Systemic and central nervous system metabolic alterations in Alzheimer’s disease
Metabolic alterations, related to cerebral glucose metabolism, brain insulin resistance, and age-induced mitochondrial dysfunction, play an important role in Alzheimer’s disease (AD) on both the systemic and central nervous system level. To study the extent and significance of these alterations in AD, quantitative metabolomics was applied to plasma and cerebrospinal fluid (CSF) from clinically well-characterized AD patients and cognitively healthy control subjects. The observed metabolic alterations were associated with core pathological processes of AD to investigate their relation with amyloid pathology and tau-related neurodegeneration.
In a case-control study of clinical and biomarker-confirmed AD patients (n = 40) and cognitively healthy controls without cerebral AD pathology (n = 34) with paired plasma and CSF samples, we performed metabolic profiling, i.e., untargeted metabolomics and targeted quantification. Targeted quantification focused on identified deregulated pathways highlighted in the untargeted assay, i.e. the TCA cycle, and its anaplerotic pathways, as well as the neuroactive tryptophan and kynurenine pathway.
Concentrations of several TCA cycle and beta-oxidation intermediates were higher in plasma of AD patients, whilst amino acid concentrations were significantly lower. Similar alterations in these energy metabolism intermediates were observed in CSF, together with higher concentrations of creatinine, which were strongly correlated with blood-brain barrier permeability. Alterations of several amino acids were associated with CSF Amyloidβ1–42. The tryptophan catabolites, kynurenic acid and quinolinic acid, showed significantly higher concentrations in CSF of AD patients, which, together with other tryptophan pathway intermediates, were correlated with either CSF Amyloidβ1–42, or tau and phosphorylated Tau-181.
This study revealed AD-associated systemic dysregulation of nutrient sensing and oxidation and CNS-specific alterations in the neuroactive tryptophan pathway and (phospho)creatine degradation. The specific association of amino acids and tryptophan catabolites with AD CSF biomarkers suggests a close relationship with core AD pathology.
Our findings warrant validation in independent, larger cohort studies as well as further investigation of factors such as gender and APOE genotype, as well as of other groups, such as preclinical AD, to identify metabolic alterations as potential intervention targets.
KeywordsAlzheimer’s disease Metabolomics Energy metabolism Tryptophan pathway CSF AD biomarkers
Central nervous system
CSF/serum albumin ratio
In Alzheimer’s disease (AD), glucose hypometabolism is considered a typical feature of the disease at clinical stages, indicating the loss of neuronal function in specific brain regions . Cerebral glucose hypometabolism, characterized by impaired glucose uptake and utilization related to brain insulin resistance [2, 3], and progressive mitochondrial dysfunction with aging  have both been recently associated with AD and suggest involvement of energy metabolism alterations in AD pathophysiology. Importantly, these alterations in early AD may occur both at the central nervous system (CNS) and the systemic level and play a role in clinical disease progression [5, 6]. Despite these observations, the extent and significance of CNS and systemic metabolic alterations in AD remain poorly understood. Therefore, further and in-depth characterization of metabolic alterations to unravel potential new targets for therapeutic intervention is needed. Metabolomics is a powerful phenotyping technology, which allows to systematically identify and quantify the active small molecule-metabolite complement of cells, tissues, or biofluids and provide a sensitive and highly specific multiparametric measure of disease phenotype at the molecular level [7, 8, 9, 10, 11, 12, 13, 14].
A few recent metabolomics data-driven studies in subjects with clinically defined AD support the view of AD as an energy metabolism and metabolic signaling disorder by describing metabolic alterations in amino acid, acylcarnitine, sphingolipid, and lipid metabolism [6, 15, 16, 17, 18, 19]. While many alterations were observed, the results of these studies are inconsistent regarding the amplitude and direction of the alterations and limited by the lack of quantitative data and shortcomings in study design. Importantly, the link between systemic and CNS changes and their relationships with the AD core brain pathology, i.e., amyloid accumulation and tau pathology, remain largely unexplored.
