Effects of neuroactive metabolites of the tryptophan pathway on working memory and cortical thickness in schizophrenia

A number of tryptophan metabolites known to be neuroactive have been examined for their potential associations with cognitive deficits in schizophrenia. Among these metabolites, kynurenic acid (KYNA), 5-hydroxyindole (5-HI), and quinolinic acid (QUIN) are documented in their diverse effects on α-7 nicotinic acetylcholine receptor (α7nAChR) and/or N-methyl-D-aspartate receptor (NMDAR), two of the receptor types thought to contribute to cognitive impairment in schizophrenia. In this study, serum levels of KYNA, 5-HI, and QUIN were measured in 195 patients with schizophrenia and in 70 healthy controls using liquid chromatography-tandem mass spectrometry; cognitive performance in MATRICS Consensus Cognitive Battery and cortical thickness measured by magnetic resonance imaging were obtained. Patients with schizophrenia had significantly lower serum KYNA (p < 0.001) and QUIN (p = 0.02) levels, and increased 5-HI/KYNA (p < 0.001) and QUIN/KYNA ratios (p < 0.001) compared with healthy controls. Multiple linear regression showed that working memory was positively correlated with serum 5-HI levels (t = 2.10, p = 0.04), but inversely correlated with KYNA concentrations (t = −2.01, p = 0.05) in patients. Patients with high 5-HI and low KYNA had better working memory than other subgroups (p = 0.01). Higher 5-HI levels were associated with thicker left lateral orbitofrontal cortex (t = 3.71, p = 2.94 × 10−4) in patients. The different effects of 5-HI and KYNA on working memory may appear consistent with their opposite receptor level mechanisms. Our findings appear to provide a new insight into the dynamic roles of tryptophan pathway metabolites on cognition, which may benefit novel therapeutic development that targets cognitive impairment in schizophrenia.


Introduction
The tryptophan pathway has been increasingly targeted in drug discovery efforts for treating cognitive impairments in neuropsychiatric conditions including schizophrenia 1,2 . One of the metabolites in this pathway is kynurenic acid (KYNA), which has attracted considerable interest, as it is a noncompetitive antagonist of α-7 nicotinic acetylcholine receptor (α7nAChR) 3,4 that has been linked to cognitive functions, in particular working memory and attention 5,6 . Accumulated evidences suggest that schizophrenia is associated with abnormal α7nAChR-mediated neurotransmission 7,8 and cortical KYNA contributes to cognitive impairment through the α7nAChR mechanism 3,[9][10][11] . KYNA also antagonizes the glycine-binding site of the N-methyl-Daspartate receptor (NMDAR) 4,12 and NMDAR dysfunction has been linked to cognition, especially working memory deficits in schizophrenia [13][14][15][16] .
5-HI and KYNA may also diverge on their roles in cognition, as they may modulate γ-aminobutyric acid (GABA)ergic neurotransmission differently, as 5-HI facilitates GABAergic transmission via excitation of α7nAChR 22 , but KYNA concentration-dependently reduces GABA levels by modulating α7nAChR function 23 . GABAergic dysfunction has also been consistently linked to working memory deficits in schizophrenia [24][25][26] .
Notably, a third metabolite in the tryptophan pathway, quinolinic acid (QUIN), is an agonist of NMDAR 27,28 . QUIN is an excitotoxin that promotes neurodegeneration and neuroinflammation, and may negatively impact cognition [29][30][31] . Therefore, 5-HI, KYNA, and QUIN are three neuroactive metabolites within the tryptophan pathway, which may potentially influence cognition, although their in vivo combined effects are likely complex and are currently unknown. The effort of targeting the tryptophan pathway for treating cognitive deficits in schizophrenia should first understand the independent vs. joint effects of these neuroactive metabolites. Accordingly, we tested the hypothesis that working memory in schizophrenia may be dynamically dependent on the balance among 5-HI, KYNA, and QUIN. As these metabolites are known to influence α7nAChR and NMDAR 32,33 , which are distributed in high densities in the cortex 34,35 , we will test a second hypothesis that they may also counteractively impact the cortical thickness in patients with schizophrenia.

