Abstract
Parkinson's disease (PD) is the second most common neurodegenerative disease in the world. Identification of PD biomarkers is crucial for early diagnosis and to develop target-based therapeutic agents. Integrative analysis of genome-scale metabolic models (GEMs) and omics data provides a computational approach for the prediction of metabolite biomarkers. Here, we applied the TIMBR (Transcriptionally Inferred Metabolic Biomarker Response) algorithm and two modified versions of TIMBR to investigate potential metabolite biomarkers for PD. To this end, we mapped thirteen post-mortem PD transcriptome datasets from the substantia nigra region onto Human-GEM. We considered a metabolite as a candidate biomarker if its production was predicted to be more efficient by a TIMBR-family algorithm in control or PD case for the majority of the datasets. Different metrics based on well-known PD-related metabolite alterations, PD-associated pathways, and a list of 25 high-confidence PD metabolite biomarkers compiled from the literature were used to compare the prediction performance of the three algorithms tested. The modified algorithm with the highest prediction power based on the metrics was called TAMBOOR, TrAnscriptome-based Metabolite Biomarkers by On–Off Reactions, which was introduced for the first time in this study. TAMBOOR performed better in terms of capturing well-known pathway alterations and metabolite secretion changes in PD. Therefore, our tool has a strong potential to be used for the prediction of novel diagnostic biomarkers for human diseases.
Similar content being viewed by others
Introduction
Parkinson’s disease (PD) is the most common neurodegenerative motor function disorder, which affects 0.1% of the population older than 40 years1. This ratio reaches 2–3% for the population older than 652. The world’s population is aging as a result of the improvements in health care and life quality, causing an increase in the PD incidence rate. The number of people with PD is expected to double between 2015 and 20403. The current treatments mainly relieve symptoms, but there is no neuroprotective treatment yet4.
The main pathological features of PD are the loss of the dopaminergic neurons in the substantia nigra region and the accumulation of α-synuclein-containing Lewy bodies in those neurons2. PD is defined as a complex disease with multiple underlying genetic and environmental factors5. Oxidative stress and mitochondrial dysfunction are the primary metabolic hallmarks of the degeneration of neurons in PD6. Gene expression profiling studies are widely used to investigate molecular mechanisms of the disease7,8,9. Numerous PD transcriptome datasets are available in databases like Gene Expression Omnibus (GEO)10.
Although clinical symptoms are widely utilized for the diagnosis of Parkinson’s disease, pathological indications of the disease appear at the molecular level several years before the clinical symptoms. Therefore, the investigation of preclinical biomarkers of PD is essential to diagnose the disease in the early stages, develop target-based therapeutical agents, and distinguish PD from other Parkinsonian diseases. Research on predicting molecular biomarkers commonly focuses on differential gene expression analysis to put forward gene biomarkers11,12. Yet, measuring gene-based biomarkers is hard compared to measuring metabolite-based biomarkers in terms of laboratory labor, cost, and time. Decrease in dopamine levels is a well-known PD metabolite biomarker13, and defining more metabolite-based biomarkers can be crucial in early diagnosis of PD before dopamine deficiency arises.
Biomarker prediction by integrative usage of computational models and omics data is an active research field14,15. TIMBR (Transcriptionally Inferred Metabolic Biomarker Response)16 is an algorithm that predicts metabolite biomarkers from the constraint-based analysis of genome-scale metabolic networks. TIMBR basically uses gene expression changes to calculate the network demands required for the secretion of metabolites, and compares them between two states. Before, it was successfully applied to predict early metabolite markers for toxicant-induced organ damages in rats17,18,19 and to analyze the changes in the secretion rates of metabolites for different experimental mouse models of PD20. Genome-scale metabolic models (GEMs) provide a mathematical platform for the integration of omics data for predicting condition specific metabolic behaviors of organisms. The publication of the first genome-scale metabolic network of Homo sapiens, Recon121, made it possible to investigate human physiology, pathology, and pharmacology22,23. The unified human GEM (Human-GEM)24 is the last released human GEM. The human GEMs are commonly used for the systematic investigation of the metabolic mechanisms of diseases25,26,27, including brain-related diseases28,29,30.
In this study, thirteen PD related transcriptome datasets were mapped on Human-GEM by the TIMBR algorithm and its modified versions to provide a powerful computational biomarker prediction approach. Transcriptome datasets from the most affected brain region in PD, substantia nigra, were chosen to better capture PD-associated metabolic alterations. Metabolites predicted to have changed secretion profiles for the majority of the datasets were considered to be potential biomarkers with a meta-analysis approach. Metabolite and pathway-based metrics were used to compare the algorithms. A reaction on–off based modification of TIMBR algorithm introduced in this study, called TAMBOOR (TrAnscriptome-based Metabolite Biomarkers by On–Off Reactions), was found to be the most powerful approach. The metabolites predicted by TAMBOOR were suggested as candidate biomarkers for PD.
Materials and methods
The methodology of this study is summarized in Fig. 1.
The basic pipeline followed in this study for predicting metabolite biomarkers by the integration of omics data and genome-scale metabolic networks.
Human-GEM
The generic genome-scale metabolic model of Homo sapiens (Human-GEM)24 version 1.11.0 was used in this study. The model consists of 13,069 reactions, 3,067 genes and 8,366 metabolites. The simulations were performed in Ham’s medium, which consists of 44 metabolites (Supplementary Table S1)24, and three additional metabolites, which are taurine, ornithine and NH3. The uptakes of those metabolites were allowed in the simulations whereas the uptakes of all other metabolites in the model were blocked by setting the lower bounds of the rates through corresponding exchange reactions to zero. Maximum glucose and oxygen uptake rates were respectively constrained as 0.32 μmol/g/min and 1.76 μmol/g/min, the values from literature for the resting state brain metabolism31,32. Maximum uptake rates of fourteen amino acids (histidine, isoleucine, leucine, lysine, methionine, phenylalanine, threonine, tryptophan, valine, tyrosine, cysteine, arginine, ornithine and taurine), NH3 and other carbon source metabolites in Ham’s medium were constrained to be 1/10 of the glucose uptake rate, based on their reported relative rates31. As an additional constraint, the biomass production rate was constrained to be at least 0.0001 μmol/g/min to ensure that macromolecule synthesis is active in all conditions.
Transcriptome data
Thirteen PD-related human transcriptome datasets presented in Table 1 were used to analyze changes in metabolite secretions in the substantia nigra region by using the biomarker prediction algorithms. Some of these datasets contain samples from different brain regions and/or different neurodegenerative diseases and expression profiles coming from distinct microarrays. Since the substantia nigra is the most affected brain region in PD, only post-mortem substantia nigra samples belonging to PD and control groups were considered. GSE8397 dataset includes expression profiles of lateral and medial substantia nigra regions separately. Therefore, 14 different comparisons from 13 datasets were evaluated in this study. The datasets include 112 control samples and 141 PD samples in total.
The datasets were downloaded from Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/) in non-normalized format. The normalizations were performed as detailed in our previous study40. Briefly, robust multi-chip average (RMA) and quantile normalization methods were applied for background correction and normalization of the microarray datasets whereas the RNA-seq datasets were normalized by the Trimmed Mean of M-values (TMM) method. Principal Component Analysis was applied to each normalized dataset to check for outliers40. Sample GSM506020 from dataset GSE20164, sample GSM208633 from dataset GSE8397, sample GSM663086 from dataset GSE26927, samples GSM509556 and GSM509557 from dataset GSE20333, samples GSM4054769 and GSM4054767 from dataset GSE136666, and samples GSM606624, GSM606625 and GSM606626 from dataset GSE20292 were determined as outliers and excluded from further analysis. GSE49036 contains PD samples including Braak stage information. Only the samples having Braak stage between 3 and 6 were assumed as PD samples.
