PreprintPDF Available

Abstract and Figures

INTRODUCTION Alzheimer’s disease (AD) is accompanied by metabolic alterations both in the periphery and the central nervous system. However, so far, a global view of AD-associated metabolic changes in brain has been missing. METHODS We metabolically profiled 500 samples from the dorsolateral prefrontal cortex. Metabolite levels were correlated with eight clinical parameters, covering both late-life cognitive performance and AD neuropathology measures. RESULTS We observed widespread metabolic dysregulation associated with AD, spanning 298 metabolites from various AD-relevant pathways. These included alterations to bioenergetics, cholesterol metabolism, neuroinflammation and metabolic consequences of neurotransmitter ratio imbalances. Our findings further suggest impaired osmoregulation as a potential pathomechanism in AD. Finally, inspecting the interplay of proteinopathies provided evidence that metabolic associations were largely driven by tau pathology rather than β-amyloid pathology. DISCUSSION This work provides a comprehensive reference map of metabolic brain changes in AD which lays the foundation for future mechanistic follow-up studies.
Content may be subject to copyright.
1
The landscape of metabolic brain alterations in
Alzheimer's disease
Richa Batra1*, Matthias Arnold2,3*, Maria A. Wörheide3, Mariet Allen4, Xue Wang5, Colette Blach2,
Allan I. Levey6, Nicholas T. Seyfried7, Nilüfer Ertekin-Taner4,8, David A. Bennett9,
Gabi Kastenmüller3, Rima F. Kaddurah-Daouk10#, Jan Krumsiek1#
for the Alzheimer’s Disease Metabolomics Consortium (ADMC)11
1Department of Physiology and Biophysics, Institute for Computational Biomedicine, Englander Institute for Precision
Medicine, Weill Cornell Medicine, New York, NY 10021, USA
2Department of Psychiatry and Behavioral Sciences, Duke University, Durham, NC, USA
3Institute of Computational Biology, Helmholtz Zentrum MünchenGerman Research Center for Environmental Health,
85764 Neuherberg, Germany
4Department of Neuroscience, Mayo Clinic Florida, Jacksonville, FL, USA.
5Department of Health Sciences Research, Mayo Clinic Florida, Jacksonville, FL, USA.
6Goizueta Alzheimer’s Disease Research Center, Emory University, Atlanta, GA, USA.
7Department of Biochemistry, Emory School of Medicine.
8Department of Neurology, Mayo Clinic Florida, Jacksonville, FL, USA.
9Rush Alzheimer’s Disease Center, Rush University Medical Center, Chicago, IL, USA
10Department of Psychiatry and Behavioral Sciences, Duke Institute for Brain Sciences and Department of Medicine,
Duke University, Durham, NC, 27708, USA.
11The full list of contributing scientists is available at https://sites.duke.edu/adnimetab/team/.
*These authors contributed equally to this work
#Corresponding authors
Contact: Jan Krumsiek (jak2043@med.cornell.edu) and Rima Kaddurah-Daouk (rima.kaddurahdaouk@duke.edu)
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
2
Abstract
We present a comprehensive reference map of metabolic brain changes in Alzheimer’s disease
(AD). In a multi-center study within the Accelerating Medicines Partnership in AD, we
metabolically profiled 500 samples from the dorsolateral prefrontal cortex (DLPFC) and 83
samples from the temporal cortex (TCX). In the DLPFC, 298 metabolites were correlated with AD-
related traits, including late-life cognitive performance and neuropathological b-amyloid and tau
tangle burden. Out of these 298 metabolites, 35 replicated in TCX and a previous study. A
conditional analysis suggests that metabolic associations with tangle burden were largely
independent of b-amyloid load in the brain. Our results provide evidence of brain alterations in
bioenergetic pathways, cholesterol metabolism, neuroinflammation, osmoregulation, and other
pathways. In a detailed investigation of the glutamate/GABA neurotransmitter pathway, we
demonstrate how integration of complementary omics data can provide a comprehensive view of
dysregulated biochemical processes. All associations are available as an interactive network at
https://omicscience.org/apps/brainmwas/.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
3
1 Introduction
Alzheimer’s disease (AD) is the most common cause of dementia, with prevalence rates expected
to increase markedly over the next decades1. It is a neurodegenerative disorder defined by the
deposition of b-amyloid and accumulation of neurofibrillary tangles of phosphorylated tau protein
in the brain2. These proteinopathies are further accompanied by other pathogenic processes
including neuroinflammation, oxidative stress, innate immune response, and neurotransmission1.
In addition, a large body of evidence implicates metabolic pathways both in the periphery and in
the central nervous system in AD3–8. Moreover, metabolic enzymes and transporters are among
the most commonly targeted proteins in pharmaceutical interventions across all diseases9,10,
emphasizing the translational potential of systematically identifying metabolic alterations.
However, until now a comprehensive reference map of metabolic brain changes related to AD,
AD-associated neuropathological manifestation, and cognitive decline has been missing.
Here we present a large, multi-center study from the Accelerating Medicine Partnership in AD
(AMP-AD) consortium, analyzing a total of 583 post-mortem brain tissue samples using broad,
non-targeted metabolomics measurements. This dataset represents, to the best of our
knowledge, the largest metabolomics study of aging brain tissue to date. In the first part of our
study, we analyzed 667 metabolites in 500 brain tissue samples from the dorsolateral prefrontal
cortex (DLPFC). This resulted in various metabolic associations with AD-related traits, including
b-amyloid and tau tangles neuropathological burden, as well as late-life cognitive performance.
We provide preliminary evidence that tau-related pathology is the main driver of metabolic
alterations, while b-amyloid-related alterations are secondary effects. We confirmed a subset of
associations in an independent set of 83 temporal cortex (TCX) brain tissue samples, profiled with
the same metabolomics platform. In addition, part of our findings overlapped with previously
reported AD-associated metabolic brain changes5. Moreover, we exemplify how the integration of
metabolomics with complementary omics data can enable a comprehensive view of biochemical
changes in AD. To this end, we investigated molecular changes downstream of the
glutamate/GABA neurotransmitters by integrating proteomics measurements from 262 matching
DLPFC samples.
Our study provides strong evidence for the metabolic alteration of bioenergetic pathways,
cholesterol metabolism, neuroinflammation, osmoregulation, neurotransmission, and other
pathways in AD pathogenesis. Further, to maximize utilization of our study by the scientific
community, we have made both our data and our findings available through the AD Knowledge
Portal and an interactive web resource at https://omicscience.org/apps/brainmwas/.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
4
2 Results
2.1 Cohort description and characteristics of brain metabolomics
data
We analyzed brain samples of 500 participants from the Religious Order Study and the Rush
Memory and Aging Project (ROS/MAP) cohorts11,12, including 352 females and 148 males, with a
mean age at death of 91 (Table 1). Following enrollment in the study, participants were evaluated
for physiological and cognitive function once per year (Figure 1). Neuropathology was assessed
after autopsy. Out of the 500 participants, 220 were diagnosed with AD (with or without a
secondary cause of dementia) at the time of death, 119 had mild cognitive impairment, 153 were
without cognitive impairment, and 8 had other forms of dementia. Samples from the dorsolateral
prefrontal cortex (DLPFC) brain region were used for untargeted metabolic profiling.
Metabolomics measurements were analyzed in relation to eight AD-related traits covering late life
cognitive assessments and postmortem pathology: Clinical diagnosis at the time of death, level
of cognition proximate to death, cognitive decline during lifetime, b-amyloid load, tau tangle load,
global burden of AD pathology (global NP), NIA-Reagan score, neuropathological diagnosis
inferred based on combination of Braak stage and CERAD score (NP diagnosis, see methods for
diagnostic criteria). A detailed description of these AD-related traits is provided in Supplementary
Table 1.
The metabolomics platform identified 667 metabolites from various chemical classes (super-
pathways) in the brain samples, including lipids (42.7%), amino acids (22.6%), nucleotides
(6.7%), carbohydrates (6.3%), cofactor and vitamins (4.3%), xenobiotics (3.7%), peptides (2.1%),
and energy-related metabolites (1.5%) (Figure 2a, Supplementary Table 2). Previous blood-
based metabolomics studies reported strong influences of medications and supplements (such
as vitamins) on metabolic profiles4. To investigate such effects in brain tissue-based metabolic
profiles, we examined influences of 103 grouped medication classes and supplements on
metabolic abundances. 552 out of 667 (82.75%) of the metabolites correlated with one or more
medications or supplements taken during lifetime. The group of medications to treat benign
prostatic hypertrophy associated with the highest number of metabolites (81 metabolites),
followed by diuretics (55 metabolites) and multivitamins (52 metabolites). A comprehensive list of
medication classes and their effect on the metabolome is provided in Supplementary Table 3.
Given these strong associations with the metabolome, medication effects excluding AD and
neurologic drugs were regressed out from the metabolic profiles for all following analyses.
Moreover, since the postmortem interval (PMI) before sample collection at autopsy may also
impact analyte levels, we investigated its effects and found that 307 metabolites associated with
PMI (Supplementary Table 4). PMI was therefore included as a covariate in all following
analyses.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
5
To obtain a baseline understanding of metabolism in brain, we used cognitively normal samples
to compute associations between measured metabolites and demographic parameters
independent of AD. This included metabolic associations with age, BMI, sex, and years of
education (as a proxy for socioeconomic status). Two metabolites, 1-methyl-5-imidazoleacetate
and N6-carboxymethyllysine, were significantly associated with age at 5% false discovery rate
(FDR). Surprisingly, there were no significant associations with sex, BMI, or education, which is
in stark contrast to findings in blood1315. Details of this baseline analysis can be found in
Supplementary Table 5.
