C-Reactive Protein and the Kynurenic Acid to Quinolinic Acid Ratio are Independently Associated with White Matter Integrity in Major Depressive Disorder

Abstract

Kynurenic acid (KynA) and quinolinic acid (QA) are neuroactive kynurenine pathway (KP) metabolites that have neuroprotective and neurotoxic properties, respectively. At least partly as a result of immune activation, the ratio of KynA to QA in the blood is reduced in major depressive disorder (MDD) and has been reported to be positively correlated with gray matter volume in depression. This study examined whether the inflammatory mediator, C-reactive protein (CRP) and the putative neuroprotective index, KynA/QA, were associated with white matter integrity in MDD, and secondly, whether any such associations were independent of each other or whether the effect of CRP was mediated by KynA/QA. One hundred and sixty-six participants in the Tulsa 1000 study with a DSM-V diagnosis of MDD completed diffusion tensor imaging and provided a serum sample for the quantification of CRP, KynA, and QA. Correlational tractography was performed using DSI Studio to map the specific white matter pathways that correlated with CRP and KynA/QA. CRP was negatively related to KynA/QA (standardized beta coefficient, SBC=−0.35 with standard error, Std.E=0.13, p < 0.01) after controlling for nine possible confounders, i.e., age, sex, body mass index (BMI), medication status, lifetime alcohol use, severity of depression, severity of anxiety, length of illness, and smoking status . Higher concentrations of CRP were associated with decreased white matter integrity (fractional anisotropy, FA) of the bilateral cingulum and fornix after controlling for the nine potential confounders (SBC= −0.43, Std.E=0.13, p = 0.002). Greater serum KynA/QA was associated with increased white matter integrity of the bilateral fornix, bilateral superior thalamic radiations, corpus callosum, and bilateral cingulum bundles after controlling for the same possible confounders (SBC= 0.26, Std.E=0.09, p = 0.005). The relationship between CRP and FA was not mediated by KynA/QA. Exploratory analyses also showed that KynA/QA but not CRP was associated with self-reported positive affect, attentiveness, and fatigue measured with the PANASX (SBCs = 0.17-0.23). Taken together, these results are consistent with the hypothesis that within a subgroup of MDD patients, a higher level of systemic inflammation alters the balance of KP metabolism but also raise the possibility that CRP and neuroactive KP metabolites represent independent molecular mechanisms underlying white matter alterations in MDD.

1. Introduction

Evidence from diverse fields suggests that in a subset of patients, inflammatory processes play a role in the pathophysiology of major depressive disorder (MDD) (Capuron and Miller, 2004; Dantzer et al., 2008; Miller and Raison, 2016). C-reactive protein (CRP), one of the most commonly-used measures of systemic inflammation in clinical practice, has been shown in multiple meta-analyses or large-scale studies to be elevated on average in MDD populations relative to healthy controls (Haapakoski et al., 2015; Horn et al., 2018; Howren et al., 2009; Osimo et al., 2019; Pitharouli et al., 2021). These data raise the question of whether the structural brain alterations associated with depression (Brandl et al., 2022; Savitz and Drevets, 2009; van Velzen et al., 2020) are related to elevations in circulating CRP. There is some evidence for this hypothesis in the form of cross-sectional associations between higher CRP concentrations and lower gray matter volume or cortical thickness in the context of depression (Doolin et al., 2018; Green et al., 2021; Han and Ham, 2021; Meier et al., 2016a; Opel et al., 2019; van Velzen et al., 2017). Less is known about the relationship between CRP and white matter abnormalities (Han and Ham, 2021) although we recently showed that higher circulating CRP concentrations were associated with widespread reductions in white matter integrity measured using quantitative anisotropy in MDD participants (Thomas et al., 2021).

The mechanistic pathways underlying these immune-brain relationships are still unclear. Inflammatory processes alter the breakdown of tryptophan into kynurenine and ultimately, the pattern of metabolism within the kynurenine pathway (KP) (Dantzer et al., 2011). We and others have previously hypothesized that the generation of neuroactive KP metabolites may be one pathway through which inflammatory processes impact brain structure and function in the context of depression (Muller et al., 2009; Myint and Kim, 2014; Savitz, 2020; Savitz et al., 2015c). Approximately 95% of tryptophan is broken down into kynurenine (Bender, 1983). Kynurenine is in turn metabolized along several different pathways to produce various metabolites often loosely termed “the kynurenines” (Figure S1). Our laboratory has principally focused on two neuroactive KP metabolites that have opposing effects on the N-methyl-D-aspartate (NMDA) receptor, the NMDA receptor antagonist, kynurenic acid (KynA), and the NMDA receptor agonist, quinolinic acid (QA) (Foster et al., 1984b; Stone and Perkins, 1981). While KynA is thought to have neuroprotective properties (Foster et al., 1984b), QA is a known neurotoxin that exerts its effects through multiple mechanisms including direct activation of the NMDA receptor, increasing glutamate release while inhibiting its reuptake by astrocytes, the formation of reactive oxygen and nitrogen species, phosphorylating tau and destabilizing the cellular cytoskeleton, inducing the expression of pro-inflammatory cytokines, and interfering with autophagy (Guillemin, 2012).

