Exploring Human Brain Metabolism via Genome-Scale Metabolic Modeling with Highlights on Multiple Sclerosis.

Introduction

Cerebral dysfunctions affect millions
of individuals across all
regions of the world and give rise to a wide range of neurological
diseases due to the structural and functional complexity of the human
brain stemming from the interactive cellular metabolism of its specific
cells including neurons and glial cells (astrocytes, microglia, and
oligodendrocytes). 1 Neurons transmit electrical
and chemical signals to different parts of the body. Astrocytes provide
regulation of the brain’s environment and metabolic support
to neurons. Microglia are resident macrophages in the central nervous
system (CNS). Oligodendrocytes produce a myelin sheath that wraps
adjacent axons of neurons. These brain-specific cells perform altered
adaptive responses to metabolic stresses and bioenergetic challenges
in health, aging, and neurodegeneration. 2 − 4

Advances in isolation
and measurement technologies have provided
a comprehensive understanding of cerebral events in health and disease
at a single-cell resolution. 5 , 6 The transcriptional
differences between brain-specific cells were also elucidated at the
single-cell level. 7 , 8 For example, the Alzheimer’s
disease (AD) risk gene APOE was downregulated in
oligodendrocytes and astrocytes; however, it was upregulated in the
microglia of patients with AD. 8 Moreover,
the transcriptional profiling of human microglia elucidated heterogeneity
between gray matter (GM) and white matter (WM) in multiple sclerosis
(MS), which is the most prevalent nontraumatic debilitating illness
affecting young individuals. 9 , 10 The pathological hallmark
of MS includes inflammatory lesions, demyelination, and damaged oligodendrocytes.
It occurs when the immune system mistakenly attacks the protective
myelin sheath covering nerve fibers, causing inflammation and damage.
This disrupts the communication between the brain and the rest of
the body, leading to various symptoms. The exact cause is unknown,
but genetic predisposition, environmental factors, and infections
may contribute. The treatment strategies such as disease-modifying
therapies, physical therapy, and symptomatic management help MS patients
to slow progression and improve quality of life. 11 In addition to oligodendrocytes, astrocytes are also recognized
as highly active players during lesion formation in MS. 12 Understanding the roles of these various brain
cell types is crucial for developing targeted therapies aimed at modulating
the immune response, protecting myelin, and promoting repair in neurodegenerative
diseases such as MS.

Computational approaches provide critical
insight into the understanding
of prokaryotic and eukaryotic systems. In this regard, genome-scale
metabolic models (GEMs), which cover the metabolic reaction network
of cellular metabolism with related genes and enzyme information,
have become a powerful tool in the studies of systems biology over
the past two decades. In addition to global human GEMs, tissue-specific
GEMs were reconstructed to investigate certain tissue metabolism in
health and disease. 13 − 17 Especially when considering the inherent difficulty of experimental
studies in the human brain, metabolic interactions and neurotransmitter
metabolism in and between astrocytes and neurons were investigated
by brain-specific GEMs at the system level. 18 − 21

In context-specific modeling,
generic human GEMs are adapted to
specific conditions, tissues, or diseases by incorporating experimental
omics data such as transcriptomics, proteomics, or metabolomics. 22 , 23 This method captures unique metabolic activity within a given biological
context. By providing an overview of metabolism at the system level,
context-specific GEMs help uncover the molecular basis of complex
phenotypes and improve personalized medicine strategies. Utilizing
a context-specific GEM can be a valuable approach for studying MS
from a systems biology perspective, providing insights into its complex
pathology and exploring potential avenues for therapeutic intervention.
The metabolic pathways dysregulated in MS, such as changes in energy
metabolism, lipid metabolism, and neurotransmitter synthesis, can
be captured by context-specific GEMs.

Current brain-specific
GEMs include astrocyte and neuron compartments.
However, microglia and oligodendrocytes also play a critical role
in brain physiology and homeostasis in health and disease. 3 , 24 − 29 Here, brain cell-specific GEMs were reconstructed for neurons, astrocytes,
microglia, oligodendrocytes, and oligodendrocyte precursor cells (OPCs)
by integrating single-cell RNA-seq data and global Human1. 7 , 30 The integration of expression data was carried out by a task-driven
integrative network inference for tissues (tINIT) algorithm. 31 Then, transport reactions between these specific
cells were added to represent their crosstalk in the brain. Then,
iHumanBrain2690 was employed to estimate the metabolic alterations
in response to changes in glucose, ketone bodies, and oxygen supplies,
occurring in several neurodegenerative diseases, and to predict reporter
metabolites in MS. The reconstruction of single-cell-level GEMs for
astrocytes, neurons, microglia, and oligodendrocytes and their integration
into a combined brain model iHumanBrain2690 decipher cell-specific
key features of different metabolic states and diseases such as neurodegenerative
and neurodevelopmental disorders as well as intracranial tumors.

