the enhanced glutamine uptake required to meet the heightened energy demands of mast cells within the tumor microenvironment. This increased glutamine consumption in cancer-associated mast cells, suggesting metabolic competition with effector immune cells, further supporting an immune-suppressive and tumor-promoting microenvironment. These findings collectively demonstrate the complex metabolic adaptations of mast cells in lung cancer and highlight their active participation in shaping the tumor microenvironment through both metabolic reprogramming and enhanced degranulation responses.

Temperature-dependent metabolic analysis

Research has demonstrated that temperature plays a significant role in cancer therapy, particularly in enhancing immune responses against tumors (48). This insight into the thermal sensitivity of cancer cells and immune functions has motivated further investigations into temperature effects on the growth rates of both cancer cells and immune cells, such as mast cells, in various microenvironments including the lung. Given the extensive reaction network in Human1 and its intricate pathway interactions, we developed the MTSA method to specifically investigate temperature effects on both cancer cell growth and mast cell behavior in the lung cancer microenvironment. This approach is particularly valuable as it provides insights into how temperature variations affect the complex interplay between lung cancer metabolism and mast cell function. Our analysis using MTSA revealed distinct biomass distribution patterns between healthy and cancerous conditions in both lung tissue and mast cells within the lung cancer microenvironment (Figures 9,10). We investigated the clinically significant temperature spectrum of 36–40 ℃, which spans from normal body temperature (37 ℃) to fever states frequently observed in cancer patients. The plots demonstrate a clear bimodal distribution in biomass production across this temperature range, with notable differences between healthy and diseased states.

Figure 9 Tissue-level biomass distribution analysis comparing healthy and cancerous lung tissue across temperature range 36–40 ℃. The plots demonstrate that cancerous tissue (red) maintains higher mean biomass values compared to healthy tissue (blue), with peak production occurring at different optimal temperatures: 37 ℃ for cancerous tissue (83.87) and 38 ℃ for healthy tissue (82.52), suggesting distinct temperature-dependent metabolic preferences between the two conditions. The distributions demonstrate a gradual decline in biomass production as temperature increases, with a notable decrease at 40 ℃ for both tissue types.

Figure 10 Temperature-dependent biomass distribution in healthy and cancerous mast cells analyzed using the MTSA method. The plots show the contrasting biomass patterns between healthy (blue) and cancerous (red) mast cells across temperatures from 36 to 40 ℃. Healthy mast cells maintained high biomass values (0.62–0.64) with slight variations across temperatures, showing particular resilience at fever-range temperatures (38–40 ℃). In contrast, cancerous mast cells maintained a consistently low biomass (0.10) across all temperature points, suggesting that the lung cancer microenvironment significantly suppresses mast cell growth and metabolic activity. This dramatic reduction in biomass indicates that the tumor microenvironment creates conditions that substantially impair mast cell proliferation. MTSA, metabolic thermodynamic sensitivity analysis.

In lung tissue samples, cancerous tissue consistently demonstrated higher mean biomass values compared to healthy tissue across all temperature points (Figure 9). Healthy tissues (blue) exhibited optimal biomass production at 38 ℃, while patient tissues showed peak biomass at 37 ℃. As temperatures increased beyond these optimal points, both tissue types exhibited a gradual decrease in biomass production. We found that both tissue types exhibited their lowest biomass production at 40 ℃, with cancerous tissues showing a steeper reduction than healthy tissues.

The analysis of mast cell populations revealed a striking contrast between healthy and cancerous cells across different temperatures (Figure 10). Healthy mast cells exhibited significantly higher mean biomass values compared to their cancerous counterparts at all temperature points. In healthy mast cells, the mean biomass started at approximately 0.64 at 36 ℃, dipped slightly to 0.62 at 37 ℃, and then recovered to stabilize around 0.63–0.64 in the fever range (38–39 ℃). At 40 ℃, the biomass remained steady at 0.63, demonstrating remarkable thermal resilience. In contrast, cancerous mast cells consistently displayed a mean biomass of approximately 0.10 across all temperatures, with minimal variation in response to temperature changes. This stark difference suggests that the tumor microenvironment substantially inhibits mast cell proliferation and metabolic activity, likely through alterations in signaling pathways or resource availability. The robust thermal response observed in healthy mast cells, particularly in the fever range, may reflect an evolutionary adaptation linked to their immune function. However, this adaptive capability appears to be lost in the cancerous state, as cancerous mast cells fail to exhibit significant biomass changes across temperatures. These findings underscore the potential influence of thermal regulation on mast cell function and its broader implications for cancer progression.

