Journal of Alzheimer’s Disease 60 (2017) 495–504 495 DOI 10.3233/JAD-170485 IOS Press Pathway Metabolites in Alzheimer’s Disease

Lasse Melvaer Giila,b,∗, Øivind Midttunc, Helga Refsumd,e, Arve Ulvikc, Rajiv Advanif , A. David Smithe and Per Magne Uelandb,g aDepartment of Internal Medicine, Haraldsplass Deaconess Hospital, Bergen, Norway bDepartment of Clinical Science, University of Bergen, Norway cBevital AS, Norway dDepartment of Nutrition, Institute of Basic Medical Sciences, University of Oslo, Norway eOPTIMA, Department of Pharmacology, University of Oxford, UK f Department of Clinical Medicine, University of Bergen, Norway gLaboratory of Clinical Biochemistry, Haukeland University Hospital, Bergen, Norway

Accepted 7 July 2017

Abstract. Background: Metabolites of , produced via the kynurenine pathway (), have been linked to Alzheimer’s disease (AD) in small cohorts with conflicting results. Objective: To compare differences in plasma kynurenine levels between AD and controls and identify potential associations with cognition. Methods: The study included 65 histopathologically-confirmed AD patients and 65 cognitively-screened controls from the Oxford Project to Investigate Memory and Ageing (OPTIMA) cohort. Cognition was assessed using the Cambridge Cognitive Examination (CamCog). Tryptophan, kynurenines, neopterin, and vitamin B6 forms were measured in plasma by liquid chromatography-tandem mass spectrometry. Non-parametric statistics, logistic regression and standardized robust regressions were applied with a false discovery rate of 0.05. Results: Tryptophan, , 3-hydroxyanthranilic acid, and were lower in AD (Odds ratios (ORs) 0.24 – 0.47; p-values <0.001 – 0.01). Pyridoxal 5’phosphate did not differ between AD and controls. Kynurenine, anthranilic acid, quinolinic acid, and markers of immune activation (neopterin, kynurenine/tryptophan ratio, and the PAr index (Pyridoxic acid/(Pyridoxal 5’phosphate + Pyridoxal)) increased with age (␤ 0.31 – 0.51; p-values <0.001 – 0.006). Xanthurenic acid decreased with age (␤: –0.42, p < 0.001). Elderly AD patients with high quinolinic acid performed worse on the CamCog test, indicated by a significant age*quinolinic acid interaction (␤ 0.21, p < 0.001). Conclusion: Plasma concentrations of several kynurenines were lower in patients with AD compared to controls. Low xanthurenic acid occurred in both AD and with aging. Inflammation-related markers were associated with age, but not AD. However, elevated QA was associated with poor cognition in older AD patients.

Keywords: Alzheimer’s disease, aging, cognition, , kynurenine pathway, quinolinic acid, vitamin B6, xanthurenic acid

INTRODUCTION accumulation of amyloid-␤ in plaques, tau in tan- gles, activation of microglia, and damage to cerebral Alzheimer’s disease (AD) is the most common microvasculature [1]. In addition to activation of cause of dementia and is characterized by cerebral the innate immune system in the brain, there is an increase in migration of peripheral immune cells to ∗ Correspondence to: Lasse Melvaer Giil, MD, Haraldsplass the brain [2]. High body mass index and type II dia- Deaconess Hospital, Mail Box 6165, Postal code 6165, 5009 Bergen, Norway. Tel.: +47 46890860; E-mail: lassegiil@gmail. betes increases the risk of developing AD; impaired com. insulin signaling and altered levels of key adipokines

