Nicotinamide as potential biomarker for Alzheimer’s disease: a translational study based on metabolomics

The metabolic routes altered in Alzheimer's disease (AD) brain are poorly understood. We performed untargeted 1 H-NMR metabolomics in hippocampus of McGill-R-Thy1-APP transgenic (Tg) rats, a model of AD-like cerebral amyloidosis. Three groups of 9 month-old rats were tested: hemizygous Tg+/-, displaying mild amyloid pathology characterized by intraneuronal amyloid β (iAβ) accumulation; homozygous Tg+/+, showing iAβ, senile plaques (extracellular Aβ deposition) and neuroinammation, and wild-type (WT). Eighteen metabolites were detected, three of them showed signicant differences among genotypes, but only two were specically assigned to a known molecule: nicotinamide adenine dinucleotide (NAD) and nicotinamide (Nam). Only Tg+/+ rats showed signicantly decreased levels of total-NAD, NADH and NAD + as compared to WT, and a signicant increase in NAD+/NADH ratio, suggesting an alteration of the redox state, alongside the reduction of all forms of NAD. Transcript levels of NAD-consuming (CD38 and PARP2) and NAD-synthesis (NMNAT2) enzymes were increased in both transgenic genotypes. Next, Nam and NAD were evaluated at the peripheral level by targeted 1 H-NMR analysis in rat plasma, where NAD/H was undetectable, Nam levels were unchanged among genotypes, but Trigonelline (a metabolic product of Nam) was reduced in Tg+/+. Finally, the translational potential of these ndings in the rat model was assessed by measuring Nam and Trigonelline levels in plasma of participants in the German longitudinal study on Aging, Cognition and Dementia (AgeCoDe), by targeted GC-EI/MS. While trigonelline was undetected, Nam was signicantly reduced in AD demented patients respect to cognitively normal participants (controls). This nding in Nam was replicated in a second independent case-control sample drawn from the same AgeCoDe. Next, the predictive value of Nam on disease progression was analyzed in AgeCoDe. Herein, NADH and NAD+ levels differed between genotypes (NADt: F(2, 6): 8.31, p NADH: 6): NAD+: 6): p = 0.05). Post-hoc analyses revealed that Tg+/+ showed signicantly lower levels of NADt, NADH and NAD+ compared with those observed in WT, while Tg+/- did not differ from either WT or Tg+/+ in any of the cases. *p<0.05; **p<0.01 (B) Mean NAD+/NADH ratio in hippocampal homogenates described in (A). Results are means ± SEM (n=3 rats per group). One-way ANOVA test resulted to be signicant (F(2, 6): 9.25; p = 0.01) and post-hoc comparisons revealed that NAD+/NADH ratio from Tg+/+ was signicantly lower than NAD+/NADH ratios observed in WT and Tg+/-. *p<0.05. (C) Transcript levels of NAD+ consumption enzymes (CD38, PARP1, PARP2 and SIRT3) and NAD+ synthesis enzyme (NMNAT2). Each bar represents the mean ± SEM of at least three independent experiments performed by triplicate for each sample normalized by Eukaryotic Translation Elongation Factor 1 Alpha 1 (EEF1A1). The mean ± SEM relative to WT (=1) is shown. Values above the dashed line (+1.5) were considered different from WT (=1). Unpaired Student t-tests showed no signicant differences between hemizygous (Tg+/-) and homozygous (Tg+/+) transgenic rats in all genes above 1.5 fold change relative to WT (CD38: t=0.2753, p=n.s.; PARP2: t=0.3588, p=n.s.; NMNAT2: t=1.038, p=n.s.).