Results and Discussion

Reconstructed Cell Type-Specific and Integrated Human Brain
GEMs

For the investigation of the brain-specific cellular
metabolism, five GEMs (iNeuron2527, iAstrocyte2303, iMicroglia1540,
iOligodendrocyte2023, and iOPC1795) were reconstructed for brain cell
types including neuron, astrocyte, microglia, oligodendrocyte, and
OPC ( Supplementary Data 1 ) based on their
single-cell RNA sequencing and tINIT algorithm. Along with 57 essential
metabolic tasks, brain cell type-specific reactions were used as essential
tasks in the reconstruction of GEMs. The highest and least number
of reactions, metabolites, and genes were identified in the iNeuron2527
and iMicroglia1540 metabolic networks ( Figure 1 a), respectively. The distribution of reactions
resulted in lipid metabolism, with the highest number of reactions
in all reconstructed GEMs ( Figure 1 b). 4565 metabolic reactions were found to be in common
in all generated metabolic models ( Figure 1 c). Metabolite distribution demonstrated
that cytosol is the top subcellular compartment with the highest number
of metabolites in all human brain cell-specific GEMs ( Figure 1 d). Metabolic task performances
of all GEMs were evaluated by 257 metabolic tasks listed by Robinson
et al. (2020). 30 iNeuron2527 and iMicroglia1540
successfully performed 218 and 169 metabolic tasks, respectively ( Table S1 ). Subsequent to the generation of single
brain cell-specific GEMs, integration of neurons, astrocytes, microglia,
and oligodendrocyte compartments into whole human brain GEM was performed
by intercellular reactions which represent metabolic trafficking among
them. The final integrated human brain GEM, called as iHumanBrain2690,
includes 27615 metabolic reactions, 20600 metabolites, and 2690 genes
( Figure 2 ).

(a) Comparison
of the number of reactions, metabolites, and genes,
(b) number of reactions in different metabolisms, (c) number of shared
reactions, and (d) distribution of metabolites in the cellular compartments
in reconstructed cell type-specific human brain GEMs.

Metabolic crosstalk in iHumanBrain2690.

Metabolic Activities in Single Brain Cells and Whole Brain iHumanBrain2690
at Resting State

The flux of each metabolic reaction was
minimized and maximized as an objective function for flux variability
analysis (FVA) ( Supplementary Data 1 ).
The flux balance analysis (FBA) was performed for single brain cells
with the maximization of biomass reaction flux as the objective function
to computationally elucidate metabolic activities in a healthy-state
human brain ( Figure 3 ) ( Supplementary Data 1 ). The maximization
of biomass reaction predicted the fluxes of growth varying between
0.0064 μmol/g tissue/min for microglia and 0.0433 μmol/g
tissue/min for OPC ( Table S2 ). Oligodendrocyte
simulation with maximization of both biomass and myelin reaction resulted
in a flux value of 0.0422 μmol/g tissue/min, which is also the
value for the maximization of only its biomass reaction flux with
zero myelin reaction flux. The maximization of only myelin reaction
brought about a flux value of 0.0193 μmol/g tissue/min. By setting
myelin reaction flux between 0 and 0.0193 μmol/g tissue/min
via upper and lower limits, the change of oligodendrocyte biomass
reaction flux with myelin reaction flux was investigated, and increased
myelin production resulted in decreased biomass flux as expected ( Figure S1 ). Two different FBA simulations were
performed with iOligodendrocyte2023 at myelin reaction fluxes of 0
and 0.0100 μmol/g tissue/min. Zero flux indicates inactive myelin
reaction and maximum biomass flux. To keep both myelin and biomass
reactions active in oligodendrocytes, myelin formation flux was set
to 0.0100 μmol/g tissue/min in which biomass reaction flux is
0.0278 μmol/g tissue/min.

Mean subsystem fluxes in reconstructed
brain cell type-specific
GEMs at resting state.

The number of nonzero fluxes (flux value >10 –6 μmol/g tissue/min) changes between 944 (microglia)
and 1664
(neuron) in healthy-state simulation ( Table S2 ). Mean subsystem fluxes were calculated based on the reactions with
nonzero fluxes, and the subsystems with a minimum of 5 nonzero fluxes
in any generated GEM resulted in 56 subsystems ( Figure 3 ) excluding transport and exchange/demand
reactions ( Table S3 ). Despite consisting
of a small number of metabolic reactions, oxidative phosphorylation
has the highest mean fluxes in all brain cell types. Fatty acid oxidation,
which includes the greatest number of total reactions, has more than
50 reactions with nonzero fluxes in neurons, astrocytes, and OPCs.