ISSN 1387-2877/17/$35.00 © 2017 – IOS Press and the authors. All rights reserved 496 L.M. Giil et al. / Kynurenine Pathway, Alzheimer’s Disease, and Aging suggest that AD is linked to metabolic dysfunction Systemic levels and activity of kynurenines outside [3]. The metabolism of tryptophan via the kynure- of the brain has been addressed in clinically- nine pathway has been linked to insulin resistance [4], diagnosed AD. The activity of KAT was reduced inflammation [5], and neurodegenerative disorders, in red blood cells in patients with AD [13]. Studies including AD [6]. investigating the plasma levels of these metabolites The major metabolic pathway of tryptophan in AD patients have reported lower Trp [14], higher (Trp) degradation, an essential amino acid, is the HK [15], and lower Trp with lower KA and higher QA kynurenine pathway. The first step is the forma- [16]. The sample sizes in these studies were relatively tion of formylkynurenine, catalyzed either by hepatic small with around 20–34 cases and 15–19 controls. In tryptophan 2,3-dioxygenase (TDO), or indoleamine- a large community-based cohort (N = 7052) plasma 2,3-dioxygenase (IDO), expressed in monocytes. levels of most kynurenines, except XA, increased Formylkynurenine is rapidly converted to kynurenine whereas plasma tryptophan decreased with age [9]. (Kyn), which is metabolized to 3-hydroxykynurenine The kynurenine pathway is of major interest in (HK) by kynurenine monooxygenase (KMO) and fur- AD, due to its links with inflammation, insulin ther to 3-hydroxyanthranilic acid (HAA) by kynure- resistance, and neuroactive metabolites. However, nine kinase (KYNase) and then to quinolinic acid cofactors, such as PLP, substrate availability, and a (QA); a precursor of nicotine adenine dinucleotide pro-inflammatory state, may influence kynurenine (NAD). Anthranilic acid (AA) is an intermediary levels in both aging and AD. We sought there- metabolite generated from Kyn by KYNase, while fore to investigate 1) the levels kynurenines in formation of (KA) from Kyn and pathologically-confirmed AD patients and screened xanthurenic acid (XA) from HK are both reactions controls, 2) the association between the kynurenine catalyzed by kynurenine aminotransferase (KAT)[7]. pathway and aging, 3) associations of kynurenines The active form of vitamin B6, pyridoxal 5’- with cognitive impairment in patients with AD, and phosphate (PLP), is a cofactor of the enzymes 4) changes in cofactors and regulators of kynurenine KYNase and KAT, while flavin adenine dinucleotide pathway in ageing and patients with AD. is a cofactor of KMO [8]. The kynurenine pathway is regulated by of the innate immune system, primarily interferon gamma (IFN-␥), which activates METHODS IDO. The activity of TDO is stimulated by corti- costeroids and Trp, and TDO is the main enzyme Study participants responsible for Trp catabolism and homeostasis [7]. Finally, the kynurenines are excreted by the kidney Sixty-five patients and 65 healthy controls were and their serum levels are thus also dependent on renal recruited from the Oxford Project to Investigate function [9, 10]. Memory and Ageing (OPTIMA). The OPTIMA Trp enters the brain by carrier-mediated transport study recruited both cognitively impaired and cogni- across the blood-brain barrier and is a precursor to tively intact elderly persons. OPTIMA started in 1988 serotonin. Thus, plasma Trp can affect cerebral sero- and the participants included in the present study were tonin synthesis. Levels of IDO and TDO are low in recruited from 1994 to 2009. Participants underwent the brain and generation of kynurenines in the brain a detailed medical history, physical examination and depends on availability of the downstream metabo- neuropsychological assessment. All 65 patients in lite Kyn. In fact, Trp, Kyn and 3-HK all cross the the present report had their diagnoses confirmed at blood-brain barrier from the periphery. The kynure- autopsy by the histopathological diagnosis of proba- nine pathway is expressed in both neuronal and glial ble (n = 3) or definite (n = 62) AD using the CERAD cells where KA is produced in astrocytes and QA criteria [17]. Control subjects were all screened cog- in microglia. KA is neuroprotective due to its action nitively on an annual basis to exclude any with cogni- as an antagonist at the N-methyl-D-aspartate receptor tive impairment, and the 51 who died were all shown (NMDAR). However, QA has neuroexcitatory effects to have normal brains by histopathology. Neuroimag- due to the activation of NMDAR. QA can further ing was performed, blood, urine, and cerebrospinal lead to lipid peroxidation and generation of reactive fluid samples were collected. The personnel han- oxygen species. KMO inhibition, reducing QA, had dling the samples were unaware of any diagnoses and beneficial effects in animal models of both AD and other clinical variables. The complete protocol for the Huntington’s disease (reviewed in [7, 11, 12]). OPTIMA study is described in other works [18]. L.M. Giil et al. / Kynurenine Pathway, Alzheimer’s Disease, and Aging 497