Here, we present a case-control study using paired plasma and cerebrospinal fluid (CSF) samples from AD patients and cognitively healthy controls characterized using combined clinical and biomarker-based criteria  to which we applied state-of-the-art metabolomic approaches. We used a thorough stepwise approach from untargeted profiling and pathway analysis to targeted absolute quantification to determine the presence and magnitude of metabolic alterations and metabolite concentration ranges. By linking systemic and CNS metabolic alterations and by addressing relationships between metabolites and markers of core AD pathology, i.e., amyloid accumulation and tau-related neurodegeneration, this study contributes to the functional understanding of AD pathophysiology.
Participants with AD were recruited among patients with cognitive impairment attending the Memory Clinics of the Department of Psychiatry and the Department of Clinical Neurosciences at Lausanne University Hospital for the diagnosis of their complaints . Control subjects were recruited by announcements and word of mouth. All participants underwent a comprehensive medical, neuropsychological, and psychosocial evaluation, as well as brain MRI or CT scans, and venous and lumbar punctures. The MRI and CT scans were used to exclude cerebral pathologies possibly interfering with the cognitive performance.
Clinical characteristics of the cohort
AD (n = 40)
Control (n = 34)
Female, n (%)
BMI, kg/m2, mean ± SD
23.83 ± 3.06
24.60 ± 4.01
Age, year, mean ± SD
74.88 ± 6.38
65.35 ± 6.17
MMSE, mean ± SD
24.40 ± 4.15
28.71 ± 1.29
CDR, mean ± SD
0.65 ± 0.30
0.00 ± 0.00
AD CSF biomarkers
Aβ1–42, pg/ml, mean ± SD
556.22 ± 115.39
979.12 ± 164.38
Tau, pg/ml, mean ± SD
715.70 ± 300.05
196.18 ± 59.72
pTau-181, pg/ml, mean ± SD
91.95 ± 23.84
42.91 ± 10.51
ApoEε4, n (%ε4)
Qalb, mean ± SD
6.69 ± 3.76
5.27 ± 1.82
AD diagnosis and cognitive assessments
The diagnosis of MCI or mild dementia of AD type was based on neuropsychological and clinical evaluation made by a consensus conference of neuropsychologists, psychiatrists, and/or neurologists prior to inclusion into the study, as described elsewhere  and detailed in Additional file 2: methods (section 1.1) together with the performed cognitive assessments.
Sample collection, APOE genotyping, and CSF AD biomarker assessment
CSF and plasma samples were obtained as previously described , and subsequently, CSF AD biomarkers Aβ1–42, tau, and pTau-181 were measured using ELISA (Fujirebio, Ghent, Belgium). The APOE genotype was determined as previously described . Brief details of both procedures are outlined in the Additional file 2: methods (section 1.2 and 1.3).
State-of-the-art untargeted and targeted metabolic profiling
Materials and detailed methods are outlined in Additional file 2: methods (section 1.4).
Following the extraction with MeOH:ACN, plasma and CSF sample extracts were subjected to LC-MS analysis using the 6550 iFunnel Q-TOF MS interfaced with 1290 UHPLC (Agilent Technologies, Basel, CH) as previously described . The data were processed using XCMS Online  and signal drift correction was applied and metabolite features showing analytical variability > 30% were removed. Putative identification was done in XCMS Online linked to METLIN metabolite database , and metabolite identities were further validated with tandem MS experiments as previously described [23, 26].
Broad-scale targeted profiling
In parallel with untargeted profiling, broad-scale targeted screening was performed with a focus on intermediates involved in multiple central carbon pathways (242 metabolites) using a 6495 iFunnel triple quadrupole system (QqQ, Agilent Technologies, Basel, CH) interfaced with the 1290 UHPLC system. Data was acquired in dynamic multiple reaction monitoring mode (dMRM, cycle time 600 ms). Data processing was done using MassHunter Quantitative Analysis (for QqQ, version B.07.01/ Build 7.1.524.0, Agilent Technologies). Signal drift correction was applied on the QC samples , and metabolites with CV > 20% were discarded.
Pathway analyses were performed using MetaboAnalyst 3.0 , and the human pathways from the Homo sapiens Kyoto Encyclopedia of Genes and Genomes (KEGG) database were used as the source of pathway topologies to deduce pathways of interest for the absolute quantification method. Pathway impact has been calculated as the sum of the importance measures (i.e., centrality measure within a given metabolic network) of the matched metabolites normalized by the sum of the importance measures of all metabolites in each pathway  (for further details, see Additional file 2: methods, section 1.4.5).