Glucose is the primary substrate in the energy metabolism of astrocytes,
neurons, microglia, and oligodendrocytes. 28 , 32 , 33 It is taken up from the blood circulation
and metabolized to pyruvate via glycolysis. Pyruvate can either be
converted to lactate in the cytosol by anaerobic respiration or enter
the mitochondria to be further metabolized via the TCA cycle and oxidative
phosphorylation. Long-chain fatty acids are transported into mitochondria
via a carnitine shuttle for β-oxidation and energy production.
Fatty acyl CoA is β-oxidized into acetyl-CoA which is metabolized
through the TCA cycle and oxidative phosphorylation to generate ATP.
Fatty acid and cholesterol biosynthesis start with acetyl-CoA to form
malonyl-CoA and acetoacetyl-CoA, respectively. 34 , 35 In fatty acid metabolism, acetyl-CoA and malonyl-CoA are converted
to palmitic acid. Then, elongation and desaturation reactions form
polysaturated and unsaturated fatty acids. In cholesterol metabolism,
acetoacetyl-CoA and acetyl-CoA are catalyzed to 3-hydroxy-3-methylglutaryl-CoA
(HMG-CoA). It is converted into mevalonate, a key intermediate in
cholesterol biosynthesis.

Subsequent to single-cell predictions,
FBA and FVA were applied
with the maximization of the glutamate/glutamine/GABA cycle between
neuron and astrocyte in the integrated iHumanBrain2690. The exchange
reaction fluxes obtained from single-cell simulations were applied
as constraints. The biomass reaction fluxes and number of reactions
with nonzero flux demonstrated similar results with the single-cell
computations ( Table S4 ). Also, the computation
of intercellular transfer reactions, including glutamate/glutamine/GABA
cycle, glycine, aspartate, alanine, and lactate reactions, resulted
in fluxes greater than 0.01 μmol/g tissue/min ( Figure 4 ). Brain-specific cells perform
unique metabolite and neurotransmitter trafficking for information
processing. Glutamate is the primary neurotransmitter and hub metabolite
linking glucose and amino acid metabolism in brain-specific cells. 36 Glutamate homeostasis and recycling between
neurons and astrocytes via the glutamate–glutamine cycle are
critical for energy metabolism and normal functioning of the brain.
Glutamine is widely distributed in CNS and is a precursor of glutamate,
aspartate, and GABA neurotransmitters. 37 It is produced from glutamate and ammonia in astrocytes and transported
into neurons, where it is converted into glutamate. GABA is also synthesized
in neurons. Both glutamate and GABA are transferred from neurons to
astrocytes. Computational analysis revealed active intercellular reaction
fluxes through the glutamate/glutamine/GABA cycle in iHumanBrain2690
( Figure 4 ). Similarly,
intercellular lactate transports among neurons, astrocytes, and oligodendrocytes
were also computed as active in the iHumanBrain2690 network. Astrocyte-
or oligodendrocyte-derived lactate can be taken up by neurons where
it is used as an energy substrate. 28 Lactate
acts as a metabolic substrate for active neurons and has been shown
to increase synaptic plasticity, decrease excitability in neurons
and neural networks, have neuroprotective effects, and lessen neuroinflammation. 38

Intercellular reaction fluxes in iHumanBrain2690 at the
resting
state.

Metabolic Alterations in Ketogenic Diet at Resting State

Rather than depending on sugar (glucose) derived from carbohydrates,
the keto diet utilizes ketone bodies, which are a fuel source produced
by the liver from stored fat. Acetoacetate and (R)-3-hydroxybutanoate
are two main ketone bodies that can substitute for glucose as alternative
energy sources in the brain metabolism. 39 ATP generation in brain cells depends on glycolysis in the cytosol
and mitochondrial TCA cycle and oxidative phosphorylation. Glucose
is broken down to pyruvate in glycolysis. Then, pyruvate is converted
into acetyl-CoA and enters the TCA cycle. Contrary to glucose, ketone
bodies are directly converted into acetyl-CoA and enter the TCA cycle
without glycolysis. These metabolic processes were investigated in
iHumanBrain2690 by enhancing glucose and ketone body uptake in silico . An increase in the acetoacetate and (R)-3-hydroxybutanoate
uptake fluxes resulted in a greater TCA cycle and lower glycolysis
subsystem fluxes than that of the glucose increase in four single
cells in the integrated brain network ( Figure 5 ). Thus, our reconstructed iHumanBrain2690
metabolic model successfully predicted main energy alterations in
the case of glucose and ketone body supplementation. Furthermore,
the response of iHumanBrain2690 to oxygen reduction pointed out a
decreased trend in the TCA cycle and glyoxylate/dicarboxylate and
oxidative phosphorylation subsystems activity and an increase in glycolysis/gluconeogenesis.
These alterations computationally demonstrate a metabolic shift from
mitochondrial aerobic metabolism to anaerobic glycolysis in the integrated
human brain model due to hypoxia. Oxygen level plays an important
role in cellular metabolism and its reduction diminishes oxidative
phosphorylation and Krebs cycle rates. 40 It is also critical for the generation of nitric oxide and reactive
oxygen species, leading to respiration rate decrease and cellular
damage, respectively. The cells adapt to prolonged hypoxia by activating
hypoxia-inducible factors. 40

Metabolic activity
alterations due to glucose and ketone body supplementations
and oxygen reduction in (a) iHumanBrain2690 subsystems, (b) single
brain cell subsystems, and (c) intercellular reactions. All y -axes demonstrate the sum of metabolic fluxes (μmol/g
tissue/min) in the related subsystems.