Global cognitive performance was measured using Statistics the Cambridge Cognitive Examination (CamCog). CamCog measures specific cognitive domain func- A p-value < 0.05 was considered statistically tions in areas of orientation, language, memory, significant. Univariate analyses were done by praxis, attention, abstract thinking, perception, and non-parametric statistics (Mann-Whitney U Test calculation [19]. The range of possible scores are (MWUT), Wilcoxon Signed Rank test (WSR) and from 0 (lowest) to 107 (highest) with a typical cut-off Chi-Square test (χ2)), since the assumption of a nor- for dementia ≤79. mal distribution was not true. Significance levels were adjusted for multiple testing at a false discov- ery rate (FDR) of 0.05 (Benjamini-Hochberg). The Ethics Benjamini-Yekutieli method [22] was used to adjust Protocols were approved by the Central Oxford the associations between metabolites and age, due to p Ethics Committee, number 1656. All participants very low -values. Q-values are reported for adjusted p provided written consent after information regarding analyses (Q-value of 0.05 = first -value < FDR). the study protocol was provided to the participants Prior to multivariate analyses, metabolites and cre- and in cases of moderate- to severe dementia, also atinine were transformed by the natural logarithm, to a caregiver. Histopathological examination of 1/square root or the Box-Cox transformation [23] to the brain was performed only in participants that achieve multivariate normality and converted to stan- provided written consent for brain donation ante- dardized variables (z-scores) to better compare effect mortem. sizes [24]. Due to significant age-mismatch, we compared unmatched- (MWUT) and groups matched on age Measurement of metabolites and gender (WSR) by 1 to 1 propensity score match- ing algorithm (nearest neighbor, random matching Blood samples were collected in tubes containing order and caliper 0.2) [25]. Using logistic regression, ethylenediamine tetraacetic acid (EDTA) from non- we introduce age, gender and creatinine as covariates fasting participants. EDTA plasma was divided in to and tested if case-control differences persisted in the small aliquots to avoid repeated freeze-thaw cycles ◦ presence of potential confounders. We first examined and then stored at –80 C until biochemical analy- if there were case-control differences in empirical sis was performed (in August 2014). Trp, Kyn, AA, confounders (substrate availability (Trp), monocyte- KA, HK, HAA, XA, QA, pyridoxal 5´-phosphate activation (neopterin), co-factor (PLP) availability (PLP), pyridoxal, 4-pyridoxic acid, riboflavin, flavin and storage time) and included confounders with mononucleotide, neopterin, and cotinine were mea- identified differences. A 10% change-in-estimate sured by Bevital AS (http://www.bevital.no) using [26], here odds ratio (OR = ((crude OR – adjusted liquid chromatography-tandem mass spectrometry OR)/crude OR)) indicated minor confounding. A (LC-MS/MS) assay [20]. The kynurenine/tryptophan ≥ ∗ change to a non-significant p-value ( 0.05), indi- ratio (KTR) was calculated and as (Trp/Kyn) 1000. cated major confounding. The PAr Index (PAr) was calculated as Pyri- Influential outliers were present in linear regres- doxic acid/(PLP + Pyridoxal). Serum creatinine sion analyses with age and CamCog as dependent was measured in the Oxford hospitals clinical variables (Cooks distance > 4/N, leverage > 2*K/N chemistry laboratory by the Jaffe photometric where K = number of covariates and N = sample size). method. To reduce the risk of type I and II errors due to influence from outliers, we used a high-efficiency, Power analysis high-breakdown point, robust estimator of linear regression (“MM estimation”) that is robust to out- Given a power of 0.8, an R2 of 0.3 with covari- liers in the presence of normally distributed residuals ates, and an alpha level of 0.05 in equally distributed [27]. The CamCog score and age distributions both groups and a predictor with a normal distribution, had strong negative skew. These were transformed to the required sample size was 116 to detect an odds a positive skew by subtracting them from their high- ratio (OR) of 2 (or OR = 0.5) with logistic regression. est score, followed by log transformation. CamCog The minimum detectable effect size, (N = 130), was was standardized and analyzed as the natural loga- OR = 1.91 (or OR = 0.52) (G*Power [21]). rithm of its errors. The age transformation described 498 L.M. Giil et al. / Kynurenine Pathway, Alzheimer’s Disease, and Aging above was again subtracted from the highest value Trp, Kyn, HAA, and QA were significantly lower and then standardized, to maintain direction. To avoid in AD in both the matched and unmatched groups breaking the assumption of multicollinearity due to while XA was significantly lower in AD in the age- correlated metabolites (R 0.19–0.80, Spearman Rho), matched group only. HK and KA were significantly each metabolite was analyzed separately in all analy- lower in the unmatched group, only. ses. Analyses were performed in Stata 14 (StataCorp. 2015. Stata Statistical Software: Release 14. College Multivariate analyses Station, TX: StataCorp LP, package: robreg), R ver- sion 3.3.2 (packages: PM Match and p.adjust) [28]. Significantly lower Trp and larger variation Graphs were generated in GraphPad Prism 7.0 for (interquartile range) in neopterin in AD could act Windows (GraphPad Software, LA Jolla California as confounders. A simple model with age, gender, USA). and creatinine (model 1) was compared to a model with age, gender, creatinine, and Trp (model 2), and RESULTS age, gender, creatinine, and neopterin (model 3). Each metabolite was entered separately in each of these Characteristics of study sample models with AD versus controls as the dependent variables. Moving from model 1 to 2 and/or 3, a ± 0.1 The AD patients were on average 6 years younger change in OR, or a p-value < 0.05 that change to than the controls (p < 0.001, Table 1). Cotinine >0.05, would indicate minor and major confounding, levels, a biomarker of recent nicotine exposure, respectively (Table 3). renal function (serum-creatinine) and sample storage Kyn was no longer significantly lower in AD times were not significantly different for AD versus patients when adjusting for Trp. Adjustment for Trp controls. resulted in weaker, but still significantly lower HAA and XA in AD patients. Adjustment for neopterin Metabolites in Alzheimer’s disease patients and resulted in relatively stronger associations with lower healthy controls Kyn, HAA and QA in AD.

To adjust for age-differences, we generated Metabolites with respect to age propensity score matched groups based on age and gender (N = 42 cases per group). The mean (M) age The results from robust, standardized regressions in the matched AD group was 78.46 years (standard with age as the dependent variable, each metabo- deviation (SD) 6.34) and 78.55 years (SD 6.84) in the lite analyzed separately, are summarized in Fig. 1 matched control group, showing no significant differ- (“smile-plot”). Gender, diagnosis, and creatinine ence (p = 0.954, t-test). The matched groups were also were covariates in all analyses. similar in gender distribution (X2 0.048, p = 0.827) Neopterin, KTR, QA, Kyn, PAr, and AA were sig- and renal function (p = 0.993, MWUT). The metabo- nificantly, positively associated with age, while XA lite levels in the unmatched and matched groups are was significantly inversely associated, after adjusting shown in Table 2. for multiple testing. The covariates were not included