Targeted quantification of tricarboxylic acid (TCA) cycle intermediates, tryptophan breakdown products, and other amino acids and acylcarnitines
Absolute quantification was performed using the 6495 QqQ mass spectrometer interfaced with the 1290 UHPLC, operated in the dMRM mode. In brief, aliquots of calibrators, plasma, or CSF were extracted by the addition of internal standard mixtures (in MeOH) after which the sample was directly injected for LC-MS/MS analysis (transitions are provided in Additional file 1: Table S1). Stable isotope-labeled analogues were used as internal standards to determine the response factor while correcting for extraction yield and matrix effect. Data processing was done using MassHunter Quantitative Analysis.
Group comparison was performed with the absolute concentration data, which was done using a parametric t-test with a p value significance cut-off 0.05 (FDR < 0.25). Additional testing was performed to assess (1) gender differences, (2) the CSF/plasma ratio of metabolite concentrations, (3) the influence of blood-brain barrier (BBB) permeability, and (4) the potential confounding effect of age and ApoE4, for which p < 0.05 was considered significant. Associations between metabolite concentrations and single CSF AD biomarker concentrations (t-tau, pTau-181, and Aβ1–42) were evaluated using simple and multiple linear regression analysis for the AD group only. Details of these analyses are outlined in Additional file 2: methods (section 1.5).
Clinical characteristics of controls and subjects with Alzheimer’s disease
For this study, n = 40 well-characterized AD patients and n = 34 cognitively healthy controls were selected (Table 1). As about 20% of the patients with a clinical criteria-based diagnosis of AD have no cerebral AD pathology , we only included AD subjects with both clinical signs of (prodromal) AD and a CSF AD biomarker profile (pTau-181/Aβ1–42 ratio > 0.078) . Conversely, a significant percentage of elderly persons without any clinical signs of AD have cerebral AD pathology . In this study, we only included control subjects who were cognitively healthy and had normal CSF AD biomarker profiles. The AD patient and control groups did not differ in male/female distribution and BMI, but the AD patients were generally older. In addition, the two populations had a different frequency of the ApoEε4 allele, a known risk factor for AD, and a marginally significant different CSF/serum albumin ratio (Qalb), considered here as a marker of blood-brain barrier permeability .
Metabolic profiling highlights disrupted core energy metabolism and tryptophan pathway alterations in Alzheimer’s disease
Untargeted profiling pointed towards significant alterations in the amino acid metabolism and energy-producing fatty acid oxidation (i.e., acylcarnitine levels) in plasma and in CSF of AD patients (Additional file 1: Table S1). These differences were confirmed by further broad-scale targeted screening that allowed us to reveal several additional changes in the levels of glycolysis and tryptophan and kynurenine pathway intermediates (Additional file 1: Table S2). Pathway over-representation combined with topology analysis, which considers the position and biological relevance of profiled metabolites within their respective pathways, showed significantly enriched tryptophan and histidine metabolism as well as beta-oxidation pathway in plasma. In CSF, enriched tryptophan and lysine metabolism were highlighted, as well as glycolysis/gluconeogenesis, the pentose phosphate pathway, and carnitine synthesis (P < 0.05, Additional file 1: Table S3). Following these results, we quantified in an absolute fashion different intermediates in the TCA cycle as a hub of energy metabolism, and its anaplerotic pathways, i.e., fatty acid oxidation and specific amino acid pathways. The downstream products of tryptophan metabolism were also quantified due to high enrichment and impact score of tryptophan metabolism in both plasma and CSF in the pathway analysis (P < 0.002, Impact > 0.22).
Acylcarnitines, the transporter variants of fatty acid oxidation intermediates that fuel the TCA cycle by generating AcetylCoA via beta-oxidation, showed significantly higher concentrations in plasma of AD patients compared to control subjects (Fig. 2, Additional file 1: Table S5). These were medium- and long-chain acylcarnitines with an acyl-chain of C6 (%diff = 31.4%, P = 0.016), C8 (%diff = 34.8%, P = 0.048), C10 (%diff = 37.0%, P = 0.029), C12 (%diff = 36.4%, P = 0.012), C14 (%diff = 34.4%, P = 0.0003), C16 (%diff = 14.2%, P = 0.009), and C18 (%diff = 21.1%, P = 0.002). In addition, the concentration of acetylcarnitine (C2) was significantly higher (%diff = 19.2%, P = 0.025), whilst the free pool of carnitine (C0) in plasma was significantly lower (%diff = − 12.4%, P = 0.026) in AD patients. In CSF, the same trend of accumulation in AD was observed for acylcarnitines with a chain length between C6 and C12, whereas the long-chain acylcarnitines were below the limit of quantification.