In addition to the TCA cycle, acetyl-CoA relates
to critical metabolic
pathways including fatty acid biosynthesis, cholesterol biosynthesis,
and β-oxidation of fatty acids. 41 Metabolic activities in cholesterol metabolism and fatty acid biosynthesis
were increased with the uptake of ketone bodies into the iHumanBrain2690
metabolic network ( Figure 5 ). Owing to the supply of acetyl-CoA derived from acetoacetate
and (R)-3-hydroxybutanoate, the sum of the absolute values of fluxes
in β-oxidation fatty acids subsystems declined in line with
the reduction in the carnitine shuttle subsystem which transports
long-chain fatty acids for β-oxidation. Ketone bodies serve
as significant brain energy sources and are precursors for the production
of amino acids and lipids, particularly cholesterol. 42 Brain ketone body metabolism is regulated by its blood
concentration and blood–brain barrier permeability. High concentrations
are observed during fasting and high-fat diets. Ketone bodies enter
the mitochondrial metabolism and are used in the ATP generation and
oxidative processes. 39 To enhance brain
energy metabolism with ketones, ketogenic therapies have the potential
to treat various neurological diseases, including epilepsy, Alzheimer’s
disease, and Parkinson’s disease. 43 , 44

Neuron and Astrocyte Support to Oligodendrocytes in Demyelination

As our simulations are in agreement with the experimental results
published in the literature, validating the correct reconstruction
of our integrated metabolic brain GEM, iHumanBrain2690, we next applied
this model to myelin degradation to understand the key systemic effects
on brain metabolism. Damage to the protective myelin sheath around
nerve cells is a common feature of demyelinating diseases. To investigate
demyelination metabolism via genome-scale metabolic modeling, the
myelin formation flux in the oligodendrocyte compartment of the iHumanBrain2690
was reduced from 0.010 μmol/g tissue/min to zero at 10% intervals.
The metabolic flux alterations were investigated in four brain cells
using the minimization of metabolic adjustment (MOMA) approach interrelatedly.
Not only oligodendrocyte metabolism but also neuron, astrocyte, and
microglia metabolisms were altered in the iHumanBrain2690 network.

Neurotransmitters play a crucial role in brain homeostasis in health
and disease. The flux alterations of intercellular reactions computationally
demonstrated neurotransmitters and metabolites trafficking between
single brain cells due to demyelination ( Figure 6 ). Glutamate and GABA are primary excitatory
and inhibitory neurotransmitters in CNS, respectively. In demyelination,
glutamate and GABA transfers from neuron to astrocyte were decreased,
and glutamate transport into microglia was increased in the iHumanBrain2690
network ( Figure 6 ).
Experimental studies reported altered levels of these neurotransmitters
and suggested glutamate and GABA metabolisms as therapeutic targets
in the most common demyelinating disease MS. 45 Moreover, in MS, when astrocytes are stimulated by cytokines, their
homeostatic and metabolic functions are reduced, leading to impaired
glutamate uptake. This impairment can result in excitotoxicity and
cause a metabolic disconnect from axons/neurons due to a decreased
release of lactate. 12 Glutamine transport
from astrocytes into neurons was also reduced ( Figure 6 ). Hence, the total flux of glutamate/glutamine/GABA
cycle between astrocyte and neuron was declined due to demyelination.
The flux of neurotransmitter aspartate into astrocytes was decreased
in computational analysis. Dietary supplements of d -aspartate
enhanced synaptic plasticity reserve and decreased fatigue in patients
with progressive MS. 46

Change of metabolic fluxes
(μmol/g tissue/min) among neurons,
astrocytes, microglia, and oligodendrocytes in the iHumanBrain2690
network due to demyelination. All x -axes represent
the myelin reaction flux (μmol/g tissue/min) in oligodendrocytes.

Neurons and astrocytes provide N-acetyl- l -aspartate (NAA)
and lactate to oligodendrocytes ( Figure 6 ). NAA is a notable metabolite that transfers
from neurons to oligodendrocytes. NAA, which is synthesized from acetyl-CoA
and aspartate, is converted to acetate and aspartate. They are utilized
in fatty acid synthesis, protein synthesis, and different metabolic
processes in the cells. NAA production flux in neurons and transport
into oligodendrocytes were elevated in myelin degradation using the
iHumanBrain2690 metabolic network. The conversion of NAA into aspartate
and acetate in oligodendrocytes was also increased, and NAA-derived
aspartate and acetate participated in cellular processes due to demyelination.
Experimental studies demonstrated reductions in NAA levels in GM and
WM due to axonal injury in MS. 47 Lactate
transport among brain cells was also altered due to demyelination
in oligodendrocytes. Lactate transport from astrocyte to oligodendrocyte
was enhanced, and its transport from oligodendrocyte to neuron was
declined. These demonstrated a requirement for lactate in oligodendrocytes
for cellular activities in demyelination. Serum lactate levels in
patients with MS were measured higher than that of healthy controls
and it was suggested as a potential biomarker in MS. 48

