Changes in Kynurenine Pathway Metabolites and Brain Structure in Response to Behavioral Activation Therapy

Abstract

Behavioral activation therapy (BA) is an efficacious treatment for major depressive disorder (MDD). However, the neurobiological mechanism of action is unclear. The present study is a secondary, multimodal analysis of data from two BA clinical trials, integrating previously reported kynurenine pathway (KP) measures with structural MRI. We tested (a) whether BA is associated with pre-post changes in gray matter volume (GMV) within a priori regions of interest (hippocampus, amygdala, and striatum), and (b) whether BA-associated changes in KP metabolites including kynurenic acid (KynA), quinolinic acid (QA), and KynA/QA relate to concomitant changes in GMV. Twenty-seven participants with mood and anxiety disorders and clinically significant depressive symptoms underwent 10 weeks of BA with pre- and post-treatment blood sampling and MRI. KP metabolites were measured with high-performance liquid chromatography and tandem mass spectrometry. Data were analyzed using linear mixed effects and regression models controlling for age, sex, BMI, therapy group, and intracranial volume (for MRI outcomes). All reported p-values were FDR-corrected. BA was associated with a pre- to post-treatment reduction in left putamen volume (β = −0.11, p < 0.05). At baseline, QA was negatively correlated with bilateral caudate volumes (left: r = −0.47, p < 0.05; right: r = −0.49, p < 0.05). Pre-post BA increases in KynA (β = 0.183, p < 0.01) and KynA/QA (β = 0.159, p < 0.05) were associated with increases in right hippocampal volume, and KynA changes were also positively associated with increases in right amygdala volume (β = 0.234, p < 0.05). Furthermore, QA increases were associated with GMV increases in both the right hippocampus (β = 0.148, p < 0.05) and nucleus accumbens (β = 0.224, p < 0.05). Exploratory analyses revealed additional associations between other KP metabolites (e.g., 3hydroxykynurenine, tryptophan) and volumetric changes. Baseline KP metabolites and baseline GMV did not predict concurrent symptom change. In targeted prospective tests, larger pre- to post-treatment QA decreases predicted lower depression and anxiety symptoms at 3 months, whereas ΔGMV within a priori ROIs did not predict 3-month symptoms. Our findings suggest that BA is associated with changes in KP metabolism and GMV within affective neural circuitry, and that changes in QA may predict longer-term clinical outcomes. However, given the modest sample size and absence of a control group, these findings should be considered preliminary and cannot establish whether observed changes reflect BA-specific effects or nonspecific effects of time. Future research using randomized clinical designs, larger samples, and multimodal approaches will be critical to delineate the role of KP metabolites in supporting neuroplasticity with BA specifically, and psychotherapy more generally.

Introduction

Major depressive disorder (MDD) is a leading cause of disability worldwide and is associated with both psychological symptoms and neurobiological alterations, including disrupted immune signaling (Miller, Maletic, & Raison, 2009; Irwin & Miller, 2007). Psychotherapeutic interventions such as behavioral activation therapy (BA) – a structured intervention focused on enhancing engagement in meaningful and reinforcing activities (Dimidjian et al., 2011; Martell, Dimidjian, & Herman-Dunn, 2021) – have demonstrated efficacy (Butler et al., 2006; Dimidjian et al., 2006). However, substantial heterogeneity in treatment response remains a clinical challenge (Dimidjian et al., 2006), with only about 40% of patients showing significant improvement and more than half relapsing a few years after psychotherapy (Cuijpers et al., 2021; Steinert et al., 2014). These findings provide incentive for identifying biological markers that predict or track therapeutic response to move toward personalized care. Further, understanding how psychotherapeutic interventions influence brain and immune function may clarify mechanisms of symptom change and allow for the refinement of treatment protocols.

Structural imaging studies have found that individuals with depression show morphometric differences relative to non-psychiatric comparison groups, including gray matter volume (GMV) reductions in key cortico-limbic structures such as the hippocampus (Campbell et al., 2004; McKinnon et al., 2009; Savitz & Drevets, 2009; Schmaal et al., 2015; Videbech & Ravnkilde, 2004), amygdala (Hamilton, Siemer, & Gotlib, 2008; Savitz & Drevets, 2009), and striatum (Koolschijn et al., 2009; Savitz & Drevets, 2009). Psychotherapy has been associated with structural changes in brain regions implicated in mood and cognitive control. In particular, cognitive behavioral therapies (CBTs) have been shown to promote both increases in GMV within limbic regions such as the amygdala and nucleus accumbens, as well as decreases in hippocampus (Meng et al., 2021; Redlich et al., 2025; Seminowicz et al., 2013). These previous findings suggest that therapy may facilitate neuroplasticity across circuits involved in emotion regulation and cognitive control. Although the precise neurobiological mechanisms remain unclear, it is plausible that psychotherapy-induced neuroplasticity involves inflammatory and metabolic pathways (Savitz, 2020).