Elevated QA has been implicated in multiple inflammatory and neurological disorders (Georgin-Lavialle et al., 2016; Heyes et al., 2001; Lee et al., 2017; Lim et al., 2017; Lovelace et al., 2016; Thirtamara-Rajamani et al., 2017; Zadori et al., 2018). Because QA is a precursor of nicotinamide adenine dinucleotide (NAD+), which as an electron donor plays a critical role in oxidative phosphorylation and glycolysis, increased metabolism down the QA pathway is thought to be an adaptation to help meet the increased energy needs of activated immune cells (Savitz, 2020). At a certain threshold, however, the enzyme that converts QA to nicotinic acid mononucleotide and ultimately NAD+ (quinolinic acid phosphoribosyltransferase, QPRT) may become saturated leading to the toxic accumulation of QA (Braidy et al., 2011; Jones et al., 2015). The link between inflammation and cellular energy use may explain why we and others have consistently reported decreases in circulating concentrations of KynA relative to QA (KynA/QA) in individuals with depression compared with healthy controls (Bartoli et al., 2021; Doolin et al., 2018; Myint et al., 2007; Paul et al., 2022; Savitz et al., 2015a; Savitz et al., 2015d; Schwieler et al., 2016; Wurfel et al., 2017). Similar changes have also been found in the cerebrospinal fluid (CSF) (Bay-Richter et al., 2015) and postmortem brain (Steiner et al., 2011) of suicide attempters and severely depressed individuals, respectively.

Thus, a prevailing hypothesis is that depression - or perhaps an inflammatory subtype of depression - is characterized by an imbalance in KP metabolism in favor of QA, and that this imbalance has a plethora of adverse neurobiological consequences ranging from excitotoxicity to impaired neuroplasticity (Guillemin, 2012; Muller et al., 2009; Myint and Kim, 2014; Savitz, 2020; Tsuchiyagaito et al., 2021). This hypothesis is based in part on animal studies demonstrating that QA expression is increased in the brain following various types of neuroinflammatory insults and that damage to the brain can be mitigated by blocking metabolism down the QA pathway with the use of kynurenine monooxygenase (KMO) inhibitors (Breda et al., 2016; Campesan et al., 2011; Chiarugi et al., 2001; Chung et al., 2009; O’Farrell et al., 2017; Sundaram et al., 2020; Zakhary et al., 2020; Zwilling et al., 2011). Consistent with these preclinical studies, we previously found that reductions in KynA relative to QA-pathway metabolites were associated with reduced amygdalar and hippocampal volumes in individuals with MDD and bipolar depression (Savitz et al., 2015b; Savitz et al., 2015c). Similar findings were subsequently reported by other groups (Doolin et al., 2018; Paul et al., 2022). Moreover, we extended these results to the medial prefrontal cortex (mPFC) (Meier et al., 2016a) and also showed that serum KynA/QA was positively correlated with hippocampal volume in concussed football players with symptoms of depression (Meier et al., 2016b). In contrast, little is known about the relationship between KP metabolism and white matter integrity. To our knowledge, only one study has addressed this relationship in the context of mood disorders. Poletti and colleagues found that lower serum KynA was associated with reduced integrity (elevated radial diffusivity and mean diffusivity) of several white matter tracts in participants with bipolar disorder (BD) although QA was not measured in this study (Poletti et al., 2018).

This study had three main goals. First, to replicate our previous report of a negative relationship between circulating CRP concentrations and white matter integrity using Fractional Anisotropy (FA) derived from a single b-value acquisition. Second, to evaluate the relationship between serum KynA/QA and white matter integrity. Third, to determine if the putative effects of CRP and KynA/QA on white matter integrity are independent of each other or if KynA/QA mediates the relationship between CRP and white matter integrity. The main hypotheses were (1) that there would be a negative relationship between CRP and white matter integrity; (2) that there would be a positive relationship between KynA/QA and white matter integrity, and (3) that KynA/QA would mediate the relationship between CRP and white matter integrity. Given the dearth of studies in this area we opted for a whole brain approach with appropriate statistical correction for multiple comparisons rather than prespecified region of interest (ROI) analyses. The study was conducted using correlational tractography (Yeh et al., 2021), a novel tractography modality that tracks the voxelwise correlation between FA and study variables of interest. The statistical significance was calculated using a nonparametric inference that takes multiple comparisons into account (Yeh et al., 2016). In exploratory analyses, the relationships between CRP, KynA/QA and affect measured with the positive and negative affect schedule-expanded form (PANASX) (Watson and Clark, 1994) were evaluated. We also tested whether CRP had a moderating effect on the relationship between KynA/QA and FA and whether PANASX scores were related to FA measures (see supplemental data).

