A multi-tissue type genome-scale metabolic network for analysis of whole-body systems physiology

Abstract

Background: Genome-scale metabolic reconstructions provide a biologically meaningful mechanistic basis for the genotype-phenotype relationship. The global human metabolic network, termed Recon 1, has recently been reconstructed allowing the systems analysis of human metabolic physiology and pathology. Utilizing high-throughput data, Recon 1 has recently been tailored to different cells and tissues, including the liver, kidney, brain, and alveolar macrophage. These models have shown utility in the study of systems medicine. However, no integrated analysis between human tissues has been done.

Results: To describe tissue-specific functions, Recon 1 was tailored to describe metabolism in three human cells: adipocytes, hepatocytes, and myocytes. These cell-specific networks were manually curated and validated based on known cellular metabolic functions. To study intercellular interactions, a novel multi-tissue type modeling approach was developed to integrate the metabolic functions for the three cell types, and subsequently used to simulate known integrated metabolic cycles. In addition, the multi-tissue model was used to study diabetes: a pathology with systemic properties. High-throughput data was integrated with the network to determine differential metabolic activity between obese and type II obese gastric bypass patients in a whole-body context.

Conclusion: The multi-tissue type modeling approach presented provides a platform to study integrated metabolic states. As more cell and tissue-specific models are released, it is critical to develop a framework in which to study their interdependencies.

Figure 1

General properties and characteristics of individual cell-specific metabolic reconstructions. (A) We modeled three cell-specific metabolic networks of human tissues: hepatocyte from liver, myocyte from skeletal muscle, and adipocyte of adipose tissue. (B) The tissue specific properties are broken up into three main sections. The first describes the topological and knowledge base characteristics of the metabolic networks. The second shows the in vitro growth rate and the required non-glucose carbon amount to maintain that growth rate. This growth rate was used as a maintenance function to model biological turnover. The third section details the energy and oxidative capacities of the networks. Note that the growth rate was set as a constraint for these simulations. (C) The biomass maintenance functions for the three metabolic networks were built based on the individual dry cell weight compositions. The AM is primarily composed of lipids while the MM is of protein. The HM has a more balanced composition with protein, glycogen, and lipids being the major components.

Figure 2

Workflow for building the cell-specific reconstructions. Human Recon 1 [5] was utilized as the starting point for modeling human metabolism. The genome sequence annotation upon which Recon 1 was built was updated, and all incorrectly lumped and balanced reactions were corrected. Multiple reaction and enzyme databases, including UniProt, were then used to build a draft model. The draft model was then manually curated using literature and multiple high-throughput data sets. In addition, important metabolic pathways that were not in Recon 1 (e.g. hepatic ketogenesis) were added accordingly to the cell-specific reconstructions.

Figure 3

Comparison of the metabolic reactions in the three cell-specific reactions. (A) The three cell-specific reconstructions have a similar number of reactions in most subsystems. It is interesting to note that the AM has much fewer amino acid metabolic reactions and the MM fewer lipid metabolic reactions. The reaction number difference is in accordance with the cell composition differences. The HM has the most reactions in most subsystems due to the liver's varied metabolic states. For comparison, the number of reactions in each subsystem is also provided for Recon 1. (B) The three cell-specific reconstructions share the most intracellular metabolic reactions that seem to be a "core" group of reactions required for most human cells. The rest of the reactions are mostly exclusive. The AM and MM have very little reaction similarity that is not also found in the HM.

Figure 4

Schematic of the multi-tissue modeling approach. The three cell-specific reconstructions are combined into a multi-tissue model by connecting them all to a new blood compartment. Metabolites enter the model through the extra-system through exchange reactions. Metabolites are then imported into the different cells through gene associated intercellular transporters and/or free diffusion. For differentiating the cell-specific models, all reactions in the model were annotated with [a], [h], [m], and [bl] for the AM, HM, MM and blood compartment, respectively.

Figure 5

The Alanine and Cori cycles of human metabolism. (A) The Alanine and Cori cycles are methods for peripheral tissues to receive glucose under nutrient limited situations. Gluconeogenic substrates (e.g. alanine and lactate) are released from peripheral tissues and absorbed by the liver to produce glucose. The glucose is then returned to the peripheral tissues for their metabolic requirements. (B) The flux spans for the HM and MM under individual or integrated simulations are shown. The multi-tissue modeling approach has a constraining effect on the HM and MM models (see Table 1). GLC = glucose, ALA = alanine, PYR = pyruvate, LAC = lactate.

Figure 6

The absorptive state of human metabolism. (A) In the absorptive state, food is digested and absorbed primarily as glucose and amino acids. The schematic shows influx of glucose into the blood and the modeling-defined fractions of intake into each of the three models. Essential amino acids and fatty acids were also provided. For the multi-tissue simulations, the AM model stores triacylglycerol, the MM stores protein and glycogen, and the HM stores glycogen. Some of the glucose delivered to the HM was converted to fatty acids that are transported to the AM for triacylglycerol production. (B) The flux spans for the three cell-specific reconstructions individually and when integrated are shown. Integration had a constraining effect on the HM and MM models, but had an opposite effect on the AM. This was due to fatty acid production by the HM that was then transported to the AM. GLC = glucose, GGN = glycogen, AA = amino acids, FAs = fatty acids.

Figure 7

Workflow and characteristics of context-specific models. (A) Two context-specific multi-tissue metabolic networks were built using post-absorptive exchange constraints, gene expression data, the GIMME algorithm, and flux variability analysis. The two models detailed the metabolism of obese and type II diabetes obese individuals based on the prolonged starvation multi-tissue simulation. (B) We compared the reaction activity of the two context-specific models. First, we looked at how reactions were expressed based solely on the gene expression data (left column). Second, we looked at reaction activity by determining the flux variability of the two context-specific models from the expression data and GIMME algorithm (right column).