One candidate for such a mechanism is the kynurenine pathway (KP), the primary route for tryptophan metabolism and a critical mediator of inflammation. The KP yields a diverse array of metabolites with neuroactive properties. Quinolinic acid (QA) is a glutamatergic N-methyl-D-aspartate (NMDA) receptor agonist that can have pro-inflammatory and neurotoxic effects (Savitz, 2016; Schwarcz et al., 2012), while kynurenic acid (KynA) acts as an NMDA antagonist and has been associated with neuroprotective effects (Savitz, 2020; Tanaka et al., 2022). Under inflammatory conditions, tryptophan metabolism is preferentially shunted toward QA production (O’Connor et al., 2009; Zunszain et al., 2012), reducing KynA levels and leading to an imbalance that is consistently associated with mood disorders (Meier et al., 2016; Myint et al., 2007; Savitz et al., 2015d). Dysregulation of the KP, particularly elevated QA and reduced KynA, has been linked to both depressive symptomatology and alterations in brain structure and function (Paul et al., 2022; Savitz et al., 2015b).

In previous work, we showed that BA decreased QA and increased KynA/QA (Savitz et al., 2022). In independent cross-sectional samples, we further observed that a neuroprotective profile among clinically depressed populations was associated with increases in GMV of the hippocampus, amygdala, and basal ganglia – regions frequently implicated in mood and motivation (Savitz et al., 2015a; Savitz et al., 2015c). These findings raise the possibility that the balance between neuroprotective and neurotoxic kynurenine metabolites may underlie or accompany structural brain changes induced by psychotherapy, motivating the current study’s primary focus on examining how changes associated with BA in KynA, QA and KynA/QA relate to structural changes in these regions of interest (ROIs). We consider two non-mutually exclusive pathways. In one pathway, BA may reduce stress and inflammation and shift tryptophan metabolism away from the QA/3-HK branch and/or toward KynA, which could facilitate hippocampal-striatal plasticity observable as GMV change. In a complementary pathway, BA may directly re-engage reward circuitry and support circuit plasticity, leading to symptom improvement and subsequent normalization of immune activation and peripheral KP indices. In the current study, we test whether changes in KP metabolites during a 10-week BA intervention were associated with concurrent changes in GMV of pre-defined cortico-limbic and striatal ROIs, including the hippocampus, amygdala, caudate, putamen, and nucleus accumbens. We also examined whether baseline levels of KP metabolite were associated with subsequent structural brain changes. Given only two time points (pre- and post-therapy), the present design cannot determine the causal directionality of relationships. Moreover, both KP metabolites and MRI-derived GMV are best conceptualized as context-dependent biomarkers rather than inherently “beneficial” or “harmful” in a uniform sense. However, based on prior findings, we hypothesized that increases in KynA/QA would be associated with increased GMV in hippocampus, amygdala, and basal ganglia, consistent with the idea that a shift toward neuroprotective balance promotes structural integrity (Savitz, 2020).

Methods

Study Design and Participants

This study combined data collected as part of two larger clinical trials focused on identifying neuroimaging and behavioral predictors of treatment response. The first was a non-randomized, unblinded clinical trial in which all enrolled participants had clinically significant depression and completed BA (ClinicalTrials.gov: #NCT02602340). The second was a randomized, unblinded clinical trial in which participants with clinically significant symptoms of generalized anxiety disorder (GAD) were randomized to complete either BA or exposure-based therapy (Santiago et al., 2020; ClinicalTrials.gov: #NCT02807480). Both trials used an identical 10-week group BA protocol, were conducted at the same site, and shared identical inclusion and exclusion criteria apart from diagnostic entry (MDD vs. GAD). For the present analyses, we included participants with a current MDD diagnosis or a current GAD diagnosis who also met a priori criteria for clinically significant depressive symptoms (PHQ-9 ≥ 9) and who received BA. Thus, all participants received the same intervention and met a comparable depressive symptom threshold, irrespective of primary anxiety diagnosis. To account for any residual differences between the parent studies, a binary covariate indicating primary diagnosis (MDD or GAD) was included in all KP-GMV and clinical outcome models. For analyses that involved clinical response or prospective symptom outcomes, we also included primary diagnosis as a covariate to account for potential study-level differences. Supplementary Figure 1 includes a CONSORT flow diagram that traces sample selection from initial consent through analytic inclusion for each parent trial.

Participants were recruited from outpatient mental health clinics and the general community through electronic and print advertisements. All participants completed baseline and post-treatment clinical, behavioral, and neurobiological assessments surrounding 10 weeks of group-based BA. A total of 27 individuals completed the BA intervention and both pre- and post-treatment blood draws and magnetic resonance imaging (MRI) scans. Inclusion criteria included a Patient Health Questionnaire-9 (PHQ-9) score of 9 or higher, indicating clinically significant depressive symptoms, and a diagnosis of MDD or GAD as determined by the Mini International Neuropsychiatric Interview (M.I.N.I.). Exclusion criteria included a history of psychotic disorders, bipolar disorder, recent substance use disorder, or contraindications to MRI. Table 1 shows the demographic and clinical characteristics, as well as baseline descriptive statistics for KP metabolites and immune markers of the completers. Baseline characteristics were compared between studies and reported in Supplementary Table 1.