Metabolic activities in demyelination were computationally
investigated
by decreasing myelin formation flux in oligodendrocytes, and the sum
of metabolic fluxes resulted in various subsystem responses in different
brain cell types in iHumanBrain2690 ( Figure 7 ). The total metabolic fluxes of glycolysis
and oxidative phosphorylation subsystems were enhanced in all brain
cell types. However, in contrast to responses of astrocytes, microglia,
and oligodendrocytes, TCA cycle metabolic activity was only decreased
in the neuron compartment of iHumanBrain2690. The brain of MS patients
demonstrates impaired glucose metabolism. 49 An experimental study on MS animal model astrocytes revealed a significant
increase in glycolytic and TCA cycle genes. 50 Also, a change in metabolism toward oxidative phosphorylation was
suggested and the contribution of MS astrocytes to brain inflammation
was hypothesized due to TCA cycle upregulation. 50 In the study of in vitro oligodendrogliopathy model using
human CNS-derived oligodendrocytes, the majority of ATP was generated
in aerobic glycolysis that helps protein and lipid synthesis needed
to form myelin. 51 Reduced glycolytic ATP
production would worsen myelin withdrawal and promote cell survival. 51

Change of metabolic activities in brain cell types in
iHumanBrain2690
due to demyelination. All y -axes demonstrate the
sum of metabolic fluxes (μmol/g tissue/min) in the related subsystems.

Reporter Metabolite Analysis: MS Study

Oligodendrocytes
are the primary targets in MS, where the immune system attacks and
damages the myelin they produce, leading to demyelination and disruption
of nerve signal transmission. Astrocytes contribute to the formation
of glial scars in areas of demyelination, which can inhibit repair
processes and axonal regeneration. Microglia become activated in MS,
contributing to inflammation and further damage to myelin and neurons
through the release of proinflammatory cytokines and reactive oxygen
species. Neurons can suffer secondary damage due to demyelination
and inflammation, leading to axonal loss and neurodegeneration, which
correlate with the progression of disability in MS. Thus, comprehending
the functions and alterations of distinct brain cell types (oligodendrocytes,
astrocytes, microglia, and neurons) is imperative in the advancement
of tailored treatments intended to regulate the immune system, protect
myelin, and stimulate repair in diseases such as MS.

In the
present study, taking MS as a neurodegenerative disorder example,
reporter metabolite analysis (RMA) was performed to understand metabolic
reprogramming/regulation in response to disease-related changes in
the brain. RMA identifies the key metabolites that are most significantly
affected by transcriptional changes in their associated gene expressions
in response to different perturbations. 52 Our analysis revealed hot spots (important reporter metabolites)
in microglia and oligodendrocytes ( Supplementary Data 2 ). Regardless of their subcellular localization, 219
and 257 metabolites (p-value <0.05) were identified as reporter
metabolites in GM and WM microglia, respectively. L-carnitine was
found to be the top reporter metabolite in GM and arachidonate in
the WM region. L-carnitine was only predicted in GM; however, arachidonate
was predicted in both regions. Similar to L-carnitine, leukotriene
E4 and N-acetyl-LTE4 were only identified as reporter metabolites
in GM microglia. GM and WM samples have 82 common reporter metabolites
and a large number of them including arachidonate, adrenic, behenic,
decanoic, linoleic, palmitic, oleic, hexanoic, octanoic, and valeric
acids were involved in fatty acid-related metabolites in both GM and
WM microglia.

Over-representation analysis (ORA) of reporter
metabolites demonstrated
that biosynthesis of unsaturated fatty acids is the most significantly
over-represented one in GM microglia ( Figure 8 a) and metabolism of xenobiotics by cytochrome
P450 in WM microglia ( Figure 8 b) in MS patients. Butanoate metabolism, nicotinate and nicotinamide
metabolism, fatty acid degradation, ascorbate and aldarate metabolism,
glyoxylate and dicarboxylate metabolism and fatty acid biosynthesis
in GM microglia and biosynthesis of unsaturated fatty acids, pantothenate
and CoA biosynthesis, sphingolipid metabolism, and linoleic acid metabolism
in WM microglia were also significantly over-represented ( p -value <0.025) in MS patients. ORA of common metabolites
in gray and white matter MS pointed out significant enrichment ( p -value <0.025) in the biosynthesis of unsaturated fatty
acids, fatty acid biosynthesis, α-linolenic acid metabolism,
butanoate metabolism, and fatty acid degradation ( Figure S2 ).

Over-representation and pathway analysis of reporter metabolites
in (a) GM and (b) WM microglia and (c) oligodendrocytes due to MS.