2. Methods

2.1. Participants

This study included 166 participants aged 18–55 years who received a DSM-5 diagnosis of MDD (with or without comorbid anxiety). Participants were drawn from the first half of the Tulsa 1000 (T1000) study based on the availability of diffusion weighted imaging data and kynurenine metabolite data (Victor et al., 2018). There were 244 MDD subjects in this cohort. KP metabolite data were available from 190 out of 244 MDD participants. An additional 24 MDD participants who did not have sufficient diffusion weighted imaging data were excluded from the analysis. All participants were evaluated in person with the Mini International Neuropsychiatric Inventory (MINI) (Sheehan et al., 1998) by trained clinical psychiatric interviewers. Data were collected between January 2015 and February 2017. Participants were recruited from the Laureate Psychiatric Clinic and Hospital, other local behavioral and mental health providers, and through newspaper, flyer, online, radio, and other media advertisements in the Tulsa metropolitan area. Exclusion criteria included comorbid psychiatric disorders (except for anxiety disorders), substance use disorders (except alcohol use disorder), significant or unstable medical conditions (including cardiovascular, gastrointestinal, endocrine, neurological, hematological, rheumatological or metabolic disorders), a history of moderate-to-severe traumatic brain injury, a history of autoimmune disorders (except hypothyroidism), a positive urine drug screen, a body mass index (BMI) <17 or >38 kg/m², and general MRI exclusion criteria [details in (Victor et al., 2018)]. Approval for the study was obtained from the Western Institutional Review Board (#194919), and written informed consent was obtained from all participants.

2.2. Behavioral Data

Other than the structured psychiatric clinical interview with the MINI, all participants completed the Patient-Reported Outcomes Measurement Information System (PROMIS) (Gershon et al., 2010) scales for depression and anxiety, the Customary Drinking and Drug Use Record (CDDR) structured interview for lifetime alcohol use (Brown et al., 1998), and the positive and negative affect schedule-expanded form (PANASX) (Watson and Clark, 1994) for specific emotional states. The PANASX scale includes 60 items that describe different feelings and emotions. Participants are required to “Indicate to what extent you have felt this way during the past few weeks.” The items are grouped into the following four subgroups and subscales: general dimension scales (negative affect, positive affect), basic negative emotion scales (fear, hostility, guilt, sadness), basic positive emotion scales (joviality, self-assurance, attentiveness), and other affective states (shyness, fatigue, serenity, surprise). Participant demographic and behavioral data are summarized in Table 1.

Table 1.

Demographic and behavioral data of the participant sample.

	Subjects included in analyses (n=166)		Total MDD sample (n=244)
General Demographic Data	Mean	SD	Mean	SD
Age	36.99	11.58	35.77	11.41
Sex (Male %)	30.70	-	27.50	-
BMI a	28.28	5.28	28.61	5.41
Medicated (%) b	65.70	-	68.40	-
Depression severity c	60.93	7.88	61.55	7.28
Anxiety severity d	62.68	6.15	62.71	6.65
Number of episodes e	3.88	3.27	4.01	3.37
Alcohol use f	4.99	2.54	5.11	2.56
Smoker (%) g	26.0	-	29.9	-
Emotional States h	Mean	SD	Mean	SD
General negative affect	23.38	7.26	23.66	7.65
General positive affect	24.65	7.13	24.49	7.25
Fear	13.15	4.72	13.19	4.81
Hostility	12.23	4.33	12.36	4.47
Guilt	15.57	6.25	16.03	6.32
Sadness	14.17	4.91	14.15	4.90
Joviality	17.22	5.89	17.18	5.80
Self-assurance	13.28	4.45	12.95	4.35
Attentiveness	11.01	3.20	10.97	3.28
Shyness	8.76	3.78	8.87	3.91
Fatigue	13.20	4.00	13.12	3.94
Serenity	6.72	2.45	6.81	2.52
Surprise	5.13	2.13	5.04	2.04
Ethnicity	%		%
Asian	1.20	-	0.80	-
Black	7.80	-	7.80	-
Hispanic	4.20	-	3.70	-
Native American	15.10	-	15.6	-
White	68.70	-	69.5	-
Other	3.00	-	2.50	-

^aBMI = body mass index;

^bmeasured by ordered categories. ^b Medicated is defined as participants with MDD taking psychotropic medications.

^cPROMIS depression T score.

^dPROMIS anxiety T score.

^eMeasured by MINI interview. Participants with over 10 episodes were treated as having had 10 episodes.

^fLog-transformed lifetime alcohol usage was used. Data obtained from CDDR interview.

^gSelf-reported regular smoker.

^hMeasured by the positive and negative affect schedule-expanded form (PANASX).