Table 1.

Demographic and Clinical Characteristics at Baseline (open-label n = 19; randomized n = 8; total n = 27).

Age (mean (SD))	34.0 (12.8)
Sex = Male (N, %)	10 (37.0)
Educational attainment (%)
Less than high school education	1 (3.7)
High school or GED	5 (18.5)
Some college, no degree	10 (37.1)
Associate’s degree	2 (7.4)
Bachelor’s degree	9 (33.3)
Race (%)
Asian American	0 (0)
Black or African American	4 (14.8)
Native American	1 (3.7)
Other	4 (14.8)
White	18 (66.7)
Body Mass Index (mean (SD))	26.1 (5.3)
Clinical
Baseline PROMIS Depression (mean (SD))	61.6 (2.6)
Baseline PROMIS Anxiety (mean (SD))	59.23 (2.6)
Baseline BDI (mean (SD))	23.4 (8.2)
Baseline SDS (mean (SD))	16.3 (6.4)
MDD, current episode N (%)	20 (74.1)
MDD lifetime N (%)	27 (100)
Panic disorder current N (%)	1 (3.7)
Agoraphobia current N (%)	6 (22.2)
Social anxiety disorder N (%)	6 (22.2)
PTSD current N (%)	1 (3.7)
GAD current N (%)	13 (48.1)
Kynurenine Pathway Metabolites (mean (SD))
Tryptophan (μM)	53.56 (10.1)
Kynurenine (μM)	1.56 (0.4)
Kynurenic Acid (ng/mL)	34.39 (14.0)
3-Hydroxykynurenine (ng/mL)	26.8 (8.2)
Quinolinic Acid (nmol/L)	389.36 (82.58)
Cytokine Assays (mean (SD))
Interleukin-6 (pg/mL)	0.23 (0.18)
Tumor Necrosis Factor (pg/mL)	1.01 (0.72)
Interleukin-1 Receptor Antagonist (pg/mL)	409.24 (255.52)
Medications
Antidepressants N (%)	11 (40.7)
SNRI N (%)	1 (3.7)
Atypical N (%)	5 (18.5)
SSRI N (%)	5 (18.5)
Benzodiazepines N (%)	0 (0)
Mood Stabilizer N (%)	0 (0)
Antipsychotics N (%)	0 (0)

Note: Participants in both trials met criteria for clinically significant depressive symptoms (PHQ-9 ≥ 9). See table for diagnostic composition; SD = standard deviation; PROMIS = Patient-Reported Outcomes Measurement Information System; BDI = Beck Depression Inventory; SDS = Sheehan Disability Scale; MDD = major depressive disorder; PTSD = post-traumatic stress disorder; GAD = generalized anxiety disorder; SNRI = serotonin-norepinephrine reuptake inhibitor; SSRI = selective serotonin reuptake inhibitor.

Written, informed consent was obtained from all participants, and the study was approved by the Western Institutional Review Board (WIRB). All procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. Participants were compensated for their time completing assessments but were not compensated for time spent completing therapy.

Intervention

Participants completed 10 weeks of BA (90-minute sessions) in a structured group format according to a manual developed by RLA and CM, informed by prior BA treatment guides (e.g., Martell, Dimidjian, and Herman-Dunn’s clinician’s guide, 2010; 2nd ed. 2021). The intervention aimed to increase engagement in meaningful, reinforcing, and rewarding activities and reduce avoidance, individualized for each participant. Sessions were led by trained therapists and participants were asked to complete homework between sessions to monitor one’s activities and mood, implement the various BA concepts, and engage in BA activities. Treatment fidelity was monitored via independent ratings of randomly selected sessions (see prior work for further details on interventions and clinical outcomes; Berg et al., 2023; Santiago et al., 2020; Savitz et al., 2022).

Clinical Outcome Measure

Depressive symptoms were assessed using the PROMIS Depression computer-adaptive test (PROMIS-D; Cella et al., 2010). The computer-adaptive test draws from the PROMIS Depression item bank and uses item response theory to adaptively administer items to achieve efficient and precise measurement with low respondent burden. PROMIS-D scores are reported on a T-score metric (mean = 50, SD = 10; higher scores indicate more severe depressive symptoms) referenced to U.S. normative data. Items assess core depressive symptoms (e.g., negative mood, anhedonia/low positive affect, hopelessness/self-worth, and social withdrawal) over the past 7 days. PROMIS emotional distress item banks demonstrate strong reliability and measurement precision across a wide range of symptom severity (Pilkonis et al., 2011). PROMIS-D has also been used in randomized clinical trials, and prior work suggests that a 3- to 4-point change on the PROMIS-D T-score metric represents a reasonable minimally important difference for interpreting treatment effects (Kroenke et al., 2020). Therefore, in sensitivity analyses, we additionally defined responders using a minimally important difference criterion of a ≥3-point decrease in PROMIS-D. In the parent trials, PROMIS-D was administered at baseline, weekly during treatment, and post-treatment and served as the harmonized primary outcome; the Beck Depression Inventory-II (BDI-II) was also administered at baseline and post-treatment as a secondary outcome. The Beck Depression Inventory-II (BDI-II) was also administered at baseline and post-treatment and served as a secondary outcome measure.