Discussion

Our study presents key contributions to understanding lung cancer metabolism through the development of genome-scale metabolic models for both lung tissue and mast cells in the context of cancer. These comprehensive models, built upon the extensive Human1 framework, provide unique insights into the metabolic reprogramming that occurs in lung cancer and its immune microenvironment. Our multi-level approach, which integrates gene expression, reaction fluxes, and pathway analysis, offers a powerful platform for investigating the complex metabolic interactions between cancer cells and immune components.

Our cell-type deconvolution analysis revealed a significant reduction in resting mast cells within cancerous tissues, providing the foundation for our subsequent metabolic investigations. The analysis demonstrated substantial alterations in immune cell populations, particularly highlighting the critical role of mast cells in the tumor microenvironment. This finding motivated our development of specialized metabolic models to understand the functional implications of these cellular changes.

The machine learning analysis of our metabolic models, combined with our multi-level analytical approach, revealed critical alterations in several key pathways. Most significantly, we discovered a specialized metabolic adaptation in lung cancer cells involving the selective upregulation of four specific amino acids (valine, isoleucine, histidine, and lysine) in the aminoacyl-tRNA pathway. The synthesis reactions for these amino acids showed approximately 15% higher flux values in cancerous tissues compared to healthy tissue. This preferential utilization suggests a precise metabolic strategy employed by cancer cells to meet their elevated energy requirements. By simultaneously analyzing gene, reaction and pathway importances, we were able to identify not only which amino acids were upregulated but also understand the broader context of how this selective usage supports cancer cell proliferation and survival. In addition, our multi-level analysis not only revealed previously known dysregulated genes, reactions and pathways in lung cancer, but also discovered new important dysregulations like NUP42, NUP133, and NUP188 genes functions, which are not well studied in lung cancer.

The subsequent analysis of mast cell metabolism in the tumor microenvironment revealed striking adaptations, including enhanced histamine transport and increased glutamine consumption. These changes suggest a shift toward a more active state in cancer-associated mast cells. The identification of key transport-related genes, including the SLCO family members and SLC transporters, highlights the critical role of metabolite transport in mast cell function within the tumor environment. These findings provide new perspectives on how mast cells adapt their metabolism to support or potentially inhibit tumor progression.

A major methodological advancement of our study is the development of the MTSA method, which addresses a critical gap in cancer metabolism research. By integrating thermodynamic principles with GSM, MTSA enables systematic investigation of temperature-dependent metabolic changes. This approach is particularly significant, as it allows for the quantitative prediction of how thermal fluctuations—arising from fever, inflammation, or therapeutic interventions—may influence both cancer progression and immune cell function. The ability of MTSA to analyze multiple metabolic pathways while accounting for temperature effects provides a powerful framework for studying the intricate relationship between thermal conditions and cellular metabolism in cancer. This innovative approach has broad applicability beyond the scope of our current study. The MTSA method can be readily adapted to investigate metabolic responses to temperature changes in various fields of biology, including different organisms, other cancer types, and diverse immune cell populations. Its versatility makes it a valuable tool for exploring temperature-dependent metabolic phenomena in fields such as immunology, microbiology, plant biology, and environmental science. Applying MTSA to our models revealed distinct temperature-dependent metabolic patterns that differentiate cancerous from healthy states. Notably, while healthy lung tissue exhibited peak biomass production at 38 ℃, cancerous tissue maintained higher biomass production at 37 ℃, suggesting fundamental differences in temperature-dependent metabolic regulation. More strikingly, cancer-associated mast cells showed a dramatic reduction in biomass production across all temperatures (36–40 ℃), indicating a profound metabolic impairment in these immune cells within the tumor microenvironment. These findings highlight the potential impact of temperature fluctuations on both tumor progression and immune function.