2.3. CRP

Serum from morning blood samples was isolated following standard laboratory procedures and stored at −80 °C. CRP serum concentrations were analyzed in duplicate with the V-PLEX Neuroinflammation Panel-1 Human Kits on a Meso Quickplex SQ120 instrument (Meso Scale Diagnostics, Maryland, USA). The lowest level of quantification (LLOQ) was 27.6 pg/mL and the intra-and inter-assay coefficients of variation were 2.30% and 10.04%, respectively. CRP concentrations were log-transformed, and one participant was removed from CRP related analyses due to an absolute value larger than three standard deviations from the mean. No samples were below the LLOQ.

2.4. Kynurenine Pathway Metabolites

Serum was isolated from morning blood samples following standard laboratory procedures and stored at −80°C. Serum KynA and QA concentrations were measured by Keystone Bioanalytical, Inc with high-performance liquid chromatography and tandem mass spectrometry (MS/MS) detection using their standard protocols. The lowest level of quantitation (LLOQ) and intra-assay percentage of the coefficient of variation for KynA and QA were 1ng/mL, 3.76% and 1ng/mL, 4.76%, respectively.

2.5. MRI Data Acquisition and Preprocessing

Diffusion MRI scans were acquired using two identical 3.0T scanners (GE Discovery MR750) with brain-dedicated receive-only 32 element coil arrays optimized for parallel imaging (Nova Medical, Inc.). The diffusion-weighted imaging (DWI) data were acquired using a single-shell acquisition with 60 diffusion encoding directions (b value = 1000 s/mm², TR/TE = 9000/83.6 ms, with acquisition and reconstruction matrix = 128 × 128, field of view (FOV) = 25.6 × 25.6 cm, slice thickness = 2 mm, without interslice spacing, 73 axial slices, acceleration factor R = 2 in the phase encoding direction) and 8 no diffusion-weighted images (b value = 0 s/mm²) acquired at the beginning of the scan. The total acquisition time was 10 min and 50 s. DWI data were preprocessed using the FMRIB Software Library tool (FSL, version 6.0, https://fsl.fmrib.ox.ac.uk/fsl) and the DSI Studio (August 30, 2021 build, http://dsi-studio.labsolver.org). The FSL ‘eddy’ tool was used to estimate and correct eddy current-induced distortions and gross participant movement (Andersson and Sotiropoulos, 2016). The quality of the dataset was assessed using the eddy QC tools (Bastiani et al., 2019). Slices with signal loss caused by participant movement coinciding with the diffusion encoding were detected and replaced by predictions made by means of a Gaussian process (Andersson et al., 2016). The quality control criteria were set as an average absolute volume to volume head motion of <3 mm, or total outliers <5%. Skull stripping was performed for each participant using FSL-BET (Smith, 2002). The diffusion data were reconstructed in the MNI space with the default settings of DSI Studio. A diffusion sampling length ratio of 1.25 was used, and the output resolution was 2 mm (Yeh and Tseng, 2011; Yeh et al., 2010). The reconstructed results of each subject were also inspected. Diffusion tensor model fitting and fractional anisotropy (FA) value were computed for each subject at the voxel level using DSI Studio and served as a white matter integrity index in the following analyses.

2.6. Correlational Tractography (Primary and Secondary Analyses)

Correlational tractography (Yeh et al., 2021) is a novel diffusion MRI analytic approach that integrates a multivariable linear regression model and fiber tracking to identify the subcomponents of the white matter tracts that show association with a variable of interest, and allows statistical inference to be reported as a false discovery rate (FDR) via a permutation test (Yeh et al., 2016). In the current study, correlational tractography was performed using DSI Studio to map the specific white matter pathways that were correlated with CRP and KynA/QA. For the primary analyses, a multiple regression model was used to identify the association between CRP or KynA/QA and FA value at the level of each voxel, controlling for age, sex, and BMI. Local voxels exceeding a t-statistic threshold of 2.5 for an effect of CRP or KynA/QA on FA were selected, and fiber tracking was performed via a deterministic fiber tracking algorithm (Yeh et al., 2013). This deterministic fiber tracking algorithm allows for crossing fibers within voxels, which helps to reduce false-positive connections and ensure an excellent valid connection rate (i.e., 92%) (Maier-Hein et al., 2017; Zalesky et al., 2016). Track trimming was set with a default value of four iterations. A length threshold of at least 20 voxels was used to identify associated white matter tracts. Bootstrap resampling with 4,000 randomized permutations was used to estimate the null distribution of track length and provide false discovery rates (FDR). The cerebellum was masked out from the analyses to avoid spurious findings due to partial scan coverage of the cerebellum. For more detailed methodology documentation, please see (Yeh et al., 2016).

To tease apart the effect of KynA and QA on FA, we performed a secondary analysis, i.e., two additional multiple regression models using either KynA concentration or QA concentration were performed at the level of each voxel, controlling for age, sex, and BMI. All other correlation tractography procedures such as the parameter settings and statistical threshold remained identical to the primary analyses.

