In the current study, we examined the associations among white matter FW, peripheral inflammation, and depressive symptoms in adolescents with MDD, testing whether inflammatory cytokines might play a moderating role in the relationship between altered white matter FW and depressive symptoms. To investigate this, we collected 3-T multi-shell dMRI data and measured 10 pro-/anti-inflammatory peripheral cytokine levels in medication-free adolescents with MDD and healthy controls. Our hypothesis was that inflammatory cytokines would play a moderating role in the relationship between altered white matter FW and depressive symptoms in medication-free adolescents with MDD.
A total of 84 HC and 63 patients diagnosed with MDD were recruited for the cross-sectional of this study. The ethics committees of the Affiliated Brain Hospital of Guangzhou Medical University approved this study (ethical approval number: 2020055), and the written informed consent was obtained from all participants and their parents or legal guardians.
Both MDD patients and HC underwent a structured clinical interview based on the diagnostic and statistical manual of mental disorders, fifth edition. MDD patients met specific inclusion criteria, including being between the ages of 12 and 18, scoring ≥ 17 on the 17-item Hamilton Rating Scale for Depression (HAMD), and being medication-free for at least 4 weeks. All MDD patients were recruited at the outpatient department of the Affiliated Brain Hospital of Guangzhou Medical University. Age and education level-matched HC were recruited through targeted advertising efforts from local community. Exclusion criteria for all participants were a history of developmental disorders, tic disorders, or attention deficit hyperactivity disorder, any other mental illness (e.g., bipolar disorder, posttraumatic stress disorder, and schizophrenia), neurological disorders or other major physical illnesses, immune-inflammatory disorders, and contraindications to MRI scanning. For HC, additional exclusion criteria comprised any history or current diagnosis of MDD or other psychiatric conditions.
The HAMD is divided into five symptom dimensions. The Anxiety/Somatization dimension includes six items related to mental and physical anxiety, gastrointestinal symptoms, general symptoms, hypochondriasis, and self-awareness (Items 10, 11, 12, 13, 15, 17). The Retardation dimension consists of four items addressing depressed mood, work and interest, psychomotor retardation, and sexual symptoms (Items 1, 7, 8, 14). The Cognitive Disturbance dimension includes three items related to feelings of guilt, suicidal thoughts, and agitation (Items 2, 3, 9). The Sleep Disruption dimension assesses difficulty falling asleep, shallow sleep, and early awakening (Items 4, 5, 6). Finally, the Weight dimension evaluates weight loss with a single item (Item 16). Anxiety symptoms were assessed using the Hamilton Anxiety Rating Scale (HAMA) scale.
Upon enrollment in the study, whole blood samples were collected from the participants. All blood samples were collected after a minimum 4-hour fasting period, which was consistent with previous study [16]. Fasting status was confirmed with each participant. Plasma was separated within 30 min of collection using standard centrifugation protocols (3000 rpm for 10 min at 4 °C) and immediately stored at -80 °C until analysis. Plasma was then separated, aliquoted into Eppendorf tubes, and stored at -80 °C for further analysis. The levels of ten peripheral cytokines (interferon gamma [IFN-γ], IL-10, IL-1b, IL-2, IL-4, IL-6, IL-8, Tumor Necrosis Factor-alpha [TNF-α], C-reactive protein [CRP], and Complement component 4 [C4]) were detected by the human high sensitivity T cell magnetic bead panel (Millipore, Billerica, MA, USA, HSTCMAG-28SK) and human neurodegenerative disease magnetic bead panel 2 (Millipore, HNDG2MAG-36K) with the Luminex Magpix-based assay (Luminex corporation), following the manufacturer's instructions. Data generated from the assay were evaluated against a cubic curve fitting and corrected for background readings using Millipore Analyst 5.1 Software (EMD Millipore, Billerica, MA) [17]. In order to achieve normality for statistical analysis, natural log-transformation was applied to all peripheral cytokine values. The intra- and inter-assay coefficients of variation were below 10 and 15%, respectively.
Participants were scanned using a Siemens Magnetom Prisma 3.0 T MRI Scanner, equipped with a 64-channel head coil, at the Magnetic Resonance Center of Affiliated Brain Hospital of Guangzhou Medical University. Multi-shell dMRI was acquired in the posterior to anterior direction with 64 gradient directions at b = 1000 s/mm, and 2000 s/mm, and 10 interleaved b = 0 s/mm images. An additional 64 gradient directions at b = 3000 s/mm2 were collected for tractography studies. However, this b = 3000 shell was excluded from all free water modeling calculations to circumvent non-Gaussian diffusion effects that could bias the two-compartment model estimation. To correct for susceptibility-induced distortions, 10 images (b0) with b = 0 s/mm were collected with anterior to posterior phase encoding.
