Metabolomics is a rapidly emerging “omics” that establishes disease-specific signatures of perturbations in hundreds of metabolites, reflecting alterations in multiple networks affected in the disease. Application of metabolomics for the diagnosis of AD is very attractive. Profiling in plasma and CSF could be used to establish metabolic signatures for the purposes of an accurate diagnosis, including the early clinical (e.g., MCI) or preclinical stages, or within individuals to monitor disease progression and therapeutic efficacy. Metabolomics can be conducted in biofluids, such as plasma or CSF, with high throughput and relatively low cost. Recently, metabolomic profiling was used to assess cross-sectional alterations in either CSF or plasma samples from individuals with different clinical severity of AD 
. However, these studies used diverse metabolomics platforms, biofluids (CSF or plasma), and differed in the range of identified metabolites, thus limiting efforts of comparing the results. In the present study, we applied a non-targeted mass spectrometry–based metabolomic profiling to determine global changes in metabolites and various putative metabolic pathways in CSF and plasma from the same individuals in relationship to AD progression. To our knowledge, this is the first study that (1) evaluates progressive metabolic changes in CN, MCI and AD subjects in both CSF and plasma, (2) establishes to what extent metabolic changes in lumbar CSF are reflected in plasma, (3) identifies unique and common metabolic pathways specifically affected by AD severity in plasma and CSF, and (4) validates plasma as a reliable biofluid for metabolic studies of brain-related disorders. We found that approximately 30% of the metabolic pathways altered in the CSF in MCI patients vs. CN, and ~60% in AD patients vs. CN, were also affected in plasma from the same individuals (, ). The number of affected pathways in CSF and plasma increased with disease severity. Thus, in AD patients the total of affected pathways increased by 50% in CSF and doubled in plasma compared to MCI patients (, ). However, in MCI and AD individuals the number of affected pathways was always greater in plasma, thus reflecting changes in organs other than brain that are associated with AD (–). Our data demonstrate that signatures in CSF and plasma have significant overlap, and most of the pathways affected early in MCI continue to be altered in AD subjects. In both CSF and plasma in MCI and AD groups the perturbed canonical pathways included those related to energy metabolism and mitochondrial function; lipid biosynthesis, trafficking and metabolism; amino acid biosynthesis and metabolism; neurotransmitter biosynthesis and metabolism; and hormone biosynthesis and metabolism (–). However, we also identified pathways that were specifically affected in either plasma or CSF (–).
One of the prominent dysfunctions in AD is a progressive failure of neuronal networks and neurotransmitter systems. Results of the extensive studies in multiple animal and cellular models of AD suggest that synaptic malfunction and synaptic loss occur prior to the development of Aβ plaques and neurofibrillary tangles 
. These synaptic alterations are directly associated with deteriorated synaptic strength and synaptic plasticity, including long-term potentiation 
. Acetylcholine, noradrenalin, dopamine and serotonin neurotransmitter systems are primarily affected in AD with subsequent loss of associated neurons 
. Consistent with that, we have found prominent, early changes in tryptophan biosynthesis in both CSF and plasma of MCI and AD patients (–). Tryptophan is a precursor for serotonin, melatonin, and niacin synthesis 
. Therefore, not surprisingly, we found alterations in the serotonin/melatonin pathway in CSF of both MCI and AD patients, and in plasma of MCI individuals (, ). These data are in agreement with recent studies indicating that loss of serotonergic neurons correlates with AD severity, memory impairment, and neuropsychiatric symptoms 
, and that melatonin protects against Aβ toxicity in cellular and animals models of AD 
. Additional significant changes in neurotransmitter metabolism were observed in the acetylcholine pathway in CSF from AD individuals and in gamma amino butyric acid (GABA) pathway in plasma from MCI patients (, ). This is in agreement with the previously reported changes detected in post-mortem CSF samples from patients with confirmed AD using targeted metabolomics 
. We also found that in addition to tryptophan, multiple amino acid metabolic pathways were progressively affected in CSF and plasma in AD and MCI patients compared to CN (, ). While only L-arginine and tryptophan pathways were altered in both plasma and CSF of MCI patients (), the number of pathways equally affected in CSF and plasma of AD patients considerably increased and included beta-alanine, aspartate and aspargine, alanine, L-cysteine, L-methionine, methionine-cysteine-glutamate along with L-arginine and lysine metabolism (). Our findings are in a good agreement with alterations in amino acids measured in CSF of AD patients using variety of methods 
. It is important to note that pattern of alterations in amino acid pathways in plasma was very similar to the observed in CSF in both MCI and AD cohorts (, ). However, significant alterations in the lysine pathway were detected only in the plasma in MCI and AD individuals (, ). It is known that lysine being a strictly ketogenic amino acid, is also required for the synthesis of L-carnitine. L-Carnitine is the only transporter of fatty acids to mitochondria to be metabolized with production of energy. Indeed, previous study demonstrated that levels of carnitine were lower in CSF from MIC-AD and AD patients than in CSF from non-AD subjects 
. In our study, perturbations of the lysine metabolic pathway most accurately differentiated CN from the MCI and AD groups in plasma ().