2.7. Exploratory Analysis and Sensitivity Analysis

Once the tracts that showed a significant association with CRP or KynA/QA ratio were identified, the mean FA value was extracted to perform the following analyses. First, to explore the association between CRP or KynA/QA and the mean FA from the identified tracts and the specific emotional states, a robust linear regression model was performed with emotional states as the outcome while controlling for age, sex, and BMI. Second, to examine the robustness of the association between CRP or KynA/QA and FA, several additional variables that could theoretically influence CRP or KynA/QA or cause white matter structure change, or both, were selected as potential confounders (also known as principles of confounder selection) (VanderWeele, 2019). In addition to age, sex and BMI, the selected variables included medication status (defined as whether subjects were taking psychotropic medication), depression severity, anxiety severity, number of episodes (obtained from MINI interview), lifetime alcohol use (obtained from CDDR interview), and smoking status. A robust linear regression model was used to test whether the association between CRP or KynA/QA and FA would be sensitive to these confounders by adding these variables in the regression model.

2.8. Mediation Analysis

We used the R package “mediation” (Tingley et al., 2014) to test the hypothesis of whether KynA/QA ratio mediates the association between CRP concentration and FA. One thousand bootstrapping samples were used to estimate the average total (direct) effect and mediation (indirect) effect. All the potential confounders (i.e., age, sex and BMI, medication status, depression severity, anxiety severity, number of episodes, lifetime alcohol use, smoking status) were added as covariates in the mediation analysis.

3. Results

3.1. Associations between CRP and KP Metabolites

Linear regression models controlling for aforementioned confounders (i.e., age, sex and BMI, medication status, depression severity, anxiety severity, number of episodes, lifetime alcohol use, smoking status) were used to test the associations between CRP and KP metabolites. CRP concentration was inversely associated with KynA/QA (standardized beta coefficient, SBC=−0.35 with a standard error, Std.E=0.13, p < 0.01). This association was driven by the positive correlation between CRP and QA (SBC = 0.35, Std.E=0.12, p < 0.01). There was no significant correlation between CRP and KynA (SBC = −0.05, Std.E=0.13, p = 0.70).

3.2. Primary Analyses: Tracts Correlated with CRP

Consistent with our previous report, correlation tractography analysis using FA as an index of white matter integrity revealed that several white matter tracts were negatively correlated with CRP concentration (total number of tracts =1228, mean white matter tract length = 50.5 mm, FDR < 0.05). Specifically, higher CRP concentrations were associated with decreased white matter integrity (FA value) among the bilateral fornix and bilateral cingulum bundles (Figure 1.A). We extracted the mean FA value from these identified tracts to estimate the effect size by using a robust linear regression model controlling for three well known confounders in diffusion tensor imaging (DTI) studies, i.e., age, sex, and BMI (Figure 1.B). The estimated standardized beta coefficient (SBC, equivalent to Cohen’s D) was −0.42 with a standard error (Std.E) = 0.09. Thus, for every standard deviation increase in the log-transformed CRP value, the FA in these regions was decreased by 0.42 standard deviations. There were no significant positive associations between CRP concentration and FA value.

Figure 1.

White matter tracts with FA values inversely correlated with log-transformed CRP concentration. (A). 3D rendering of identified tracts that showed a significant negative association with CRP concentration. (B). Scatter plot of mean FA extracted from the negative associated tracts and CRP concentration. Note that the mean FA value shown in panel (B) is after regressing out the effects of age, sex, and BMI.

3.3. Primary Analyses: Tracts Correlated with KynA/QA

As shown in Figure 2A, correlation tractography revealed that several white matter tracts were positively correlated with KynA/QA (total number of tracts =1120, mean white matter tract length = 46.2 mm, FDR < 0.05). Specifically, individuals with greater KynA/QA showed increased white matter integrity (FA value) among the bilateral fornix, bilateral superior thalamic radiations, corpus callosum (body and tapetum portions), and bilateral cingulum bundles (Figure 2.A, Figure S2). We used these tracts as a mask to extract the mean FA value to estimate the effect size by using a robust linear regression model controlling for age, sex, and BMI (Figure 2.B). The SBC was 0.19 with Std.E = 0.08. Thus, for every standard deviation increase in KynA/QA, the FA in these regions was increased by 0.19 standard deviations. There were no significant negative associations between FA and KynA/QA.

Figure 2.

White matter tracts with FA values positively correlated with kynurenic acid/quinolinic acid (KynA/QA). (A). 3D rendering of identified tracts that showed a significant positive association with KynA/QA. (B). Scatter plot of mean FA extracted from the positively associated tracts and KynA/QA. Note that the mean FA value shown in panel (B) is after regressing out the effects of age, sex, and BMI.