Total samples
N = 500
Sex
Female
352 (70.4%)
Male
148 (29.6%)
Postmortem interval (hours)
6.6 (5.2, 8.7)
BMI
25.2 (22.5, 28.2)
Years of education
16.00 (13.00, 18.00)
Age at death
91 (87, 95)
APOE4 alleles
0
374 (75%)
1
121 (24%)
2
5 (1%)
Clinical diagnosis at death
AD
220 (44%)
Mild cognitive impairment (MCI)
119 (23.8%)
No cognitive impairment (NCI)
153 (30.6%)
Other
8 (1.6%)
Format: N (%) or median (IQR)
Table 1. ROS/MAP cohort overview. Postmortem interval refers to the time between death and sample preservation.
BMI = body mass index. IQR = interquartile range, i.e., middle 50% of the data.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
6
Figure 1: Study overview. 500 ROS/MAP participants were included in this analysis. For each participant, data was
available on cognitive assessments during lifetime, postmortem AD brain pathology, and brain metabolic profiles from
the dorsolateral prefrontal cortex (DLPFC) region. Metabolic profiles were investigated for associations with AD-related
traits and a metabolite interaction network was inferred using a Gaussian graphical model (GGM). Associations were
tested for replication in 83 temporal cortex samples from the Mayo Clinic brain bank cohort and compared to a
previously published brain-based study. Finally, various pathways previously implicated in AD were metabolically
characterized, and a detailed metabolomic/proteomic characterization of the glutamate/GABA neurotransmitter
pathway was generated.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
7
2.2 AD is associated with widespread metabolic alterations in brain
To assess AD-related metabolic changes, we computed statistical associations between
metabolic profiles and the 8 AD-related traits. All statistical models accounted for AD-related
confounders (age, sex, years of education, BMI, and copies of APOE4) as well as postmortem
interval. A total of 298 out of 667 metabolites (44.7%) were significantly associated with one or
more AD traits at 5% false discovery rate (FDR). 80 out of the 298 metabolites showed unique
associations with just one of the traits. A total of 218 metabolites associated with more than one
trait, which is likely due to high correlations across traits (Figure 2b, Supplementary Figure 1).
The majority of the 298 metabolites was associated with one of three AD traits: Cognitive decline
(n = 201), tau tangles (n = 188), and global burden of pathology (n = 183) (Figure 2c).
Interestingly, only 34 metabolites associated with b-amyloid, which was the lowest number of
associations among the eight AD traits. Furthermore, we observed that 159 out of the 298
metabolites (53.4%) were associated with both premortem parameters and postmortem
pathological assessments. All statistical results are provided in Supplementary Table 6. Sex-
based stratified analysis revealed that 29 of the 298 metabolites (10%) showed associations with
at least one trait that were significantly modulated by sex (Supplementary Table 7), and APOE4-
stratified analysis showed that associations of 77 metabolites (26%) were influenced by APOE4
status (Supplementary Table 8).
To illustrate the strength of the observed associations, we provide two of the most significant
associations in the dataset as examples: Glycerophosphoethanolamine (GPE) levels positively
associate with cognitive decline (FDR: 7.05e-13, Figure 2d, left) and N-acetylglutamate
negatively associate with global AD pathology (FDR: 3.59e-08, Figure 2d, right). GPE levels
were higher with lower cognitive abilities, which corroborates previously published findings16. N-
acetylglutamate levels showed lower levels with higher AD pathology load.
The 298 metabolites that associated with AD-related traits were distributed across all super-
pathways, including 113 (37.92%) within the largest super-pathway of lipids, followed by amino
acids with 78 (26.17%) associations, and the rest in remaining six super-pathways (Figure 2e).
At the more fine-grained sub-pathway level, metabolic associations were distributed across 72
out of the 101 sub-pathways covered in the data (Supplementary Figure 2).
We statistically inferred a metabolic network and annotated it with effect directions and the lowest
adjusted p-value across the eight AD traits (Figure 2f). The network is based on a Gaussian
graphical model (GGM), which corresponds to a data-driven representation of biochemical
pathways17,18. GGMs have previously been used to systematically investigate various trait effects
on the metabolome13,19. To further explore our findings networks for each AD trait are available
as a Cytoscape file (Supplementary File 1), as well as an interactive online version at
https://omicscience.org/apps/brainmwas/.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
8
Taken together, this analysis revealed global metabolic changes with respect to various AD-
related clinical and neuropathological traits. These alterations encompass all measured metabolic
super-pathways, highlighting the massive impact of the disease on brain metabolism.
Figure 2: Overview of metabolic associations with AD. a, Metabolites measured in brain samples are distributed
across various metabolic classes, referred to as “super-pathways” throughout the manuscript. b, Kendall correlations
across the eight AD-related traits. c, A total of 298 metabolites were associated with at least one of the eight AD-related
traits. d, Examples of two metabolites with the lowest adjusted p-values. Note that the traits were discretized (median-
split) for visualization. e, Distribution of metabolic associations across super-pathways. f, Gaussian graphical model of
metabolites. Metabolites are colored based on the negative log10 of the lowest adjusted p-value across AD-related traits
multiplied with the direction of the respective effect estimate. Global NP = Global burden of AD neuropathology. NP
Diagnosis = postmortem diagnosis based on Braak stage and CERAD score.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
9
2.3 Conditional analysis suggests tau pathology as driver of
metabolic changes in brain
According to the b-amyloid hypothesis of AD, b-amyloid is key to AD pathogenesis20. It is
considered to influence the accumulation of tangles of phosphorylated tau as well as tangle-driven
pathogenesis21. As a result, b-amyloid has been the focus of most therapeutic approaches2226.
However, recent evidence suggests that tau tangles might be acting independent of b-amyloid27.
To identify metabolic signatures specific to β-amyloid and tau tangles we performed conditional
analyses by adjusting for the respective other neuropathology (Figure 3a). In our standard
association analysis, i.e., without accounting for β-amyloid load, 188 metabolites were associated
with tau tangle load. 119 out of these 188 associations were still significant after accounting for
β-amyloid load in the statistical model. While 34 metabolites associated with β-amyloid load in the
standard association analysis, only one remained significant after accounting for tau tangle load.
Details of the standard and conditional analysis are available in Supplementary Table 6 and
Supplementary Table 9, respectively. Taken together, this analysis suggests that metabolic
associations of tau tangles are largely independent of β-amyloid load, while metabolic
associations of β-amyloid load are confounded by tau tangle load.
To corroborate this finding with another omics layer, we performed the same analysis on
proteomics profiles. In the standard association analysis, 695 proteins associated with tau tangle
load, i.e., without accounting for β-amyloid load. 252 out of these 695 were still associated with
tau tangle load after accounting for β-amyloid load in our statistical model. While 265 proteins
were associated with β-amyloid load in the standard association analysis, only 68 of these
remained correlated after accounting for tau tangle load (Figure 3b). Details of the proteomics
standard and conditional analysis are available in Supplementary Table 10.
Taken together, metabolic associations were more widespread for tau tangles and less dependent
on β-amyloid load, which was partially confirmed by a similar trend in the proteomics data.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
Figure 3: Comparison of standard and conditional analyses of
b
-amyloid and tau tangles. Tangle-associated
signals appeared largely independent of b-amyloid signals, while b-amyloid signals were more strongly dependent on
tau tangle signals. a,b Overlap of tangle- and β-amyloid-associated metabolites and proteins, respectively, with and
without adjusting for the respective other neuropathology.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
2.4 AD-associated metabolic alterations overlap across
independent brain studies
To strengthen confidence in our findings, we performed replication analysis using a metabolomics
dataset from a Mayo Clinic brain bank cohort. In addition, we compared our results to a previously
published brain metabolomics study based on samples from the Baltimore Longitudinal Study of
Aging (BLSA)5. Detailed replication results of Mayo can be found in Supplementary Table 11
and published BLSA results used for comparison can be found in Supplementary Table 12.
In the Mayo data, 83 temporal cortex brain samples were used for untargeted metabolic profiling,
including 63 AD patients and 20 controls. Of the 8 AD-related traits used in the discovery phase
with ROS/MAP cohort, neuropathology-based diagnosis was the only matching trait available in
this cohort. Individual measures of neuropathology were not comparable between cohorts, and
cognitive assessments were not available for the Mayo cohort. A total of 257 metabolites of the
298 significant in ROS/MAP cohort were measured in the Mayo cohort. 30 of these 257
metabolites were associated with AD in both datasets, i.e., with AD diagnosis in the Mayo cohort
and with at least one of the eight AD-related traits in ROS/MAP (Figure 4a), all of which showed
consistent effect directions.
In the BLSA study5, 43 samples from the inferior temporal gyrus (ITG) and middle frontal gyrus
(MFG) brain regions were used for targeted metabolic profiling. The study identified 130
metabolites, of which the authors focused on 26, which were further categorized into different
biochemical groups. In their analysis, 9 out of 26 metabolites associated with AD diagnosis. All
26 metabolites were measured in our study, 17 of these 26 associated with AD-related traits, and
6 of these 17 were among the 9 metabolites associated in their study (Figure 4b). Of those 6
metabolites, 5 had the same effect directions, while cysteine was found to be upregulated in the
BLSA study and downregulated in ROS/MAP.