Table 1 Characteristics of study cohort Alzheimer’s disease Controls Mdn, % IQR Mdn, % IQR pa Age, years 74.3 15.1 81.6 8.6 <0.001 Gender, % female 43 51 0.380 CamCog, scoreb 63.0 25.0 99.0 5.5 <0.001 Cotinine, nmol/L 0.6 2.3 0.6 4.6 0.914 Creatinine, ␮mol/Lc 100.0 23.0 95.0 24.0 0.565 Storage time, y 13.5 5.0 12.7 3.0 0.850 Mdn, Median; IQR, Interquartile range. aP-values from Pearson Chi-Square test (gender), else = Mann-Whitney U Test. bThe range of the CamCog score is from 0 (lowest possible) to 107 (highest possible). The usual cut-off for dementia is 79 points and below (98.5% of healthy controls above, 81.4% of AD patients below). cUsing the Modification of Diet in Renal Disease (MDRD) equation, 65% of cases and 69% of controls had a glomerular filtration rate between 30 and 60, but no participants had < 30. L.M. Giil et al. / Kynurenine Pathway, Alzheimer’s Disease, and Aging 499

Table 2 Metabolite concentrations in AD and controlsa Non age-matched Age-matchedb Metabolite AD Controls Qc AD Controls Qd Trp, ␮mol/L 57.5 [11.2] 62.3 [15.1] 0.006∗ 54.6 [16.2] 62.7 [10.7] 0.002∗ Kyn, ␮mol/L 1.8 [0.5] 2.0 [0.8] 0.006∗ 1.83 [0.5] 2.1 [0.8] 0.036∗ HK, nmol/L 40.7 [17.9] 51.6 [26.7] 0.002∗ 44.7 [21.3] 51.7 [31.0] 0.068 KA, nmol/L 48.6 [26.3] 58.6 [28.6] 0.006∗ 48.5 [50.2] 57.4 [34.0] 0.149 XA, nmol/L 14.1 [8.8] 15.9 [9.4] 0.149 13.4 [10.0] 16.5 [10.3] 0.027∗ AA, nmol/L 16.4 [6.4] 19.1 [6.5] 0.079 16.9 [6.0] 19.2 [30.7] 0.175 HAA, nmol/L 28.7 [11.0] 38.3 [14.1] <0.001∗∗ 28.6 [12.1] 39.2 [15.8] 0.002∗ QA, nmol/L 465 [230] 565 [370] 0.002∗ 502 [265] 629 [434] 0.048∗ KTR, – 32.3 [9.0] 32.3 [13.0] 0.570 34.0 [8.0] 33.5 [17] 0.680 Neop, nmol/L 17.8 [7.5] 18.9 [10.2] 0.084 18.2 [43.7] 19.1 [14.0] 0.468 PLP, nmol/L 59.0 [58.7] 60.0 [48.4] 0.738 49.6 [44.2] 58.5 [58.5] 0.570 PAr, – 0.6 [0.4] 0.6 [0.4] 0.905 0.61 [0.4] 0.6 [0.6] 0.802 Trp, Tryptophan; Kyn, Kynurenine; KA, Kynurenic acid; AA, Anthranilic acid; HK, 3-hydroxykynurenine; XA, Xanthurenic acid; HAA, 3-Hydroxyanthranilic acid; QA, Quinolinic acid; KTR, (Kynurenine/Tryptophan)∗1000; Neop, Neopterin; PLP, Pyridoxal 5’-phosphate; PAr, PAr Index (Pyridoxic acid/(PLP + Pyridoxal)); ␮mol/L, micromoles per liter; nmol/L, nanomoles per liter. aNumbers are given as median and interquartile range in parenthesis. bThe cases were matched on age and gender by propensity score matching. cp-value obtained by Mann-Whitney U test and adjusted for multiple testing (Benjamini-Hochberg at 0.05). dp-value obtained by Wilcoxon signed rank test and adjusted for multiple testing (Benjamini-Hochberg at 0.05). ∗q-value < 0.05, ∗∗q-value < 0.001.

Table 3 Multivariate analyses of kynurenines in Alzheimer’s disease versus controlsa Model 1, simpleb Model 2, tryptophanb Model 3, neopterinb Metabolites OR p OR ORc p OR OR p Trp 0.29 <0.001 0.29 – <0.001 0.29 0% <0.001 Kyn 0.49 0.009 0.69 –0.41 0.246d 0.39 0.20 0.005 XA 0.35 <0.001 0.51 –0.46 0.027 0.34 0.03 <0.001 HAA 0.24 <0.001 0.33 –0.38 <0.001 0.20 0.17 <0.001 QA 0.47 0.012 0.45 0.04 0.016 0.36 0.23 0.005 OR, odds ratio; p, p-value; creat., Creatinine; Trp, Tryptophan; Kyn, Kynurenine; XA, Xanthurenic acid; HAA, 3-Hydroxyanthranilic acid; QA, Quinolinic acid. aLogistic regressions with Alzheimer’s disease versus controls as the outcome (see b, below). bModel 1 = age, gender and the metabolite listed. Model 2 & 3 = Model 1 + tryptophan or neopterin. cMinor confounding = OR > +/– 0.1. OR = (crude (model 1) OR – adjusted OR (model 2 or 3))/crude OR. dMajor confounding = change in the p-value to p ≥ 0.05. in the graph (male (␤: 0.11, p = 0.119), AD (␤: –0.85, Pathway and result summary p < 0.001) and creatinine (␤: –0.85, p < 0.001), results with Trp as the metabolite). To facilitate understanding of the findings, a simplified pathway diagram was constructed with summaries of significant findings (Fig. 2). Cognitive performance