Findings
Alzheimer's disease (AD) is a progressive neurodegenerative proteinopathy chracaterized by deposition of amyloid β (Aβ) and hyperphosphorylated tau protein in the brain of patients. Associated to the cognitive impairments and neurodegeneration, AD brains show metabolic disturbances [2]. However, the metabolic routes altered are poorly understood. As the metabolic pathways are evolutionarily conserved, the metabolic pro les carried out in transgenic (Tg) animal models of AD could be directly translated into human studies [3]. A promising animal model is the McGill-R-Thy1-APP rat [4] expressing the human amyloid precursor protein (APP) with the Swedish and Indiana mutations responsible for familial AD in humans. The hemizygous Tg+/-rats display intraneuronal Aβ accumulation and mild cognitive and behavioral impairments, representing an excellent model that resembles early stages of AD [5]. The homozygous Tg+/+ rats show the full AD-like-amyloid pathology, accompanied by neuroin ammation and cognitive impairment, re ecting stages of late AD [4]. While the Tg rat model has been extensively used to explore stages of AD pathology and validation of experimental therapeutic candidates, studies linking the metabolic pro le in hippocampus and plasma in association with the degree of amyloid pathology are still lacking. Consequently, this study aimed to characterize metabolic abnormalities in the hippocampus and plasma of homo-and hemizygous McGill-R-Thy1-APP rats by using untargeted metabolomics. Promising ndings in the rat were followed up in human plasma samples exploring their potential utility as AD biomarkers.
To determine Aβ-associated shifts in brain metabolites, we rst performed a highly sensitive multiplex ELISA to quantify total Aβ levels within the hippocampus of a sub-set of Tg rats (n=3-7). The median  (Fig.1A), three showed signi cant differences between genotypes. However, only two of them were assigned to a speci c known metabolite: NAD and nicotinamide (Nam) (Fig.1B). Standard runs analysis con rmed that changes were in NAD and not in its related metabolite NADP (Fig.1C). As previously described for these key molecules in the NAD salvagepathway, Nam and NAD levels showed an inverse relationship (Fig.1D). Since NMR analysis cannot differentiate between NADH and NAD+, we quanti ed total NAD (NADt), NADH and NAD+ by a colorimetric kit. One-way ANOVA tests indicated that NADt, NADH and NAD+ levels differed between genotypes. Post-hoc analyses revealed that Tg+/+ showed signi cantly lower levels of all NAD forms compared with those observed in WT. While the Tg+/-rat showed an intermediate level for NAD forms, these levels did not reached signi cance neither with WT nor with Tg+/+ ( Fig.2A). These observations suggest that a threshold of Aβ accumulation is required to affect NAD metabolism. Comparison of NAD+/NADH ratios between genotypes were signi cant in the one-way ANOVA test with NAD+/NADH ratio signi cantly higher in the Tg+/+ compared to WT and Tg+/- (Fig.3B). This observation suggests a clear alteration of the redox state in the brains of the Tg+/+ rats which is probably still incipient in the Tg+/-rat to reach signi cance. We also evaluated wether changes in gene expression of NAD+consuming and NAD+-generating enzymes might explain changes observed in Nam and NAD+ [6]. RT-qPCR experiments revealed that in both Tg rats genotypes the mRNA levels of CD38 (member of the cyclic ADP-ribose synthase family) and PARP2 (member of the ADP-ribose transferases family) were increased more than 1.5-fold as compared to WT, whereas transcript levels of PARP1 and SIRT3 (sirtuin) were unaffected (Fig.2C). As to NAD+-generating enzymes, gene expression analysis showed that mRNA levels of NMNAT2 were increased 1.5-fold in both Tg rats genetypes when compared to WT (Fig.2C). The NMNAT2 gene is a NAM-mononucleotide adenylyltransferase mainly expressed in brain. Based on these results, both the NAD+-consuming and the NAD+-generating pathways seem to be activated in AD brain suggesting a potential redox disturbances following the ongoing amyloid pathology. While central disturbance in NAD+ metabolism in Tg rats was observed, its translation to peripheral tissue was unclear.
Consequently, 1 H-NMR spectroscopy in plasma from these rats was performed. While NAD/H was undetectable, one-way ANOVA tests did not reveal signi cant difference in their Nam plasma levels (F(2, 16): 1.77, p>0.05). Interestingly, after a database analysis and 1 H-1 H-TOCSY con rmation, we detected levels of Trigonelline (a product of Nam metabolism) [7] in the plasma of the rats that were signi cantly reduced in Tg+/+ compared to both Tg+/-and WT rats (Fig.3A).