Participants
The study recruited 195 patients with schizophrenia (all in-patients) and 70 healthy controls (HCs). Patients who met the diagnostic criteria of schizophrenia according to the Structured Clinical Interview of the Diagnostic and Statistical Manual of Mental Disorders-IV were enrolled during their hospitalization at the Beijing Huilongguan Hospital. HCs were recruited through local advertisements at nearby communities and were excluded if they had a history of psychiatric disorders or psychosis among their first-degree relatives. Participants were also excluded if they had a lifetime history of any other Axis I disorders, head trauma, current or previous substance or alcoholism dependence (except nicotine), or systemic diseases including neurological disorders, organic brain disorders, and unstable medical illnesses. One hundred and twelve patients were antipsychotic medication-free at the time of admission and blood sample collection; 19 patients were on a first-generation antipsychotic alone or combined with second-generation antipsychotics; and the remaining patients were on the following second-generation antipsychotics: risperidone (14), clozapine (5), olanzapine (5), and aripiprazole (4). Thirty-six patients were on two or more second-generation antipsychotics and four patients were also on sodium valproate. Patients received antipsychotic medications once hospitalized and, most initially, medication-free patients were on antipsychotic medications during brain imaging, which was on average 5.5 ± 2.3 days after the initial blood draw. The current antipsychotic medication dose (based on time of blood draw) was calculated as a chlorpromazine (CPZ)-equivalent dose 36,37 (Table 1). This study was approved by the ethics committee and institutional review board of Beijing Huilongguan Hospital. All participants provided written informed consent.

Clinical and cognitive assessments
Clinical symptoms were evaluated by the Positive and Negative Syndrome Scale (PANSS) in patients of schizophrenia by one of the three attending psychiatrists. The inter-rater intra-class correlation coefficient among the raters was above 0.80. Cognitive function was assessed using the validated Chinese version of the MATRICS Consensus Cognitive Battery (MCCB) [38][39][40] . The MCCB contains assessments of seven cognitive domains: Working Memory, Speed of Processing, Attention and Vigilance, Verbal Learning, Visual Learning, Reasoning and Problem Solving, and Social Cognition. Initial scores were converted to domain scores. Impaired working memory was evidenced a pervasive and the core of schizophreniarelated cognitive disability 41,42 ; thus, our primary focus was on the working memory domain.

Biochemistry
Blood samples were collected with BD Vacutainer serum tubes in the morning after overnight fasting and centrifuged immediately at 4°C for 10 min at 3000 r.p.m.
Serum was then aliquoted into separate tubes and stored at −80°C. High-performance liquid chromatography was used to separate serum KYNA, 5-HI, and QUIN, and tandem mass spectrometry was then performed to quantify them using standard protocols. Details of the 5-HI assay, quality control, and validations were in Supplementary Material 1. Details of the KYNA and QUIN assays were described as previous reports [43][44][45] . The intratrial coefficient of variations for high, median, and low quality were 1.2%, 2.7%, and 1.5% for 5-HI; 4.0%, 12.5%, and 8.8% for KYNA; and 4.1%, 6.2%, and 7.4% for QUIN, respectively. We further calculated the 5-HI/KYNA ratio and the QUIN/KYNA ratio to explore their putative agonistic/antagonistic effects on α7nAChR and NMDAR, respectively.

Image processing
Structural images were collected in 153 schizophrenia patients and 65 HCs using a 3.0 Tesla Prisma MRI scanner (Siemens, Germany) and a 64-channel head coil at the MRI Research Center of the Beijing Huilongguan Hospital. Parameters of the three-dimensional magnetization prepared rapid acquisition gradient echo sequence were as follows: echo time = 2.98 ms, inversion time = 1100 ms, repetition time = 2530 ms, flip angle = 7°, field of view = 256 × 224 mm 2 , matrix size = 256 mm × 224 mm, and thickness/gap = 1/0 mm covering the whole brain. Participants used earplugs and foam pads to reduce magnetic machine noise and head movement, and who were reminded not to move their head during the procedure. A computer-connected monitor was acquired to detect head movement distance. The automated and validated segmentation were conducted by using Free-Surfer v5.3 (http://surfer.nmr.mgh.harvard.edu/) 46 . Thirty-four cortical gray matter regions in each hemisphere were extracted as regions of interest for statistical analysis according to the Desikan-Killiany atlas 47,48 .