3.4. Secondary Analyses: Tracts Correlated with KynA concentration or QA concentration

In an effort to delineate how KynA versus QA concentrations affected the FA value, we also performed correlation tractography using KynA or QA concentrations individually, controlling for age, sex, and BMI. As shown in Figure 3, correlation tractography identified a small number of white matter tracts that were positively correlated with KynA concentration (Figure 3A, total number of tracts = 19, mean white matter tract length = 50.9 mm, SBC = 0.19, Std.E = 0.08, FDR < 0.05), and no significant negative associations between FA and the KynA concentration. Further, correlation tractography revealed that a small number of white matter tracts were negatively correlated with QA concentration (Figure 3B, total number of tracts = 62, mean white matter tract length = 53.1 mm, SBC = −0.15, Std.E = 0.08, FDR < 0.05), and no significant positive associations between FA and QA concentration were found.

Figure 3.

White matter tracts with FA values associated with kynurenic acid (KynA) concentration or quinolinic acid (QA) concentration. (A). 3D rendering of identified tracts that showed a significant positive association between FA and KynA. No significant negative associations between FA and KynA were found. (B). 3D rendering of identified tracts that showed a significant negative association between FA and QA. No significant positive associations between FA and QA were found.

3.5. Sensitivity to Potential Confounders

As sensitivity analyses, in addition to age, sex, and BMI, we examined whether other potential confounders would explain the observed effect of CRP concentration and KynA/QA on FA. These additional confounders included depression severity, anxiety severity, medication status, number of depressive episodes, lifetime alcohol use, and smoking status. We found that the negative association between CRP concentration and FA remained significant after regressing out all nine potential confounders (SBC = −0.43, Std.E = 0.13, p = 0.002). The mean FA along the identified tracts that was negatively associated with CRP concentration in Figure 1A was used as the outcome in the regression model. That is, the bilateral fornix and bilateral cingulum bundles. Similarly, the positive association between KynA/QA and FA also remained significant after regressing out all nine potential confounders (SBC = 0.26, Std.E = 0.09, p = 0.005). The mean FA along the identified tracts that was associated with KynA/QA ratio in Figure 2A was used as the outcome in the regression model. That is, the bilateral fornix, bilateral superior thalamic radiations, corpus callosum (body and tapetum portions), and bilateral cingulum bundles.

3.6. Mediation Analysis

The bilateral fornix and cingulum bundles showed association with both CRP and KynA/QA. Therefore, the mean FA extracted from these white matter tracts was used as outcomes to perform mediation analysis. We tested whether the negative association between CRP concentration and the FA value in the bilateral fornix and cingulum bundles was mediated by KynA/QA ratio by using a causal mediation analysis while controlling for all measured confounders (i.e., age, sex and BMI, medication status, depression severity, anxiety severity, number of episodes, lifetime alcohol use, smoking status). The indirect effect through the KynA/QA pathway was not significant (Figure 4). Thus, our results did not support this hypothesis.

Figure 4.

Path diagram of the mediation model. The figure shows the standardized beta coefficient (SBC) and the p value of total effect (direct effect) of CRP on FA as well as the mediation effect (indirect effect) of CRP on FA through the KynA/QA path. As shown in the figure, the mediation effect was not significant.

3.7. Exploratory Analyses: Relationships between CRP, KynA/QA, FA, and Emotional States

Robust linear regression models with age, sex, and BMI as covariates were used to explore whether there were any associations between CRP or KynA/QA, mean FA value from the identified overlapping white matter tracts (the bilateral fornix and cingulum bundles), and each of the specific emotional states. The SBC, Std.E, and p values are summarized in Table 2. There were no significant associations between CRP concentration, mean FA value and the specific emotional states (indexed by subscales of the PANASX). The KynA/QA ratio was positively associated with general positive affect (SBC = 0.170, Std.E = 0.084, p_uncorrected = 0.042), attentiveness (SBC = 0.226, Std.E = 0.083, p_uncorrected = 0.007), and negatively associated with fatigue (SBC = −0.173, Std.E = 0.085, p_uncorrected = 0.042).

Table 2.

Association between CRP, KynA/QA, FA, and Emotional States.

	Association with CRP			Association with KynA/QA			Association with FA
Emotional States	SBC	Std.E	puncorrected	SBC	Std.E	puncorrected	SBC	Std.E	puncorrected
General negative affect	−0.046	0.095	0.629	−0.023	0.080	0.773	0.049	0.079	0.534
General positive affect	−0.017	0.106	0.875	0.170	0.084	0.042*	0.057	0.087	0.516
Fear	0.081	0.106	0.447	−0.065	0.087	0.457	0.041	0.087	0.635
Hostility	−0.034	0.091	0.711	−0.048	0.076	0.530	0.003	0.076	0.966
Guilt	−0.073	0.098	0.456	0.089	0.082	0.285	−0.021	0.082	0.798
Sadness	−0.068	0.107	0.525	0.005	0.087	0.957	−0.017	0.088	0.852
Joviality	0.130	0.101	0.197	0.049	0.084	0.562	−0.004	0.084	0.959
Self-assurance	−0.030	0.102	0.769	0.111	0.081	0.171	0.011	0.085	0.893
Attentiveness	0.023	0.098	0.811	0.226	0.083	0.007*	0.075	0.082	0.367
Shyness	−0.093	0.102	0.364	0.121	0.082	0.143	0.096	0.083	0.249
Fatigue	0.088	0.102	0.390	−0.173	0.085	0.042*	−0.111	0.085	0.192
Serenity	0.118	0.101	0.243	0.041	0.083	0.623	0.050	0.086	0.563
Surprise	−0.004	0.095	0.965	0.069	0.077	0.371	0.130	0.077	0.093

