An atlas of human metabolism

Abstract

Genome-scale metabolic models (GEMs) are valuable tools to study metabolism and provide a scaffold for the integrative analysis of omics data. Researchers have developed increasingly comprehensive human GEMs, but the disconnect among different model sources and versions impedes further progress. We therefore integrated and extensively curated the most recent human metabolic models to construct a consensus GEM, Human1. We demonstrated the versatility of Human1 through the generation and analysis of cell- and tissue-specific models using transcriptomic, proteomic, and kinetic data. We also present an accompanying web portal, Metabolic Atlas (https://www.metabolicatlas.org/), which facilitates further exploration and visualization of Human1 content. Human1 was created using a version-controlled, open-source model development framework to enable community-driven curation and refinement. This framework allows Human1 to be an evolving shared resource for future studies of human health and disease.

Fig. 1.. Overview of Human1 generation and curation.

A simplified illustration of the key steps involved in the generation of Human1 from HMR2, Recon3D, and iHsa. The bottom of the diagram represents the ongoing open-source curation of Human1 using input from databases, literature, other models, and the scientific community. The four side panels provide further detail into selected Human1 features: extensive reaction mass and charge balancing to achieve 100% stoichiometric consistency; incorporation of new enzyme complex information; mapping model components to standard database identifiers; version-controlled and open-source model curation framework. In the bar graphs in the upper left panel, “Balanced” reactions represent the number of mass-balanced reactions, “Consistent” metabolites are the number of stoichiometrically consistent metabolites, and “R3D model” is the model version of Recon3D.

Fig. 2.. Highlighted features provided by the Metabolic Atlas web portal.

A collection of screen captures from Metabolic Atlas, illustrating key features such as 2D and 3D metabolic network maps. A zoomed inset shows a subset of the endoplasmic reticulum compartment map, from which further information on components such as reactions, enzymes, or metabolites can be accessed in the GEM Browser. Interaction partner graphs are dynamically generated for any given enzyme or metabolite in Human1, which show the connectivity to other metabolites and enzymes based on their associated reactions.

Fig. 3.. Structural and functional comparison of cancer- and healthy tissue-specific GEMs.

(A) Visualization of differences in models’ reaction content using a tSNE projection to two dimensions based on the Hamming similarity. See Fig. S5 for individual point labels. (B) Heatmap showing pairwise comparisons of reaction content between GEMs specific to healthy liver (CHOL-NT, LIHC-NT, and Liver-GTEx), blood, and their corresponding cancers (CHOL, LIHC, and LAML). (C) Relative subsystem coverage (number of reactions present in a model that are associated with the given subsystem) compared among GEMs of liver and liver tumors. Only subsystems with at least a 10% deviation from mean subsystem coverage among the models are shown. (D) Summary of metabolic task performance by the healthy and cancerous liver models, showing only the tasks that differed in at least one of the models. (E) Comparison of relative subsystem coverage between LAML- and blood-specific GEMs, showing only subsystems with at least a 10% deviation between the two models. (F) A summary of the five metabolic tasks that could be completed by the LAML GEM but failed in the healthy blood GEM. ROS, reactive oxygen species; GSL, glycosphingolipid; FA, fatty acid; [p], peroxisomal compartment; DHA, docosahexaenoic acid.

Fig. 4.. Predicted gene essentiality among different cell lines and human GEMs.

(A) Schematic illustration of the generation of cell-line-specific GEMs from HMR2, Recon3D, and Human1, and subsequent prediction of gene essentiality based on the GEMs’ ability to perform basic metabolic tasks. Genes predicted to be essential by the GEMs were compared to experimental measures of gene essentiality (45, 49) obtained from CRISPR knockout screens. (B) Comparison of gene essentiality predictions among the three reference GEMs and their 5 derivative cell line models with CRISPR screen results from Hart et al. (45). Left: Average accuracy, specificity, and sensitivity of predictions across the 5 cell lines for each reference GEM, with error bars representing the standard error of the mean. Right: Comparison of the Matthews Correlation Coefficient (MCC) of the predictions for each of the reference GEMs and cell lines. The “All” category indicates genes found to be essential in all 5 cell lines. (C) Comparison of gene essentiality predictions among the three reference GEMs and their 621 derivative cell line models with CRISPR screen results from the DepMap database (49).

Fig. 5.. Generation and analysis of human ecGEMs.

(A) Graphical representation of the pipeline used to construct NCI-60 cell-line specific ecGEMs from Human1. (B) Cumulative distribution of flux variability among reactions in HOP62-GEM and ecHOP62-GEM. Only the ~3,200 reactions that carried a flux of >10⁻⁸ mmol/gDW h when optimizing biomass production in HOP62-GEM were included in the plot. Distributions for all 11 cell lines are shown in Fig. S12. (C) Comparison of predicted with measured exchange fluxes (log₁₀-transformed absolute flux values) for the 11 cell-specific ecGEMs, where only the set of metabolites present in the growth medium (Ham’s medium) was specified. Different colored markers represent the different cell lines. Metabolites whose fluxes were systematically under- or over-predicted among the different models are labeled in circles, whereas the other ~78% lie within the shaded oval. Note that metabolites along the bottom of the plot have a predicted flux of zero but are shown here as having the absolute minimum measured value to avoid logarithm of zero. (D) Boxplots showing the relative error in predicted growth rate for the 11 cell-specific ecGEMs and non-ecGEMs. “Unbounded” indicates that the solutions are effectively unbounded and therefore have unquantifiable (infinite) error. Colored markers on the x-axis denote the exchange constraints that were cumulatively added to the models when making predictions. “Media” indicates that only the metabolites present in the growth medium were specified, without constraining their exchange rates. “Glucose”, “Lactate”, and “Threonine” indicate that the exchange flux for those metabolites in the model were constrained to the measured value.