Statistical analysis
Normality of raw data was assessed with normal Q-Q plots and Shapiro-Wilkinson tests. Serum 5-HI, KYNA, QUIN levels, and 5-HI/KYNA ratio all deviated from normal distribution; all were normalized after natural logarithm (ln) or square root (sqrt) transformation. Ttests or χ 2 -test were used to compare the demographic data of patients with schizophrenia and HCs. Separate univariate analyses of covariance (ANCOVAs) were used to determine between-group differences in metabolites, MCCB and PANSS subdomain, and total scores, with sex and age as covariates. Multiple linear regression model was used to test whether working memory score was associated with the serum levels of 5-HI, KYNA, and QUIN in schizophrenia patients, adjusted for age and sex. Significant but opposite directions of associations of the metabolites with working memory would imply opposite roles. For those metabolites having significant opposite effects on working memory, we further divided patients into high-(concentration equal to or above the median) and low-(concentration below the median) level subgroups, and compared their working memory performance, adjusted for age and sex. Associations between cortical thickness of 68 cortical regions and serum levels of 5-HI, KYNA, and QUIN were further explored by multiple linear regression adjusted by age and sex in patients. These analyses were also repeated in HCs. The effects of smoking status and psychotropic medications on the three metabolites were also explored. Significance was set at p < 0.05 in all tests.

Relationship of working memory with 5-HI, KYNA, and QUIN
The multiple linear regression model using the three metabolites as predictors was significant (F = 3.18, p = 0.01), wherein working memory was significantly predicted by higher level of 5-HI (t = 2.10, p = 0.04) and lower KYNA (t = −2.01, p = 0.05) ( Table 2). The results were not statistically significant, although 5-HI (t = 0.94, p = 0.35) and KYNA (t = −1.48, p = 0.14) had the same trend in the medication-free group, similar to findings in the total-patient group. One possible reason is that each divided group may have lower statistic power than the total-patient group. The results were still significant (F = 10.11, p < 0.001) when we added smoke and CPZ equivalent in the model, and 5-HI (t = 2.18, p = 0.03) and KYNA (t = −2.24, p = 0.03) have the same opposite trends as before. The model was insignificant in HCs (model p = 0.83). We also explored other MCCB domains and total score but found no significant results in either patients or controls (Supplementary Table 2). The relationships of working memory with 5-HI/KYNA and QUIN/KYNA were also explored. 5-HI/KYNA was positively associated with working memory (t = 2.74, p = 0.007) in patients, but not in HCs (Fig. 2). There was no significant correlation with QUIN/KYNA in either patients or controls.
We further divided the patients into high-and low-level in four subgroups based on medium splits for 5-HI and KYNA: high 5-HI/low KYNA, low 5-HI/low KYNA, high 5-HI/high KYNA, and high 5-HI/low KYNA (Fig. 3). The overall ANCOVA was significant (F = 3.81, p = 0.01). Post-hoc tests showed that patients with high 5-HI and low KYNA had the best working memory performance compared with the other three subgroups (p = 0.01-0.004); however, the other three subgroups did not significantly differ. The model was not significant in HCs (F = 0.20, p = 0.89).

Relationship of cortical thickness with 5-HI, KYNA, and QUIN
Multiple cortical regions showed significantly reduced cortical thickness in patients compared to HCs (Supplementary Table 3). We next explored the associations of 5-HI, KYNA, and QUIN with thickness of 68 cortical regions after controlling for sex and age. Only the left lateral orbitofrontal cortex (LOFC) showed significantly positive association with the 5-HI levels (t = 3.71, p = 2.94 × 10 −4 ) after Bonferroni correction for multiple comparisons (0.05/68 = 7.35 × 10 −4 ) (Fig. 4). However, we also observed nominally significant (p < 0.05 uncorrected) positive associations between 5-HI and the left and right frontal poles, left insula, right caudal-middle frontal, right lateral orbitofrontal, and right lingual areas in patients with schizophrenia (p = 0.03-0.004), whereas the right lateral occipital region had a nominally negative association with 5-HI (p = 0.05) (Supplementary Table 4). In medication-free patients, 5-HI was still correlated with LOFC thickness (r = 0.39, p = 1.4 × 10 −4 ). There were no  significant associations between cortical thickness and 5-HI in HCs. No significant associations between cortical thickness and KYNA, QUIN, or their ratios were found in patients or controls after correction for multiple comparisons (Supplementary Table 4).