Overall, 35 of the 298 associations identified in ROS/MAP were confirmed with consistent effect
directions in either the Mayo or the BLSA cohort.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
Figure 4: Overlap across independent cohorts and brain-regions. a, The Mayo and ROS/MAP cohorts have 30
metabolic associations with consistent effect directions in common. b, The BLSA and ROS/MAP cohorts have 5
metabolic associations with consistent effect direction in common (green and brown). Cysteine showed inconclusive
effect directions, with a positive association with AD in the BLSA cohort and a negative association with AD in the
ROS/MAP cohort.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
2.5 Metabolic alterations further characterize pathways previously
implicated in AD
To examine the contribution of metabolic alterations in previously reported AD-related pathogenic
processes, we selected four pathway groups for further exploration (Figure 5). Comprehensive
functional annotations based on Metabolon’s sub-pathways are available in Supplementary
Figure 2. For each pathway group discussed below, we included metabolites associated with at
least one of the eight AD-related traits.
Bioenergetic pathways: Bioenergetic dysregulation is a hallmark of AD, which has been
demonstrated using different technologies, from PET neuroimaging to deep molecular profiling
such as metabolomics and proteomics studies2831. In our analysis, key metabolites from
bioenergetic pathways, including glycolysis, branched-chain amino acid (BCAA) metabolism, and
mitochondrial b-oxidation, were found to be deregulated in AD. This included positive correlations
of the glycolytic metabolites glucose, glycerate, glucose 6-phosphate, and 1,5-anhydroglucitol;
the BCAAs valine, isoleucine, and leucine, as well as their 1-carboxyethyl conjugates and
degradation products b-hydroxyisovalerate, and 3-hydroxyisobutyrate; and the acylcarnitines
isobutyrylcarnitine (C4), tiglyl carnitine (C5), 2-methylbutyrylcarnitine (C5), glutarylcarnitine
(C5-DC), and 5-dodecenoylcarnitine (C12:1). High abundances of these metabolites resemble
observations in blood metabolic profiles of individuals with type 2 diabetes and stand in contrast
to blood-based studies in AD, which reported negative associations of, e.g., BCAA levels with
AD4,32. Together, these results are in line with the hypothesis that AD might represent a “type 3”
diabetes that selectively affects the brain33.
Cholesterol metabolism and sterol pathway: The strongest genetic risk for AD is exerted by
variants of the APOE gene, a lipoprotein involved in cholesterol transport and metabolism3436.
Our findings provide further evidence for the AD-associated significance of this pathway, with
degradation products of cholesterol showing positive correlations with AD-related traits. These
degradation products include -hydroxy-3-oxo-4-cholestenoic acid (7-HOCA), which has been
described as a CSF-based marker for blood-brain-barrier integrity37; 4-cholesten-3-one, a product
of cholesterol oxidation and isomerization through bacterial enzymes38; and 7-hydroxycholesterol,
a precursor for bile acids. Notably, cholesterol itself did not show any significant associations,
indicating potential dysfunctional cholesterol clearance rather than a direct role of cholesterol in
AD. This hypothesis is further supported by previous studies where we observed a significant
increase of secondary bile acids in AD7,33,39.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
Neuroinflammation and oxidative stress: Neuroinflammation is a central pathogenic feature of
AD and is accompanied by the production of reactive oxygen species leading to oxidative stress40.
AD has been associated with both lipid mediators of inflammatory processes as well as immune
response, including eicosanoids, and molecules involved in the antioxidant defense, such as
glutathione4143. In line with these findings, we observed significant positive correlations of
metabolites in the glutathione pathway with AD, indicating an upregulated antioxidant response.
Significant metabolites included 4-hydroxy-nonenal-glutathione, a marker for detoxification of lipid
peroxidation through glutathione S-transferases (GSTs)44; cysteinylglycine disulfide, a
degradation product of oxidized glutathione42; and ophthalmate, an endogenous analog of hepatic
glutathione (GSH) and potential marker for GSH depletion45. Moreover, pro-inflammatory
eicosanoids showed positive associations with AD, including 15-oxoeicosatetraenoic acid
(15-KETE), which has been linked to GST inhibition46, and 12-hydroxy-heptadecatrienoic acid
(12-HHTrE), overall providing further molecular evidence for active inflammatory processes in
AD. In contrast, anti-inflammatory long-chain omega-3 polyunsaturated fatty acids (PUFAs), such
as eicosapentaenoate (EPA) and docosahexaenoate (DHA)47,48, were negatively associated with
AD.
Osmoregulation. Osmolytes are a class of molecules that primarily sustain cell integrity49. They
have been suggested to play a neuroprotective role in AD by activating mTOR-independent
autophagy signaling to inhibit accumulation of b-amyloid plaques50. Osmolytes also affect protein
folding51, and their therapeutic potential has been discussed in AD as well as other
neurodegenerative proteinopathies52. Moreover, osmolyte imbalances can impact neuronal
hyperexcitation by influencing neurotransmitter uptake49. In our analysis, we observed positive
associations of several osmolytes with AD, including 2-aminoadipate, arginine,
glycerophosphorylcholine (GPC), myo-inositol, serine, and urea, whereas betaine was negatively
associated with the disease. As these observations are based on bulk tissue metabolomics, it
remains unclear if these metabolites are deregulated within or outside of the cell. Nevertheless,
the strong statistical significance underlying these associations suggest an important role of
osmoregulation in AD which warrants further investigation.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
Figure 5: Pathway-level metabolic associations with AD-related traits. The highlighted biological processes have
previously been implicated in AD. Our data provides a metabolic characterization of the alterations of these pathways
in the AD brain: Cholesterol metabolism has an established connection to late-onset AD through APOE4, the major
genetic risk factor for the disease. Bioenergetic dysregulation is one of the earliest detectable changes in the central
nervous system in AD and has also been described in the periphery. Inflammation and oxidative stress have been
reported to synergistically affect AD pathogenesis. Osmoregulation affects various aspects of AD pathology, including
protein folding, neural excitation, and autophagy.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
2.6 Integration of complementary omics provides comprehensive
view of biochemical cascade downstream of neurotransmitters
As a detailed showcase of the complex, biochemical interconnections in brain omics data, we
selected a biochemical cascade downstream of the neurotransmitters glutamate and gamma
aminobutyric acid (GABA). An elevated synaptic excitatory/inhibitory (E/I) ratio of these
neurotransmitters has been linked to hyperexcitability and cognitive impairment observed in
AD53,54 . Furthermore, given GABA’s positive correlation with efficient working memory within the
DLPFC region55, it is of high significance to investigate GABA-related deregulation in this region.
We compiled biochemical steps of metabolites and enzymes downstream of glutamate using
known reactions from the public database pathbank56 (Figure 6). Notably, the cascade does not
contain the routes from GABA to glutamate or from putrescine to GABA due to a lack of coverage
of metabolites along those pathways.
Based on proteomics profiles available for 262 matching brain samples, we performed a targeted
association analysis of AD-related traits and proteins that are enzymatically involved in this
pathway cascade (Supplementary Table 13). Significant metabolic and proteomic associations
with at least one of the eight AD-related traits were annotated on the respective molecules within
the cascade.
The pathway cascade starts with glutamate, which was positively associated with AD in our data.
Excitatory glutamatergic synapses involving N-methyl D-aspartate receptors (NMDAR) have
previously been targeted by memantine to treat severe AD57. Glutamate is the precursor of the
inhibitory neurotransmitter GABA, which we found to be negatively associated with AD.
Interestingly, protein abundance of glutamate decarboxylase (GAD2), which catalyzes the
production of GABA from glutamate was also negatively associated with AD pathology in our data.
This negative association provides a potential explanation for the imbalance between the two
neurotransmitters. Glutamate metabolism is directly connected to the urea cycle, in which
ornithine, arginine, and urea were positively associated with AD. Urea buildup to neurotoxic levels
has been observed in postmortem brains of Huntington’s disease and has furthermore been
linked to dementia58. Inhibition of arginase (ARG2) has been suggested to reduce the production
of urea59. Arginase was positively associated with AD in our data; it catalyzes the conversion of
arginine to ornithine, with urea as a byproduct. Urea cycle further feeds into the polyamine
pathway, in which putrescine was negatively associated with AD, while spermidine and spermine
were not significantly associated with AD. Putrescine promotes the clearance of apoptotic cells
via efferocytosis60, a mechanism affected in AD and other neurodegenerative diseases61. The
enzyme S-methyl-5'-thioadenosine phosphorylase (MTAP) links the polyamine pathway to
methionine metabolism, in which methionine, methionine sulfoxide, s-adenosylmethionine, and
s-adenosylhomocysteine were positively associated with AD, with concordant changes in protein
levels of respective enzymes MTAP, mitochondrial peptide methionine sulfoxide reductase
(MSRA) and methionine adenosyltransferase (MAT2A). In a previous study, we have shown that
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
higher levels of methionine in CSF were associated with AD8. Methionine acts as an antioxidant
by forming methionine sulfoxide and is a precursor of s-adenosylmethionine, which is a key methyl
donor in brain cells and involved in the synthesis of the neurotransmitters dopamine, epinephrin,
and serotonin via the folate cycle62.