Previous metabolomics studies detected changes in the levels of the neurotransmitter norepinephrine (NE) and its major metabolite 3-methoxy-4-hydroxy phenylglycol (MHPG) involved in noradrenalin
neurotransmitter system 
. Loss of NE would be expected as a result of neuronal loss in locus coeruleus established in AD patients 
. Interestingly, Czech et al reported increased levels of NE in CSF of AD subjects suggesting a compensatory mechanism where surviving neurons have higher secretion of NE 
. However, another study reported the opposite – a decrease in NE levels and an increase in MHPG in the CSF of AD subjects 
. We were able to identify both metabolites in the CSF of CN, MCI and AD subjects; however, levels of these metabolites were not significantly changed between our study groups. It will be interesting to determine whether the discrepancy between the results could partially be explained by the impact of the post mortem changes or the origin of the CSF samples – ventricular post-mortem CSF 
vs. lumbar, as in our study and 
, or the different metabolomics platforms utilized in all of the above-mentioned studies.
Two additional pathways that were significantly and specifically affected in plasma by AD severity included polyamine metabolism and aminoacyl-tRNA biosynthesis in the cytoplasm (). Alterations in the levels of polyamines found in the brain tissue of AD patients have been linked to the abnormal regulation of calcium flux, glutamate receptor function, and excitotoxicity 
. The major metabolites affected in that pathway in plasma of MCI and AD individuals included, but were not limited to, arginine, tryptophan, proline, lysine, glutamine, GABA, and urea. Aminoacyl-tRNA synthetases (AARS) and translation factors are key enzymes required for protein biosynthesis. The exact mechanisms associated with altered AARS pathway in AD remain unknown. However, recently AARS was linked to the biosynthesis of the dinucleotide polyphosphates 
, which play an important role as neurotransmitters, stimulate GABA release in peripheral and central nervous system, and are involved in the response to oxidative stress and metabolic changes 
A number of studies have proposed the role for mitochondrial dysfunction in the early pathogenesis of AD 
. We have previously demonstrated the presence of metabolic signatures of energetic stress and mitochondrial dysfunction in brain tissue from three transgenic mouse models of familial AD (FAD) 
. Consistent with our previous findings, the current study also identified multiple pathways related to energy metabolism and mitochondrial function that were already significantly affected in MCI cohort. The TCA (Krebs) cycle was most affected in both the CSF and plasma of MCI and AD patients (, ). We also found significant alterations in saturated fatty acid metabolism in the plasma of MCI patients and in the CSF of AD patients, and fatty acid omega oxidation in the plasma of AD patients (–). Additional pathways related to altered brain energetics included pyruvate, mitochondrial ketone bodies, glycolysis and gluconeogenesis, and were affected in CSF and/or plasma from MCI and AD patients supporting the role for mitochondrial dysfunction in early AD. Among additional pathways correlated with AD progression was the urea cycle 
. We have found that alterations in urea cycle were detected only in the CSF of MCI subjects but in both, CSF and plasma of AD patients (–). Changes in the urea cycle correlate well with previously reported changes in body fluids of AD patients 
and could be used to discriminate between those MCI patients who do and do not progress to AD ().