For responder analyses, we operationalized depressive-symptom response using a Reliable Change Index (RCI) approach applied to PROMIS-D T-scores (Zahra & Hedge, 2010), consistent with prior work using PROMIS outcomes in BA samples (Savitz et al., 2022). The RCI quantifies whether an individual’s pre- to post-treatment change exceeds measurement error given the scale’s reliability: $S{E}_{\mathit{\text{diff}}}=\sqrt{2}\times S{D}_{pre}\times \sqrt{(1-REL)}$. Participants were classified as responders if they demonstrated a decrease (improvement) in PROMIS-D of at least 1.96 × SE_diff from pre- to post-treatment. All others were classified as non-responders. In the present sample, SE_pre = 5.74 and SE_diff = 4.37, yielding a reliable improvement threshold of a ≥8.57-point decrease in PROMIS-D T-score. As a sensitivity analysis reflecting clinically meaningful improvement, we additionally defined responders using a minimally important difference criterion of a ≥3-point decrease in PROMIS-D.

Anxiety symptoms were assessed using the PROMIS Anxiety computer-adaptive test (PROMIS-A), which was administered and scored analogously to PROMIS-D (mean = 50, SD = 10; higher scores indicate greater anxiety severity). Items assess core anxiety symptoms (e.g., fear, anxious misery/worry, and hyperarousal) over the past 7 days. PROMIS-A was used as a secondary clinical outcome in prospective analyses predicting anxiety symptoms at 3-month follow-up.

Kynurenine Pathway Metabolite and Cytokine Assays

Blood samples were collected via venipuncture at pre- and post-treatment visits between 7:00 AM and 11:00 AM. Serum samples were collected using BD Vacutainer serum tubes, processed following the BD Vacutainer standard protocol, and subsequently stored at −80°C.

In addition to our primary focus on kynurenic acid (KynA), quinolinic acid (QA), and the KynA/QA ratio, we also conducted exploratory analyses including other metabolites within the KP pathway: tryptophan (TRP), kynurenine (KYN), 3-hydroxykynurenine (3-HK), KYN/TRP ratio, KynA/KYN ratio, and KynA/3-HK ratio, given their relevance to downstream KP metabolism. Concentrations of all KP metabolites were measured blind to timepoint by Charles River Laboratories using high-performance liquid chromatography coupled with tandem mass spectrometry. Established protocols were followed for quantification, with the following the lowest levels of quantification (LLOQs) and intra-assay coefficients of variation (CVs): TRP (3 μM, 1.4%), KYN (0.225 μM, 1.6%), KynA (7.5 μM, 6.6%), 3-HK (5 nM, 3.9%), and QA (50 nM, 8.6%). KynA samples (n=5) that fell below the detection threshold were imputed at 90% of the LLOQ.

Interleukin-6 (IL-6) and tumor necrosis factor (TNF) were quantified blind to visit using a V-PLEX Panel (Meso Scale Diagnostics). The LLOQs were 0.07 pg/mL (IL-6) and 0.02 pg/mL (TNF). Samples were run in duplicate, with intra-assay CVs of 7.28% (IL-6) and 2.15% (TNF). IL-1RA was measured using a solid-state ELISA (R&D Systems) with an LLOQ of 31.2 pg/mL and an intra-assay CV of 2.44%.

MRI Acquisition

MRI data were acquired using a pair of identical GE MR750 3.0T MRI scanners. Structural imaging was performed using a T1-weighted Magnetization Prepared Rapid Acquisition Gradient Echo (MPRAGE) sequence. Imaging parameters included a 240mm field of view (FOV) a matrix size of 256×256, slice thickness of 1mm producing 0.938×0.938×1mm voxels, 186 axial slices, repetition time (TR) of 5ms, echo time (TE) of 1.964ms, inversion times (TI) of 725ms, and a flip angle of 8°.

MRI Preprocessing and Analysis

Structural MRI data were processed using FreeSurfer analysis suite version 6.0.0 (http://surfer.nmr.mgh.harvard.edu). The automated preprocessing pipeline (recon-all) was employed, consistent with our previous work (Savitz et al., 2015b; Savitz et al., 2015c). This pipeline generated estimates of GMV and thickness, as well as cortical volume and thickness. Quality assurance was conducted by trained raters to make sure there were no gross reconstruction errors. Cortical thickness was calculated as the closest distance from the gray/white boundary to the gray/CSF boundary at each vertex across the cortex (Fischl & Dale, 2000). This approach involves reconstructing the cortical surface and computing thickness using a surface-based method rather than voxel-based morphometry, allowing for improved spatial accuracy and inter-subject alignment. The Destrieux 2009 atlas, which defines 148 cortical ROIs (Destrieux et al., 2010), was used to extract regional thickness and area measures. Thickness values were averaged within each ROI for statistical analysis. For subcortical structures, volumetric estimates were derived using FreeSurfer’s probabilistic segmentation of 40 subcortical ROIs (Fischl et al., 2002). All analyses were conducted separately for the left and right hemispheres to account for potential lateralization effects. For the present analyses, we focused on ROIs implicated in affective and motivational processes, including the hippocampus, amygdala, nucleus accumbens, caudate, and putamen. All FreeSurfer volumetric variables were z-scored prior to analysis; therefore, timepoint coefficients are expressed in SD units (y-standardized effects).