Overall, our analysis provides an integrated, multi-omics view of neurotransmitter-related
changes known to play a role in the pathogenesis of AD.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
Figure 6: Metabolic changes downstream of the neurotransmitters glutamate/GABA. This multi-omics cascade
starts with biochemical process involving conversion of glutamate into GABA within glutamate metabolism. Glutamate
metabolism feeds into the urea cycle by conversion of glutamate to ornithine. Urea buildup to neurotoxic levels has
been observed in postmortem brains of Huntington’s disease and has furthermore been linked to dementia. Urea cycle
connects to polyamine metabolism via conversion of ornithine into putrescine. Putrescine promotes the clearance of
apoptotic cells via efferocytosis, a mechanism affected in AD. Polyamine metabolism connects to methionine
metabolism though methionine salvage pathway. Methionine acts as an antioxidant and is a precursor of s-
adenosylmethionine, which is a key methyl donor in brain cells and involved in the synthesis of neurotransmitters
dopamine, epinephrin, and serotonin via folate cycle.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
3 Discussion
In this work, we provide a global view of metabolic changes in brain related to Alzheimer’s
disease. Our study is based on broad untargeted metabolomics profiles of 500 brain tissue
samples from the DLPFC, covering 667 metabolites from various biochemical classes. We
demonstrated that in cognitively normal individuals, age, sex, education, and BMI did not show
major effects on brain metabolites. These limited associations of brain metabolites with
demographics and socioeconomic status stand in contrast to the strong associations seen with
blood metabolic profiles1315. Conversely, intake of medications had major effects on brain
metabolome, as observed in blood metabolic profiles4, highlighting the importance to account for
effects of pharmaceuticals.
In the subsequent association analysis, we found that 298 out of the 667 metabolites correlated
with at least one of the eight investigated AD-related traits, covering cognition and several
neuropathological parameters. We confirmed 30 of our associations using independent samples
from the Mayo Clinic brain bank cohort. Additionally, 5 associations were confirmed using a study
on the BLSA cohort5. Two pathways, urea cycle and glutathione metabolism, were associated
with AD in all three cohorts. This overlap was observed despite the substantial differences in
sample sizes, profiled brain regions, study designs, and clinical parameters. We thus conclude
that the 35 metabolites and two pathways are high confidence AD-related metabolic signals in
brain tissue, and the metabolic associations unique to our ROS/MAP study need further
validation. Of note, we observed significant modulation of metabolic associations through sex and
APOE4 status, which is concordant with previous findings in blood-based metabolomics data3.
We explored our findings in the context of various functional processes that have been previously
implicated in AD, including bioenergetic pathways, cholesterol metabolism, neuroinflammation,
and osmoregulation. Our study extends the view on these AD-related pathways through metabolic
alterations in brain. Of these processes, metabolic alterations of osmoregulation within the central
nervous system have, to the best of our knowledge, so far not been studied in detail. Osmolytes
participate in multiple critical processes associated with neurogenerative diseases including
protein folding63, autophagy50, and hyperexcitation of neurons49. While our observations might to
some extent be confounded by, e.g., systematic differences in the hydration status of AD patients
prior to death, the large number and strong significance of associations within this class suggests
a potentially functional link to pathomechanisms in AD.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
Moreover, we investigated detailed biochemical alterations downstream of glutamate and GABA.
Imbalances of these neurotransmitters have previously been associated with hyperexcitability and
cognitive impairment in AD53,54. In our study, the excitatory neurotransmitter glutamate was
positively associated with AD, while the inhibitory neurotransmitter GABA was negatively
associated with AD. To investigate the downstream effects of this excitatory/inhibitory imbalance,
we explored the metabolic and enzymatic changes in the biochemical cascade starting from the
conversion of glutamate to GABA, connecting glutamate to urea cycle, polyamine metabolism,
and methionine metabolism. Our study shows how the integration of metabolomics with
proteomics provides a comprehensive overview of biochemical changes downstream of these
neurotransmitters. Moreover, to the best of our knowledge, this is the first reporting of low levels
of GABA in AD within the DLPFC region. DLPFC is associated with working memory in
individuals55, which becomes impaired during AD pathogenesis64. Thus, GABA levels within the
DLPFC region have been of considerable interest to the AD community64, which is corroborated
by our results.
Addressing the complex interplay of β-amyloid deposition and tau tangles in AD, we performed a
conditional statistical analysis. In our data, 97% of the β-amyloid-associated metabolites were
dependent on tau tangle load, while only 36.7% of the tangle-associated metabolites were
dependent on β-amyloid load. Our study thus provides preliminary evidence that the metabolic
component of tangle-driven pathogenesis is independent of β-amyloid, which is in line with recent
literature that suggests that tau accumulation might be independent of β-amyloid27. This finding
may also suggest that metabolic changes in the brain are mostly later events in the pathologic
cascade of AD65 and closer temporally to tau pathology, neurodegeneration and cognitive decline
than to b-amyloid accumulation. Further supporting this, the largest number of associations in
ROS/MAP were detected with cognitive decline, an event deemed to be at the later stages of the
pathologic cascade of events in AD66.
Despite the many novel insights into metabolic alterations in brain observed in AD, our study has
some limitations. First, cross-sectional studies cannot assess the causal direction of the identified
associations. That is, an observed metabolic change in AD could be a factor directly contributing
to disease development, or it could be a downstream effect of the pathological changes in brain.
The true effect direction can only be determined in mechanistic follow-up studies or by genetic
causality analysis such as Mendelian randomization67, for which our study did not have the
necessary statistical power. Second, postmortem tissue samples are prone to substantial
biological and technical variation, as seen in the association of 307 out of 667 metabolites with
postmortem interval (PMI), i.e., the time between death and sample preservation. Despite the
statistical correction for PMI interval, degradation of certain metabolites until sample preservation
is a factor that cannot be controlled in this type of study.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
Follow-up studies will be needed to build upon our findings, to complete the picture of
dysregulated metabolism and pathological pathways in the Alzheimer’s disease brain. In
particular, the wide availability of multi-omics datasets will provide a more holistic picture of the
molecular changes associated with the disease6870. The integration of proteomics data into the
glutamate/GABA pathway exploration in our study represents a pilot analysis in this direction;
however, large-scale studies with an “ome-wide” integration of the (epi-)genome, transcriptome,
proteome, and metabolome are required to further elucidate the mechanistic basis of AD
pathogenesis and outline potential treatment options. To enable these efforts, we published raw
and processed metabolomics data through the AD Knowledge Portal provide all analysis codes,
as well as the interactive reference catalog of hundreds of associations reported in this study in
an accompanying web portal to the research community.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
4 Methods
4.1 Cohorts, clinical data, and neuropathological data
ROS/MAP cohort: The Religious Order Study (ROS) and Rush Memory and Aging Project (MAP)
cohorts11,12 are two longitudinal, clinicopathologic studies conducted by the Rush Alzheimer’s
Disease Center. ROS started in 1994 with the recruitment of individuals from religious
communities across the United States. MAP started in 1997 with the recruitment of individuals
from a wide range of backgrounds and socio-economic statuses from northeastern Illinois. Both
cohorts were approved by an institutional review board of Rush University Medical Center. Both
studies focus on older individuals who agreed to longitudinal clinical analysis and brain donation
after death. All participants signed an informed consent, an Anatomic Gift Act, and a repository
consent to allow their data and biospecimens to be shared. Following enrollment in the study,
participants were evaluated for physical and cognitive function annually. After death, pathologic
assessment was performed. 514 samples from DLPFC brain region were used for metabolomics
profiling, along with associated metadata, including medications taken during lifetime, age at
death, sex, BMI, postmortem interval, APOE genotype status, education history, cognitive scores
during lifetime, cognitive decline (computed-based on longitudinal cognitive scores), clinical
diagnosis at death, b-amyloid and tau protein load in brain tissue, global burden of AD
neuropathology (mean of neuritic plaques, diffuse plaques, and neurofibrillary tangles), NIA-
Reagan score, Braak stage and CERAD score. Neuropathological diagnosis was derived using
the following criteria: AD case status was assigned where Braak stage was ≥ 4 and CERAD score
was ≤ 2; control case status was assigned where Braak stage was ≤ 3 and CERAD score was
3. All clinical parameters have previously been described in detail71.
Mayo Clinic cohort: 84 samples from the temporal cortex brain region were obtained from the
Mayo Clinic Brain Bank. Details on this cohort have been provided in previous studies72,73. All
samples received diagnoses at autopsy following neuropathologic evaluation. Briefly, 64 samples
had neuropathologic diagnosis of AD with Braak ³4.0 and 20 control samples had Braak £3 and
without any neurodegenerative diagnoses. All 84 samples were from North Americans of
European descent with ages at death ³60 for AD and ³53 for controls.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
Cohort differences: Three cohorts were used in this publication - ROS/MAP for discovery, Mayo
brain clinical cohort for replication and a published Baltimore Longitudinal Study of Aging (BLSA)
based study5 for comparison. These cohorts have fundamental differences: (a) Participant
recruitment. BLSA is an aging study, Mayo Clinic samples are from an archival brain bank with
neuropathologic diagnoses of AD and control, while ROS/MAP recruited older people. (b) AD-
related traits. Mayo has diagnosis determined by neuropathology and BLSA has diagnosis
determined based on neuropathology and cognitive conditions. ROS/MAP records several
neuropathological as well as cognitive scores. (c) Unlike the other two cohorts, ROS/MAP collects
various lifetime variables longitudinally, including cognitive scores, lifestyle, medications taken by
participants. (d) Sample sizes were lower in BLSA (43), and Mayo (84), compared to ROS/MAP
(514). (e) Different brain regions were profiled. BLSA sampled frontal and temporal gyrus, Mayo
the superior temporal gyrus of the temporal cortex, and ROS/MAP the dorsolateral prefrontal
cortex (DLPFC).
4.2 Metabolomics profiling
Brain metabolic profiles were measured using the untargeted metabolomics platform from
Metabolon Inc. Briefly, tissue samples were divided into four fractions; two for ultra-high
performance liquid chromatography-tandem mass spectrometry (UPLC-MS/MS; positive
ionization), one for UPLC-MS/MS (negative ionization), and one for a UPLC-MS/MS polar
platform (negative ionization). Peaks were quantified using the area-under-the-curve in the
spectra. To account for run-day variations, peak abundances were normalized by their respective
run-day medians. Compounds were identified using an internal spectral database. A detailed
description of all experimental procedures can be found in supplementary information.