Biomarker prediction via TIMBR and alternative algorithms
The transcriptionally inferred metabolic biomarker response (TIMBR) algorithm16 was used here to predict biomarker potential of extracellular metabolites in PD patients. Briefly, transcriptome data from controls and PD patients were separately mapped on the human genome-scale metabolic network, Human-GEM, using GPR (gene-protein-reaction) rules defined in the model for each reaction. In this mapping, expression values of genes were assigned to reactions as follows: For the reactions controlled by a single enzyme, the gene expression value of the responsible gene was directly assigned to the reaction. The maximum of the expression values of responsible genes was assigned to the reactions controlled by isoenzymes, which are different enzymes catalyzing the same reaction independently. The minimum of the expression values was assigned to the reactions controlled by enzyme complexes, which consist of at least two enzymes that catalyze the reaction together. The expression values assigned to the reactions with this approach were referred to as reaction scores. Then, for each metabolite with a defined secretion reaction in the model, a linear-programming based optimization was performed to predict a flux distribution that satisfied a high secretion flux through the metabolite (Fig. 2). The TIMBR algorithm solves an optimization problem with the minimization of the weighted sum of fluxes, where the weights of the objective function are calculated as fold changes of transcriptome-based reaction scores from the GPR rules (control/disease for disease case, and disease/control for control case simulations). With this weight-based optimization strategy, reactions with the higher objective function coefficient values (lower expression values in the case of interest) were forced to take lower flux values. In this way, a flux vector is predicted such that it is correlated with the upregulation/downregulation information at mRNA level. The algorithm we implemented here includes some modifications to the original algorithm16. The modified parts of TIMBR were detailed in our previous study20.
Modified TIMBR and TAMBOOR approaches for predicting metabolite biomarkers based on the gene expression changes. (a) Reaction weights are calculated as fold changes of transcriptome-based reaction scores obtained by mapping transcriptome data on the genome-scale metabolic model. The minimization of the weighted sum of fluxes is used as the objective function to predict flux distribution correlated with the upregulation/downregulation information. 90% of the maximum secretion flux through the related metabolite is used as a constraint in the optimization. (W: weight, S: reaction score, V: flux vector of reactions, Vmetabolite: the secretion rate of the metabolite of interest and Xmetabolite: the network demand) (b) Toy models to show how flux distributions differ based on the reaction score fold changes in control and disease states. The weight-based minimization strategy forces fluxes to pass through reactions with lower objective function coefficient values. (c) Network demand and production score calculations in modified TIMBR and TAMBOOR approaches. The summation of the internal fluxes is calculated as the network demand in the modified TIMBR whereas the number of active reactions is used to represent the network demand in TAMBOOR. Then, network demand differences between two conditions are used to calculate production scores.
In the TIMBR algorithm, the sum of weighted reaction fluxes is minimized (Eq. 1) by setting the lower boundary of the metabolite secretion reaction of interest to 90% of its maximum rate (Eq. 2). In this way, it is possible to calculate minimum network demand for the production of each metabolite. In this optimization, irreversible form of the metabolic model obtained by using convertToIrreversible function in the COBRA Toolbox41 was used to calculate flux values. The weighted network demand is calculated as a result of the optimization.
In Eq. (1), W is the weight vector consisting of the reaction weights as fold changes from transcriptome data, and V is flux vector of reactions. In Eq. (2), Vmetabolite is the secretion rate of the metabolite of interest. In Eq. (3), Xmetabolite is the network demand. Based on these equations, network demands were calculated for both PD and control conditions separately for each metabolite. Human-GEM was used with the described constraints in the previous section to apply the TIMBR algorithm. Network demand differences between the two conditions were used to calculate production scores as in Eq. (4)16. The metabolites with a high production score were considered as biomarker metabolites with increased production potential in PD, whereas the metabolites with a low production score were accepted as metabolites with reduced secretion in PD.
The alternative versions of TIMBR considered here differs in terms of how network demand (Xmetabolite) is represented. We used two different alternatives here (Fig. 2). The first one, also applied in our previous study20, uses the sum of the internal fluxes as the network demand required for the production of the metabolite as shown in Eq. (5), where Vinternal is the internal fluxes.
Assuming a correlation between enzyme levels and fluxes, a lower value of the sum of fluxes in the disease condition compared to the control condition implies a lower total enzyme requirement for the high-level production of a metabolite in the disease condition (Eq. (2)). This can be interpreted as a higher tendency towards the production of the metabolite in the disease condition. The original TIMBR formulation (Eq. (3)), on the other hand, multiplies each flux with the transcriptome-based fold changes.
As a second alternative, Xmetabolite was calculated based on the number of active reactions in the flux distributions predicted by the optimization defined in Eq. (1). Reactions with flux values higher than 10–5 were assumed to be “on” (active) reactions, while the others were assumed to be “off” (inactive). Xmetabolite was calculated as shown in Eq. (6).
In this alternative approach, we compared the path lengths for the high-level production of a metabolite in disease and control cases. A shorter path means a low number of enzymes. Thus, the shorter path in the disease state for the production of a given metabolite can be interpreted as an indication of higher production potential for that metabolite since its production is possible with a smaller number of enzymes (Fig. 2). The production score calculation and interpretation steps for this new modified version are the same as other versions of TIMBR as explained above. This reaction on–off based approach (Eq. (6)) led to the most promising results among the three versions (see results), and it was named as TrAnscriptome-based Metabolite Biomarkers by On–Off Reactions (TAMBOOR).
Metabolite-based pathway enrichment analysis
Metabolite enrichment analysis was performed using the Metabolites Biological Role (MBROLE) 2.0 online tool42 (https://csbg.cnb.csic.es/mbrole2/) for metabolites whose secretion behavior were predicted to be changed in PD. A merged list consisting of metabolites whose secretion rates were predicted to be increased or decreased in at least 7 comparisons out of 14 comparisons for the substantia nigra region was utilized for the enrichment analysis. All metabolites analyzed in this study were given in Supplementary Information S1 with the corresponding number of times they were predicted as increased/decreased. The Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway database43 library was selected to identify significantly affected metabolic pathways. The KEGG IDs of metabolites were used to apply the enrichment analysis based on the KEGG pathway database library in MBROLE.
Results and discussion
In Human-GEM, 1036 metabolites can be secreted in silico based on our simulations on maximizing metabolite productions. For each of the fourteen PD-control comparisons, the production scores of 1036 secreted metabolites in the model were calculated with TIMBR and the two modified versions (Fig. 2) to predict potential increase or decrease in their secretion rates in the disease condition. Metabolites detected in the first 25th percentile of production score ranking were assumed to have a high probability of secretion in PD patients while metabolites detected in the last 25th percentile were considered to have a high probability of secretion in controls. These metabolites were defined as candidate biomarkers for the disease.
Comparative evaluation of biomarker prediction approaches
Biomarker prediction powers of the three approaches (original TIMBR, modified TIMBR and TAMBOOR) were initially compared based on three metrics: the number of true predictions based on well-known PD metabolite biomarkers, the number of PD-related pathways in the metabolite enrichment analysis, and the number of unique metabolites included in the PD-related enriched pathways.