Discussion
This study examined the effects of 5-HI, KYNA, and QUIN on working memory and cortical structures, based on preclinical evidence of their effects on α7nAChR and NMDAR. In patients with schizophrenia, KYNA and QUIN, but not 5-HI, were lower than in HCs. Working memory was significantly influenced by 5-HI and KYNA but in opposite direction in the patients. Patients exhibiting high 5-HI and low KYNA concentrations had better working memory performance than the other subgroups. Finally, 5-HI level was positively associated with cortical thickness of the left orbitofrontal cortex, which was significantly associated with working memory performance in the patients. 5-HI is a relatively potent tryptophan metabolite that increases glutamate release and the function of GABA interneurons 21 , and in high dose can cause convulsion 49 . 5-HI precursor indole is transformed from tryptophan by tryptophanase in indole-producing bacteria of human gut 50,51 . Diverse oxygenase could degrade indole to indole derivatives 52,53 such as 5-HI. The formation of 5-HI can be observed by incubating rat liver homogenates with its direct precursor indole, supporting an endogenous production of 5-HI 21 . In rats with hepatic encephalopathy, indole produced by gut bacteria is absorbed and metabolized into several metabolites including 5-HI, which may accumulate in the blood and brain 17,21,54 . Earlier clinical research has suggested that blood 5-HI may provide treatment effect monitoring for hyperactive behaviors, provided some initial support for its potential clinical utility 55 . Previous rodent studies indicated that 5-HI activates both presynaptic and postsynaptic α7nAChR Fig. 4 The relationship of serum levels of 5-HI with cortical thickness in patients with schizophrenia. The left and right frontal poles, left insula, right caudal-middle frontal, left and right lateral orbitofrontal, and right lingual areas were positively and significantly associated with 5-HI at p = 0.03-0.0003, and the right lateral occipital region was inversely associated with 5-HI at p = 0.05, but only the lateral orbitofrontal cortex (arrow) was significant after correction for multiple comparisons. A Cortical regions were colored according to t-scores by multiple linear regression. The color bar represents t-scores. B Scatter plot between 5-HI and the left lateral orbitofrontal cortex (arrow in A) in patients with schizophrenia.
to mediate glutamate release 18,19,22 , which may facilitate working memory. In addition, 5-HI increases acetylcholine (ACh) efficacy though Ca 2+ currents, indicating that 5-HI and ACh may cooperate to influence α7nAChR 18 . Regardless of whether our finding here is related to α7nAChR, the data provided the first evidence in humans, supporting the hypothesis that 5-HI may facilitate working memory performance in patients with schizophrenia.
Reduced serum KYNA concentration in schizophrenia was highly replicable in several recent reports [56][57][58] , which were not consistent with several earlier studies 59,60 . Our results are also inconsistent with studies showing elevated KYNA in the post-mortem brains of patients with schizophrenia 4,45,61 . As KYNA normally does not pass the BBB 62 , it is unclear whether the observed inverse correlation between serum KYNA levels and working memory is due to these brain mechanisms; as such, the KYNArelated results should be viewed with caution. However, impaired BBB has been proposed in patients with schizophrenia 63,64 , which may explain this significant correlation only in patients but not in controls. Moreover, evidences showed peripheral kynurenine can pass through BBB 62,65 and may influence brain functions 2,66 . Preclinical studies have demonstrated that KYNA negatively affects α7nAChR-dependent presynaptic mechanisms in the prefrontal cortex and disrupts local GABAergic synaptic signaling 67 . Knocking out kynurenine aminotransferase II that decreases endogenous brain KYNA can increase working memory performance 68,69 . At high concentrations, KYNA is a competitive antagonist of NMDAR 70,71 ; however, at low concentrations, KYNA has a more potent inhibitory effect on α7nAChR activation than on NMDAR 4 . Furthermore, reducing KYNA increases 5-HIdependent activation of α7nAChR 19 . Therefore, our data appear as corroborating these preclinical data on the opposing 5-HI vs. KYNA effects, showing that patients having high 5-HI and low KYNA was associated with the best working memory.