In CNS, triggering receptor expressed on myeloid
cells 2 (TREM2)
regulating the cellular activity in microglia in the degradation of
lipid-based components is highly expressed in microglia and is strongly
found in demyelinating plaques in MS patients. 53 A variety of lipid- and lipid-interacting ligands, such
as sphingomyelin, oxidized phospholipids, phosphatidylcholine, and
APOE, activate TREM2. The formation of sphingosine 1-phosphate (S1P),
a ligand for the S1P receptor family produced by microglia, is stimulated
by inflammatory lesions. The elimination of highly inflammatory and
toxic small molecules that build up at demyelinating sites is made
possible by receptor-mediated endocytosis. Meanwhile, microglial phagocytosis
clears large molecular debris such as myelin and dead cells produced
during lesion growth, which may otherwise hinder repair. 53

Primary progressive multiple sclerosis
(PPMS), a type of MS, leads
to a continuous decline in brain and nerve function without significant
periods of remission or relapse. RMA resulted in 122 reporter metabolites
( p -value < 0.05) in oligodendrocytes ( Supplementary Data 2 ). Carbamoyl-phosphate, acetate,
and ornithine were computed as top reporter metabolites ( p -value < 0.025). ORA of reporter metabolites demonstrated significant
enrichment ( p -value < 0.025) in tyrosine metabolism,
ascorbate and aldarate metabolism, arginine and proline metabolism,
glycolysis/gluconeogenesis, riboflavin metabolism, arginine biosynthesis,
and histidine metabolism in PPMS oligodendrocyte ( Figure 8 c). Oligodendrocyte results
have 26 and 11 common reporter metabolites with GM and WM microglia,
respectively ( Figure S2 ). Nine of them,
including indoleacetate, d -glucarate, indole-3-acetaldehyde,
3-amino-propanal, and acetyl-CoA, are shared reporter metabolites
in both oligodendrocyte and microglia MS.

Arachidonate was identified
as a reporter metabolite in our study.
The arachidonic acid is metabolized to protanoids, prostacyclin, and
thromboxane via cyclooxygenases (COXs). Lipoxygenases (LOXs) convert
arachidonic acids into leukotrienes and hydroxyeicosatetraenoic acids
(HETEs). COX-2 and 5-LOX inhibitors provide a potential strategy in
MS treatment by targeting the arachidonic acid pathway. 54 Several short-, medium-, and long-chain fatty
acids were found as reporter metabolites in MS. Fatty acids are in
relation to critical metabolic processes such as immune cell polarization,
differentiation, and function, blood–brain integrity, inflammation,
degeneration, and remyelination. 55 Supplementation
with different fatty acids demonstrated alleviation of symptoms in
MS animal models. Another reporter metabolite, identified in the present
study, is L-carnitine, which is necessary for the transport of long-chain
fatty acids into mitochondria. In a pilot study, serum carnitine level
was detected significantly lower in MS patients with fatigue compared
with patients without fatigue, and it was suggested that higher dietary
carnitine intake may provide benefits for MS-related fatigue. 56 In the cuprizone-induced demyelinating rat model
study, L-carnitine treatment demonstrated the neuroprotective effect
and significantly improved abnormalities in rat sciatic nerves. 57

Conclusions

The computational analysis gives clues
about the metabolic behaviors
in the human brain, which is one of the most complex organs in the
body because of the interactive cellular metabolism of its specific
cells including neurons and glial cells. Here, the reconstructed brain
cell-specific (iNeuron2527, iAstrocyte2303, iMicroglia1540, iOligodendrocyte2023,
and iOPC1795) and integrated brain (iHumanBrain2690) GEMs allowed
us to investigate human brain metabolism at the system level. The
metabolic fluxes in many subsystems of the core carbon, amino acid,
and fatty acid metabolisms were predicted by FBA at the resting state.
The changes in glucose, ketone bodies, and oxygen were observed to
result in notable metabolic alterations in glycolysis/gluconeogenesis,
the TCA cycle, and glyoxylate/dicarboxylate and oxidative phosphorylation
subsystems and intercellular reactions.

Neurological diseases
are considered critical perturbations leading
to dysfunction of brain metabolism and cerebral activity. One of them
is MS, a chronic disease of CNS, characterized by inflammation, demyelination,
and axonal damage. Inflammatory microglia induce astrocytes to become
highly reactive and neurotoxic. These reactive astrocytes can harm
CNS by producing reactive oxygen species and saturated lipids and
may disrupt metabolic regulation and glutamatergic signaling at synapses,
resulting in excitotoxicity. Thus, the multifaceted nature of these
neurological diseases requires the reconstruction of several single
brain cell-specific GEMs, which further provide opportunities for
researchers to reconstruct a two- or three-compartment integrated
human brain network with the specific brain cells of interest to investigate
metabolic interactions in and between them. Our simulations with the
integrated metabolic brain GEM, iHumanBrain2690, are in agreement
with the experimental results published in the literature, validating
its correct reconstruction and allowing its use in the interpretation
of key cell-specific characteristics of various metabolic states and
diseases, such as intracranial tumors and neurodegenerative and neurodevelopmental
disorders.