4.3 Data preprocessing
ROS/MAP and Mayo metabolomics: Metabolites with over 25% missing values were filtered
out, leaving 667 out of an original 1,055 metabolites for ROS/MAP and 664 out of 827 for Mayo.
Probabilistic quotient normalization was applied to correct for sample-wise variation74, followed
by log2 transformation. Remaining missing values were imputed using a k-nearest-neighbor-
based algorithm19. Outlier samples in the data were removed using the local outlier factor
method75 implemented in the R package bigutilsr. To account for remaining irregularly high or low
single concentrations, values with absolute abundance above q = abs(qnorm(0.0125/n)), with n
representing the number of samples, were set to missing. This formula finds the cutoff for values
with less than 2.5% two-tailed probability to originate from the same normal distribution as the
rest of the measurement values, after applying a Bonferroni-inspired correction factor (division by
sample size). These new missing values were then imputed by another round of the k-nearest-
neighbor algorithm.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
ROS/MAP proteomics: Proteomics data was downloaded from the AMP-AD Knowledge Portal
(https://adknowledgeportal.synapse.org), details of proteomic profiling and data processing can
be found in the original publication76. Briefly, data were log2-transformed and corrected for batch
effects using ‘median polish’ approach. In our analysis, proteins with over 25% missing values
were filtered out and remaining missing values were imputed using a k-nearest-neighbor-based
algorithm19. Outliers were treated with same approach as the metabolomics data (see above).
Medication correction: For the ROS/MAP cohort, all prescription and over the counter
medications were collected at each study visit. To account for influences of these medications on
metabolomics and proteomics, a linear stepwise backward selection approach was used4. All
preprocessing steps were performed using the maplet R package77.
4.4 Differential analysis of metabolites and proteins
ROS/MAP: Five outliers identified by the local outlier factor method, six samples with missing
medication information, 1 sample with missing BMI and 2 samples with missing APOE genotype
status were removed from further analysis. Therefore, after preprocessing, 500 samples were
used for metabolic analysis and 262 matching samples were used for proteomic analysis.
Metabolite and protein associations were computed using generalized linear models with the traits
as response variables and molecule levels as predictors. For statistical analysis, the following
transformations were made: Square root of b-amyloid load and binarized NIA-Reagan score (0
low likelihood of AD, 1 high likelihood of AD). For the association analysis with clinical diagnosis,
8 non-AD related dementia samples were removed. Appropriate link functions were used
according to the respective variable types, i.e., identity link function for continuous traits (regular
linear regression for b-amyloid, tau tangles, global burden of pathology, cognition levels, cognitive
decline), logit for binary traits (logistic regression for NIA-Reagan score and NP diagnosis), and
probit for the ordinal trait (ordinal regression for clinical diagnosis after death). All models
accounted for confounding effects of age, sex, BMI, postmortem interval, number of years of
education, and number of APOE ε4 alleles. Notably, age, sex, years of education did not show
much influence on metabolic profiles of cognitively normal samples, but are known confounders
of AD3, justifying the correction in the models. To account for multiple hypothesis testing, p-values
were corrected using the Benjamini-Hochberg (BH) method78. Cognitive decline and cognition
levels are inversely related to AD, and thus the direction of association was reversed for those
two traits after statistical analysis.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
Mayo: One AD sample with missing APOE genotype status was removed from further analysis.
Therefore, after preprocessing, 63 AD and 20 control samples with complete information on age
at death, APOE ε4 allele status, and sex were used for our analysis. For replication, metabolites
that associated with any of the eight AD-related traits in ROS/MAP cohort at 5% FDR were
selected. The analysis was performed using two subsequent logistic regressions with diagnosis
as outcome. The first model was built without any confounder correction. To account for multiple
hypothesis testing, p-values were corrected using the Benjamini-Hochberg (BH) method78.
Metabolites with adjusted p-values < 0.05 were selected for the second model. The second model
was built with confounders sex, number of APOE ε4 alleles, and age at death. Metabolites with
nominal p-values < 0.05 in the second model were considered replicated. All analyses were
performed using the maplet R package77.
4.5 Stratified analysis
To determine the influence of sex and APOE ε4 status on metabolic associations, we performed
a stratified analysis per factor (sex and APOE ε4 status) for each AD-related trait. Metabolites
significant at 5% FDR were selected to compute within-group (male/female, APOE ε4+/ APOE ε4-)
metabolic associations with AD-related traits. b estimates across groups per metabolite were
compared using z-scores13, defined as
𝑧 = #
b
!"#$%&!"
b
!"#$%'
#
$%!"#$%&'&$%!"#$%''
, where b
'()*+,
and b
'()*+-
are
the b coefficients from the linear regressions performed in the two groups, and
𝑠𝑒'()*+,
and
𝑠𝑒'()*+-
are the corresponding standard errors. Z-scores are approximately standard normally
distributed and were thus used to compute p-values using a normal distribution. Any metabolite
with a nominal p-value < 0.05 was considered significantly different within the respective group.
4.6 Metabolic network inference
To infer the metabolite-metabolite interaction network, a partial correlation-based Gaussian
graphical model (GGM) was computed using the GeneNet R package79. P-values of partial
correlations were corrected using the Bonferroni method. Partial correlations with adjusted
p-values < 0.05 were used for network construction between metabolites. To annotate the
metabolic network with AD associations, a score was computed for each metabolite/trait
combination as follows:
𝑝$.)(% = 𝑑 ∗ (−1 ∗ log,/
(
𝑝. 𝑎𝑑𝑗
)), where
𝑝. 𝑎𝑑𝑗
is the adjusted p-value of
the model, and d is direction (-1/1) of metabolite association based on test statistic (positive or
negative correlation with AD-related trait). To aggregate the signal across the traits, an overall
score was defined as the pscore with maximum absolute value. This overall score was used to color
the nodes in GGM in Figure 2f and the online supplement.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
Data availability
The data used in this paper can to be obtained from two sources: (1) Metabolomics data for the
ROS/MAP and Mayo cohorts, clinical data for the Mayo cohort, and proteomics data for the
ROS/MAP cohort are available via the AD Knowledge Portal (https://adknowledgeportal.org). The
AD Knowledge Portal is a platform for accessing data, analyses, and tools generated by the
Accelerating Medicines Partnership (AMP-AD) Target Discovery Program and other National
Institute on Aging (NIA)-supported programs to enable open-science practices and accelerate
translational learning. The data, analyses and tools are shared early in the research cycle without
a publication embargo on secondary use. Data is available for general research use according to
the following requirements for data access and data attribution
(https://adknowledgeportal.org/DataAccess/Instructions). For access to content described in this
manuscript see: http://doi.org/10.7303/syn26401311. (2) The full complement of clinical and
demographic data for the ROS/MAP cohort are available via the Rush AD Center Resource
Sharing Hub and can be requested at https://www.radc.rush.edu.
An interactive network view of AD associations from this study can be found at
https://omicscience.org/apps/brainmwas/.
All R scripts to generate the tables and figures of this paper are available at
https://github.com/krumsieklab/ad-brain-landscape.
Acknowledgements
This work was done as part of the National Institute of Aging’s Accelerating Medicines Partnership
for AD (AMP-AD) and was supported by NIH grants 1U19AG063744, 1R01AG069901-01A1,
U01AG061357, P30AG10161, P30AG72975, R01AG15819, R01AG17917, U01AG46152,
U01AG61356, RF1AG058942, RF1AG059093, and U01AG061359. The results published here
are in whole or in part based on data obtained from the AD Knowledge Portal
(https://adknowledgeportal.org).
The Religious Orders and the Rush Memory and Aging studies were supported by the National
Institute on Aging grants P30AG10161, R01AG15819, R01AG17917, U01AG46152, and
U01AG61356. The NIA also supported the Alzheimer Disease Metabolomics Consortium which
is a part of the NIA’s national initiatives AMP-AD and M2OVE-AD (R01 AG046171, RF1
AG051550, and 3U01 AG061359-02S1). We thank the participants of ROS and MAP for their
essential contributions and the gifts of their brains to these projects. All subjects gave informed
consent.
The Mayo Clinic samples are part of the RNAseq study data led by Dr. Nilüfer Ertekin-Taner,
Mayo Clinic, Jacksonville, FL as part of the multi-PI U01 AG046139 (MPIs Golde, Ertekin-Taner,
Younkin, Price). Samples were provided from the following sources: The Mayo Clinic Brain Bank.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
Data collection was supported through funding by NIA grants P50 AG016574, R01 AG032990,
U01 AG046139, R01 AG018023, U01 AG006576, U01 AG006786, R01 AG025711, R01
AG017216, R01 AG003949, NINDS grant R01 NS080820, CurePSP Foundation, and support
from Mayo Foundation.
RB thanks her colleagues from Krumsiek lab for fruitful discussions and support in this work.
Author contributions
RB, MArnold, GK, JK designed the computational and statistical methods, performed the analysis,
interpreted the results, and drafted the manuscript. MAW created the interactive online
supplement. DAB is PI of the ROS/MAP study and provided samples as well as phenotypic data.
NET, XW, MAllen provided samples and phenotypic data on the Mayo clinic cohort and
contributed to the analysis of those samples. CB curated and managed phenotypic data. AIL,
NTS provided the preprocessed proteomics data for matching ROS/MAP samples and
contributed to the analysis of those samples. All authors read and reviewed the manuscript. RK-
D acquired funding and is the overall PI of the Alzheimer's disease metabolomics consortium.