The first metric is the number of true predictions, which is determined based on the number of comparisons in which the well-known PD-related changes in metabolite secretion rates were truly predicted. These changes are increase in lactate44, glutamate45,46 and decrease in dopamine47, eumelanin2 and serotonin secretions48. The number of comparisons in which the well-known PD-related metabolite production changes were correctly predicted was given in Fig. 3a for each algorithm. Metabolites predicted as biomarkers in at least 7 comparisons out of 14 comparisons were assumed to have a high potential for being biomarkers and considered in further analysis. TIMBR and TAMBOOR could predict increase in lactate secretion and decrease in dopamine and serotonin secretions successfully whereas the modified TIMBR approach could predict only the decrease in eumelanin level.
Predictions of the well-known PD-associated metabolite secretion changes and PD-associated pathways by three algorithms. (a) The well-known PD-associated metabolite secretion changes and the number of comparisons in which these metabolite secretion changes were correctly predicted. (b) Significantly enriched PD-associated pathways and the number of related metabolites whose secretion rates were predicted to be altered by the given approach.
The second and third metrics are related to enriched pathways. In order to compare metabolite enrichment analysis results, seven PD-related pathways, which are Parkinson’s disease, tyrosine metabolism, oxidative phosphorylation, TCA cycle, pentose phosphate pathway, glutathione metabolism and phenylalanine metabolism, were selected based on KEGG pathway annotation. Metabolite biomarker lists predicted by each algorithm were subjected to enrichment analysis to identify if they are commonly involved in these PD-related pathways. Significantly enriched PD-associated pathways were presented in Fig. 3b, with the number of pathway metabolites whose secretion rates were predicted to be altered. All of the seven PD-related pathways were found to be significantly enriched with the metabolites captured by the TAMBOOR approach. Parkinson’s disease, tyrosine metabolism, oxidative phosphorylation and phenylalanine metabolism terms were enriched based on the original TIMBR predictions. Only Parkinson’s disease and tyrosine metabolism terms were enriched when the metabolite enrichment analysis was applied to the metabolites predicted by the modified TIMBR. Different pathways identified to be enriched can be associated with the same altered metabolite. The main purpose of the biomarker prediction approaches applied here is to predict as many reliable metabolite biomarkers as possible. Therefore, the number of unique metabolites included in all of the significantly enriched PD-related pathways is also a critical metric for metabolite-level biomarker predictions, and it was used as the third metric in our analysis.
For a further examination of the predictive power of the three approaches, metabolites predicted as high-potential biomarkers by each approach were searched in the literature to determine whether the predicted changes are compatible with experimental studies. The predicted high-potential biomarkers supported by at least two experimental studies were assumed to be high-confidence biomarkers. A total of 25 metabolite-level changes were identified as high-confidence biomarkers, as listed in Table 2. The list of high-confidence biomarkers and the list of well-known PD-associated metabolite level changes (5 metabolites in Fig. 3a) add up to 30 metabolites in total, which were compared with the predicted metabolites by the three approaches to calculate statistical metrics: precision and recall. The number of predictions, true positive values, and calculated statistical metrics for each approach are given in Table 3. Precision was used to get information about how many positive predictions overlap with the known positives (Fig. 3a and Table 2) whereas recall was used to get information about how many known positives could be correctly predicted. The calculated recall values show that TAMBOOR can predict more than half of the known positives.
The results of the metrics used to compare different biomarker prediction approaches were represented in Fig. 4 for each algorithm. The highest number of true predictions, the highest number of PD-related enriched pathways, the highest number of unique metabolites included in the PD-related enriched pathways, and the highest precision and recall values were obtained by TAMBOOR. Based on these metrics, TAMBOOR was suggested as the most powerful approach to predict metabolite biomarkers for PD.
Metrics for the comparison of the biomarker prediction approaches. (a) Statistical metrics. (b) Metrics based on well-known metabolic changes.
The performance of computational prediction approaches can differ based on the properties of the used model and data. In our previous work20, the changes in the metabolite secretion rates for different experimental mouse models of PD were successfully predicted with the modified TIMBR approach while it did not perform well in this study. Whereas a brain-specific metabolic network model of mouse (iBrain674-Mm) with 992 reactions controlled by 674 genes was used in that study for the production potentials of 44 secreted metabolites, the model used in the current study, Human-GEM, enabled the screening of 1036 secreted metabolites. The logic of the modified TIMBR relies on the sum of internal fluxes, and Human-GEM consists of numerous metabolite transport reactions between cell compartments. The sum of the fluxes of those transport reactions may have dominated the overall sum for the modified TIMBR approach, obscuring the differences between controls and patients in a larger metabolic network model.
Altered metabolite productions in PD based on TAMBOOR predictions
Extracellular levels of 118 metabolites were predicted to be increased in at least 7 comparisons with TAMBOOR, whereas levels of 156 metabolites were predicted to be decreased in at least 7 comparisons out of 14 comparisons (Supplementary Information S1). The decrease in dopamine production, the well-known metabolic change in PD, was detected in 10 of 14 comparisons. Also, the level of dopamine precursors, tyrosine and L-dopa, were predicted to be decreased for 9 and 7 comparisons respectively. In addition, an increase in lactate production and a decrease in serotonin level, other well-known changes in PD, were detected in 9 and 7 comparisons respectively.
TAMBOOR predicted 14 out of 25 high-confidence biomarkers, with 11 of them predicted only by TAMBOOR. Three of them are salsolinol, 1,2-dehydrosalsolinol, and 4-methyl-2-oxopentanoate (ketoleucine), whose productions were predicted to be increased in 10 comparisons. Salsolinol is a dopamine-derived compound. Its structure is very similar to 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP), a compound used to create experimental PD models. It can be described as a neurotoxin that contributes to the loss of dopaminergic neurons in PD60. Increased salsolinol levels were detected in the cerebrospinal fluid of PD patients with dementia97,98. Besides, the increase in salsolinol levels was associated with visual hallucinations in PD59. 4-methyl-2-oxopentanoate, another metabolite predicted by TAMBOOR, inhibits mitochondrial complex I, and its levels were reported to be increased in the plasma and cerebrospinal fluid (CSF) of PD patients in metabolome studies73,74. The productions of acetone, formate, urea and biliverdin were also predicted to be increased in 7–8 comparisons. Elevated serum levels of acetone and formate were reported in PD by a metabolome study63. Both metabolites have roles in mitochondrial functions, which are affected in PD. Urea levels have also been reported as elevated for different brain regions in PD patients with dementia68. It was demonstrated that increased serum level of biliverdin is correlated with reactive oxygen species (ROS) level and disease severity69. Productions of carnosine, nicotinamide and calcitriol (vitamin D) were predicted to be decreased in 9, 8 and 7 comparisons, respectively. A reduced level of carnosine was suggested as a diagnostic marker for AD91. Carnosine administration was tested as a treatment strategy in a mouse model of PD, and the results showed that daily carnosine intake can reduce PD symptom progression92. It was shown that experimentally created nicotinamide deficiency caused dopamine reduction, cell loss, and consequently parkinsonism90. It was also reported that elevated levels of nicotinamide N-methyltransferase reduces nicotinamide level and consequently Complex I activity in idiopathic PD89. Vitamin D deficiency was observed in PD patients and calcitriol, which is the active form of vitamin D, was shown to be neuroprotective for PD animal models87,88.