However, why patients have reduced serum KYNA but there is an inverse correlation between KYNA and working memory is difficult to interpret. Similar to our finding, significantly low peripheral serum kynurenine and/or KYNA levels have been found in diverse cohorts of patients [56][57][58]72 . Meanwhile, higher KYNA in the brain is thought to impair cognitive functions 2 , which appears consistent with our finding of an inverse correlation between peripheral KYNA and working memory. We believe that the difficulty to explain the seemingly contradictory findings is in part due to the lack of a good explanation on why peripheral kynurenine and/or KYNA are reduced in schizophrenia, while brain KYNA appears increased in schizophrenia, compared with that in HCs. Further complicating the issue is that about 80% of kynurenine and KYNA in the blood were bound to albumin or other circulating binding proteins 62,73 , which may lead to differential availability of free kynurenine and KYNA, and our study is limited by not measuring specifically the free serum kynurenine or KYNA. Basic neuroscience effort to simultaneously assessing central and peripheral kynurenine and KYNA, while invasively studying the potential mechanism underlying the central-peripheral metabolite relationships may be needed.
We observed that serum QUIN concentration was significantly lower in patients with schizophrenia than in HCs. Previous studies attempted to identify QUIN-related abnormalities in schizophrenia but largely failed to show a significant differences in QUIN levels between patients and controls in post-mortem brain tissues 45 , blood 57 , or CSF 74 . However, QUIN normally also does not pass the BBB 62 and our analysis did not show significant relationship between QUIN and working memory or other clinical and cognitive measures.
An association between lower 5-HI to thinner cortical thickness of the LOFC in patients is intriguing. Previous researches indicated that the orbitofrontal cortex supports working memory 75 , specifically encoding gustatory 75 , emotional 76 , and abstract information 77 . The LOFC is particularly important for reward learning 78 , a process closely related to working memory 79 . We consider the finding linking 5-HI to the thickness of the LOFC a further supportive evidence of a potential cognition enhancement effect of this metabolite in patients. Also some evidences showed relationships between kynurenine and subcortical volumes in mental disorders that KYNA/ 3-hydroxykynurenine were positively correlated to the hippocampal volume in bipolar disorder 80 and negatively correlated with the left hippocampal activity in major depressive disorder 81 . We further explored the associations between subcortical volumes, and KYNA and KYNA/QUIN. However, there was no significant results in patients or in HCs (Supplementary Material 6).
The study has a number of limitations. We did not test whether the effects of these metabolites occurred through α7nAChR, NMDAR, or other receptor mechanisms in the brain, although 5-HI is lipophilic and may be BBB permeable 20 . However, KYNA and QUIN do not normally pass the BBB, making the blood-based finding or the lack thereof difficult to explain. We also did not measure the 5-HI precursor indole (Fig. 1) to rule out its contributions to the observations here. However, indole itself was thought not to interact with GABAergic or ionotropic glutamate receptors 17 . Furthermore, we measured 5hydroxyoxindole (5-HOI), another metabolite from the indole and oxindole branch (Fig. 1), to confirm that the 5-HI results reported here is distinct from 5-HOI (Supplementary Material 7). We also did not measure the gut flora that possibly influence the concentration of indole and further related to 5-HI level. The inflammatory markers were not tested in the present study, which may limit the interpretation of relationship between metabolites, as high inflammatory activity can lead to elevations of kynurenine and KYNA [82][83][84] . We also did not measure diet and physical activities that may have an effect on peripheral levels of KYNA 85,86 , cortical thickness 87 , and cognition 88 , which is another limitation. Despite these limitations, the observed pattern of higher working memory in patients with high 5-HI and low KYNA appears consistent with the directions of their mechanism of action 19 .
In summary, our findings suggest that there appears multiple potential mechanisms by which the tryptophan pathway is relevant to the cognitive performance in schizophrenia, including the indole branch whose relationship to schizophrenia has not been previously studied. Therefore, the finding that 5-HI and KYNA may have opposing effects on working memory among patients with schizophrenia is new and is consistent with preclinical evidences on their counteractive mechanisms. Our data may provide new insight into potential targets in the tryptophan pathway in our effort to develop novel therapeutic strategy for treating working memory and other cognitive impairment in schizophrenia.