Cognitive performance was tested using robust DISCUSSION regression by MM estimation. The outcome variable CamCog was transformed by the natural logarithm In our study, we have shown that metabolites of of errors made on the CamCog test, and potential the kynurenine pathway have different associations effect modification by age was tested by introducing with AD and age. In patients with AD, Trp, Kyn, interaction terms in the regressions (Table 4). HAA, XA, and QA were all significantly lower than in After multiple testing was adjusted for, the interac- controls; however, Kyn was not lower when Trp was tion between age and QA was significantly associated adjusted for. Aging was associated with an increase with more errors on CamCog assessment. The associ- in Kyn, AA, and QA. XA was decreased, and this ations with the covariates gender and creatinine were reduction also occurred in AD. Aging was associ- non-significant. ated with a pro-inflammatory state, indicated by high 500 L.M. Giil et al. / Kynurenine Pathway, Alzheimer’s Disease, and Aging

The lower Trp in AD identified in our study is consistent with previous reports [14–16]. However, there are discrepancies between studies regarding downstream metabolites. While we found reduced QA, this metabolite has been previously reported to be increased in AD [16]. The same study also reported reduced KA, which was not supported by our find- ings, neither was the previously described increase in HK [15]. Lower HAA and XA have to our knowl- edge not been previously described in AD patients. A comparison between this study and previous reports can be found in Table 5. In our study, age is poten- tially a major confounder of kynurenine metabolite levels. Similar relations with age have been observed in a large community cohort where renal function was another important determinant of kynurenine levels [9]. Our study has a comparative advantage in the rigorous diagnosis, and sample size is larger than in previous studies. This allowed us to adjust for poten- tial confounders, which had influence on the model. This illustrates the need for larger cohorts that can Fig. 1. Strength of metabolite-age associations (“Smile plot”). better adjust for the aforementioned confounders and ∗-log(p-value) = the negative logarithm, base 10, of the p-value investigate relationships to cognition in more com- (1 is p = 0.1, 2 is p = 0.01, 3 is p = 0.001, 4 is p = 0.0001, plex models. etc.). Standardized regression coefficients = regression coefficients (␤) when all variables = z-scores. The strongest associations Trp, HAA, XA, and QA were lower in AD patients. in a smile plot are seen in the upper, outer corners. FDR The low Trp in AD patients may reflect low pro- (false discovery rate), significance threshold with the Benjamini- tein intake, which was not available in our study, but Yekuteli correction at 0.05; PLP, Pyridoxal 5’-phosphate; PAr, malnutrition would in theory worsen with increas- PAr Index (Pyridoxic acid/(PLP + Pyridoxal)); Neopt, Neopterin; Trp, Tryptophan; Kyn, Kynurenine; KA, Kynurenic acid; AA, ing cognitive impairment. Lower levels of XA and Anthranilic acid; HK, 3-hydroxykynurenine; XA, Xanthurenic HAA could partly be explained by lower substrate acid; HAA, 3-Hydroxyanthranilic acid; QA, Quinolinic acid; KTR, ∗ availability (Trp). Kyn was lower in AD patients, (Kynurenine/Tryptophan) 1000; Creat, Creatinine. Estimates of but not when adjusted for Trp. Kyn, HAA, and QA covariates (not shown) on age with Trp: Male (␤: 0.11, p = 0.119), AD (␤: –0.85, p < 0.001), creatinine (␤: –0.85, p < 0.001). were even relatively lower in AD, when adjusting for the larger variation in neopterin among patients with AD. Our observation may suggest low activ- KTR, neopterin, and the PAr index, also considered ity of IDO and/or TDO in AD patients, which may to be an inflammatory marker [29]. High QA among reflect low pro-inflammatory activity relative to a older patients with AD was associated with more low bioavailability of Trp. Nonetheless, HAA, XA, errors on CamCog assessment. Notably, the inflam- and QA were all still reduced after adjustment for matory markers KTR, neopterin and the PAr index Trp and neopterin. There was no difference in levels were not associated with AD. of PLP between AD and controls. A clinically rele- Strengths of the study include the systematic vant issue is if the reduced availability of Trp in AD examination of the histopathologically-diagnosed affects the therapeutic response to selective-serotonin OPTIMA participants, measurement of many reuptake inhibitors. Similarly, it would be of great metabolites in the kynurenine pathway, and impor- interest to determine if decreased plasma levels of tant confounders. However, the possibility of Kyn, a precursor to the NMDA agonist QA and the residual confounding cannot be excluded. The main NMDA antagonist KA, could influence the therapeu- weaknesses were an age-mismatch between the tic response to the anti-dementia drug memantine, an AD and control group, and a relatively low sample NMDA antagonist [30]. size. Effect modifications should be interpreted with Kynurenines (Kyn, AA, and QA) increased with caution in small samples and must be validated in a age together with markers of inflammation (KTR, subsequent patient population. neopterin, and PAr). The immune response to L.M. Giil et al. / Kynurenine Pathway, Alzheimer’s Disease, and Aging 501