In addition to the well-known metabolite level changes and the high-confidence biomarker list, several other metabolites predicted by TAMBOOR were associated with Parkinson's metabolism and neurodegeneration, and, they can be considered as candidate biomarkers. The increase in the production of hexanoic acid (hexanoylcarnitine), a medium-chain acylcarnitine, was predicted in 11 of 14 comparisons. In an experimental study99, the contribution of hexanoylcarnitine accumulation to oxidative stress in rat brains was demonstrated. Hexanoylcarnitine accumulation debilitates the antioxidant defense of the brain by decreasing glutathione levels. Hexanoylcarnitine increase has also been reported as an age-related change in rat brain medial prefrontal cortex100. Consequently, hexanoic acid/hexanoylcarnitine can be considered as a candidate biomarker for PD. Acetyl-threonine production was predicted to be increased in 10 comparisons. In a study, almost a two-fold change in acetyl-threonine level was detected between control cells and cells exposed to 100 nM rotenone, a chemical used to create experimental PD models101. It is also known that acetyl-threonine has roles in neuronal growth and different neuronal functions102. Therefore, a change in its level is supposed to be related to neurodegenerative diseases. The productions of heparan sulfate proteoglycans, glycan, glycochenodeoxycholate and leukotriene D4 (LTD4) were predicted to be increased in 9 comparisons in this study. Cell surface heparan sulfate proteoglycans promote neuronal internalization of α-synuclein and also other amyloid forming proteins in neurodegenerative diseases by contributing to neuronal binding and uptake of protein fibrils103. Therefore, the predicted increase in heparan sulfate proteoglycans production is in line with the main pathology of PD, α-synuclein aggregation. Glycan was suggested as a potential biomarker for neurological diseases104. An elevated level of glycochenodeoxycholate was reported for mild cognitive impairment (MCI) and Alzheimer's disease (AD)105. Leukotriene D4 is a subgroup of cysteinyl leukotrienes produced by the oxidation of arachidonic acid. Cysteinyl leukotrienes, inflammatory lipid mediators, have been associated with neurodegenerative diseases, including PD106. An increase in xanthurenate level was reported in serotonin deficiency107. The decrease in serotonin levels is one of the well-known metabolic changes in PD. Therefore, predicting increased xanthurenate production in 8 comparisons is consistent with the experimental studies.
Production of histamine was predicted to be decreased in 11 comparisons. Histamine has important roles in modulating striatal synaptic transmission and behavior108. Since striatum is one of the most affected brain regions in PD, reduced histamine secretion in PD should be confirmed in further studies. Production of chloride was predicted to be decreased in 10 comparisons. Chloride is an electrolyte that has an essential role in controlling neuronal excitability. Low levels of chloride have been associated with dyskinesia in PD patients109. Cysteinyl-glycine and arginine productions were predicted to be decreased in 8 comparisons. Reduced cysteinyl-glycine was detected as a marker for levodopa-induced oxidative stress in PD patients110. Arginine level in the CSF of PD patients was reported to be low compared to the controls84.
Metabolites that were predicted to have altered secretion behavior in PD were searched on Human Metabolome Database (HMDB)111 to check if they were previously associated with any disease. Many of them were found to be linked to colorectal cancer in different studies. In recent years, impairments in the gut and neurological disorders have been closely associated with the gut-brain axis theory. It has been known that gut microbiota affects both PD and colorectal cancer (CRC) pathogenesis. Different studies have reported that PD patients have a significantly decreased risk for CRC112,113.
Most of our predictions are compatible with the experimental evidence in the literature. However, there are also some predictions conflicting with literature reports. For instance, a decline was reported for most of the amino acids, including lysine and taurine, in PD84,114. Contrarily, lysine and taurine productions were predicted to be increased by TAMBOOR. Most of the metabolites whose levels were predicted to be decreased by TAMBOOR are the metabolites of arachidonic acid metabolism, such as prostaglandins, lipoxins, hydroxyeicosatetraenoic acids (HETEs), and thromboxane A2. We could not link all of those alterations directly to PD based on literature survey. However, it was reported that the daily intake of arachidonate increases the PD risk115. Also, it was suggested that prostaglandins promote protein aggregation in PD57. Therefore, a decrease in the levels of arachidonic acid metabolism-related metabolites is not expected. Fe3 + production was predicted to be decreased for 8 comparisons, although elevated iron levels were associated with PD116.
Most of the TAMBOOR-based predictions with strong evidence could not be predicted by the original and modified TIMBR approaches. Decreased dopamine and serotonin production and increased lactate production capacities could not be predicted by the modified TIMBR. Decreased production capacity of dopamine precursors, tyrosine and L-dopa, was only detected by TAMBOOR. Modified TIMBR and original TIMBR, on the other hand, wrongly predicted increased production of L-dopa. Vitamin D deficiency, nicotinamide deficiency, reduced carnosine level, which is previously suggested as a diagnostic biomarker for neurodegenerative diseases, and elevated urea and salsolinol levels, which are previously detected as a metabolic change in neurodegenerative diseases including PD, were also only predicted by TAMBOOR. On the other hand, there are also some high-confidence metabolite level changes having strong literature evidence such as an increase in hydroxide level, and a decrease in glutathione and Apolipoprotein A1 levels, which could not be captured by TAMBOOR, but correctly predicted by the other algorithms.
GEM-based prediction algorithms like TIMBR and TAMBOOR have some limitations. For example, they assume that reaction activities are directly correlated with the expression levels of genes coding for the enzymes that catalyze the reactions. Another assumption is the scoring of the secretion tendency of a metabolite based on the number of active reactions or the weighted sum of fluxes through active reactions. Although these assumptions are biologically relevant, they cannot be exactly true for all cases. Still, it is important to develop a GEM-based metabolite biomarker prediction algorithm with a high predictive power to pave the way for novel PD biomarkers that can be useful for early diagnosis. The most common molecular-level biomarker prediction strategy in literature is discovering gene biomarkers using differential gene expression analysis. However, TIMBR and TAMBOOR predict metabolite biomarkers, which are easier to measure compared to gene biomarkers by taking samples from blood or CSF.
Pathway-based changes in PD based on TAMBOOR predictions
Functional enrichment analysis was applied to the metabolites predicted to have altered secretion by TAMBOOR to identify metabolic pathways significantly enriched with those metabolites. Many PD-related and brain function-related terms were detected as significantly enriched (The False Discovery Rate (FDR) < 0.05). The list of significantly enriched pathways is given in Table 4 (All enrichment results are available in Supplementary Information S2). The comprehensive literature survey presented in this section shows that the majority of the enriched terms/pathways are associated with PD.
Phenylalanine metabolism, tyrosine metabolism, glutathione metabolism, and, arachidonic acid metabolism terms were listed as significantly enriched pathways in Table 4. Phenylalanine and tyrosine are precursors for the synthesis of catecholamines, including dopamine synthesis. Besides, dopaminergic neuron loss is one of the main pathological features of PD65. Hence, significant alterations in phenylalanine, tyrosine, and catecholamine metabolisms are expected in PD. Glutathione acts as an antioxidant and a redox regulator in brain metabolism. Glutathione level reduction is known as the first sign of oxidative stress during PD progression117. Therefore, affected glutathione metabolism is also one of the expected results for PD. Intermediate products of arachidonic acid metabolism have essential roles in the central nervous system, such as neurotransmitter transfer, synaptic signaling, and neuronal firing. Altered arachidonic acid metabolism was associated with different brain-related disorders, including PD118. Studies have reported that arachidonic acid consumption increases the risk of PD115 and suggested enzymes in the arachidonic acid metabolism as drug targets119.