It is worth noting that for the majority of measured metabolites, the observed differences were more pronounced in women than in men as illustrated in Additional file 1: Figure S1.
Correlations of altered metabolites in CSF with BBB permeability
Metabolite alterations associated with CSF biomarkers of core AD pathology
Distinct systemic and CNS pathway alterations related to AD were observed in this case-control study applying a thorough stepwise metabolomics approach in concomitantly collected plasma and CSF samples from well-characterized subjects with AD and cognitively healthy controls. Amino acids were decreased, and fatty acid-oxidation metabolites and TCA cycle intermediates were increased in plasma of AD patients compared to control subjects. In their CSF, the concentrations of tryptophan pathway metabolites and creatinine were increased. Specific alterations were related to amyloid while others were associated with tau pathology and neuronal injury as measured by CSF biomarkers.
The alternative angle of viewing AD as an energy metabolism and metabolic signaling disorder has recently evolved following advancements in analytical methods and new findings on the disease’s pathophysiology [32, 33]. While decreased neuronal glucose metabolism and associated altered bioenergetics are recognized as a common feature in AD, its extent and relationships with the “core” pathological processes of AD, i.e., amyloid pathology and tau-related neurodegeneration, necessitate further investigation [1, 34, 35, 36]. Decreased glucose sensing by the brain in AD could signal a fasted state to the body and lead to compensatory activation of alternative sources to fuel the TCA cycle, such as amino and fatty acids . In our study, global lower plasma concentrations of amino acids in AD patients compared to controls indeed suggest that readily available amino acids could have been used to replenish the TCA cycle  either by forming TCA cycle intermediates (glucogenic pathway) or by forming acetylCoA (ketogenic pathway, Fig. 2, ). In addition to the pool of free amino acids, fatty acid oxidation can fuel the TCA cycle via production of acetylCoA. Our results showed significantly higher concentrations of the carnitine forms of main fatty acid oxidation intermediates in plasma of AD patients compared to control subjects, i.e., long-chain acylcarnitines (LCACs, from C6 to C18) and acetylcarnitine (C2). This increase could be related to incomplete oxidation of acyl-CoA intermediates resulting in their retroconversion to acylcarnitine for the transport and release to the plasma, to avoid adverse toxic effects of their accumulation in mitochondria . This fuel efflux (i.e., acetylcarnitine and LCACs) is assumed to occur when the fuel delivery exceeds energy generation capacity of the TCA cycle . This is in accordance with our results showing higher concentrations of TCA cycle intermediates in plasma and CSF of AD patients. Taken together, the observed alterations in the energy metabolism hub (TCA cycle) and its anaplerotic pathways, amino acid, and fatty acid oxidation, both in plasma and CSF, imply disrupted nutrient sensing and oxidation and thus energy homeostasis in AD. These alterations appear to be of systemic origin and are reflected in CSF depending on increased BBB permeability, which is supported by the observed significant positive correlation between the CSF concentrations of these metabolites and QAlb in AD patients (Fig. 4). Moreover, several amino acids, i.e., arginine, citrulline, isoleucine, leucine, ornithine, phenylalanine, and tryptophan, were negatively associated with CSF Aβ1–42 concentrations, thus with higher cerebral amyloid burden (Fig. 5). This is in line with previous literature where inclusion of CSF amino acids of the one-carbon metabolism in a prediction-model improved diagnostic accuracy [20, 40]. In comparison, the associations of plasma levels of amino acids with the AD CSF biomarkers were weaker, except for lysine and taurine with CSF pTau-181.