The diffusion sequence was acquired with the following settings: repetition time = 2500 ms, echo time = 83 ms, field of view = 220 mm, slice thickness = 2 mm, voxel size = 2 × 2 × 2mm.
During scanning, all dMRI images were visually screened by a trained technician (L.S.) to identify any abnormal radiological or structural features. The Statistical Parametric Mapping software (http://www.fil.ion.ucl.ac.uk/spm) was employed to register all dMRI images to their first b0 images for further validation. No participants were excluded from subsequent analysis based on these screenings. For the dMRI data, image intensities were initially normalized using the mean b0 image. The b0-inhomogeneity distortion was corrected using two opposite phase-encoded images and the "topup" tool in FSL [18]. Additionally, the "eddy" tool in using FSL 6.0.6.5 was employed to correct eddy-current induced field inhomogeneities and head motion for each image volume in a single resampling step [19].
Free-water modeling and tensor estimation were performed using DIPY (version 1.10.0) (http://nipy.org/dipy/index.html), with the two-compartment free-water model fitted to preprocessed diffusion data using dipy.reconst.fwdti. FreeWaterTensorModel with default parameters. In each voxel, the signal was fitted to a two-compartment model, consisting of a FW compartment (isotropic tensor) and a tissue compartment (FW-corrected tensor) [4]. The FW measure represents the relative contribution of FW in each voxel, ranging from 0-1. The tensor of the tissue compartment reflects the tissue microstructure after removing the signal contributed by FW. This study evaluated both the FW component and FAt. Notably, unlike most previous studies, the model was estimated from multi-shell diffusion imaging data, which offers enhanced stability and robustness compared to single-shell data commonly used in dMRI studies [20].
Voxel-wise analysis was performed using an automated Tract-Based Spatial Statistics (TBSS) pipeline [21]. Initially, individual fractional anisotropy (FA) maps were nonlinearly aligned to a standard space using a target image, which was selected as the most representative FA image (designated with the flag '-n'). This option is recommended for studies involving adolescents and young children. The chosen target image belonged to a 15-year-old participant from the MDD group. Following image registration, a cross-subject mean FA image was generated, which guided the creation of the white matter (WM) tract "skeleton". The threshold for the skeleton was set at FA > 0.2 to include major WM pathways while excluding peripheral tracts that are more susceptible to partial volume effects and inter-subject variability. Subsequently, each subject's FW and FAt data were projected onto the group skeleton for voxel-wise analysis.
For statistical analysis, the randomize function within FSL was utilized to conduct permutation-based nonparametric statistics with 5000 permutations [22]. Significant differences were identified using a p value image, where p < 0.01, corrected for multiple comparisons across space through threshold-free cluster enhancement (TFCE) [23]. Anatomical locations were identified using the ICBM-DTI-81 atlas [24].
Continuous variables were expressed as the mean and standard deviation (SD) or the median and interquartile range (IQR), while categorical variables were presented as frequencies and proportions. Demographic and clinical variables were analyzed using independent t-tests, the Mann-Whitney U-test, and Chi-square tests, respectively. The normality of the data was assessed using the Shapiro-Wilk test. Statistical significance was defined as p < 0.05.
A general linear model (GLM) was employed to examine the main effect of diagnosis (HC or MDD) on individual variation in peripheral cytokine levels, while controlling for age and sex as confounding factors in Model 1. Model 2 was additionally adjusted for smoking status (yes or no), drinking status (yes or no), somatic comorbidities (with or without), and family history (yes or no) to assess stability based on Model 1. Each of the ten peripheral cytokines was tested independently, and Bonferroni correction was applied to account for multiple testing.
Between-group comparisons of white matter dMRI parameters, specifically FW and FAt, were conducted using Randomise in FSL. A voxel-wise GLM analysis was performed to assess the equality in FW and FA between the MDD and HC groups while accounting for age and sex as covariates. TFCE was applied for multiple comparison corrections across all skeleton voxels.
Partial correlation was used to test for associations between white matter microstructure (mean FW values extracted from voxels showing significant between-group differences) and depressive symptoms (HAMD scores). We averaged FW values across significant voxels based on: (1) the global nature of systemic inflammation effects on brain tissue; and (2) increased statistical power while reducing multiple comparison burden. Then, we created general linear models that tested the main effects of cytokine levels and white matter microstructure and their interaction. Significant interactions were further analyzed using simple slopes [25], and the Johnson-Neyman method was employed to identify the significant region of moderation [26].
We repeated all regression analyses controlling for additional confounding variables, and removing outliers to test whether the result was robust. Given the small proportion of male patients in our dataset (9 males, 9.5%), we performed the analysis excluding them. We also use HAMA scores, instead of HAMD scores, as outcome variable to test the symptom specificity.