Mitochondrial dysfunctions in energy metabolism and oxidative stress are important factors in PD pathogenesis6. The terms oxidative phosphorylation, pentose phosphate pathway, and ATP-binding cassette transporters, presented as significantly enriched pathways in Table 4, are also directly related to energy metabolism and oxidative stress. ATP-binding cassette (ABC) transporters are highly active in the brain, and their roles in some neurodegeneration-related processes were reported120. Riboflavin is an essential B vitamin in the nervous system because of its role in myelin synthesis. The oxidative state impairments related to riboflavin deficiency have been reported. Riboflavin intake was tested as a treatment for PD and other neurodegenerative diseases121. Also, thiamine is another type of vitamin B whose deficiency is related to PD95. Both riboflavin and thiamine metabolisms were identified in the enrichment analysis (Table 4).
PD is typically characterized by tremors and muscle stiffness. The term vascular smooth muscle contraction in the enrichment analysis results (Table 4) is directly related to this symptom. Vascular smooth muscle cell dysfunction was considered as a crucial marker to interpret alterations in vascular function at the onset and progression of neurodegenerative diseases, including Parkinson’s disease122. Also, the relationship between vascular smooth muscle contraction and PD was demonstrated in another enrichment-based analysis123. Pantothenate and CoA biosynthesis is also one of the significantly enriched pathways. Variations in pantothenate kinase (PKAN) causes interruptions in coenzyme A (CoA) biosynthesis and consequent iron accumulation in the brain. This phenomenon is related to many neurodegenerative clinical symptoms, including rigidity, loss of ambulation, and cognitive and visual impairment124.
Alanine, aspartate and glutamate metabolism was found to be enriched in the enrichment analysis results as shown in Table 4. The relation between the changes in glutamate metabolism and neuronal metabolic dysfunctions is known to play an important role in the pathophysiology of PD125. Glutamate is an excitatory transmitter whose excess levels are neurotoxic. Mechanisms for reducing glutamate toxicity, like removing excessive synaptic glutamate by excitatory amino acid transporter (EATT), are known to be affected in PD45. Glutamate aspartate transporter is an important EATT for the synaptic reuptake of glutamate126. Thus, it can be concluded that the enriched alanine, aspartate and glutamate metabolism term is directly related to altered glutamate toxicity and transport mechanism in PD. Epithelial cell signaling in Helicobacter pylori infection was reported as an enriched pathway in Table 4. It was previously reported that bacterial overgrowth and Helicobacter pylori infection in the small intestine affected motor fluctuations by interfering with the absorption of antiparkinsonian drugs127. The enriched term, epithelial cell signaling in Helicobacter pylori infection, is most probably related to this phenomenon. Asthma is another significantly enriched term in the enrichment analysis. It was also previously reported that asthma patients are at higher risk of developing PD in later life128.
Conclusion
The constraint-based modelling approaches are effectively used to investigate disease metabolism by integrating omics data. Here, we compared the prediction capabilities of the original TIMBR, the modified TIMBR and TAMBOOR approaches to identify potential metabolite biomarkers for Parkinson's disease using 13 different transcriptome datasets. A high-confidence metabolite biomarker list for Parkinson's disease was compiled by a literature survey for the predicted metabolite level changes by different algorithms. Different metrics covering the prediction of well-known and high-confidence metabolite level changes and the prediction of metabolites belonging to PD-related metabolic pathways were used to measure biomarker prediction abilities of the three approaches. TAMBOOR reached the highest scores for these metrics.
Metabolites predicted to have altered secretion rates by TAMBOOR and enriched pathways corresponding to these metabolites are mainly related to PD-related metabolisms. Majority of the predicted alterations are consistent with the literature. However, an experimental study that reports secretion rates of all of the metabolites predicted in this study is not available for healthy and disease subjects. Well-designed comprehensive metabolome analyses will be valuable for a complete validation of our results.
The metabolites whose secretion behavior were predicted to be changed by TAMBOOR are candidate biomarkers. Moreover, the metabolites predicted to have increased production capacity can be potential drug targets while the metabolites predicted to have decreased production capacity have the potential to be used as supplementary therapeutics in PD. Those potentials should be validated with experimental studies to identify novel diagnostic biomarkers and treatment strategies for PD.
Data availability
The normalized datasets are available from the corresponding author on reasonable request.
References
Dawson, T. M. & Dawson, V. L. Molecular pathways of neurodegeneration in Parkinson’s disease. Science (80-. ). 302, 819–822 (2003).
Poewe, W. et al. Parkinson disease. Nat. Rev. Dis. Prim. 3, 17013 (2017).
Dorsey, E. R. & Bloem, B. R. The Parkinson pandemic—A call to action. JAMA Neurol. 75, 9–10 (2018).
Mao, Q., Qin, W., Zhang, A. & Ye, N. Recent advances in dopaminergic strategies for the treatment of Parkinson’s disease. Acta Pharmacol. Sin. 41, 471–482 (2020).
Deng, H., Wang, P. & Jankovic, J. The genetics of Parkinson disease. Ageing Res. Rev. 42, 72–85 (2018).
Meiser, J., Weindl, D. & Hiller, K. Complexity of dopamine metabolism. Cell Commun. Signal. 11, 34 (2013).
Zhang, Y., James, M., Middleton, F. A. & Davis, R. L. Transcriptional analysis of multiple brain regions in Parkinson’s disease supports the involvement of specific protein processing, energy metabolism, and signaling pathways, and suggests novel disease mechanisms. Am. J. Med. Genet. Part B Neuropsychiatr. Genet. 137, 5–16 (2005).
Riley, B. E. et al. Systems-based analyses of brain regions functionally impacted in Parkinson’s disease reveals underlying causal mechanisms. PLoS One 9, e102909 (2014).
Dumitriu, A. et al. Gene expression profiles in Parkinson disease prefrontal cortex implicate FOXO1 and genes under its transcriptional regulation. PLoS Genet. 8, e1002794 (2012).
Edgar, R., Domrachev, M. & Lash, A. E. Gene Expression Omnibus: NCBI gene expression and hybridization array data repository. Nucleic Acids Res. 30, 207–210 (2002).
Magare, A. C. & Patel, M. S. Biomarkers identification for Parkinson’s disease using machine learning. in 2021 International Conference on Artificial Intelligence and Machine Vision (AIMV) 1–6 (2021).
Wang, C., Chen, L., Yang, Y., Zhang, M. & Wong, G. Identification of potential blood biomarkers for Parkinson’s disease by gene expression and DNA methylation data integration analysis. Clin. Epigenet. 11, 1–15 (2019).
Havelund, J. F., Heegaard, N. H. H., Færgeman, N. J. K. & Gramsbergen, J. B. Biomarker research in Parkinson’s disease using metabolite profiling. Metabolites 7, 42 (2017).
Shiri Stempler, K. Y. & Ruppin, E. Integrating transcriptomics with metabolic modeling predicts biomarkers and drug targets for Alzheimer’s disease. PLoS One 9 (2014).
Ozerov, I. V. et al. In silico pathway activation network decomposition analysis (iPANDA) as a method for biomarker development. Nat. Commun. 7, 1–11 (2016).
Blais, E. M. et al. Reconciled rat and human metabolic networks for comparative toxicogenomics and biomarker predictions. Nat. Commun. 8, 14250 (2017).
Pannala, V. R. et al. Metabolic network-based predictions of toxicant-induced metabolite changes in the laboratory rat. Sci. Rep. 8, 1–18 (2018).
Pannala, V. R. et al. Genome-scale model-based identification of metabolite indicators for early detection of kidney toxicity. Toxicol. Sci. 173, 293–312 (2020).
Rawls, K. D. et al. Genome-scale characterization of toxicity-induced metabolic alterations in primary hepatocytes. Toxicol. Sci. 172, 279–291 (2019).