Our brain networks can also be combined with pathogen
metabolic
networks to decipher pathogen–host interactions in the human
brain computationally. Consequently, single and integrated human brain
GEMs can help in the elucidation of metabolic activities in health
and neurological diseases for the development of novel therapeutic
strategies. Experimental and clinical studies are needed for computational
approaches, and collaborative studies may help expedite the development
of effective treatment strategies.

Materials and Methods

Reconstruction of Brain Cell-Specific GEMs

GSE67835
which investigates transcriptional differences in human brain cells
including neurons, astrocytes, microglia, oligodendrocytes, and oligodendrocyte
precursor cells (OPCs) was obtained from the Gene Expression Omnibus
database. 7 , 58 GSE67835 covers single-cell RNA-seq of 466
cell samples from healthy brains. The data set is given with raw read
values for each sample. Raw read values were normalized to transcript
per million reads (TPM) using transcript lengths in the human genome
from Lopes et al. where the gene lengths were calculated from the
BioMart Web site based on the human reference genome (GRCh38.p12). 59 , 60 The normalization is performed by taking both gene length and sequencing
depth into consideration in the TPM calculation. 61

Human1 metabolic network was used as a generic human
GEM. 30 The task-driven integrative network
inference for tissues (tINIT) algorithm was selected to integrate
Human1 with gene expression data to generate brain cell-specific GEMs. 31 Mean values of TPM were calculated, and a cutoff
value of 1 was used for the tINIT algorithm ( Figure S3 ). The major advantage of the tINIT algorithm is that all
predefined metabolic tasks are satisfied by the automatically generated
GEMs. Robinson et al. (2020) defined 57 essential metabolic tasks
that are necessary for human cell viability. 30 In addition to predefined essential metabolic tasks, cell-specific
metabolic tasks obtained from the literature for astrocytes, neurons,
microglia, and oligodendrocytes were included in the GEM extraction
via the tINIT algorithm.

The previous human brain models 19 , 62 are valuable
sources of metabolic tasks for astrocytes and neurons since they were
manually reconstructed to investigate metabolic interactions in and
between astrocytes and neurons using literature support. The production
of glutamate from glutamine and the conversion of glutamate into 4-aminobutyrate
(GABA) were added as metabolic tasks in neurons. The production of
glutamine from glutamate was added to the astrocyte, and the conversion
of GABA into glutamate was defined for both neuron and astrocyte compartments.
The reaction of serine into glycine and its reverse reaction were
added to neurons and astrocytes, respectively. The conversion of branched-chain
amino acids (leucine, isoleucine, and valine) was defined in both
neuron and astrocyte lists to form α-ketoisocaproate, α-keto-β-methylvalerate,
and α-keto-isovalerate, respectively. Dopamine production from
levodopa was defined in neurons, and its conversion into norepinephrine
was included in both astrocyte and neuron metabolic tasks. Adrenaline
and acetylcholine productions were added to the neuron list.

The conversion of glutamine into glutamate was included in microglia
GEM reconstruction. 25 N-acetyl- l -aspartate (NAA) synthesis was defined in neurons and its degradation
was added to the oligodendrocyte metabolic list. 63 The main idea behind defining additional brain cell-specific
metabolic tasks is to include related reactions and their metabolites
(neurotransmitters) in the integration and crosstalk of brain-specific
cells. Different from other tasks, myelin sheath formation reaction,
considered as essential metabolic task in oligodendrocytes, does not
exist in Human1, which is used as a generic human model with the tINIT
algorithm. Therefore, the myelin formation reaction was defined by
its major components including cholesterol, galactosylceramide, phosphatidylcholine,
phosphatidylethanolamine, and sphingomyelin. 64 Along with its exchange and transport reactions, it was added to
the Human1 model before the oligodendrocyte-specific GEM reconstruction.
Additional metabolic tasks were not defined for OPCs and its GEM was
generated with 57 essential metabolic tasks.

Integration of Neuron, Astrocyte, Microglia, and Oligodendrocyte
GEMs into iHumanBrain2690

To understand the metabolic interaction
in the human brain among neurons, astrocytes, microglia, and oligodendrocytes,
four brain cell type-specific GEMs (iNeuron2527, iAstrocyte2303, iMicroglia1540,
and iOligodendrocyte2023) were integrated with the intercellular exchange
reactions. To label reactions, subsystems, subcellular compartments,
and metabolites, “_ N ,″ “_A,”
“_M,” and “_O” suffixes were used for
neuron, astrocyte, microglia, and oligodendrocyte compartments, respectively.
Subsequent to the conversion of RAVEN-compatible GEMs into COBRA-compatible
structures via “ravenCobraWrapper” function, the integration
was performed by the “mergeTwoModels” function in the
COBRA Toolbox. 65 , 66 First, neuron and astrocyte models
were combined, and then, microglia and oligodendrocyte models were
sequentially added to this two-compartment model to generate a four-compartment
brain model. The reactions representing intercellular neurotransmitter
exchange and metabolic crosstalk ( Table S5 ) among these brain cells were added based on available human brain
GEMs and literature support using “addReaction” function.
In addition to intercellular reaction available in iMS570, 19 glutamine transfer from astrocyte to microglia
and glutamate transfer from neuron to microglia were defined. 67 Intercellular lactate transport among these
four brain cells and NAA transfer from neuron to oligodendrocyte were
added in accordance with the literature. 25 , 68