4. Discussion

This investigation examined the relationship between CRP, KynA/QA, and white matter integrity in a group of individuals with MDD. There were three main findings. First, serum CRP concentrations were inversely associated with white matter integrity (FA) of the cingulum and fornix. Second, individuals with a higher concentration of serum KynA/QA, a putative neuroprotective index (Savitz et al., 2015c), showed greater integrity of several midline white matter tracts, including the corpus callosum, cingulum bundles, fornix, and thalamic radiations. Third, contrary to our hypothesis, the association between CRP and FA of the cingulum and fornix was not mediated by KynA/QA.

The inverse association between CRP and FA is broadly consistent with our previous report in the same sample of MDD participants in which we used a different metric of white matter integrity, i.e. quantitative anisotropy, as the outcome measure (Thomas et al., 2021). The negative relationship between CRP and FA of the cingulum and fornix reported here is also partially consistent with previous work reporting associations between higher concentrations of inflammatory cytokines and FA reductions in brain regions such as the corpus callosum, corona radiata, and the inferior/superior fronto-occipital fasciculi in MDD or BD populations (Benedetti et al., 2016; Comai et al., 2022; Lim et al., 2021; Sugimoto et al., 2018).

The second main finding was the positive association between KynA/QA and FA values of the corpus callosum, cingulum bundles, fornix, and thalamic radiations. Secondary analyses revealed that the relationship was due to both KynA and QA in a subset of these white matter tracts. It is noteworthy that significantly fewer white matter tracts were identified when using KynA or QA concentration individually, suggesting that KynA/QA is a more sensitive index than KynA or QA alone. QA has a similar potency to glutamate at the NMDA receptor but it remains in the synaptic cleft for a longer period of time due to less-efficient reuptake, and therefore has stronger excitotoxic effects (Foster et al., 1984a). Additionally, QA can augment glutamate release by neurons and inhibit its uptake by astrocytes leading to excessive glutamate concentrations and excitotoxicity (Guillemin, 2012). In contrast, KynA blocks the neurodegenerative and epileptogenic effects of QA when administered in the rat brain (Foster et al., 1984b). At least in vitro, QA also acts as a gliotoxin by inducing apoptosis of astrocytes and can potentiate the release of proinflammatory mediators (Ting et al., 2009). Oligodendroglia, the cells responsible for myelination of axons, are particularly vulnerable to inflammation, perhaps explaining why reductions in numbers or density of oligodendrocyte cells are one of the replicated findings in postmortem studies of depression (Mechawar and Savitz, 2016). If circulating concentrations of KynA/QA reflect the relative abundance of neuroprotective and neurotoxic KP metabolites in the brain parenchyma, then this may explain the association between KynA/QA and white matter integrity. Nevertheless, it should be noted that although the concentration of QA in the blood correlates significantly with QA in the CSF (r~0.5), the same cannot be said of KynA (Haroon et al., 2020; Paul et al., 2022; Sellgren et al., 2019). However, as discussed elsewhere (Savitz, 2022), the issue is complicated by the fact that the concentration of biomarkers in the CSF do not always correlate significantly with their concentration in the brain parenchyma (Gadad et al., 2021). Thus, although there is no consensus regarding the neurophysiological significance of serum measures of KP metabolites, a growing number of studies report correlations between these metabolites and brain structure or function (Chen et al., 2021; Comai et al., 2022; DeWitt et al., 2018; Doolin et al., 2018; Lee et al., 2017; Meier and Savitz, 2022; Paul et al., 2022; Poletti et al., 2018; Savitz, 2020). Ultimately, experimental studies that manipulate the KP will be needed to resolve this question.