Table 4 Cognitiona in patients with AD according to kynureninesb: Effect modification by age Model 1: Simplec Model 2: Interaction with aged Metabolite*age ␤ p ␤ p ␤ p Trp –0.14 0.401 –0.15 0.605 –0.01 0.605 Kyn 0.03 0.760 0.17 0.145 0.21 0.092 HK 0.21 0.029 0.23 0.017 0.12 0.235 KA 0.02 0.780 0.054 0.509 0.15 0.007 XA 0.09 0.361 0.09 0.317 0.04 0.641 AA 0.10 0.174 0.17 0.019 0.16 0.066 HAA –0.10 0.218 –0.04 0.782 0.11 0.460 QA 0.10 0.465 0.23 0.057 0.21 <0.001∗ KTR 0.12 0.273 0.13 0.277 0.17 0.085 Neopt –0.02 0.779 0.06 0.467 0.14 0.177 PLP –0.09 0.235 –0.09 0.261 –0.01 0.803 PAr 0.05 0.343 0.13 0.072 0.13 0.112 Trp, Tryptophan; Kyn, Kynurenine; KA, Kynurenic acid; AA, Anthranilic acid; HK, 3-hydroxykynurenine; XA, Xanthurenic acid; HAA, 3-Hydroxyanthranilic acid; QA, Quinolinic acid; KTR, (Kynurenine/Tryptophan)∗1000; Neopt, Neopterin; PLP, Pyridoxal 5’-phosphate; PAr, PAr Index (Pyridoxic acid/(PLP + Pyridoxal)); ␤, standardized regression coefficient. aCamCog was the outcome after the variable was inversed and log transformed so that higher is worse, see c,d. bAll analyses with standardized robust linear regression by MM estimation at default efficiency of 85%. cIndependent variables age, gender, creatinine and metabolite (as listed). dIndependent variables age, gender, creatinine, age, metabolite and age*metabolite (as listed). ∗Significant after adjustment for multiple testing (Benjamini-Hochberg), adjusted for all p-values in model 2.

Fig. 2. Pathway diagram and summary of findings. ∗Measured metabolites are in bold, related but unmeasured metabolites are in normal and smaller font. Enzymes: IDO, Indolamine deoxygenase; TDO, Tryptophan deoxygenase; KAT, Kynurenine aminotransferases (I, II, III); KMO, Kynurenine monooxygenase; KATs, Kynurenine aminotransferases; 3-HAO, 3-hydroxyanthranilic acid 3,4-dioxygenase; spont., sponta- neous; NAD+, nicotine adenine dinucleotide; KTR, Kynurenine-tryptophan ratio; HK, 3-hydroxykynurenic acid; HAA, 3-hydroxyanthranilic acid. pathogens is impaired with age, while aging is asso- been associated with aging [33, 34]. A study on ciated with a pro-inflammatory state. Monocytes metabolite levels in monocytes after administration and macrophages play a key role in age-associated of IFN-␥ showed that the most significantly increased immune activation [31]. Neopterin is mainly syn- kynurenines were QA and KTR [35]. Taken together, thesized by activated monocytes and [32] and has these data suggest that monocyte activation might 502 L.M. Giil et al. / Kynurenine Pathway, Alzheimer’s Disease, and Aging

Table 5 Comparison with previous studies Studies Greilberger [14] Schwarz l [15]a Gulaj [16] Current studyb Cases, n 16 20 34 65 Controls, n 15 19 18 65 Diagnosis Clinical Clinical Clinical Neuropathology Trp ↓ ns ↓↓ Kyn ns ns ns ↓ HK na ↑ ns ns KA na ns ns ns XA na na ↑↓ AA na na ns ns HAA na na na ↓ QA na ns ↑↓ Neopt ns na na ns KTR ↑ na ↑ ns n, number of participants; ↓, lower in AD; ↑, higher in AD; ns, no significant difference; na, not measured; Trp, Tryptophan; Kyn, Kynurenine; KA, Kynurenic acid; AA, Anthranilic acid; HK, 3-hydroxykynurenine; XA, Xanthurenic acid; HAA, 3-Hydroxyanthranilic acid; QA, Quinolinic acid; KTR, (Kynurenine/Tryptophan)∗1000; Neopt, Neopterin. aControl group had subjective cognitive impairment, were younger and had more males. bThe diagnosis was based on both a clinical diagnosis and postmortem examination of the brain. lead to increased kynurenine levels with age. XA was an increase in QA, AA, KTR, Kyn, neopterin, and the the only kynurenine that showed consistent reduc- PAr index, all related to a pro-inflammatory state, and tions in AD and aging. This is of interest, as the XA was reduced in both AD and with aging. Despite incidence of AD is closely related to age [36]. XA the overall reduction in QA, older AD patients with is a proposed scavenger of free radicals [37] and high QA had poorer cognitive performance. Further oxidative stress increases in aging [33] and AD clinical studies on the kynurenine pathway in AD are [38]. It is possible that XA is consumed due to warranted. elevated levels of free radicals. Interestingly, XA acti- vates metabotropic glutamate receptors (mGlu2/3) ACKNOWLEDGMENTS and XA was protective against psychosis in an ani- mal model. It is reduced in the plasma of patients with We thank the patients and participants of the and their first-degree relatives [39]. A OPTIMA study, and their families and caregivers future perspective could be the study of the psychi- for making this study possible. We also thank atric consequences of reduced XA in AD. Dr. Catherine Joachim for the neuropathology and There were no direct associations between global all support staff in OPTIMA for their help. The cognitive scores and any of the kynurenines, after original OPTIMA cohort was supported by grants multiple testing was adjusted for. However, high from Bristol-Myers Squibb, Medical Research Coun- QA was associated with poor performance on the cil and the Charles Wolfson Charitable Trust. We CamCog test in older patients with AD, but not in thank Haraldsplass Deaconess Hospital for funding a younger patients. QA is a known cerebral excitotoxin, part time research position for the first author. activating NMDA receptors and increasing the phos- Authors’ disclosures available online (http://j-alz. phorylation of tau [40]. Plasma QA is not expected to com/manuscript-disclosures/17-0485r1). enter the brain to any large extent [41]. However, there is impairment of the blood-brain barrier with both age REFERENCES [42] and AD [43], and serum QA levels are correlated with cerebrospinal fluid levels of QA in multiple scle- [1] Querfurth HW, LaFerla FM (2010) Alzheimer’s disease. N Engl J Med 362, 329-344. rosis [44]. Whether QA crosses into the brain in the [2] Heppner FL, Ransohoff RM, Becher B (2015) Immune presence of AD- and age-related blood-brain barrier attack: The role of inflammation in Alzheimer disease. Nat impairment should be addressed in future studies. Rev Neurosci 16, 358-372. In summary, lower plasma concentrations of Kyn, [3] Cai H, Cong WN, Ji S, Rothman S, Maudsley S, Martin B (2012) Metabolic dysfunction in Alzheimer’s disease and HAA, XA, QA, and Trp were found in AD patients related neurodegenerative disorders. Curr Alzheimer Res 9, compared to healthy controls. Aging was associated 5-17. L.M. Giil et al. / Kynurenine Pathway, Alzheimer’s Disease, and Aging 503