These discoveries open several promising avenues for future research and therapeutic interventions. The selective upregulation of specific aminoacyl-tRNA synthesis pathways suggests potential therapeutic targets, particularly in combination with existing treatments that disrupt protein synthesis in cancer cells. Additionally, the temperature-dependent metabolic patterns identified through MTSA highlight the potential for temperature-based therapeutic strategies, which could influence the balance between cancer cell growth and immune cell function.

Our genome-scale metabolic models and the MTSA method provide valuable tools for advancing cancer metabolism research. While our computational approach offers comprehensive metabolic insights, it relies on steady-state assumptions that may not fully capture the dynamic nature of cancer metabolism. Additionally, the use of bulk tissue data, despite our deconvolution methods, may not completely account for cellular heterogeneity within the tumor microenvironment. While the GSE18842 dataset provides high-quality paired gene expression profiles, it lacks comprehensive clinical and molecular annotation including treatment history, smoking status, and key driver mutations (e.g., EGFR, KRAS, ALK, TP53). These factors are known to significantly influence metabolic programs in lung cancer, and their absence limits our ability to perform stratified analyses that could enhance the interpretability and translational relevance of our findings. The lack of mutation status information particularly restricts our understanding of how different molecular subtypes might exhibit distinct metabolic reprogramming patterns. Future studies incorporating comprehensive clinical and molecular annotation would be valuable for validating and extending our findings across different lung cancer subtypes and patient populations. Nevertheless, our multi-level approach successfully identified meaningful metabolic signatures that warrant further investigation. Future experimental validation, particularly focusing on the selective amino acid utilization patterns and temperature-dependent metabolic adaptations of both cancer cells and mast cells, could help overcome these limitations and pave the way for novel therapeutic strategies. Furthermore, while the metabolic biomarkers identified in this study show promise for enhancing early detection and monitoring of lung cancer progression, their clinical utility needs to be validated in larger patient cohorts.

Through the integration of GSM and advanced computational techniques, our study provides novel insights into lung cancer metabolism and the role of mast cells in the tumor microenvironment. Our multi-level approach has revealed previously unknown metabolic adaptations, particularly in amino acid utilization, while the introduction of MTSA has opened new possibilities for understanding temperature-dependent metabolic regulation in cancer.

Our study has several limitations that should be acknowledged. First, our genome-scale metabolic models rely on steady-state assumptions that may not fully capture the dynamic nature of cancer metabolism. Cancer cells exist in continuously evolving microenvironments with fluctuating nutrient and oxygen levels, which our models cannot completely represent. Second, despite employing sophisticated deconvolution methods, our use of bulk tissue data may not adequately account for the significant cellular heterogeneity within tumor microenvironments. This heterogeneity includes diverse cancer cell subpopulations and varying states of immune cells that could impact metabolic profiles.

While our computational approach successfully identified meaningful metabolic signatures, experimental validation will be essential. Future studies should focus on validating the selective amino acid utilization patterns and temperature-dependent metabolic adaptations we observed in both cancer cells and mast cells. Additionally, although the metabolic biomarkers identified show promise for enhancing early detection and monitoring of lung cancer progression, their clinical utility requires validation in larger, more diverse patient cohorts.

Conclusions

Our GSM and machine learning analysis of lung cancer and mast cells has revealed significant metabolic reprogramming with potential clinical implications. Through our multi-level analytical approach, we identified several key metabolic signatures that distinguish cancerous from healthy states. Firstly, our cell-type deconvolution analysis demonstrated a significant reduction in resting mast cells within cancerous tissues, providing clear evidence of altered immune cell composition in the lung cancer microenvironment.