Related to these energy metabolism alterations, creatinine was significantly increased in CSF of AD patients, negatively correlated with Qalb, and positively associated with CSF Aβ1–42. As a by-product of the high energy storage metabolite phosphocreatine , the observed higher concentrations of creatinine in CSF in AD may be a result of excessive phosphocreatine usage (followed by degradation) and/or disrupted creatine-phosphocreatine shuttle  in the conditions of inadequate glucose supply. The negative correlation of creatinine with BBB permeability (Fig. 4) implies that this process takes place in the CNS. The potential dysregulation of this process is further illustrated by the negative correlation between creatine and creatinine in both plasma and CSF (Spearman’s rho 0.46, p = 0.003 in plasma and − 0.33, p = 0.037 in CSF) in AD patients, suggesting that creatinine is produced at the expense of creatine.
Our results also highlighted the CNS-specific deregulation of the tryptophan-kynurenine pathway, with significantly higher concentrations of kynurenic acid and quinolinic acid in CSF of AD patients (Fig. 3). Both of these tryptophan metabolites were previously reported to be specifically associated with neuroinflammation in CNS diseases, including AD [43, 44, 45, 46, 47]. While kynurenic acid was reported as putatively neuroprotective , quinolinic acid is considered to be neurotoxic  and found to be increased in AD in model systems , although this was not consistently confirmed in humans . Our results showed that these tryptophan catabolites were also significantly associated with core AD pathology, i.e., the putatively neuroprotective kynurenic acid was associated with lower cerebral beta-amyloid burden (higher CSF Aβ1–42 levels), whilst the neurotoxic quinolinic acid was associated with increased tau hyperphosphorylation and neuronal injury. Along with our results, previous work [49, 50] suggested the tryptophan pathway to be implicated in cerebral AD pathology and might be a possible target for disease modifying interventions.
Importantly, exploratory analysis in our study indicates more significant metabolic alterations in female AD subjects. The female susceptibility to AD has been highlighted in a very recent study , although the underlying mechanisms of how sex modifies AD risk are poorly understood. Different findings suggest that the profound age-related metabolic and hormonal changes in female (i.e., estrogen loss) exacerbate the peripheral and brain insulin signaling dysfunction leading to reduced glucose metabolism [52, 53]. In our study, age was slightly unbalanced between the AD patients and controls and we observed correlations of several metabolites with age. However, correction for age did not significantly change the observed difference between AD patients and control subjects, except for the acylcarnitines C14, C16, and cis-aconitate in plasma and kynurenic acid in CSF (Additional file 1: Table S8). Furthermore, the presence of the ApoE4 allele did not influence our observations (no interaction effect) as evaluated using ANOVA, except for creatinine in plasma (Pinteraction = 0.02) and asparagine in CSF (Pinteraction = 0.005, Additional file 1: Table S9). No difference was observed in fatty acid oxidation among individuals with different APOE genotype.
In the present study, the quantitative metabolite data acquired in paired plasma and CSF samples combined with clinical diagnosis criteria, AD CSF biomarker data, and clinical metadata allowed us to identify and quantify metabolic alterations in AD and associate them with distinct AD pathologies (amyloid pathology (Aβ1–42), neuronal injury (tau), and tau hyperphosphorylation (pTau-181)), whilst deriving information on the most likely origin of these alterations (systemic or CNS). However, it is possible that the observed metabolites are derived elsewhere, such as in the gut microbiota, which is of particular importance for tryptophan catabolism that is regulated via a highly interconnected loop involving gut microbiota .
To our knowledge, only one other metabolomics-led study in AD reported on both plasma and CSF  but relied only on untargeted discovery approach and relative comparisons without targeted quantification (i.e., validation). Although the relatively small sample size and the selection of subjects with both the clinical presentation and the presence of AD pathology (as indicated by CSF biomarkers) may be considered as limitations of this study, its quantitative character and paired investigation of plasma and CSF samples represent its asset when compared to large and heterogenous multicentric studies. Independent, larger cohort studies would allow for validation of these findings and further addressing relationships with factors such as gender and APOE genotype. Furthermore, the inclusion of other groups, in particular of subjects with normal cognition and an AD CSF biomarker profile, i.e., with preclinical AD, would enable the verification of early presence of the observed metabolic dysregulations. In a longitudinal setting, the relation between the energy metabolism alterations observed in this study and the known reduced glucose metabolism in the presymptomatic stages of AD, as well as its evolution with disease progression can be studied. This would allow for the recommendation of a new set of potentially powerful small molecule biomarkers for AD diagnosis and, more importantly, the identification of a potential target pathway(s) for prevention interventions.