An additional strength of our study is the application of UPLC-ToF-MS – based metabolomics, which allows for the detection of relative changes in multiple metabolites associated with AD clinical severity, including lipids 
. Utilization of MS in conjunction with gas chromatography (GC) and high performance liquid chromatography (HPLC) systems has recently become very popular. However, application of UPLC-MS has an advantage in that it does not require deconjugation and derivatization steps before analysis 
. Moreover, samples are processed at low temperatures relatively to the temperatures used with GC-MS, which allows for the detection of labile sterols that at high temperatures could be unstable 
. Previous studies identified perturbations in the levels of phosphatidylcholine, plasmalogens, sphingomyelins and sterols in plasma of subjects with AD 
. A decrease in desmosterol and the desmosterol/cholesterol plasma ratio measured using metabolomics technology was also proposed as a sensitive marker for AD 
. However, it is well known that AD patients lose weight, which results in lower lipid levels. Therefore, it is important to determine the specificity of lipid subclasses affected with AD severity rather than an overall lipid flux. In agreement with the previous studies, we have found that pathways related to cholesterol biosynthesis and metabolism, cholesterol and sphingolipids transport, lipid metabolism and other phospholipid and plasmalogen pathways were significantly altered in both CSF and plasma from MCI and AD patients (, ). The number of altered pathways related to lipid biosynthesis and metabolism was progressively increased in AD patients relative to MCI (). Discrimination between MCI and AD patients demonstrates that metabolic signature of altered cholesterol metabolism was prevalent in plasma samples and of altered phosphatidylinositol metabolism in CSF (). Moreover, our study identified CSF PGE2 biosynthesis and metabolism as one of the key pathways that varied with AD severity (). Implication of PGE2 in neural injury in AD is well documented, and includes modulation of protein-lipid interactions, trans-membrane and trans-synaptic signaling 
. It was shown that levels of PGE2 measured in the CSF of control, MCI and AD patients enrolled in the longitudinal study inversely correlate with AD severity: PGE2 was higher in patients with mild memory impairment, but lower in those with more advanced AD 
Disorder in the hypothalamic-pituitary-adrenal (HPA) axis with increased cortisol levels in CSF and plasma is also well established for AD patients; and increased cortisol levels in CSF from AD patients have been recently demonstrated using metabolomic profiling 
. Our data confirmed that the pathway related to the cortisol biosynthesis from cholesterol was significantly affected in both CSF and plasma from AD patients (). However, we also found that cortisone biosynthesis and metabolism was among the pathways that, along with PGE2, most accurately separated the clinical groups in CSF (). Among pathways that were uniquely affected in plasma of AD patients were those related to obesity and type II diabetes mellitus (). This is an important observation taking in consideration the data demonstrating that type II diabetes mellitus is associated with an increased risk of cognitive dysfunction and dementia, and needs to be explored in future studies 
Together, the present results, utilizing comprehensive metabolic profiling, in AD and MCI subjects confirmed previously reported observations and identified novel metabolic signatures in CSF and plasma that vary with the clinical severity of AD. Interestingly, the fact that patients were on multiple medications did not impact our ability to obtain data with significant differences between groups (Table S10
). This could partially be explained by the overlap of the medication among all three groups (Table S10
, bold font depicts same medication) or the fact that medication does not significantly affect the disease progression. Further, current metabolomics approaches, in addition to measuring metabolites originated from endogenous cellular metabolism, also detect those derived exogenously from drugs, food, and cosmetics. However, we were still able to observe robust group separation supporting high sensitivity of the approach. A strength of the study is in utilizing UPLC-ToF-MS to detect changes in broad variety of metabolites that reflect the complexity of metabolic networks altered in AD. The accuracy of our findings was also enhanced by the precise and consistent selection of participants in order to closely match experimental groups on demographic factors. However, limitations of the study also warrant consideration. One limitation was the small sample size of 15 participants per clinical group. While we still achieved robust group separation, additional studies are necessary to validate our findings in larger cohorts. It will also be important to examine the effect of sex, as the participants included in this analysis were primarily men. Lastly, future studies will need to assess the specific changes in identified pathways to shed light on disease mechanisms along with assaying the longitudinal changes in the pathways and metabolites as indicators of disease progression, especially at the early pre-clinical stages.
Taken together, our studies demonstrate that overlapping alterations in several known and unknown metabolites and various putative metabolic pathways could be detected using non-targeted mass spectrometry–based comprehensive metabolic profiling in CSF and plasma in MCI and AD individuals in respect to AD severity. The agreement of our data with previously reported changes in metabolites and metabolic pathways associated with AD or MCI and identified using multiple analytical approaches offers further support for metabolomics analysis of plasma and CSF samples for AD diagnosis. Our results suggest that additional studies with targeted metabolomics could identify specific panels of metabolites. Furthermore, the significant similarity of affected pathways based on changes in plasma and CSF metabolites and canonical pathways supports the notion that plasma closely depicts biochemical fingerprints of brain changes in AD and MCI individuals. Our data validate plasma as reliable source for metabolomic profiling and suggest that metabolomics is a valuable tool for the identification of molecular mechanisms involved in the etiology of AD and novel therapeutic targets.