The selective upregulation of specific amino acids (valine, isoleucine, histidine, and lysine) in the aminoacyl-tRNA pathway of lung cancer cells represents a precise metabolic adaptation that supports the increased energy demands of rapidly proliferating cancer cells. Concurrently, the downregulation of the CYP27A1-mediated bile acid synthesis pathway indicates a significant metabolic shift away from cholesterol metabolism in lung cancer. These specific alterations in metabolic pathways could serve as potential biomarkers for early detection and monitoring of lung cancer progression.

Our analysis of mast cell metabolism revealed enhanced histamine transport and increased glutamine consumption in the tumor microenvironment, suggesting a shift toward immunomodulatory functions that may contribute to tumor progression. Additionally, our novel MTSA demonstrated impaired biomass production in cancerous mast cells across physiological temperatures (36–40 ℃), highlighting metabolic vulnerabilities that could be exploited therapeutically.

These findings collectively provide a comprehensive understanding of the metabolic interplay between lung cancer cells and mast cells, offering new perspectives for targeted therapeutic interventions. The identified metabolic adaptations and the development of the MTSA method represent significant advancements in cancer metabolism research with potential applications in other cancer types and biological systems. Future experimental validation of these computational findings could lead to the development of novel diagnostic tools and therapeutic strategies for lung cancer.

Acknowledgments

We gratefully acknowledge the Holland Computing Center (HCC) at the University of Nebraska-Lincoln for providing high-performance computing resources and technical support that contributed to the results presented in this research.

Footnote

Reporting Checklist: The authors have completed the TRIPOD checklist. Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-555/rc