Flux Balance Analysis (FBA), Flux Variability Analysis (FVA),
and Minimization of Metabolic Adjustment (MOMA)

FBA and FVA
were used to predict metabolic fluxes in single-cell (iNeuron2527,
iAstrocyte2303, iMicroglia1540, iOligodendrocyte2023, and iOPC1795)
and integrated (iHumanBrain2690) metabolic networks in healthy-state
human brain using IBM Cplex Optimizer in MATLAB. 69 , 70 The constraints ( Table S6 ) were obtained
from previous human brain metabolic networks for neurons and astrocytes
and implemented by setting upper and lower flux limits. 19 , 62 Pentose phosphate pathway fluxes were set to 5% and 6% of the glucose
uptake flux for neurons and astrocytes, respectively. Microglia and
oligodendrocytes, similar to astrocytes, are also classified as glial
cells. Therefore, the upper bound flux constraints for microglia and
oligodendrocytes were used the same as astrocyte constraints and their
lower flux bounds were set to zero ( Table S6 ). To avoid free metabolite uptake and infeasible solution, the lower
bound of the rest of the exchange reactions was set to −0.01
μmol/g tissue/min and the upper bounds were set to 1000 μmol/g
tissue/min not to limit metabolite release out of the cell. Maximization
of biomass reaction fluxes was used as the first objective function
in all single brain cell simulations in FBA. Owing to the possibility
of multiple optima, minimization of the squared sum of intracellular
fluxes was carried out as the second optimization by fixing the maximum
value of biomass reaction flux obtained from the first optimization.
Mean subsystem fluxes, number of reactions carrying nonzero fluxes
(flux value >10 –6 μmol/g tissue/min), and
total number of reactions in the related subsystem were visualized
using the SRplot online tool. 71 FVA was
performed by minimizing and maximizing the flux through each reaction
to determine the range of possible reaction fluxes. 70

In the iHumanBrain2690 metabolic network, maximization
of the glutamate/glutamine/GABA cycle between neuron and astrocyte
was used as the objective function in FBA. 19 , 62 The objective value was added to the stoichiometric matrix, and
the second quadratic optimization was carried out to obtain objective
values with minimum fluxes. Intercellular GABA flux between the neuron
and astrocyte was set to 25% of glutamate–glutamine cycle flux.
The exchange reaction fluxes estimated from single-cell simulations
were used as constraints in the flux analysis of an integrated human
brain model. Oligodendrocytes with active myelin formation (flux value
of 0.01 μmol/g tissue/min) were used in the integrated brain
model analysis. To simulate metabolic changes due to demyelination,
myelin reaction flux was reduced from 0.010 μmol/g tissue/min
to zero, and the MOMA approach was applied by minimizing the Euclidean
distance between the flux distributions. 72 For the in silico analysis of metabolic alterations
in the ketogenic diet via iHumanBrain2690, acetoacetate and (R)-3-hydroxybutanoate
uptake fluxes of four brain cells were increased to 0.050 μmol/g
tissue/min, and glucose was increased to 0.210 μmol/g tissue/min.
For cerebral hypoxia, oxygen uptake fluxes were decreased by 0.100
μmol/g tissue/min. Subsystem activities in the four single brain
cells in iHumanBrain2690 were calculated as the sum of the absolute
value of metabolic fluxes in the corresponding subsystem. Glucose
and ketone body supplemented and oxygen reduction conditions were
compared with the resting state.

Reporter Metabolite Analysis (RMA)

Two different gene
expression data sets ( GSE111972 and GSE211739 ) were obtained from
the Gene Expression Omnibus (GEO) database to elucidate reporter metabolites
in microglia and oligodendrocytes due to MS. 52 , 58 GSE111972 covers the transcriptional profile of human microglia
in gray matter (GM) and white matter (WM) regions in MS. 9 GSE211739 includes transcriptional changes of
mature oligodendrocytes which were differentiated from human-induced
pluripotent stem cell lines derived from primary progressive MS and
healthy donors. 73 The MS microglia and
oligodendrocyte samples were statistically compared with the corresponding
control samples. The p-values were computed for each gene. The results
were used with iMicroglia1540 and iOligodendrocyte2023 GEMs for the
identification of reporter metabolites in microglia and oligodendrocytes
due to MS. The computation was performed by the “reporterMetabolites”
function in the RAVEN toolbox. 65 The reporter
metabolites ( p -value < 0.05) were used in metabolite
set enrichment and pathway topology analyses based on KEGG human metabolic
pathways via MetaboAnalyst 6.0, which is one of the most commonly
used tools for metabolomics data analysis. 74 , 75 Hypergeometric test was applied in the enrichment analysis and relative-betweenness
centrality in pathway topology analysis.