Statistical Analyses

To test the effect of BA on brain structure, we used linear mixed-effects models with two timepoints (pre- vs. post-treatment) as a fixed effect and participant as a random effect. These models were applied separately to each ROI to test for treatment-related volumetric changes. We then conducted multiple linear regression models to test associations between pre-to-post changes in KP metabolites and corresponding changes in brain structure, adjusting for covariates. Change variables (Δ) for KP metabolites, GMV, and PROMIS-D were computed as post-treatment minus pre-treatment values, such that negative values reflect decreases over time and positive values reflect increases. In a prior study with this cohort (Savitz et al., 2022), baseline KP metabolites did not predict changes in depression severity. Accordingly, the present analyses did not repeat these prospective tests, but instead focused on examining associations of KP metabolites with structural brain changes. Standardized coefficients are reported. Separate linear regression models were also used to test whether baseline GMV predicted changes in depression symptom severity over the course of BA, as measured by the PROMIS-D scale. We prospectively tested whether pre- to post-treatment changes in KP metabolites predicted clinical status (PROMIS-D and PROMIS-A scores) at 3-month follow-up.

To link biological change to behavioral response, we compared pre- to post-treatment change in GMV between responders and non-responders. For each a priori ROI, ΔGMV was regressed on a binary responder status to estimate group differences. We additionally examined intercorrelations among KP metabolites and baseline brain volumes using Pearson correlation. We provide matrices summarizing all baseline associations among primary (KynA, QA, KynA/QA) and exploratory KP metabolites with GMV across each ROI in the Supplementary Material (Supplementary Tables 2 and 3). We further assessed the relationship between changes in KP metabolites and structural brain measures (GMV, gray matter thickness, cortical volume, and cortical thickness) using linear regression models, adjusting for covariates. We also examined cytokine markers (IL-1RA, IL-6, TNF) in exploratory analyses. For each cytokine, we tested (1) pre-post change (Δ) associations with ΔKP metabolites, ΔGMV (a priori ROIs), and Δsymptoms (PROMIS-D, PROMIS-A), and (2) baseline cytokine levels predicting ΔKP and ΔGMV.

Bayesian Information Criterion (BIC) was used to determine the most parsimonious set of covariates to include in all primary and exploratory analyses. Across models, BIC consistently indicated that age, sex, (and total intracranial volume for MRI outcomes) should be included as covariates. To account for multiple comparisons, Benjamini-Hochberg false discovery rate (FDR) correction was systematically applied within each pre-specified family of analyses. For the primary analyses, which focused on KynA, QA, and KynA/QA ratio, FDR correction was applied across all models testing these three primary metabolites. For exploratory analyses, FDR correction was applied separately within each metabolite-ROI set (e.g., all models testing 3-HK with ROIs). All reported p-values reflect FDR-corrected values. Statistical analyses were performed using R statistical software package, version 4.1.0 (R Core Team, 2021).

Results

Effect of BA on Brain Structure

We used LMEs to examine the effects of BA on GMV in each of the five predefined ROIs: hippocampus, amygdala, nucleus accumbens, caudate, and putamen. Only the left putamen showed a statistically significant change over time, with greater volume observed at pre-treatment relative to post-treatment (β = −0.11, p < 0.05). There were nonsignificant volume reductions in GMV post-therapy in 7/10 ROIs –the left hippocampus, left nucleus accumbens, right putamen, bilateral amygdala, and bilateral caudate. Full model results are provided in Supplementary Table 4.

Baseline GMV and KP Metabolites as Predictors of Treatment Outcome

Baseline GMV of the ROIs did not predict treatment response, as indexed by changes in PROMIS-D scores over the course of BA. As reported previously in prior analyses from this cohort (Savitz et al., 2022), baseline concentrations of KP metabolites were not significantly associated with changes in depressive symptom severity following BA. Pre- to post-treatment changes in GMV within the a priori ROIs were not associated with pre- to post-treatment changes in PROMIS-D or PROMIS-A scores. Directions of effects were mixed and confidence intervals wide, consistent with limited power (see Supplementary Table 5). To relate biological change to behavioral response, we compared pre- to post-treatment changes in GMV between responders and non-responders. In the analytic sample, 8 participants met responder criteria and 19 were classified as non-responders. Using the ≥3-point minimally important difference criterion, 16 participants were classified as responders and 11 as non-responders. Responder status was not associated with ΔGMV in any a priori ROI (see Supplementary Table 6). Results were substantively unchanged when responder status was defined using the ≥3-point minimally important difference criterion. Similarly, parallel models using pre-post change in BDI-II scores yielded no significant associations between KP metabolites or GMV and symptom change. In prospective models, greater pre- to post-treatment decreases in QA at 10 weeks predicted lower PROMIS-D and PROMIS-A scores at 3 months (β = −0.064, p < 0.01; β = −0.083, p < 0.01). These effects remained when additionally adjusting for baseline symptoms and relevant covariates. Although pre- to post-treatment decreases in QA were associated with concurrent decreases in GMV in select regions, changes in GMV within these regions did not predict changes in depression or anxiety symptoms. No other KP metabolites showed significant prospective effects. In parallel models testing whether pre- to post-treatment change in GMV predicted 3-month PROMIS outcomes across the a priori ROIs, no ROI-level associations survived for either depression or anxiety. In exploratory models incorporating cytokines, neither ΔIL-1RA, ΔIL-6, nor ΔTNF showed significant associations with ΔKP, ΔGMV, or Δsymptoms after FDR correction; likewise, baseline cytokine levels did not predict ΔKP or ΔGMV (see Supplementary Tables 7–11).