Competing interests
R.K-D., MArnold, GK are (through their institutions) inventors on key patents in the field of
metabolomics including applications for Alzheimer disease. R.K-D. holds equity in Metabolon Inc.,
a metabolomics technologies company. This platform was used in the current analyses. R.K-D.
formed Chymia LLC and PsyProtix, a Duke University biotechnology spinout aiming to transform
the treatment of mental health disorders. JK, MArnold, and GK hold equity in Chymia LLC and IP
in PsyProtix.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
References
1. Hampel, H. et al. Blood-based biomarkers for Alzheimer disease: mapping the road to the
clinic. Nat Rev Neurol 14, 639652 (2018).
2. Ross, C. A. & Poirier, M. A. Protein aggregation and neurodegenerative disease. Nat. Med.
10, S10 (2004).
3. Arnold, M. et al. Sex and APOE epsilon4 genotype modify the Alzheimer’s disease serum
metabolome. Nat Commun 11, 1148 (2020).
4. Toledo, J. B. et al. Metabolic network failures in Alzheimer’s disease: A biochemical
road map. Alzheimer’s Dement. 13, 965984 (2017).
5. Mahajan, U. V. et al. Dysregulation of multiple metabolic networks related to brain
transmethylation and polyamine pathways in Alzheimer disease: A targeted metabolomic
and transcriptomic study. PLoS Med. 17, e1003012 (2020).
6. MahmoudianDehkordi, S. et al. Altered bile acid profile associates with cognitive
impairment in Alzheimer’s disease-An emerging role for gut microbiome. Alzheimers
Dement 15, 7692 (2019).
7. Nho, K. et al. Altered bile acid profile in mild cognitive impairment and Alzheimer’s disease:
Relationship to neuroimaging and CSF biomarkers. Alzheimers Dement 15, 232244
(2019).
8. Kaddurah-Daouk, R. et al. Alterations in metabolic pathways and networks in Alzheimer’s
disease. Transl Psychiatry 3, e244 (2013).
9. Fuente, M. de la et al. Enzyme Therapy: Current Challenges and Future Perspectives. Int.
J. Mol. Sci. 22, 9181 (2021).
10. Robertson, J. G. Enzymes as a special class of therapeutic target: clinical drugs and modes
of action. Curr. Opin. Struct. Biol. 17, 674679 (2007).
11. Bennett, D. A. et al. Overview and findings from the rush Memory and Aging Project. Curr
Alzheimer Res 9, 646663 (2012).
12. Bennett, D. A., Schneider, J. A., Arvanitakis, Z. & Wilson, R. S. Overview and findings from
the religious orders study. Curr Alzheimer Res 9, 628645 (2012).
13. Krumsiek, J. et al. Gender-specific pathway differences in the human serum metabolome.
Metabolomics 11, 18151833 (2015).
14. Stephens, C. R. et al. The Impact of Education and Age on Metabolic Disorders. Front.
Public Heal. 8, 180 (2020).
15. Mittelstrass, K. et al. Discovery of Sexual Dimorphisms in Metabolic and Genetic
Biomarkers. PLOS Genet. 7, e1002215 (2011).
16. Blusztajn, J. K., Lopez Gonzalez-Coviella, I., Logue, M., Growdon, J. H. & Wurtman, R. J.
Levels of phospholipid catabolic intermediates, glycerophosphocholine and
glycerophosphoethanolamine, are elevated in brains of Alzheimer’s disease but not of
Down’s syndrome patients. Brain Res 536, 240244 (1990).
17. Krumsiek, J., Suhre, K., Illig, T., Adamski, J. & Theis, F. J. Gaussian graphical modeling
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
reconstructs pathway reactions from high-throughput metabolomics data. BMC Syst Biol
5, 21 (2011).
18. Krumsiek, J. et al. Mining the Unknown: A Systems Approach to Metabolite Identification
Combining Genetic and Metabolic Information. PLOS Genet. 8, e1003005 (2012).
19. Do, K. T. et al. Characterization of missing values in untargeted MS-based metabolomics
data and evaluation of missing data handling strategies. Metabolomics 14, (2018).
20. Hardy, J. A. & Higgins, G. A. Alzheimer’s disease: The amyloid cascade hypothesis.
Science (80-. ). 256, 184185 (1992).
21. Hardy, J. & Selkoe, D. J. The amyloid hypothesis of Alzheimer’s disease: Progress and
problems on the road to therapeutics. Science (80-. ). 297, 353356 (2002).
22. Salloway, S. et al. Two Phase 3 Trials of Bapineuzumab in Mild-to-Moderate Alzheimer’s
Disease. N. Engl. J. Med. 370, 322333 (2014).
23. Ostrowitzki, S. et al. A phase III randomized trial of gantenerumab in prodromal Alzheimer’s
disease. Alzheimer’s Res. Ther. 9, (2017).
24. Honig, L. S. et al. Trial of Solanezumab for Mild Dementia Due to Alzheimer’s Disease. N.
Engl. J. Med. 378, 321330 (2018).
25. Henley, D. et al. Preliminary Results of a Trial of Atabecestat in Preclinical Alzheimer’s
Disease. N. Engl. J. Med. 380, 14831485 (2019).
26. Egan, M. F. et al. Randomized Trial of Verubecestat for Prodromal Alzheimer’s Disease.
N. Engl. J. Med. 380, 14081420 (2019).
27. van der Kant, R., Goldstein, L. S. B. & Ossenkoppele, R. Amyloid-β-independent regulators
of tau pathology in Alzheimer disease. Nat. Rev. Neurosci. 21, 2135 (2020).
28. Wang, W., Zhao, F., Ma, X., Perry, G. & Zhu, X. Mitochondria dysfunction in the
pathogenesis of Alzheimer’s disease: recent advances. Mol. Neurodegener. 2020 151 15,
1–22 (2020).
29. Swerdlow, R. H. Mitochondria and Cell Bioenergetics: Increasingly Recognized
Components and a Possible Etiologic Cause of Alzheimer’s Disease. Antioxid. Redox
Signal. 16, 1434 (2012).
30. Dematteis, G. et al. Proteomic analysis links alterations of bioenergetics, mitochondria-ER
interactions and proteostasis in hippocampal astrocytes from 3xTg-AD mice. Cell Death
Dis. 2020 118 11, 116 (2020).
31. Sonntag, K.-C. et al. Late-onset Alzheimer’s disease is associated with inherent changes
in bioenergetics profiles. Sci. Reports 2017 71 7, 113 (2017).
32. Tynkkynen, J. et al. Association of branched-chain amino acids and other circulating
metabolites with risk of incident dementia and Alzheimer’s disease: A prospective study in
eight cohorts. Alzheimer’s Dement. 14, 723733 (2018).
33. Monte, S. M. de la & Wands, J. R. Alzheimer’s Disease Is Type 3 DiabetesEvidence
Reviewed. J. diabetes Sci. Technol. 2, 1101 (2008).
34. Coon, K. D. et al. A High-Density Whole-Genome Association Study Reveals That APOE
Is the Major Susceptibility Gene for Sporadic Late-Onset Alzheimer’s Disease. J. Clin.
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
Psychiatry 68, 00 (2007).
35. Corder, E. H. et al. Gene dose of apolipoprotein E type 4 allele and the risk of Alzheimer’s
disease in late onset families. Science (80-. ). 261, 921923 (1993).
36. Husain, M. A., Laurent, B. & Plourde, M. APOE and Alzheimer’s Disease: From Lipid
Transport to Physiopathology and Therapeutics. Front. Neurosci. 0, 85 (2021).
37. Saeed, A. et al. -hydroxy-3-oxo-4-cholestenoic acid in cerebrospinal fluid reflects the
integrity of the blood-brain barrier. J. Lipid Res. 55, 313 (2014).
38. Kerber, T. & Vrielink, A. The role of hydrogen atoms in redox catalysis by the flavoenzyme
cholesterol oxidase. Methods Enzymol. 634, 361377 (2020).
39. MahmoudianDehkordi, S. et al. Altered bile acid profile associates with cognitive
impairment in Alzheimer’s disease-An emerging role for gut microbiome. Alzheimers
Dement 15, 7692 (2019).
40. Leng, F. & Edison, P. Neuroinflammation and microglial activation in Alzheimer disease:
where do we go from here? Nat. Rev. Neurol. 17, 157172 (2021).
41. Borkowski, K. et al. Serum metabolomic biomarkers of perceptual speed in cognitively
normal and mildly impaired subjects with fasting state stratification. Sci. Reports 2021 111
11, 112 (2021).
42. Wu, G., Fang, Y. Z., Yang, S., Lupton, J. R. & Turner, N. D. Glutathione Metabolism and
Its Implications for Health. J. Nutr. 134, 489492 (2004).
43. Allen, M. et al. Glutathione S-transferase omega genes in Alzheimer and Parkinson
disease risk, age-at-diagnosis and brain gene expression: An association study with
mechanistic implications. Mol. Neurodegener. 7, (2012).
44. Singhal, S. S. et al. Antioxidant role of glutathione S-transferases: 4-Hydroxynonenal, a
key molecule in stress-mediated signaling. Toxicol. Appl. Pharmacol. 289, 361370 (2015).
45. Dello, S. A. W. G. et al. Systematic review of ophthalmate as a novel biomarker of hepatic
glutathione depletion. Clin. Nutr. 32, 325330 (2013).
46. Snyder, N. W. et al. 15-oxoeicosatetraenoic acid is a 15-hydroxyprostaglandin
dehydrogenase-derived electrophilic mediator of inflammatory signaling pathways. Chem.
Biol. Interact. 234, 144 (2015).
