Computational prediction of changes in brain metabolic fluxes during Parkinson's disease from mRNA expression

Abstract

Background: Parkinson's disease is a widespread neurodegenerative disorder which affects brain metabolism. Although changes in gene expression during disease are often measured, it is difficult to predict metabolic fluxes from gene expression data. Here we explore the hypothesis that changes in gene expression for enzymes tend to parallel flux changes in biochemical reaction pathways in the brain metabolic network. This hypothesis is the basis of a computational method to predict metabolic flux changes from post-mortem gene expression measurements in Parkinson's disease (PD) brain.

Results: We use a network model of central metabolism and optimize the correspondence between relative changes in fluxes and in gene expression. To this end we apply the Least-squares with Equalities and Inequalities algorithm integrated with Flux Balance Analysis (Lsei-FBA). We predict for PD (1) decreases in glycolytic rate and oxygen consumption and an increase in lactate production in brain cortex that correspond with measurements (2) relative flux decreases in ATP synthesis, in the malate-aspartate shuttle and midway in the TCA cycle that are substantially larger than relative changes in glucose uptake in the substantia nigra, dopaminergic neurons and most other brain regions (3) shifts in redox shuttles between cytosol and mitochondria (4) in contrast to Alzheimer's disease: little activation of the gamma-aminobutyric acid shunt pathway in compensation for decreased alpha-ketoglutarate dehydrogenase activity (5) in the globus pallidus internus, metabolic fluxes are increased, reflecting increased functional activity.

Conclusion: Our method predicts metabolic changes from gene expression data that correspond in direction and order of magnitude with presently available experimental observations during Parkinson's disease, indicating that the hypothesis may be useful for some biochemical pathways. Lsei-FBA generates predictions of flux distributions in neurons and small brain regions for which accurate metabolic flux measurements are not yet possible.

Fig 1. Diagram for the workflow of the Lsei-FBA approach.

Flow diagram of the steps to predict metabolic fluxes for the normal brain (green boxes) and for diseased brain based on gene expression data (pink boxes) described in the Methods section. For the normal brain, the flux distribution was computed from a reconstructed model of cerebral central carbon metabolism. For the diseased brain, mRNA gene expression fold changes were first computed for patients with Parkinson’s disease (PD) versus a control group. An initial flux estimate for the diseased brain is computed for each reaction in the network by multiplying gene expression fold changes with the FBA flux predictions for the normal brain. The final flux estimate is solved subject to i) forward flux in irreversible reactions, ii) maintaining the balance of fluxes during chronic disease and iii) a least squares cost function to minimize the sum of the squared deviations between the initial and the final flux estimate.

Fig 2. Flux distribution in healthy brain and during Parkinson’s disease.

Flux distribution in healthy brain (A) and during Parkinson’s disease in the substantia nigra and dopaminergic neurons (B, average from eight data sets), averaged value for frontal cortex, BA9, putamen and cerebellum (C) and value for globus pallidus internus region (D) in μmol g (wet) brain^-1 min^-1. Black numbers, flux during normal condition; green numbers, flux decreased during PD and red numbers, increased from the normal condition. Note that for clarity not all separate biochemical steps are plotted: oxaloacetate is for instance first transaminated to aspartate before being transported across the mitochondrial membrane as part of the malate-aspartate shuttle. GLC, glucose; G3P, glyceraldehyde 3-phosphate; RU5PD, ribulose-5-phosphate; PYR, pyruvate; LAC, lactate; CIT, citrate; AKG, alpha-ketoglutarate; SUCC, succinate; MAL, malate; OAA, oxaloacetate; GLU, glutamate; GLN, glutamine, GABA, 4-aminobutanoate (synonym of gamma-aminobutyrate); O2, oxygen; OxPhos, oxidative phosphorylation. Flux values from GLC to RU5PD and from RU5PD to G3P represent 6-carbon units leaving the GLC pool rather than 3-carbon units entering the G3P pool.