Intercorrelations among KP Metabolites and Brain ROIs at Baseline

We examined intercorrelations among KynA/QA, KynA, QA, and structural brain volumes in the ROIs at baseline. Results are summarized in Supplementary Table 2. As shown, QA was negatively correlated with bilateral caudate volumes (left: r = −0.47, p < 0.05; right: r = −0.49, p < 0.05). There were no additional significant correlations between the primary outcomes of interest. Exploratory analyses revealed several significant associations with other KP metabolites (Supplementary Table 3). Notably, 3-HK, a precursor of QA, was negatively correlated with bilateral putamen volumes (left: r = −0.51, p < 0.01; right: r = −0.50, p < 0.01).

Associations Between Pre- to Post-Therapy Changes in KP Metabolites and Changes in Structural Brain ROIs

Because pre- to post-treatment GMV changes were not associated with symptom change, the following analyses were conducted to examine whether within-person changes in KP metabolites covaried with structural remodeling independent of clinical response. Increases in KynA/QA were significantly associated with greater increases in right hippocampus volume (β = 0.159, p < 0.05). Increases in KynA alone were also associated with greater increases in right hippocampus volume (β = 0.183, p < 0.01) and right amygdala volume (β = 0.234, p < 0.05). Similarly, increases in QA were associated with increases in right hippocampal (β = 0.148, p < 0.05) and right nucleus accumbens (β = 0.224, p < 0.05) volumes. To better illustrate these effects, scatterplots of significant metabolite – ROI associations are presented in Figure 1.

Figure 1.

Associations between changes in kynurenine pathway metabolites and gray matter volume in ROIs

Note: All change (Δ) variables were calculated as post-treatment minus pre-treatment; negative values indicate decreases over time, and positive values indicate increases. Regression lines reflect standardized linear regression models adjusted for age, sex, BMI, therapy group, and intracranial volume.

Right hippocampus: ΔKynA (β = 0.183, p < 0.01), ΔKynA/QA (β = 0.159, p < 0.05), ΔQA (β = 0.148, p < 0.05).

Right amygdala: ΔKynA (β = 0.234, p < 0.05).

Right nucleus accumbens: ΔQA (β = 0.224, p < 0.05).

All p-values are FDR-corrected.

GMV = gray matter volume; KynA = kynurenic acid; QA = quinolinic acid; KynA/QA = KynA/QA ratio.

Baseline KP Metabolites as Predictors of Structural Brain Changes

There were no significant findings when examining the relationship between primary KP metabolites at baseline and changes in structural brain volume within our primary ROIs. Exploratory analyses revealed significant associations between baseline concentrations of TRP and changes in GMV. More specifically, higher baseline TRP levels were associated with increases in right caudate volume (β = 0.191, p < 0.01) and decreases in right hippocampal volume (β = −0.213, p < 0.01).

Discussion

This study investigated whether GMV was altered by BA and how therapy-associated changes in KynA, QA and KynA/QA were related to GMV in cortico-limbic and striatal regions. There were four main findings. First, there was a trend for decreased GMV in a priori ROIs over time, though only decreases in the left putamen were statistically significant. Second, baseline QA was significantly (inversely) associated with bilateral caudate volume. Third, contrary to expectations, increases in both KynA and QA were significantly associated with greater increases in right hippocampal and right amygdala volumes. Finally, in prospective models, larger pre- to post-treatment decreases in QA at 10 weeks predicted lower depression and anxiety at 3-month follow-up. We discuss each of these findings below.

BA was not associated with widespread changes in brain structure. Only the left putamen among our predefined ROIs exhibited pre- post-therapy volume changes, showing reduced volume after therapy. This finding aligns with prior research demonstrating that psychotherapy, including CBT, can elicit regionally specific effects on brain morphology (Tai et al., 2021). The putamen is densely innervated by dopaminergic projections from the substantia nigra and plays an important role in motivation and reward processing (O’Doherty et al., 2004; Park et al., 2017). These domains are frequently disrupted in depression, particularly in the form of anhedonia and amotivation, and directly targeted by BA (Dimidjian et al., 2011). Elevated inflammation has been associated with reduced dopamine synthesis, release, and receptor availability in the striatum, contributing in particular to the manifestation of anhedonia (Felger & Treadway, 2017). This mechanistic pathway suggests that motivational deficits in depression may stem in part from inflammation-induced dopaminergic suppression, which BA may ameliorate through re-engagement of reward systems.