Peer Review File: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-555/prf

Funding: This work was supported by the National Institutes of Health (NIH) MIRA Award (No. 5R35GM143009).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-555/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Li C, Lei S, Ding L, et al. Global burden and trends of lung cancer incidence and mortality. Chin Med J (Engl) 2023;136:1583-90. [Crossref] [PubMed]
2. Herath S, Sadeghi Rad H, Radfar P, et al. The Role of Circulating Biomarkers in Lung Cancer. Front Oncol 2021;11:801269. [Crossref] [PubMed]
3. Komi DEA, Mortaz E, Amani S, et al. The Role of Mast Cells in IgE-Independent Lung Diseases. Clin Rev Allergy Immunol 2020;58:377-87. [Crossref] [PubMed]
4. Marichal T, Tsai M, Galli SJ. Mast cells: potential positive and negative roles in tumor biology. Cancer Immunol Res 2013;1:269-79. [Crossref] [PubMed]
5. Oldford SA, Marshall JS. Mast cells as targets for immunotherapy of solid tumors. Mol Immunol 2015;63:113-24. [Crossref] [PubMed]
6. Maciel TT, Moura IC, Hermine O. The role of mast cells in cancers. F1000Prime Rep 2015;7:09.
7. Batta I, Patial R, Sobti RC, et al. Computational Biology in the Discovery of Biomarkers in the Diagnosis, Treatment and Management of Cardiovascular Diseases. Cardiol Cardiovasc Med 2024;8:405-14.
8. Passi A, Tibocha-Bonilla JD, Kumar M, et al. Genome-Scale Metabolic Modeling Enables In-Depth Understanding of Big Data. Metabolites 2021;12:14. [Crossref] [PubMed]
9. Razmpour T, Tabibian M, Roohi A, et al. GAN-Enhanced Machine Learning and Metabolic Modeling Identify Reprogramming in Pancreatic Cancer. bioRxiv. 2025;2025–06. doi: 10.1101/2025.06.01.657244.
10. Baloni P, Dinalankara W, Earls JC, et al. Identifying Personalized Metabolic Signatures in Breast Cancer. Metabolites 2020;11:20. [Crossref] [PubMed]
11. Zhang C, Aldrees M, Arif M, et al. Elucidating the Reprograming of Colorectal Cancer Metabolism Using Genome-Scale Metabolic Modeling. Front Oncol 2019;9:681. [Crossref] [PubMed]
12. Larsson I, Uhlén M, Zhang C, et al. Genome-Scale Metabolic Modeling of Glioblastoma Reveals Promising Targets for Drug Development. Front Genet 2020;11:381. [Crossref] [PubMed]
13. Zhang B, Shi H, Wang H. Machine Learning and AI in Cancer Prognosis, Prediction, and Treatment Selection: A Critical Approach. J Multidiscip Healthc 2023;16:1779-91. [Crossref] [PubMed]
14. Robinson JL, Kocabaş P, Wang H, et al. An atlas of human metabolism. Sci Signal 2020;13:eaaz1482. [Crossref] [PubMed]
15. Newman AM, Steen CB, Liu CL, et al. Determining cell type abundance and expression from bulk tissues with digital cytometry. Nat Biotechnol 2019;37:773-82. [Crossref] [PubMed]
16. Zur H, Ruppin E, Shlomi T. iMAT: an integrative metabolic analysis tool. Bioinformatics 2010;26:3140-2. [Crossref] [PubMed]
17. Eling N, Richard AC, Richardson S, et al. Correcting the Mean-Variance Dependency for Differential Variability Testing Using Single-Cell RNA Sequencing Data. Cell Syst 2018;7:284-294.e12. [Crossref] [PubMed]
18. Bordbar A, Lewis NE, Schellenberger J, et al. Insight into human alveolar macrophage and M. tuberculosis interactions via metabolic reconstructions. Mol Syst Biol 2010;6:422. [Crossref] [PubMed]
19. Galli SJ, Borregaard N, Wynn TA. Phenotypic and functional plasticity of cells of innate immunity: macrophages, mast cells and neutrophils. Nat Immunol 2011;12:1035-44. [Crossref] [PubMed]
20. Chen L, Deng H, Cui H, et al. Inflammatory responses and inflammation-associated diseases in organs. Oncotarget 2018;9:7204-18. [Crossref] [PubMed]
21. Katsoulis-Dimitriou K, Schmidt J, Roder M, et al. Ingestion of mast cell granules boosts macrophages and drives atypical programming. J Allergy Clin Immunol 2025;156:774-89. [Crossref] [PubMed]
22. Orth JD, Thiele I, Palsson BØ. What is flux balance analysis? Nat Biotechnol 2010;28:245-8. [Crossref] [PubMed]
23. Michaelis L, Menten ML, Johnson KA, et al. The original Michaelis constant: translation of the 1913 Michaelis-Menten paper. Biochemistry 2011;50:8264-9. [Crossref] [PubMed]
24. Noor E, Bar-Even A, Flamholz A, et al. An integrated open framework for thermodynamics of reactions that combines accuracy and coverage. Bioinformatics 2012;28:2037-44. [Crossref] [PubMed]
25. Mahadevan R, Schilling CH. The effects of alternate optimal solutions in constraint-based genome-scale metabolic models. Metab Eng 2003;5:264-76. [Crossref] [PubMed]