The results in the brain and plasma of the rat prompted us to explore whether these ndings can be translated to humans. Herein, we focused our analysis on plasma because they might offer a promising alternative for biomarker in blood. Consequently, levels of Trigonelline and Nam were measured in human plasma samples derived from the longitudinal study AgeCoDe using GC-EI-MS. First, we compared whether plasma levels of 68 participants with AD dementia (AD) showed statistical differences compared to 93 cognitively normal (CN) participants. While Trigonelline was not detected in human plasma, Nam levels were signi cantly reduced in cases compared to CN (odd ratio (OR)=0.67, p=0.02, Fig.3B). In an independent replication sample drawn from AgeCoDe, including 96 CN and 29 AD, Nam showed a protective effect (OR=0.93) which, however, did not reach signi cance (p=0.7, Fig.3B). The meta-analysis of both samples con rmed the protective effect of plasma levels of Nam (OR=0.76, p=0.04, Fig.3B). In a second step, we explore whether Nam plasma levels, measured at baseline, are associated with the time to conversion to AD. Consequently, participants were included in the analysis if they have available data on plasma levels of Nam and converted to AD at any of the next three follow-ups for which data was  (Fig.3C). This analysis showed that only the high tertile of Nam showed a protective effect progression to AD (HR=0.73, p=0.04). However, we also observed that the hazard ratio (HR) is not proportional over time (curves intersect). Thus, while a person with Nam levels in plasma within the high tertile has 27% risk reduction of progressing to AD within the next 2.5 years, this HR is lost afterwards. Supporting this nding, we observed that only participants progressing to AD at FU1 showed signi cantly lower levels of Nam compared with CN (p=0.04, Fig.3D). NAD+ and related metabolites are critical compounds essential to adaptive stress responses and cell survival. Considering the number of enzymes and transcription factors sensitive to the redox potential, NAD/H redox state acquires pathophysiological relevance for aging and neurodegenerative diseases [8][9]. While several studies in mouse models for AD have shown the relevance of the NAD(P)+/NAD(P)H homoestasis in the brain, especially in hippocampus and cortex [10][11][12], few reports have been published on the role of the NAD(P)+/NAD(P)H homoestasis in the McGill-R-Thy1-APP rats. Herein, an in vivo 1 H-MRS study in McGill-R-Thy1-APP rats identi ed a complex metabolite differences in hippocampus and frontal cortex of the transgenic rats [13]. In addition, several isolated cortical regions were analyzed in aged animals (15-month-old) using 1 H-and 13 C NMR spectroscopy and HP-LC showing decreased levels of NAD+ in the cingulate cortex of Tg rats [14]. Our study now provides additional supporting evidence indicating that hippocampal Aβ burden could determine the degree of NAD+ shift in McGill-R-Thy1-APP rat brain. This information is useful because it offers a different perspective on the Aβ-mediating mechanisms involved in brain energy dysfunction observed in AD.
From a translational point of view, experimental evidence supports that NAD+ augmentation can revert cognitive de cits in AD mouse model [15,16]. Furthermore, research has shown that plasma levels of NAD+ decrease signi cantly with age in humans [17]. These animal and human data have fueled several clinical trials of NAD+ precursors which, however, produced inconsistent results [18,19]. Importantly, our report showed that a metabolic disturbance detected in McGill-R-Thy1-APP rat brain is also found in rat plasma. This nding offers the possibility to use NAD+ metabolites as peripheral biomarkers for AD. Our study identi ed a signi cant lower level of plasma Nam in AD patients compared to healthy control. This difference is also seen at least 1.5 years before the patients progressed to AD. Thus, Nam levels in plasma may also serve as biomarker for progression to AD. However, it remains to be determined when Nam level increases in plasma before dementia is diagnosed.
Our study has also limitations. For example, rats showed decrease levels of Trigonelline in plasma, whereas human showed reduced levels of Nam. Although little is known about the pathway connecting Nam and Trigonelline, their metabolism seems to be a linked. Trigonelline is the methylation product of nicotinic acid, and this latter metabolite is in constant equilibrium with Nam within the vitamin B3 metabolism [20]. Hence, disequilibrium in this pathway may be re ected in either Trigonelline or Nam levels.
In summary, this is the rst report showing a signi cant decrease of Nam plasma levels in people with AD that is observed couple of years before conversion, thereby suggesting its potencial use as biomarker for progression. Further studies in larger cohorts are now needed to con rm our ndings and the potential use of Nam as peripheral biomarkers for AD.