47. Calder, P. C. Omega-3 Fatty Acids and Inflammatory Processes. Nutrients 2, 355 (2010).
48. Song, C., Li, X., Kang, Z. & Kadotomi, Y. Omega-3 Fatty Acid Ethyl-Eicosapentaenoate
Attenuates IL--Induced Changes in Dopamine and Metabolites in the Shell of the
Nucleus Accumbens: Involved with PLA2 Activity and Corticosterone Secretion.
Neuropsychopharmacol. 2007 323 32, 736744 (2006).
49. Wilson, C. S. & Mongin, A. A. Cell Volume Control in Healthy Brain and Neuropathologies.
Curr. Top. Membr. 81, 385455 (2018).
50. Mueed, Z., Mehta, D., Rai, P. K., Kamal, M. A. & Poddar, N. K. Cross-Interplay between
Osmolytes and mTOR in Alzheimer’s Disease Pathogenesis. Curr. Pharm. Des. 26, 4699
4711 (2020).
51. Street, T. O., Bolen, D. W. & Rose, G. D. A molecular mechanism for osmolyte-induced
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
protein stability. Proc. Natl. Acad. Sci. 103, 1399714002 (2006).
52. Kushwah, N., Jain, V. & Yadav, D. Osmolytes: A possible therapeutic molecule for
ameliorating the neurodegeneration caused by protein misfolding and aggregation.
Biomolecules 10, (2020).
53. Lauterborn, J. C. et al. Increased excitatory to inhibitory synaptic ratio in parietal cortex
samples from individuals with Alzheimer’s disease. Nat. Commun. 2021 121 12, 115
(2021).
54. Bi, D., Wen, L., Wu, Z. & Shen, Y. GABAergic dysfunction in excitatory and inhibitory (E/I)
imbalance drives the pathogenesis of Alzheimer’s disease. Alzheimer’s Dement. 16, 1312
1329 (2020).
55. Yoon, J. H., Grandelis, A. & Maddock, R. J. Dorsolateral Prefrontal Cortex GABA
Concentration in Humans Predicts Working Memory Load Processing Capacity. J.
Neurosci. 36, 1178811794 (2016).
56. Wishart, D. S. et al. PathBank: A comprehensive pathway database for model organisms.
Nucleic Acids Res. 48, D470D478 (2020).
57. Wang, R. & Reddy, P. H. Role of Glutamate and NMDA Receptors in Alzheimer’s Disease.
J. Alzheimer’s Dis. 57, 10411048 (2017).
58. Handley, R. R. et al. Brain urea increase is an early Huntington’s disease pathogenic event
observed in a prodromal transgenic sheep model and HD cases. Proc. Natl. Acad. Sci. U.
S. A. 114, E11293E11302 (2017).
59. Ovsepian, S. V. & O’Leary, V. B. Can Arginase Inhibitors Be the Answer to Therapeutic
Challenges in Alzheimer’s Disease? Neurotherapeutics 15, 10321035 (2018).
60. Yurdagul, A. et al. Macrophage Metabolism of Apoptotic Cell-Derived Arginine Promotes
Continual Efferocytosis and Resolution of Injury. Cell Metab. 31, 518-533.e10 (2020).
61. Boada-Romero, E., Martinez, J., Heckmann, B. L. & Green, D. R. The clearance of dead
cells by efferocytosis. Nat. Rev. Mol. Cell Biol. 21, 398414 (2020).
62. Hensley, K. & Denton, T. T. Alternative functions of the brain transsulfuration pathway
represent an underappreciated aspect of brain redox biochemistry with significant potential
for therapeutic engagement. Free Radic. Biol. Med. 78, 123134 (2015).
63. Kumar, R. Role of naturally occurring osmolytes in protein folding and stability. Arch
Biochem Biophys 491, 16 (2009).
64. Mandal, P. K., Kansara, K. & Dabas, A. The GABAWorking Memory Relationship in
Alzheimer’s Disease. J. Alzheimer’s Dis. Reports 1, 43 (2017).
65. Jack, C. R. et al. Tracking pathophysiological processes in Alzheimer’s disease: An
updated hypothetical model of dynamic biomarkers. Lancet Neurol. 12, 207216 (2013).
66. 2021 ALZHEIMER’S DISEASE FACTS AND FIGURES. Alzheimer’s Association.
67. Holmes, M. V., Ala-Korpela, M. & Smith, G. D. Mendelian randomization in cardiometabolic
disease: Challenges in evaluating causality. Nat. Rev. Cardiol. 14, 577599 (2017).
68. Hampel, H. et al. Precision pharmacology for Alzheimer’s disease. Pharmacol. Res. 130,
331365 (2018).
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
69. Hampel, H. et al. Omics sciences for systems biology in Alzheimer’s disease: State-of-the-
art of the evidence. Ageing Res. Rev. 69, (2021).
70. Olivier, M., Asmis, R., Hawkins, G. A., Howard, T. D. & Cox, L. A. The need for multi-omics
biomarker signatures in precision medicine. Int. J. Mol. Sci. 20, (2019).
71. Mostafavi, S. et al. A molecular network of the aging human brain provides insights into the
pathology and cognitive decline of Alzheimer’s disease. Nat. Neurosci. 21, 811819
(2018).
72. Allen, M. et al. Human whole genome genotype and transcriptome data for Alzheimer’s
and other neurodegenerative diseases. Sci. Data 3, (2016).
73. Allen, M. et al. Conserved brain myelination networks are altered in Alzheimer’s and other
neurodegenerative diseases. Alzheimer’s Dement. 14, 352366 (2018).
74. Dieterle, F., Ross, A., Schlotterbeck, G. & Senn, H. Probabilistic quotient normalization as
robust method to account for dilution of complex biological mixtures. Application in1H NMR
metabonomics. Anal. Chem. 78, 42814290 (2006).
75. Breunig, M. M., Kriegel, H.-P., Ng, R. T. & Sander, J. LOF. Proc. 2000 ACM SIGMOD Int.
Conf. Manag. data - SIGMOD ’00 93104 (2000) doi:10.1145/342009.335388.
76. Johnson, E. C. B. et al. Large-scale proteomic analysis of Alzheimer’s disease brain and
cerebrospinal fluid reveals early changes in energy metabolism associated with microglia
and astrocyte activation. Nat. Med. 2020 265 26, 769780 (2020).
77. Chetnik, K. et al. maplet: An extensible R toolbox for modular and reproducible
metabolomics pipelines. Bioinformatics 0–0 (2021)
doi:10.1093/BIOINFORMATICS/BTAB741.
78. Benjamini, Y. & Hochberg, Y. Controlling the False Discovery Rate: A Practical and
Powerful Approach to Multiple Testing. J. R. Stat. Soc. Ser. B 57, 289300 (1995).
79. Schäfer, J. & Strimmer, K. A shrinkage approach to large-scale covariance matrix
estimation and implications for functional genomics. Stat. Appl. Genet. Mol. Biol. 4, 130
(2005).
.CC-BY-NC-ND 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted November 19, 2021. ; https://doi.org/10.1101/2021.11.15.468698doi: bioRxiv preprint
Article
Full-text available
Across the world a dementia case is diagnosed every three seconds. Alzheimer’s disease (AD) causes 50–60% of these cases. The most prominent theory for AD correlates the deposition of amyloid beta (Aβ) with the onset of dementia. Whether Aβ is causative remains unclear due to findings such as the recently approved drug Aducanumab showing effective clearance of Aβ, but not improving cognition. New approaches for understanding Aβ function, are therefore necessary. Here we discuss the application of optogenetic techniques to gain insight into AD. Optogenetics, or genetically encoded, light-dependent on/off switches, provides precise spatiotemporal control to regulate cellular dynamics. This precise control over protein expression and oligomerization or aggregation could provide a better understanding of the etiology of AD.
Article
Full-text available
: This paper presents maplet, an open-source R package for the creation of highly customizable, fully reproducible statistical pipelines for metabolomics data analysis. It builds on the SummarizedExperiment data structure to create a centralized pipeline framework for storing data, analysis steps, results, and visualizations. maplet's key design feature is its modularity, which offers several advantages, such as ensuring code quality through the maintenance of individual functions and promoting collaborative development by removing technical barriers to code contribution. With over 90 functions, the package includes a wide range of functionalities, covering many widely used statistical approaches and data visualization techniques. Availability: The maplet package is implemented in R and freely available at https://github.com/krumsieklab/maplet.
Article
Full-text available
Cognitive decline is associated with both normal aging and early pathologies leading to dementia. Here we used quantitative profiling of metabolites involved in the regulation of inflammation, vascular function, neuronal function and energy metabolism, including oxylipins, endocannabinoids, bile acids, and steroid hormones to identify metabolic biomarkers of mild cognitive impairment (MCI). Serum samples (n = 212) were obtained from subjects with or without MCI opportunistically collected with incomplete fasting state information. To maximize power and stratify the analysis of metabolite associations with MCI by the fasting state, we developed an algorithm to predict subject fasting state when unknown (n = 73). In non-fasted subjects, linoleic acid and palmitoleoyl ethanolamide levels were positively associated with perceptual speed. In fasted subjects, soluble epoxide hydrolase activity and tauro-alpha-muricholic acid levels were negatively associated with perceptual speed. Other cognitive domains showed associations with bile acid metabolism, but only in the non-fasted state. Importantly, this study shows unique associations between serum metabolites and cognitive function in the fasted and non-fasted states and provides a fasting state prediction algorithm based on measurable metabolites.