26. Li F, Yuan L, Lu H, et al. Deep learning-based k cat prediction enables improved enzyme-constrained model reconstruction. Nature Catalysis 2022;5:662-72.
27. Son S, Stevens MM, Chao HX, et al. Cooperative nutrient accumulation sustains growth of mammalian cells. Sci Rep 2015;5:17401. [Crossref] [PubMed]
28. Mathew M, Nguyen NT, Bhutia YD, et al. Metabolic Signature of Warburg Effect in Cancer: An Effective and Obligatory Interplay between Nutrient Transporters and Catabolic/Anabolic Pathways to Promote Tumor Growth. Cancers (Basel) 2024;16:504. [Crossref] [PubMed]
29. De Leon-Oliva D, González-Prieto P, De Castro-Martinez P, et al. Revisiting the biological role of the Warburg effect: Evolving perspectives on cancer metabolism. Pathol Res Pract 2025;273:156151. [Crossref] [PubMed]
30. Jin J, Byun JK, Choi YK, et al. Targeting glutamine metabolism as a therapeutic strategy for cancer. Exp Mol Med 2023;55:706-15. [Crossref] [PubMed]
31. Zhou L, Zhang Q, Zhu Q, et al. Role and therapeutic targeting of glutamine metabolism in non‑small cell lung cancer Oncol Lett 2023;25:159. (Review). [Crossref] [PubMed]
32. Yilmaz B, Karabay AZ. Food Additive Sodium Benzoate (NaB) Activates NFκB and Induces Apoptosis in HCT116 Cells. Molecules 2018;23:723. [Crossref] [PubMed]
33. Yin Y, He M, Huang Y, et al. Transcriptomic analysis identifies CYP27A1 as a diagnostic marker for the prognosis and immunity in lung adenocarcinoma. BMC Immunol 2023;24:37. [Crossref] [PubMed]
34. Nong JS, Zhou X, Liu JQ, et al. Nucleoporin 107 is a prognostic biomarker in hepatocellular carcinoma associated with immune infiltration. Cancer Med 2023;12:10990-1009. [Crossref] [PubMed]
35. Roy A, Narayan G. Oncogenic potential of nucleoporins in non-hematological cancers: recent update beyond chromosome translocation and gene fusion. J Cancer Res Clin Oncol 2019;145:2901-10. [Crossref] [PubMed]
36. Alghamdi RA, Al-Zahrani MH. Integrated bioinformatics analyses identifying key transcriptomes correlated with prognosis and immune infiltrations in lung squamous cell carcinoma. Saudi J Biol Sci 2023;30:103596. [Crossref] [PubMed]
37. Schroader JH, Handley MT, Reddy K. Inosine triphosphate pyrophosphatase: A guardian of the cellular nucleotide pool and potential mediator of RNA function. Wiley Interdiscip Rev RNA 2023;14:e1790. [Crossref] [PubMed]
38. Bian M, Huang S, Yu D, et al. tRNA Metabolism and Lung Cancer: Beyond Translation. Front Mol Biosci 2021;8:659388. [Crossref] [PubMed]
39. Sivanand S, Vander Heiden MG. Emerging Roles for Branched-Chain Amino Acid Metabolism in Cancer. Cancer Cell 2020;37:147-56. [Crossref] [PubMed]
40. Chiang JYL, Ferrell JM. Bile Acid Metabolism in Liver Pathobiology. Gene Expr 2018;18:71-87. [Crossref] [PubMed]
41. Schulte RR, Ho RH. Organic Anion Transporting Polypeptides: Emerging Roles in Cancer Pharmacology. Mol Pharmacol 2019;95:490-506. [Crossref] [PubMed]
42. Bhutia YD, Ganapathy V. Glutamine transporters in mammalian cells and their functions in physiology and cancer. Biochim Biophys Acta 2016;1863:2531-9. [Crossref] [PubMed]
43. Tonazzi A, Giangregorio N, Console L, et al. The Mitochondrial Carnitine Acyl-carnitine Carrier (SLC25A20): Molecular Mechanisms of Transport, Role in Redox Sensing and Interaction with Drugs. Biomolecules 2021;11:521. [Crossref] [PubMed]
44. Longo V, Catino A, Montrone M, et al. Controversial role of mast cells in NSCLC tumor progression and angiogenesis. Thorac Cancer 2022;13:2929-34. [Crossref] [PubMed]
45. Sagi-Eisenberg R. Biogenesis and homeostasis of mast cell lysosome related secretory granules. Front Cell Dev Biol 2025;13:1603999. [Crossref] [PubMed]
46. Zhao XO, Sommerhoff CP, Paivandy A, et al. Mast cell chymase regulates extracellular matrix remodeling-related events in primary human small airway epithelial cells. J Allergy Clin Immunol 2022;150:1534-44. [Crossref] [PubMed]
47. Thangam EB, Jemima EA, Singh H, et al. The Role of Histamine and Histamine Receptors in Mast Cell-Mediated Allergy and Inflammation: The Hunt for New Therapeutic Targets. Front Immunol 2018;9:1873. [Crossref] [PubMed]
48. Abreu MM, Chocron AF, Smadja DM. From cold to hot: mechanisms of hyperthermia in modulating tumor immunology for enhanced immunotherapy. Front Immunol 2025;16:1487296. [Crossref] [PubMed]