The direction of the change was unexpected but could conceivably reflect synaptic pruning or shifts in glial composition. When considering changes over the course of weeks, MRI-based GMV alterations are more likely to reflect neuroplastic processes such as dendritic spine turnover, synaptogenesis, or glial changes, rather than maladaptive atrophy (May, 2011). Indeed, short-term longitudinal studies have shown that structural MRI can capture dynamic, experience-dependent brain plasticity following behavioral interventions or learning (Sagi et al., 2012), suggesting that the observed changes in GMV in the current study may index adaptive restructuring rather than degeneration. This potential adaptive plasticity is consistent with prior literature suggesting subtle structural remodeling, including region-specific reductions in GMV with psychotherapy (Garrett et al., 2021). Volume reductions in this region could indicate a normalization of overactive cortico-striatal loops that underlie behavioral patterns that maintain or exacerbate depression, which are targets of BA’s emphasis on increased engagement in adaptive activities (Dimidjian et al., 2011). BA encourages individuals to identify and commit to meaningful activities, aiming to disrupt maladaptive behavioral routines and re-engage motivational circuits (Dimidjian et al., 2011). Conceivably, structural reductions in the putamen may reflect reduced reliance on avoidance or maladaptive behaviors. Notably, the volumetric decrease in the left putamen was not associated with KP metabolite changes, suggesting that the structural plasticity of the left putamen may not be primarily mediated by the KP - at least within the 10-week treatment window. While not statistically significant, the direction of change across most ROI also trended toward decreased volume. Collectively, these findings suggest that BA does not produce widespread GMV increases, but may instead promote subtle or circuit-dependent remodeling.

The second main finding was that at baseline, higher circulating concentrations of QA were negatively correlated with left and right caudate volumes while 3-HK was inversely associated with bilateral putamen volume. These results replicate some of our previous reports in independent samples and are broadly consistent with the hypothesis that 3-HK and QA, an NMDA agonist, can exert neurotoxic effects in NMDA receptor-rich subcortical regions (Savitz et al., 2015b; Savitz et al., 2015c). The striatum exhibits significant expression of NMDA receptors, particularly in the nucleus accumbens but with intermediate levels in the caudate and putamen (Monaghan & Cotman, 1985). In contrast, KynA and KynA/QA were not associated with GMV in our primary ROIs although weak positive trends were observed in the hippocampus and amygdala.

Third, we observed that within-person changes in KP metabolites were associatd with changes in GMV in limbic and striatal regions. More specifically, increases in both KynA, KynA/QA, and QA were significantly associated with increases in right hippocampal volume after BA. KynA was also associated with increased in amygdala volume while QA was also associated with increases in volume of the nucleus accumbens. However, given that these associations did not map onto clinical improvement, they should not be interpreted as markers of therapeutic response. KynA, an NMDA receptor antagonist with demonstrated anticonvulsant and neuroprotective effects (Croucher, Collins, & Meldrum, 1982; Schwarcz, Collins, Parks, 1982), has been shown to reduce excitotoxicity (Foster et al., 1984). QA, an NMDA agonist is neurotoxic at high enough concentrations. NMDA receptors are integral to synaptic plasticity mechanisms, such as long-term potentiation (LTP) and long-term depression (LTD), which underlie learning and memory processes. Nevertheless, the finding that both KynA and QA were positively associated with brain volume in overlapping regions complicates a straightforward interpretation as KynA and QA are usually considered to exert opposite effects. First, although QA was associated with increased right hippocampal volume, so too was KynA/QA suggesting that the positive association between ΔQA and ΔGMV may be reflective of upstream changes in KP metabolism. Second, there was a general pattern of decline in GMV from pre-to-post therapy. In this context, the direction of the expected correlations between ΔKynA, ΔQA and ΔGMV is unclear because of the counter-intuitive effect of BA on GMV in some brain regions. Accordingly, these GMV findings are best viewed as non-specific indicators of concurrent neuroplastic processes. In contrast, only decreases in QA showed prospective associations with both depression and anxiety symptoms at 3-month follow-up. This pattern underscores reductions in QA, rather than changes in GMV, as the finding most directly linked to clinical outcome in the present study.

Fourth, BA was associated with pre- to post-treatment decreases in QA, and greater QA reductions predicted lower depression and anxiety symptoms at 3 months. This prospective association was specific to QA and was not accounted for changes in GMV, further supporting QA reduction as a clinically meaningful signal independent of structural brain change. This pattern suggests that decreases in QA may be a marker of a biologically meaningful therapeutic change that supports more enduring improvement. Notably, symptom change was not related to GMV change, suggesting that immunometabolic shifts and structural remodeling may follow partly independent trajectories during recovery. Taken together, these findings highlight QA reduction as a candidate marker of therapeutic response.