Article
Full-text available
In recent years, enzymes have risen as promising therapeutic tools for different pathologies, from metabolic deficiencies, such as fibrosis conditions, ocular pathologies or joint problems, to cancer or cardiovascular diseases. Treatments based on the catalytic activity of enzymes are able to convert a wide range of target molecules to restore the correct physiological metabolism. These treatments present several advantages compared to established therapeutic approaches thanks to their affinity and specificity properties. However, enzymes present some challenges, such as short in vivo half-life, lack of targeted action and, in particular, patient immune system reaction against the enzyme. For this reason, it is important to monitor serum immune response during treatment. This can be achieved by conventional techniques (ELISA) but also by new promising tools such as microarrays. These assays have gained popularity due to their high-throughput analysis capacity, their simplicity, and their potential to monitor the immune response of patients during enzyme therapies. In this growing field, research is still ongoing to solve current health problems such as COVID-19. Currently, promising therapeutic alternatives using the angiotensin-converting enzyme 2 (ACE2) are being studied to treat COVID-19.
Article
Full-text available
Synaptic disturbances in excitatory to inhibitory (E/I) balance in forebrain circuits are thought to contribute to the progression of Alzheimer’s disease (AD) and dementia, although direct evidence for such imbalance in humans is lacking. We assessed anatomical and electrophysiological synaptic E/I ratios in post-mortem parietal cortex samples from middle-aged individuals with AD (early-onset) or Down syndrome (DS) by fluorescence deconvolution tomography and microtransplantation of synaptic membranes. Both approaches revealed significantly elevated E/I ratios for AD, but not DS, versus controls. Gene expression studies in an independent AD cohort also demonstrated elevated E/I ratios in individuals with AD as compared to controls. These findings provide evidence of a marked pro-excitatory perturbation of synaptic E/I balance in AD parietal cortex, a region within the default mode network that is overly active in the disorder, and support the hypothesis that E/I imbalances disrupt cognition-related shifts in cortical activity which contribute to the intellectual decline in AD.
Article
Full-text available
Alzheimer’s disease (AD) is a devastating neurodegenerative disorder characterized by extracellular amyloid β (Aβ) and intraneuronal tau protein aggregations. One risk factor for developing AD is the APOE gene coding for the apolipoprotein E protein (apoE). Humans have three versions of APOE gene: ε2, ε3, and ε4 allele. Carrying the ε4 allele is an AD risk factor while carrying the ε2 allele is protective. ApoE is a component of lipoprotein particles in the plasma at the periphery, as well as in the cerebrospinal fluid (CSF) and in the interstitial fluid (ISF) of brain parenchyma in the central nervous system (CNS). ApoE is a major lipid transporter that plays a pivotal role in the development, maintenance, and repair of the CNS, and that regulates multiple important signaling pathways. This review will focus on the critical role of apoE in AD pathogenesis and some of the currently apoE-based therapeutics developed in the treatment of AD.
Article
Full-text available
The pathogenesis of Alzheimer’s disease (AD), a slowly-developing age-related neurodegenerative disorder, is a result of the action of multiple factors including deregulation of Ca2+ homeostasis, mitochondrial dysfunction, and dysproteostasis. Interaction of these factors in astrocytes, principal homeostatic cells in the central nervous system, is still poorly understood. Here we report that in immortalized hippocampal astrocytes from 3xTg-AD mice (3Tg-iAstro cells) bioenergetics is impaired, including reduced glycolysis and mitochondrial oxygen consumption, and increased production of reactive oxygen species. Shotgun proteomics analysis of mitochondria-ER-enriched fraction showed no alterations in the expression of mitochondrial and OxPhos proteins, while those related to the ER functions and protein synthesis were deregulated. Using ER- and mitochondria-targeted aequorin-based Ca2+ probe we show that, in 3Tg-iAstro cells, ER was overloaded with Ca2+ while Ca2+ uptake by mitochondria upon ATP stimulation was reduced. This was accompanied by the increase in short distance (≈8–10 nm) contact area between mitochondria and ER, upregulation of ER-stress/unfolded protein response genes Atf4, Atf6 and Herp, and reduction of global protein synthesis rate. We suggest that familial AD mutations in 3Tg-iAstro cells induce mitochondria-ER interaction changes that deregulate astrocytic bioenergetics, Ca2+ homeostasis and proteostasis. These factors may interact, creating a pathogenic loop compromising homeostatic and defensive functions of astroglial cells predisposing neurons to dysfunction.
Article
Full-text available
Objective: To propose a new hypothesis that GABAergic dysfunction in excitatory and inhibitory (E/I) imbalance drives the pathogenesis of Alzheimer's disease (AD). Background: Synaptic dysfunction and E/I imbalance emerge decades before the appearance of cognitive decline in AD patients, which contribute to neurodegeneration. Initially, E/I imbalance was thought to occur first, due to dysfunction of the glutamatergic and cholinergic systems. However, new evidence has demonstrated that the GABAergic system, the counterpart of E/I balance and the major inhibitory neurotransmitter system in the central nervous system, is altered enormously and that this contributes to E/I imbalance and further AD pathogenesis. New hypothesis: Alterations to the GABAergic system, induced by multiple AD pathogenic or risk factors, contribute to E/I imbalance and AD pathogenesis. Major challenges for the hypothesis: This GABAergic hypothesis accounts for many critical questions and common challenges confronting a new hypothesis of AD pathogenesis. More specifically, it explains why amyloid beta (Aβ), β-secretase (BACE1), apolipoprotein E4 gene (APOE ε4), hyperactive glia cells, contributes to AD pathogenesis and why age and sex are the risk factors of AD. GABAergic dysfunction promotes the spread of Aβ pathology throughout the AD brain and associated cognitive impairments, and the induction of dysfunction induced by these varied risk factors shares this common neurobiology leading to E/I imbalance. In turn, some of these factors exacerbate GABAergic dysfunction and E/I imbalance. Moreover, the GABAergic system modulates various brain functions and thus, the GABAergic hypothesis accounts for nonamnestic manifestations. Furthermore, corrections of E/I balance through manipulation of GABAergic functions have shown positive outcomes in preclinical and clinical studies, suggesting the potential of the GABAergic system as a therapeutic target in AD. Linkage to other major theories: Dysfunction of the GABAergic system is induced by multiple critical signaling pathways, which include the existing major theories of AD pathogenesis, such as the Aβ and neuroinflammation hypotheses. In a new perspective, this GABAergic hypothesis accounts for the E/I imbalance and related excitotoxicity, which contribute to cognitive decline and AD pathogenesis. Therefore, the GABAergic system could be a key target to restore, at least partially, the E/I balance and cognitive function in AD patients.
Article
Full-text available
Abstract Alzheimer’s disease (AD) is one of the most prevalent neurodegenerative diseases, characterized by impaired cognitive function due to progressive loss of neurons in the brain. Under the microscope, neuronal accumulation of abnormal tau proteins and amyloid plaques are two pathological hallmarks in affected brain regions. Although the detailed mechanism of the pathogenesis of AD is still elusive, a large body of evidence suggests that damaged mitochondria likely play fundamental roles in the pathogenesis of AD. It is believed that a healthy pool of mitochondria not only supports neuronal activity by providing enough energy supply and other related mitochondrial functions to neurons, but also guards neurons by minimizing mitochondrial related oxidative damage. In this regard, exploration of the multitude of mitochondrial mechanisms altered in the pathogenesis of AD constitutes novel promising therapeutic targets for the disease. In this review, we will summarize recent progress that underscores the essential role of mitochondria dysfunction in the pathogenesis of AD and discuss mechanisms underlying mitochondrial dysfunction with a focus on the loss of mitochondrial structural and functional integrity in AD including mitochondrial biogenesis and dynamics, axonal transport, ER-mitochondria interaction, mitophagy and mitochondrial proteostasis.
Article
Alzheimer’s disease (AD) is characterized by non-linear, genetic-drivenpathophysiological dynamics with high heterogeneity in biological alterations and disease spatial-temporal progression. Human in-vivo and post-mortem studies point out a failure of multi-level biological networks underlying AD pathophysiology, including proteostasis (amyloid-β and tau), synaptic homeostasis, inflammatory and immune responses, lipid and energy metabolism, oxidative stress. Therefore, a holistic, systems-level approach is needed to fully capture AD multi-faceted pathophysiology. Omics sciences – genomics, epigenomics, transcriptomics, proteomics, metabolomics, lipidomics – embedded in the systems biology (SB) theoretical and computational framework can generate explainable readouts describing the entire biological continuum of a disease. Such path in Neurology is encouraged by the promising results of omics sciences and SB approaches in Oncology, where stage-driven pathway-based therapies have been developed in line with the precision medicine paradigm. Multi-omics data integrated in SB network approaches will help detect and chart AD upstream pathomechanistic alterations and downstream molecular effects occurring in preclinical stages. Finally, integrating omics and neuroimaging data – i.e., neuroimaging-omics – will identify multi-dimensional biological signatures essential to track the clinical-biological trajectories, at the subpopulation or even individual level.
Article
Alzheimer disease (AD) is the most common form of neurodegenerative disease, estimated to contribute 60–70% of all cases of dementia worldwide. According to the prevailing amyloid cascade hypothesis, amyloid-β (Aβ) deposition in the brain is the initiating event in AD, although evidence is accumulating that this hypothesis is insufficient to explain many aspects of AD pathogenesis. The discovery of increased levels of inflammatory markers in patients with AD and the identification of AD risk genes associated with innate immune functions suggest that neuroinflammation has a prominent role in the pathogenesis of AD. In this Review, we discuss the interrelationships between neuroinflammation and amyloid and tau pathologies as well as the effect of neuroinflammation on the disease trajectory in AD. We specifically focus on microglia as major players in neuroinflammation and discuss the spatial and temporal variations in microglial phenotypes that are observed under different conditions. We also consider how these cells could be modulated as a therapeutic strategy for AD.