Abdik, E. & Çakır, T. Systematic investigation of mouse models of Parkinson’s disease by transcriptome mapping on a brain-specific genome-scale metabolic network. Mol. Omics (2021).
Duarte, N. C. et al. Global reconstruction of the human metabolic network based on genomic and bibliomic data. Proc. Natl. Acad. Sci. 104, 1777–1782 (2007).
Sigurdsson, M. I., Jamshidi, N., Steingrimsson, E., Thiele, I. & Palsson, B. Ø. A detailed genome-wide reconstruction of mouse metabolism based on human Recon 1. BMC Syst. Biol. 4, 140 (2010).
Bordbar, A. & Palsson, B. O. Using the reconstructed genome-scale human metabolic network to study physiology and pathology. J. Intern. Med. 271, 131–141 (2012).
Robinson, J. L. et al. An atlas of human metabolism. Sci. Signal. 13 (2020).
McGarrity, S. et al. Metabolic systems analysis of LPS induced endothelial dysfunction applied to sepsis patient stratification. Sci. Rep. 8, 1–14 (2018).
Steenbergen, R. et al. Establishing normal metabolism and differentiation in hepatocellular carcinoma cells by culturing in adult human serum. Sci. Rep. 8, 1–14 (2018).
Fuhr, L. et al. The circadian clock regulates metabolic phenotype rewiring via HKDC1 and modulates tumor progression and drug response in colorectal cancer. EBioMedicine 33, 105–121 (2018).
Özcan, E. & Çakır, T. Genome-scale brain metabolic networks as scaffolds for the systems biology of neurodegenerative diseases: Mapping metabolic alterations. Syst. Neurosci. 195–217 (2018).
Moolamalla, S. T. R. & Vinod, P. K. Genome-scale metabolic modelling predicts biomarkers and therapeutic targets for neuropsychiatric disorders. Comput. Biol. Med. 125, 103994 (2020).
Varma, V. R. et al. Abnormal brain cholesterol homeostasis in Alzheimer’s disease—A targeted metabolomic and transcriptomic study. npj Aging Mech. Dis. 7, 1–14 (2021).
Çakır, T., Alsan, S., Saybaşιlι, H., Akιn, A. & Ülgen, K. Ö. Reconstruction and flux analysis of coupling between metabolic pathways of astrocytes and neurons: Application to cerebral hypoxia. Theor. Biol. Med. Model. 4, 48 (2007).
Sertbaş, M., Ülgen, K. & Çakır, T. Systematic analysis of transcription-level effects of neurodegenerative diseases on human brain metabolism by a newly reconstructed brain-specific metabolic network. FEBS Open Bio 4, 542–553 (2014).
Lesnick, T. G. et al. A genomic pathway approach to a complex disease: Axon guidance and Parkinson disease. PLoS Genet. 3, e98 (2007).
Moran, L. B. et al. Whole genome expression profiling of the medial and lateral substantia nigra in Parkinson’s disease. Neurogenetics 7, 1–11 (2006).
Zheng, B. et al. PGC-1$α$, a potential therapeutic target for early intervention in Parkinson’s disease. Sci. Transl. Med. 2, 52ra73-52ra73 (2010).
Durrenberger, P. F. et al. Selection of novel reference genes for use in the human central nervous system: A BrainNet Europe Study. Acta Neuropathol. 124, 893–903 (2012).
Xicoy, H., Brouwers, J. F., Wieringa, B. & Martens, G. J. M. Explorative combined lipid and transcriptomic profiling of substantia Nigra and Putamen in Parkinson’s Disease. Cells 9, 1966 (2020).
Simchovitz, A. et al. A lncRNA survey finds increases in neuroprotective LINC-PINT in Parkinson’s disease substantia nigra. Aging Cell 19, e13115 (2020).
Dijkstra, A. A. et al. Evidence for immune response, axonal dysfunction and reduced endocytosis in the substantia nigra in early stage Parkinson’s disease. PLoS One 10, e0128651 (2015).
Odongo, R., Bellur, O., Abdik, E. & Çakır, T. Brain-wide transcriptome-based metabolic alterations in Parkinsons disease: Human inter-region and human-experimental model correlations. Molecular Omics 9, 522–537 (2023).
Aurich, M. K., Fleming, R. M. T. & Thiele, I. MetaboTools: A comprehensive toolbox for analysis of genome-scale metabolic models. Front. Physiol. 7 (2016).
López-Ibáñez, J., Pazos, F. & Chagoyen, M. MBROLE 2.0—Functional enrichment of chemical compounds. Nucleic Acids Res. 44, W201–W204 (2016).
Kanehisa, M. & Goto, S. KEGG: Kyoto encyclopedia of genes and genomes. Nucleic Acids Res. 28, 27–30 (2000).
Henchcliffe, C. et al. Multinuclear magnetic resonance spectroscopy for in vivo assessment of mitochondrial dysfunction in Parkinson’s disease. Ann. N. Y. Acad. Sci. 1147, 206–220 (2008).
Plaitakis, A. & Shashidharan, P. Glutamate transport and metabolism in dopaminergic neurons of substantia nigra: Implications for the pathogenesis of Parkinson’s disease. J. Neurol. 247, II25–II35 (2000).
Chassain, C. et al. Cerebral glutamate metabolism in Parkinson’s disease: An in vivo dynamic 13C NMS study in the rat. Exp. Neurol. 191, 276–284 (2005).
Goldstein, D. S. Biomarkers, mechanisms, and potential prevention of catecholamine neuron loss in Parkinson disease. Adv. Pharmacol. 68, 235–272 (2013).
Mayeux, R., Stern, Y., Sano, M., Williams, J. B. W. & Cote, L. J. The relationship of serotonin to depression in Parkinson’s disease. Mov. Disord. Off. J. Mov. Disord. Soc. 3, 237–244 (1988).
Hestad, K., Alexander, J., Rootwelt, H. & Aaseth, J. O. The role of tryptophan dysmetabolism and quinolinic acid in depressive and neurodegenerative diseases. Biomolecules 12, 998 (2022).
Chang, K.-H. et al. Alternations of metabolic profile and kynurenine metabolism in the plasma of Parkinson’s disease. Mol. Neurobiol. 55, 6319–6328 (2018).
Iwaoka, K. et al. Impaired metabolism of kynurenine and its metabolites in CSF of parkinson’s disease. Neurosci. Lett. 714, 134576 (2020).
Venkatesan, D., Iyer, M., Narayanasamy, A., Siva, K. & Vellingiri, B. Kynurenine pathway in Parkinson’s disease—An update. Eneurologicalsci 21, 100270 (2020).
Sian-Hülsmann, J., Mandel, S., Youdim, M. B. H. & Riederer, P. The relevance of iron in the pathogenesis of Parkinson’s disease. J. Neurochem. 118, 939–957 (2011).
Götz, M. E., Double, K. A. Y., Gerlach, M., Youdim, M. B. H. & RIEDERERE, P. The relevance of iron in the pathogenesis of Parkinson’s disease. Ann. N. Y. Acad. Sci. 1012, 193–208 (2004).
van Rensburg, Z., Abrahams, S., Bardien, S. & Kenyon, C. Toxic feedback loop involving iron, reactive oxygen species, $α$-synuclein and neuromelanin in Parkinson’s disease and intervention with turmeric. Mol. Neurobiol. 58, 5920–5936 (2021).
Collin, F. Chemical basis of reactive oxygen species reactivity and involvement in neurodegenerative diseases. Int. J. Mol. Sci. 20, 2407 (2019).