[4] Oxenkrug G (2013) Insulin resistance and dysregulation of [21] Faul F, Erdfelder E, Lang AG, Buchner A (2007) G*Power tryptophan-kynurenine and kynurenine-nicotinamide ade- 3: A flexible statistical power analysis program for the nine dinucleotide metabolic pathways. Mol Neurobiol 48, social, behavioral, and biomedical sciences. Behav Res 294-301. Methods 39, 175-191. [5] Ueland PM, McCann A, Midttun O, Ulvik A (2017) Inflam- [22] Benjamini Y, Yekutieli D (2001) The control of the false mation, vitamin B6 and related pathways. Mol Aspects Med discovery rate in multiple testing under dependency. Ann 53, 10-27. Stat 1165-1188. [6] Myint AM (2012) Kynurenines: From the perspective of [23] Sakia RM (1992) The Box-Cox Transformation Technique: major psychiatric disorders. FEBS J 279, 1375-1385. A Review. J Roy Stat Soc. Series D (The Statistician) 41, [7] Chen Y, Guillemin GJ (2009) Kynurenine pathway metabo- 169-178. lites in humans: Disease and healthy States. Int J Tryptophan [24] Bring J (1994) How to standardize regression coefficients. Res 2, 1-19. Am Stat 48, 209-213. [8] Majewski M, Kozlowska A, Thoene M, Lepiarczyk E, Grze- [25] Austin PC (2014) A comparison of 12 algorithms for match- gorzewski WJ (2016) Overview of the role of vitamins and ing on the propensity score. Stat Med 33, 1057-1069. minerals on the kynurenine pathway in health and disease. [26] Sonis J (1998) A closer look at confounding. Fam Med 30, J Physiol Pharmacol 67, 3-19. 584-588. [9] Theofylaktopoulou D, Midttun O, Ulvik A, Ueland PM, Tell [27] Yohai VJ (1987) High breakdown-point and high efficiency GS, Vollset SE, Nygard O, Eussen SJ (2013) A community- robust estimates for regression. Ann Stat 15, 642-656. based study on determinants of circulating markers of [28] R Development Core Team (2016) R Foundation for Statis- cellular immune activation and kynurenines: The Hordaland tical Computing. Health Study. Clin Exp Immunol 173, 121-130. [29] Ueland PM, Ulvik A, Rios-Avila L, Midttun O, Gregory [10] Pawlak D, Tankiewicz A, Mysliwiec P, Buczko W (2002) JF (2015) Direct and functional biomarkers of vitamin B6 Tryptophan metabolism via the kynurenine pathway in status. Annu Rev Nutr 35, 33-70. experimental chronic renal failure. Nephron 90, 328-335. [30] Matsunaga S, Kishi T, Iwata N (2015) Memantine [11] Maddison DC, Giorgini F (2015) The kynurenine pathway monotherapy for Alzheimer’s disease: A systematic review and neurodegenerative disease. Semin Cell Dev Biol 40, and meta-analysis. PLoS One 10, e0123289. 134-141. [31] Oishi Y, Manabe I (2016) Macrophages in age-related [12] Dezsi L, Tuka B, Martos D, Vecsei L (2015) Alzheimer’s chronic inflammatory diseases. NPJ Aging Mech Dis 2, disease, astrocytes and kynurenines. Curr Alzheimer Res 12, 16018. 462-480. [32] De Rosa S, Cirillo P, Pacileo M, Petrillo G, D’Ascoli [13] Hartai Z, Juhasz A, Rimanoczy A, Janaky T, Donko T, GL, Maresca F, Ziviello F, Chiariello M (2011) Neopterin: Dux L, Penke B, Toth GK, Janka Z, Kalman J (2007) From forgotten biomarker to leading actor in cardiovascular Decreased serum and red blood cell kynurenic acid levels pathophysiology. Curr Vasc Pharmacol 9, 188-199. in Alzheimer’s disease. Neurochem Int 50, 308-313. [33] Capuron L, Geisler S, Kurz K, Leblhuber F, Sperner- [14] Greilberger J, Fuchs D, Leblhuber F, Greilberger M, Win- Unterweger B, Fuchs D (2014) Activated immune system tersteiger R, Tafeit E (2010) Carbonyl proteins as a clinical and inflammation in healthy ageing: Relevance for tryp- marker in Alzheimer’s disease and its relation to tryptophan tophan and neopterin metabolism. Curr Pharm Des 20, degradation and immune activation. Clin Lab 56, 441-448. 6048-6057. [15] Schwarz MJ, Guillemin GJ, Teipel SJ, Buerger K, Hampel [34] Oxenkrug G (2011) Interferon-gamma - inducible inflam- H (2013) Increased 3-hydroxykynurenine serum concen- mation: Contribution to aging and aging-associated trations differentiate Alzheimer’s disease patients from psychiatric disorders. Aging Dis 2, 474-486. controls. Eur Arch Psychiatry Clin Neurosci 263, 345-352. [35] Jones SP, Franco NF, Varney B, Sundaram G, Brown DA, [16] Gulaj E, Pawlak K, Bien B, Pawlak D (2010) Kynurenine de Bie J, Lim CK, Guillemin GJ, Brew BJ (2015) Expres- and its metabolites in Alzheimer’s disease patients. Adv Med sion of the kynurenine pathway in human peripheral blood Sci 55, 204-211. mononuclear cells: Implications for inflammatory and neu- [17] Mirra SS, Heyman A, McKeel D, Sumi SM, Crain BJ, rodegenerative disease. PLoS One 10, e0131389. Brownlee LM, Vogel FS, Hughes JP, van Belle G, Berg [36] Kawas C, Gray S, Brookmeyer R, Fozard J, Zonderman A L (1991) The Consortium to Establish a Registry for (2000) Age-specific incidence rates of Alzheimer’s disease: Alzheimer’s Disease (CERAD). Part II. Standardization The Baltimore Longitudinal Study of Aging. Neurology 54, of the neuropathologic assessment of Alzheimer’s disease. 2072-2077. Neurology 41, 479-486. [37] Perez-Gonzalez A, Alvarez-Idaboy JR, Galano A (2015) [18] Clarke R, Smith AD, Jobst KA, Refsum H, Sutton L, Ueland Free-radical scavenging by tryptophan and its metabolites PM (1998) Folate, vitamin B12, and serum total homocys- through electron transfer based processes. J Mol Model 21, teine levels in confirmed Alzheimer disease. Arch Neurol 213. 55, 1449-1455. [38] Galasko D, Montine TJ (2010) Biomarkers of oxidative [19] Roth M, Tym E, Mountjoy CQ, Huppert FA, Hendrie H, damage and inflammation in Alzheimer’s disease. Biomark Verma S, Goddard R (1986) CAMDEX. A standardised Med 4, 27-36. instrument for the diagnosis of in the elderly [39] Fazio F, Lionetto L, Curto M, Iacovelli L, Cavallari M, with special reference to the early detection of dementia. Br Zappulla C, Ulivieri M, Napoletano F, Capi M, Corigliano J Psychiatry 149, 698-709. V, Scaccianoce S, Caruso A, Miele J, De Fusco A, Di [20] Midttun O, Hustad S, Ueland PM (2009) Quantitative pro- Menna L, Comparelli A, De Carolis A, Gradini R, Nistico filing of biomarkers related to B-vitamin status, tryptophan R, De Blasi A, Girardi P, Bruno V, Battaglia G, Nicoletti metabolism and inflammation in human plasma by liquid F, Simmaco M (2015) Xanthurenic acid activates mGlu2/3 chromatography/tandem mass spectrometry. Rapid Com- metabotropic glutamate receptors and is a potential trait mun Mass Spectrom 23, 1371-1379. marker for schizophrenia. Sci Rep 5, 17799. 504 L.M. Giil et al. / Kynurenine Pathway, Alzheimer’s Disease, and Aging