Our results are consistent with a model in which BA increases engagement with reinforcing activities, reducing stress, and shifting tryptophan metabolism away from the “neurotoxic” branch and/or toward KynA. Via effects on glutamatergic/NMDA signaling, such immunometabolic changes could facilitate plasticity in hippocampal-striatal circuits that is observable as GMV change (Savitz et al., 2015b; Savitz et al., 2015c). However, the study design does not allow us to rule out the possibility that changes in KP metabolites play no causal role and are a secondary manifestation of clinical improvement. Nevertheless, the association between reductions in QA at 10 weeks and clinical improvement at 3 months provides some support to our hypothesis that changes in the KP may have therapeutic effects.

An interesting finding in our study was the preponderance of associations between changes in KP metabolites and changes in GMV within the right hemisphere. The reason for this lateralized effect is unclear. Nevertheless, it echoes our prior findings demonstrating significant associations between elevated neurotoxic kynurenine metabolites and right lateralized changes in cortical thickness in an independent sample of individuals with MDD (Meier et al., 2016). Such hemispheric asymmetry may indicate lateralized vulnerability or compensatory mechanisms related to emotional and cognitive processing deficits commonly observed in depression. Specifically, the right hemisphere has been implicated in emotional regulation, negative affect, and stress response --processes frequently disrupted in depression (Demaree et al., 2005). Some but not all neuroimaging studies have reported GMV reductions in the right hippocampus and amygdala among individuals with MDD (Hastings et al., 2004; Liang et al., 2023). The present right-lateralized GMV associations were observed alongside a prospective symptom signal for QA, suggesting that right-hemisphere plasticity and immunometabolic shifts may proceed in parallel yet partly independent pathways.

Several limitations must be considered. First, the modest sample size limited the power to detect subtle effects and could have contributed to false positives as well as increased uncertainty in responder versus non-responder contrasts. Second, the absence of a placebo or control group limits causal inference, as it is not possible to disentangle therapy-specific effects from non-specific factors such as time or participant expectations. Third, KP metabolites were measured in serum and may not fully capture central KP activity. Although kynurenine and 3-HK readily cross the blood brain barrier, KynA does not. Nevertheless, serum QA concentrations have been shown to correlate significantly with CSF QA levels (Haroon et al., 2020; Savitz, 2022). Fourth, structural MRI does not fully capture dynamic neural remodeling and incorporation of functional, metabolic or CSF-based measures of neuroplasticity or inflammation may have provided additional insight. Relatedly, prior work on antidepressant and electroconvulsive therapy outcomes has implicated white matter lesion burden and diffusion-derived white matter microstructure as correlates or predictors of clinical outcomes (Korgaonkar et al., 2014; Lyden et al., 2014; Taylor et al., 2013). Because the present study did not include diffusion imaging or tract-specific white matter measures, we could not evaluate whether kynurenine pathway changes track white matter plasticity within hippocampus, amygdala, or basal ganglia. Fifth, because KP metabolites and structural MRI were assessed only at pre-treatment and immediate post-treatment, we could not evaluate whether KP and GMV changs unfold on different timescales of whether the direction or magnitude of KP and GMV associations differs when biological change is assessed after a longer period of symptom improvement and behavioral consolidation. Additional within-treatment and follow-up biological assessments will be necessary to determine temporal ordering and to distinguish transient changes from more stable longer-term structural adaptations. Lastly, we did not measure changes in anhedonia which may have been more sensitive to immunometabolic modulation than depressive symptoms per se.

In summary, although few brain regions showed significant structural change following BA, the left putamen, a region critical for reward processing and motivational salience, exhibited a volume decrease. This reduction may reflect a key mechanism of BA, potentially indicating a normalization of maladaptive habit circuits and reduced reliance on rigid, avoidant behavioral patterns. Pre-post therapy increases in KynA and KynA/QA were associated with volume increases in the right hippocampus and amygdala. However, contrary to expectations, increases in QA were also associated with volume increases in the same regions. Nevertheless, the significant positive associations between KynA/QA and KynA/Kyn with hippocampal volume raise the possibility that changes linked with BA in GMV may reflect neuroplasticity that is driven by preferential metabolism down the KynA in the context of a general upregulation of the KP. Crucially, greater pre- to post-treatment decreases in QA prospectively predicted lower depression and anxiety symptoms at 3 months. These findings highlight QA as a candidate marker of therapeutic response that warrants replication in larger, well-powered longitudinal samples. Ultimately, this line of research may help refine biologically informed models of depression recovery and support the development of more personalized and targeted therapeutic strategies.

Highlights.

• BA was associated with reduced left putamen gray matter volume over 10 weeks

• KynA and KynA/QA increases were associated with greater right hippocampal volume change

• KynA increases were associated with greater right amygdala volume change

• QA increases were associated with greater right hippocampal and nucleus accumbens volume change

• Larger QA decreases predicted lower depression and anxiety symptoms at 3-month follow-up

• Findings suggest psychotherapy engages immunometabolic pathways for neuroplasticity