Regarding the neuroanatomical significance of the white matter tracts, our results are partially consistent with a DTI study in BD which found negative correlations between KynA and white matter damage of the corpus callosum, anterior thalamic radiation, and cingulum (Poletti et al., 2018) in addition to associations with the superior longitudinal fasciculus and inferior fronto-occipital fasciculus that were not observed in our study. The association between KynA/QA and the thalamic radiation is also consistent with an immunohistochemical analysis of the rat brain which showed that the anteroventral, periventricular, ventromedial and reticular nuclei of the thalamus contains numerous quinolinic acid phosphoribosyltransferase-staining glial cells (Kohler et al., 1988). Meta-analyses of DTI studies in MDD have identified white matter microstructural abnormalities in the corpus callosum (Chen et al., 2016; Wise et al., 2016) while an ENIGMA consortium study of 1305 patients and 1602 controls reported MDD-associated reductions in FA of the corpus callosum, fornix, and cingulum bundle among other regions (van Velzen et al., 2020). An imbalance between neuroprotective and neurotoxic KP metabolites may be one mechanism underlying these deleterious changes in white matter in MDD. Nevertheless, it should be noted that the effect size was small. Clearly, the balance between KynA and QA might only have a small effect on white matter integrity, but it is also possible that the effect size was attenuated because KynA/QA only affects white matter integrity in a subgroup of patients. Another possibility is that KynA/QA in the brain only modestly correlates with KynA/QA in the blood, thus, the variability between brain and serum KynA/QA ratio might reduce signal-to-noise in detecting the relationship between KynA/QA and white matter integrity. Third, there may be interactions between other factors such as diet, exercise, medication, severity, and other clinical features of the disorder that would affect the relationship between KP metabolism and brain structure.

The third finding of interest was the absence of a mediating effect of KynA/QA on the relationship between CRP and FA. This result suggests that inflammatory mediators such as CRP and neuroactive kynurenines may impact white matter structures through independent pathways if such causal relationships are indeed present. Nevertheless, although mediation analyses are often performed on cross-sectional data, this type of approach to the modeling of causal processes that unfold over time, has been shown to be at risk of generating biased parameter estimates (Maxwell and Cole, 2007; O’Laughlin et al., 2018). Thus, caution should be taken in the interpretation of this finding. Further, it is conceivable that results may have differed if inflammatory mediators other than CRP were tested in the mediation model. We focused on CRP because of its widespread use, the rich literature linking it to depression and brain structure/function, and the fact that it is usually considered to be a non-specific marker of immune activation.

In exploratory analyses we tested whether emotional states measured with the PANASX (Watson and Clark, 1994) were associated with FA, CRP or KynA/QA. Higher KynA/QA was associated with greater “attentiveness” and “general positive affect” and lower “fatigue” although these results should be interpreted with caution because they were no longer significant after correction for multiple comparisons. Attentiveness is one of the basic positive emotion subscales and is associated with alertness, concentration, and determination (Watson and Clark, 1994). General Positive Emotion is a high-order construct comprised of Joviality, Self-Assurance, and Attentiveness. Fatigue (sleepiness, tiredness, sluggishness) tends to be associated with high negative and low positive affect (Watson and Clark, 1994). The positive association between KynA/QA and positive affect is consistent with prior reports in independent MDD samples of inverse associations between KynA/QA and self-reported anhedonia (Savitz et al., 2015a; Savitz et al., 2015d) as well as the well-known anti-anhedonic properties of ketamine, which like KynA, acts as an NMDA receptor antagonist (Lally et al., 2014; Pulcu et al., 2021). This result is also in line with the report of reduced circulating KynA in participants with a melancholic subtype of MDD (Milaneschi et al., 2021). In contrast, the absence of a significant correlation between CRP and positive affect was surprising given that previous studies have demonstrated that higher CRP concentrations in patients with MDD are associated with decreased functional connectivity within the frontal-striatal reward circuitry, altered glutamate levels in the basal ganglia, and greater symptoms of anhedonia (Felger et al., 2018; Felger et al., 2016; Haroon et al., 2016). The inverse association between KynA/QA and fatigue is consistent with a recent report of reduced plasma concentrations of KynA/QA in patients with Chronic Fatigue Syndrome (CFS) relative to healthy controls (Groven et al., 2021).

Several limitations deserve mention. First, a healthy comparison group was not included because data for KP metabolites in healthy controls are not part of the Tulsa 1000 dataset. Thus, we do not know if KynA/QA was reduced in this group of MDD participants as has been demonstrated in many other studies (Bartoli et al., 2021; Doolin et al., 2018; Myint et al., 2007; Paul et al., 2022; Savitz et al., 2015a; Savitz et al., 2015d; Schwieler et al., 2016; Wurfel et al., 2017). Nevertheless, this was not the primary rationale of the study which was instead to examine the link between KynA/QA and white matter microstructural abnormalities in depression. Second, 65% of the participants were using psychotropic medications that could theoretically affect CRP, KP metabolite concentrations or FA values. However, the relationships between CRP and FA and KynA/QA and FA remained significant in sensitivity analyses which controlled for medication status. Third, as mentioned above, in any study of MDD, there are numerous potential confounds such as diet, exercise, sleep disturbance, and stage of illness. Significantly larger samples are required to tease these putative effects apart.

In sum, both higher concentrations of CRP and an imbalance in the circulating concentrations of the neuroprotective versus neurotoxic KP metabolites, KynA versus QA, were associated with reduced white matter integrity in a community sample of people with MDD. These results raise the possibility that inflammation and dysregulation of the KP are independent molecular mechanisms through which white matter damage occurs in the context of MDD. However, experimental studies will ultimately be needed in order to make such causal claims with any degree of certainty.