Famitafreshi, H. & Karimian, M. Prostaglandins as the agents that modulate the course of brain disorders. Degener. Neurol. Neuromuscul. Dis. 10, 1 (2020).
Vila, M. et al. The role of glial cells in Parkinson’s disease. Curr. Opin. Neurol. 14, 483–489 (2001).
Moser, A., Siebecker, F., Vieregge, P., Jaskowski, P. & Kömpf, D. Salsolinol, catecholamine metabolites, and visual hallucinations in L-dopa treated patients with Parkinson’s disease. J. Neural Transm. 103, 421–432 (1996).
Voon, S. M. et al. The mechanism of action of salsolinol in brain: Implications in Parkinson’s disease. CNS Neurol. Disord. Targets Former. Curr. Drug Targets-CNS Neurol. Disord. 19, 725–740 (2020).
Dostert, P., Strolin Benedetti, M. & Dordain, G. Salsolinol and the early detection of Parkinson’s disease. Early Markers Park. Alzheimer’s Dis. 93–97 (1990).
Dostert, P. et al. Dopamine-derived tetrahydroisoquinolines and Parkinson’s disease. Adv. Neurol. 60, 218–223 (1993).
Babu, G. N. et al. Serum metabolomics study in a group of Parkinson’s disease patients from northern India. Clin. Chim. Acta 480, 214–219 (2018).
Meoni, G. et al. Metabolite and lipoprotein profiles reveal sex-related oxidative stress imbalance in de novo drug-naive Parkinson’s disease patients. npj Park. Dis. 8, 14 (2022).
Zhao, H. et al. Potential biomarkers of Parkinson’s disease revealed by plasma metabolic profiling. J. Chromatogr. B 1081, 101–108 (2018).
Brun, L. et al. Clinical and biochemical features of aromatic L-amino acid decarboxylase deficiency. Neurology 75, 64–71 (2010).
Lee, E.-S., Chen, H., Hardman, C., Simm, A. & Charlton, C. Excessive S-adenosyl-L-methionine-dependent methylation increases levels of methanol, formaldehyde and formic acid in rat brain striatal homogenates: Possible role in S-adenosyl-L-methionine-induced Parkinson’s disease-like disorders. Life Sci. 83, 821–827 (2008).
Scholefield, M. et al. Severe and regionally widespread increases in tissue urea in the human brain represent a novel finding of pathogenic potential in Parkinson’s disease dementia. Front. Mol. Neurosci. 14 (2021).
Hatano, T., Saiki, S., Okuzumi, A., Mohney, R. P. & Hattori, N. Identification of novel biomarkers for Parkinson’s disease by metabolomic technologies. J. Neurol. Neurosurg. Psychiatry 87, 295–301 (2016).
Saravi, S. S. S., Saravi, S. S. S., Khoshbin, K. & Dehpour, A. R. Current insights into pathogenesis of Parkinson’s disease: Approach to mevalonate pathway and protective role of statins. Biomed. Pharmacother. 90, 724–730 (2017).
Wood, W. G., Mΰller, W. E. & Eckert, G. P. Statins and neuroprotection: Basic pharmacology needed. Mol. Neurobiol. 50, 214–220 (2014).
Martelanc, M., Žiberna, L., Passamonti, S. & Franko, M. Application of high-performance liquid chromatography combined with ultra-sensitive thermal lens spectrometric detection for simultaneous biliverdin and bilirubin assessment at trace levels in human serum. Talanta 154, 92–98 (2016).
Trupp, M. et al. Metabolite and peptide levels in plasma and CSF differentiating healthy controls from patients with newly diagnosed Parkinson’s disease. J. Parkinsons Dis. 4, 549–560 (2014).
Wuolikainen, A. et al. Multi-platform mass spectrometry analysis of the CSF and plasma metabolomes of rigorously matched amyotrophic lateral sclerosis, Parkinson’s disease and control subjects. Mol. Biosyst. 12, 1287–1298 (2016).
Qiang, J. K. et al. Plasma apolipoprotein A1 as a biomarker for Parkinson disease. Ann. Neurol. 74, 119–127 (2013).
Swanson, C. R. et al. Plasma apolipoprotein A1 associates with age at onset and motor severity in early Parkinson’s disease patients. Mov. Disord. 30, 1648–1656 (2015).
Tong, Q. et al. Reduced plasma serotonin and 5-hydroxyindoleacetic acid levels in Parkinson’s disease are associated with nonmotor symptoms. Parkinsonism Relat. Disord. 21, 882–887 (2015).
Kaiserova, M. et al. Cerebrospinal fluid levels of 5-Hydroxyindoleacetic acid in Parkinson’s disease and atypical Parkinsonian syndromes. Neurodegener. Dis. 21, 30–35 (2021).
Lim, L. et al. Lanosterol induces mitochondrial uncoupling and protects dopaminergic neurons from cell death in a model for Parkinson’s disease. Cell Death Differ. 19, 416–427 (2012).
Hu, L.-D., Wang, J., Chen, X.-J. & Yan, Y.-B. Lanosterol modulates proteostasis via dissolving cytosolic sequestosomes/aggresome-like induced structures. Biochim. Biophys. Acta (BBA)-Mol. Cell Res. 1867, 118617 (2020).
Wilson, J. M. et al. Differential changes in neurochemical markers of striatal dopamine nerve terminals in idiopathic Parkinson’s disease. Neurology 47, 718–726 (1996).
Kish, S. J. et al. Elevated γ-aminobutyric acid level in striatal but not extrastriatal brain regions in Parkinson’s disease: Correlation with striatal dopamine loss. Ann. Neurol. Off. J. Am. Neurol. Assoc. Child Neurol. Soc. 20, 26–31 (1986).
Kori, M., Aydın, B., Unal, S., Arga, K. Y. & Kazan, D. Metabolic biomarkers and neurodegeneration: A pathway enrichment analysis of Alzheimer’s disease, Parkinson’s disease, and amyotrophic lateral sclerosis. Omics. J. Integr. Biol. 20, 645–661 (2016).
Molina, J. A. et al. Decreased cerebrospinal fluid levels of neutral and basic amino acids in patients with Parkinson’s disease. J. Neurol. Sci. 150, 123–127 (1997).
Perry, T. L. & Yong, V. W. Idiopathic Parkinson’s disease, progressive supranuclear palsy and glutathione metabolism in the substantia nigra of patients. Neurosci. Lett. 67, 269–274 (1986).
Jenner, P. Altered mitochondrial function, iron metabolism and glutathione levels in Parkinson’s disease. Acta Neurol. Scand. 87, 6–13 (1993).
Luong, K. V. Q. & Nguyen, L. T. H. Vitamin D and Parkinson’s disease. J. Neurosci. Res. 90, 2227–2236 (2012).
Pignolo, A. et al. Vitamin D and Parkinson’s disease. Nutrients 14, 1220 (2022).
Parsons, R. B., Smith, S. W., Waring, R. H., Williams, A. C. & Ramsden, D. B. High expression of nicotinamide N-methyltransferase in patients with idiopathic Parkinson’s disease. Neurosci. Lett. 342, 13–16 (2003).
Williams, A. C., Cartwright, L. S. & Ramsden, D. B. Parkinson’s disease: The first common neurological disease due to auto-intoxication?. Qjm 98, 215–226 (2005).
Fonteh, A. N., Harrington, R. J., Tsai, A., Liao, P. & Harrington, M. G. Free amino acid and dipeptide changes in the body fluids from Alzheimer’s disease subjects. Amino Acids 32, 213–224 (2007).