This metabolomics study performed using paired plasma and CSF samples from two well-defined groups highlights dysregulated systemic energy metabolism in AD and CNS-specific tryptophan pathway and creatinine alterations. In plasma of AD patients, we observed higher concentrations of TCA cycle intermediates and long-chain acylcarnitines, and lower concentrations of amino acids. These alterations appear to be of systemic origin and are mirrored in the CNS as a function of BBB permeability. The associations of specific amino acid creatinine in CSF with CSF Aβ1–42 suggest their involvement in amyloid pathology. Furthermore, our findings strongly suggest that tryptophan pathway alteration in AD is CNS-specific resulting in significantly higher concentrations of the neuroprotective kynurenic acid and neurotoxic quinolinic acid in CSF. The quantified tryptophan pathway catabolites appear to be closely related with core AD pathology, i.e., amyloid accumulation and tau-related neurodegeneration. Our study demonstrates the translational potential of the pathway-oriented quantitative approach to assess in-depth systemic and CNS metabolic defects which are part of the AD pathophysiology and represent possible targets for new therapeutic interventions.
AO, GP, and JP recruited the participants and collected the clinical data and the samples. JI and JP designed the research project. VvdV and TT conducted the metabolomic analysis. HGA and HH assisted with the analytical method setup. VvdV, JI, HGA, IK, and FM processed and analyzed the data. VvdV performed the statistical analyses in collaboration with MD. VvdV, CC, JI, and JP wrote the manuscript. All authors read and approved the final manuscript.
Fondation Pierre Mercier pour la Science, the Synapsis Foundation - Alzheimer Research Switzerland (grant: 2017-PI01), and the Swiss National Science Foundation (grant: 320030L_141179) supported the current study.
Ethics approval and consent to participate
The Ethics Committee of the canton of Vaud, Switzerland, approved the study, and written informed consent was received from all participants before inclusion into the study.
Consent for publication
JP received consultation honoraria from Nestle Institute of Health Sciences, Innovation Campus, EPFL, Lausanne, Switzerland, Ono Pharma and from Fujirebio Europe. The other authors declare that they have no competing interests.
- 6.de Leeuw FA, Peeters CFW, Kester MI, Harms AC, Struys EA, Hankemeier T, et al. Blood-based metabolic signatures in Alzheimer's disease. Alzheimers Dement (Amst). 2017;8:196–207.Google Scholar
- 16.Tynkkynen J, Chouraki V, van der Lee SJ, Hernesniemi J, Yang Q, Li S, 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. Alzheimers Dement. 2018;14(6):723–33.CrossRefGoogle Scholar
- 23.Gallart-Ayala H, Konz I, Mehl F, Teav T, Oikonomidi A, Peyratout G, et al. A global HILIC-MS approach to measure polar human cerebrospinal fluid metabolome: exploring gender-associated variation in a cohort of elderly cognitively healthy subjects. Anal Chim Acta. 2018;1037:327–37.PubMedCrossRefPubMedCentralGoogle Scholar
- 32.Kim M, Snowden S, Suvitaival T, Ali A, Merkler DJ, Ahmad T, et al. Primary fatty amides in plasma associated with brain amyloid burden, hippocampal volume, and memory in the European Medical Information Framework for Alzheimer's Disease biomarker discovery cohort. Alzheimers Dementia. 2019;15(6):817–27.CrossRefGoogle Scholar
- 37.Berg JMTJ, Stryer L. Biochemistry. 5th ed. New York: W H Freeman; 2002.Google Scholar
- 46.Sharma R, Razdan K, Bansal Y, Kuhad A. Rollercoaster ride of kynurenines: steering the wheel towards neuroprotection in Alzheimer's disease. Expert Opin Ther Targets. 2018:1–19.Google Scholar
- 51.Arnold M, Nho K, Kueider-Paisley A, Massaro T, Brauner B, MahmoudianDehkordi S, et al. The Alzheimer’s Disease Metabolome: Effects of Sex and APOE ε4 genotype. bioRxiv. 2019:585455.Google Scholar
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.