[40] Lugo-Huitron R, Ugalde Muniz P, Pineda B, Pedraza- Blood-brain barrier breakdown in the aging human hip- Chaverri J, Rios C, Perez-de la Cruz V (2013) Quinolinic pocampus. Neuron 85, 296-302. acid: An endogenous neurotoxin with multiple targets. Oxid [43] Erickson MA, Banks WA (2013) Blood-brain barrier dys- Med Cell Longev 2013, 104024. function as a cause and consequence of Alzheimer’s disease. [41] Fukui S, Schwarcz R, Rapoport SI, Takada Y, Smith QR J Cereb Blood Flow Metab 33, 1500-1513. (1991) Blood-brain barrier transport of kynurenines: Impli- [44] Lim CK, Bilgin A, Lovejoy DB, Tan V, Bustamante S, cations for brain synthesis and metabolism. J Neurochem Taylor BV, Bessede A, Brew BJ, Guillemin GJ (2017) 56, 2007-2017. Kynurenine pathway metabolomics predicts and provides [42] Montagne A, Barnes SR, Sweeney MD, Halliday MR, mechanistic insight into progression. Sci Sagare AP, Zhao Z, Toga AW, Jacobs RE, Liu CY,Amezcua Rep 7, 41473. L, Harrington MG, Chui HC, Law M, Zlokovic BV (2015)