Bermúdez, M.-L., Seroogy, K. B. & Genter, M. B. Evaluation of carnosine intervention in the Thy1-aSyn mouse model of Parkinson’s disease. Neuroscience 411, 270–278 (2019).
Adiga, U. et al. Total antioxidant activity in Parkinson’s disease. Biomed. Res. 17, 145–147 (2006).
Wang, L. et al. Impact of serum uric acid, albumin and their interaction on Parkinson’s disease. Neurol. Sci. 38, 331–336 (2017).
Nguyê, L. T. H. et al. Thiamine and Parkinson’s disease. J. Neurol. Sci. 316, 1–8 (2012).
Lương, K. & Nguyễn, L. T. H. The beneficial role of thiamine in Parkinson disease. CNS Neurosci. Ther. 19, 461–468 (2013).
Naoi, M. & Maruyama, W. N-methyl (R) salsolinol, a dopamine neurotoxin, in Parkinson’s disease. Adv. Neurol. 80, 259–264 (1999).
Antkiewicz-Michaluk, L., Krygowska-Wajs, A., Szczudlik, A., Romańska, I. & Vetulani, J. Increase in salsolinol level in the cerebrospinal fluid of parkinsonian patients is related to dementia: Advantage of a new high-performance liquid chromatography methodology. Biol. Psychiatry 42, 514–518 (1997).
Tonin, A. M. et al. Disruption of redox homeostasis in cerebral cortex of developing rats by acylcarnitines accumulating in medium-chain acyl-CoA dehydrogenase deficiency. Int. J. Dev. Neurosci. 30, 383–390 (2012).
Durani, L. W. et al. Age-related changes in the metabolic profiles of rat hippocampus, medial prefrontal cortex and striatum. Biochem. Biophys. Res. Commun. 493, 1356–1363 (2017).
Willacey, C. C. W. Developing metabolomics for a systems biology approach to understand Parkinson’s disease (Leiden University, 2021).
Naseem, S. PAT-PseAAC: Prediction of acetyl threonine in protein sites with statistical moments and chou’s 5 step rule. In 2021 International Conference on Innovative Computing (ICIC) 1–9 (2021).
Holmes, B. B. et al. Heparan sulfate proteoglycans mediate internalization and propagation of specific proteopathic seeds. Proc. Natl. Acad. Sci. 110, E3138–E3147 (2013).
Hoshi, K. et al. A unique glycan-isoform of transferrin in cerebrospinal fluid: A potential diagnostic marker for neurological diseases. Biochim. Biophys. Acta (BBA)-Gen. Subj. 1861, 2473–2478 (2017).
Trushina, E. & Mielke, M. M. Recent advances in the application of metabolomics to Alzheimer’s Disease. Biochim. Biophys. Acta (BBA)-Mol. Basis Dis. 1842, 1232–1239 (2014).
Rahman, S. O., Singh, R. K., Hussain, S., Akhtar, M. & Najmi, A. K. A novel therapeutic potential of cysteinyl leukotrienes and their receptors modulation in the neurological complications associated with Alzheimer’s disease. Eur. J. Pharmacol. 842, 208–220 (2019).
Weng, R. et al. Metabolomics approach reveals integrated metabolic network associated with serotonin deficiency. Sci. Rep. 5, 1–13 (2015).
Bolam, J. P. & Ellender, T. J. Histamine and the striatum. Neuropharmacology 106, 74–84 (2016).
Mao, C. et al. Serum sodium and chloride are inversely associated with dyskinesia in Parkinson’s disease patients. Brain Behav. 7, e00867 (2017).
Müller, T. & Muhlack, S. Cysteinyl-glycine reduction as marker for levodopa-induced oxidative stress in Parkinson’s disease patients. Mov. Disord. 26, 543–546 (2011).
Wishart, D. S. et al. HMDB: The human metabolome database. Nucleic Acids Res. 35, D521–D526 (2007).
Xie, X., Luo, X. & Xie, M. Association between Parkinson’s disease and risk of colorectal cancer. Parkinsonism Relat. Disord. 35, 42–47 (2017).
Fang, H., Du, Y., Pan, S., Zhong, M. & Tang, J. Patients with Parkinson’s disease predict a lower incidence of colorectal cancer. BMC Geriatr. 21, 1–8 (2021).
Menzie, J., Pan, C., Prentice, H. & Wu, J.-Y. Taurine and central nervous system disorders. Amino Acids 46, 31–46 (2014).
Miyake, Y. et al. Dietary fat intake and risk of Parkinson’s disease: A case-control study in Japan. J. Neurol. Sci. 288, 117–122 (2010).
Zhang, X.-B., Cheng, G., Zhang, W.-J., Shen, G.-L. & Yu, R.-Q. A fluorescent chemical sensor for Fe3+ based on blocking of intramolecular proton transfer of a quinazolinone derivative. Talanta 71, 171–177 (2007).
Bharath, S., Hsu, M., Kaur, D., Rajagopalan, S. & Andersen, J. K. Glutathione, iron and Parkinson’s disease. Biochem. Pharmacol. 64, 1037–1048 (2002).
Bosetti, F. Arachidonic acid metabolism in brain physiology and pathology: Lessons from genetically altered mouse models. J. Neurochem. 102, 577–586 (2007).
Sun, A.-G., Lin, A.-Q., Huang, S.-Y., Huo, D. & Cong, C.-H. Identification of potential drugs for Parkinson’s disease based on a sub-pathway method. Int. J. Neurosci. 126, 318–325 (2016).
Katzeff, J. S. & Kim, W. S. ATP-binding cassette transporters and neurodegenerative diseases. Essays Biochem. 65, 1013–1024 (2021).
Plantone, D., Pardini, M. & Rinaldi, G. Riboflavin in neurological diseases: A narrative review. Clin. Drug Investig. 41, 513–527 (2021).
Hayes, G. et al. Vascular smooth muscle cell dysfunction in neurodegeneration. Front. Neurosci. 16 (2022).
Wang, J., Liu, Y. & Chen, T. Identification of key genes and pathways in Parkinson’s disease through integrated analysis. Mol. Med. Rep. 16, 3769–3776 (2017).
Thakur, N. et al. Rational design of novel therapies for pantothenate kinase–associated neurodegeneration. Mov. Disord. 36, 2005–2016 (2021).
Blandini, F., Porter, R. H. P. & Greenamyre, J. T. Glutamate and Parkinson’s disease. Mol. Neurobiol. 12, 73–94 (1996).
Storm-Mathisen, J. & Ottersen, O. P. Glutamate (Elsevier, 2000).
Fasano, A., Visanji, N. P., Liu, L. W. C., Lang, A. E. & Pfeiffer, R. F. Gastrointestinal dysfunction in Parkinson’s disease. Lancet Neurol. 14, 625–639 (2015).
Cheng, C.-M. et al. Risk of developing P arkinson’s disease among patients with asthma: A nationwide longitudinal study. Allergy 70, 1605–1612 (2015).
Acknowledgements
Assoc. Prof. Ali Çakmak (İstanbul Technical University) is gratefully acknowledged for his comments and suggestions on this study.
Author information
Authors and Affiliations
Contributions
T.Ç.: conceived and designed the study, interpreted results, and edited the manuscript, E.A.: designed the study, did all analyses, and wrote the manuscript.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher's note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary Information
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Abdik, E., Çakır, T. Transcriptome-based biomarker prediction for Parkinson’s disease using genome-scale metabolic modeling. Sci Rep 14, 585 (2024). https://doi.org/10.1038/s41598-023-51034-y
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41598-023-51034-y
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.