Rat model
Transgenic (Tg) McGill-R-Thy1-APP rats, harboring the human APP751 transgene with the Swedish and Indiana mutation under the control of the murine Thy1.2 promoter, were generated using the HsdBrI: WH Wistar strain [4]. Animals were provided to Fundación Instituto Leloir (FIL) by The Royal Institution for the Advancement of Learning/McGill University, Montreal, Canada, and an in-house colony was established at FIL. Rats' genotypes were determined by real time qPCR as previously described [5]. To avoid the litter effect, groups were made up of pups from 3 to 4 different litters. Homozygous (Tg+/+), hemizygous (Tg+/-), and littermates' wild type (WT) control animals were maintained in polycarbonate cages in a temperature-controlled animal facility with a 12-h dark/light cycle and allowed to consume standard diet and water ad libitum. Only 9-month-old male rats were used for experiments to avoid any potential effects of female estrus cycle. All experimental procedures were performed in accordance with the guidelines of ARRIVE and OLAW-NIH. The protocol was approved by the local animal care committee (CICUAL # A5168-01).

Hippocampal tissue and plasma collection
Rats were anesthetized with ketamine (50mg/kg) and xylacine (10mg/kg), placed under a guillotine blade, decapitated and brains quickly removed. Hippocampi were dissected on an ice-cold plate, and divided into left and right hemispheres. Each hippocampus was immediately and independently frozen in liquid nitrogen, and stored at -80 °C until used. Blood samples were collected immediately after animals' decapitation in 15 mL heparinized tubes, centrifuged at 3500 xg for 10 min. Plasma was collected, aliquoted and stored at -80 °C until use.

Human plasma samples
Plasma samples were selected from the German study on Aging, Cognition and Dementia (AgeCoDe) biobank [21]. This is a longitudinal study, where participants were recruited in primary care centers in six German cities. Inclusion criteria was to be at least 75 years old and cognitively healthy according to the general practitioner's judgment. Every ~18 months interval participants are followed-up with personal interviews and neuropsychological assessments. To date, nine follow-ups (FUs) were completed, but results from the last one are still in process. Blood samples were obtained at the 3 rd visit, processed and store at -80 0 C. For this study the 3 rd visit is considered the baseline. Controls (n=189) remained cognitively normal until the last FU analyzed, were 83.6 ± 3.1 years old, 64.0% were female and 20.6% were APOE4 carriers. Cases were participants with incident AD (n=68), 86.0 ± 3.6 years old, 64.7% female and 33.8% APOE4 carriers; and prevalent AD (n=29), 84.2 ± 3.1 years old, 75.8% female and 37.9% APOE4 carriers. Participants converting to AD in the next three visits following baseline (FU1, FU2 and FU3) were included in the analysis. At FU1 were 25 participants with age of 84.8 ± 3.5 years old, 80% women, and 28% APOE4 carriers; at FU2 were 37 participants with age of 83.6 ± 2.6 years old, 67.6% women, and 32.4% APOE4 carriers; and at FU3 were 23 participants with age of 82.7 ± 2.6 years old, 60.9% women, and 21.7% APOE4 carriers.

Expression of Aβ isoforms in rat hippocampus
To quantify human Aβ 38/40/42 MSD® V-PLEX PLUS Aβ Peptide Panel 1 kit was used following the manufacturer's instructions. Brie y, hippocampus of Tg rats were processed with illustra triplePrep kit (GE) and Individual protein samples were loaded onto MULTI-SPOT® microplates pre-coated with antibodies speci c to the C-termini of Aβ38, Aβ40 and Aβ42 and were detected with SULFO-TAG™-labeled 6E10 antibody. Light emitted upon electrochemical stimulation was read using the MSD QuickPlex SQ120 instrument. Data were analyzed using MSD Workbench 4.0 software. Values of concentration in pg/mg of total protein were expressed as median and interquartil range (IQR). Mann Whitney test was applied to assess signi cant differences between groups.

Gas chromatography Electron Impact Mass Spectrometry (GC-EI-MS)
Plasma samples were thawed on ice, and 100 ul were extracted with 900 ul of cold extraction buffer containing 40:40:20 methanol:acetonitrile:water [v:v:v]. After 30 min in an orbital mixer at 4ºC., samples were sonicated for 10 min in an ice-cooled bath-type sonicator and centrifuged for 10 min at 16000xg at 4ºC. Supernatants were collected and dried in a SpeedVac until complete dryness. Standard curves were prepared with concentrations ranging from 0.005 to 50 ug/ml of nicotinamide and trigonelline, processed in the same way as samples. Dried down samples and standards were derivatized using methoxyamine and MSTFA/FAMEs solution (N-methyl-N-trimethylsilyl-tri uoracetamid/Fatty acid methyl esters) following standard procedures [23,24]. After that, samples were analyzed in a GC-EI-MS (Q Exactive GC Orbitrap system, ThermoFisher) using a 30-m DB-35MS capillary column. Representative fragments from the GC-EI-MS analysis of Nam were extracted using TraceFinder (Version 4.1, ThermoFischer) and quanti ed using the linear range of the obtained standard curve. All analysis were performed using peak areas, transformed into Z-scores, for easier comparation among experiments.

Statistical analysis
Differences among groups in the AUC of peaks from 1 H-RMN spectroscopy experiments were analyzed by one-way ANOVA, taking p<0.05 and p<0.0001 as signi cant according to Dunnett's multiple range test. Data of protein levels of NADt/NADH and NAD+ were analyzed by one-way ANOVA tests followed by post-hoc Tukey's multiple comparisons tests. Two-tailed unpaired t-tests were used to compare transcripts expression of those genes that showed more than 1.5-fold change in their expression levels compared to WT (=1). For easier comparation among experiments of rat plasma metabolome a Z-score standardization was applied and subsequently data were analyzed by one-way ANOVA tests followed by post-hoc Tukey's multiple comparisons tests. In all cases, assumption of normality was examined using Kolmogorov-Smirnov or Shapiro-Wilk tests. A probability equal or less that 5 % was considered as signi cant. All analyses were carried out using GraphPad Prism for Windows (version 7.0). Statistical analysis and plots of data from GC-EI-MS experiments performed with human plasma were done using R-project v. 4.0.0 (https://www.R-project.org) and R-studio v.1.2.5042 (http://www.rstudio.com/). Normal distribution was visualized using qqnorm plots, and outliers (de ned as mean +/-3 standard deviation) were eliminated from the analysis (n=2). Linear regression models adjusted for sex, age and apoe4 were used to estimate the association of Nam levels vs CN in the discovery and replication experiments, as well as in FUs groups. Meta-analysis was performed using the R-package "metafor" [25] and visualized with the general function forest. The cox proportional hazards regression model, which relates time dependent variables, time dependent strata, and multiple events per subject, were performed with the Rpackage "survival" [26] and "survminer" [27]. Only samples with incident AD at FU1-3 were included, time variable was time to conversion to AD in years, and the event per subject was conversion (no=0, yes=1).
Proportional hazard assumption was tested by Schoenfeld's test, and consequently two cox regressions were performed with a split-time=2.5 years.

Declarations
Ethics approval and consent to participate The original study protocol with human subjects was approved by the local ethics committees at the following German institutions: University of Bonn; University of Hamburg; University of Duesseldorf; University of Heidelberg/Mannheim; University of Leipzig and the Technical University of Munich. Written informed consent was obtained from all participants.

Consent for publication
Not applicable Availability of data and material The datasets supporting the conclusions of this article are included within the article.

Figure 3
Plasma levels of Trigonelline as a potential peripheral biomarker of AD amyloid pathology. (A) Box plots represent the normalized NMR spectral areas of Trigonelline in plasma of WT (n=8), Tg +/-(n=5), and Tg