Study design, selection criteria, and study participants
The following case-control study was conducted at the University Hospital of Giessen, involving participants with depressive episodes (DE, n = 25) and healthy control participants (HC, n = 31). The data analyzed in this study is part of a broader investigation of depression and fatigue, encompassing various techniques, including MRI, fMRI, DTI, and EEG, across different sample groups. Previous findings from this project have been published elsewhere (Pedraz-Petrozzi et al., 2020). Data were gathered from diverse physiological (e.g., electroencephalographic recordings) and psychological (e.g., sustained attention tasks) domains to address the research questions and hypothesis proposed in this work. Alongside electroencephalographic recordings and sustained attention tasks, inflammatory parameters were assessed through blood sampling, and behavioral and psychological data were collected via questionnaires.
Inclusion criteria for participants required them to be between 18 and 65 years of age, capable of providing informed consent, and fluent in German. Individuals with major depressive disorder or a depressive episode without a concomitant psychotic episode were eligible for inclusion. Patients with depression and psychiatric comorbidities, such as personality disorders, adaptation syndrome, and post-traumatic stress disorder, were included only if their depressive episode was clinically present and predominant during the evaluation (Table S2). In the case of the participants with bipolar depression (n = 2), they were free of manic symptoms for the last six months and have had predominantly depressive episodes. Exclusion criteria for both groups consisted of insufficient knowledge of the German language and somatic or cognitive limitations that hindered participation (e.g., visual or auditory impairments). Any somatic disease or illness, except for depression or depressive episodes, were considered exclusion criteria. Table 1 provides an overview of the general baseline characteristics. Information regarding sociodemographic status, medication intake, diagnosis, and disease duration is described in the supplementary material (Tables S1-S3).
The estimated power of the total sample was calculated using G-Power (Faul et al., 2007) (N = 56, 1 − β = 0.99, f2 = 0.25, α = 0.05), and it exceeds the accepted minimum power threshold (1 − β = 0.80). This indicates that the sample size for this study design is sufficient to achieve the study objectives. This study was conducted following the Helsinki Declaration and received ethical approval from the local ethics committee of the Justus-Liebig University medical faculty. Before participating in the study, all participants provided verbal and written informed consent. Additionally, this study adheres to the ethical standards outlined by the American Psychological Association.
Table 1
– General baseline characteristics of the included participants. Values are expressed as mean (standard deviation). *Differences between groups were calculated using Pearson’s χ2-test. **Cytokines’ concentrations were log-transformed. Abbreviations: IL = interleukin, IL-1β = interleukin 1 beta, TNF-α = tumor necrosis factor-alpha.
| DE (n = 25) | HC (n = 31) | p |
Age | 31.24 (12.70) | 24.71 (5.89) | 0.02 |
Sex (f:m)* | 19:6 | 21:10 | 0.50 |
Perceived fatigue (0-160) | 95.88 (25.05) | 37.42 (23.44) | < 0.001 |
Cytokines** | | | |
IL-6 (in pg/mL) | 0.55 (0.15) | 0.49 (0.12) | 0.10 |
IL-1β (in pg/mL) | 0.60 (0.65) | 0.44 (0.37) | 0.29 |
TNF-α (in pg/mL) | 1.25 (0.13) | 1.21 (0.09) | 0.22 |
Collection of plasma samples
The plasma sample collection procedures were adapted from previous reports of a broader project with depression and fatigue, encompassing various techniques, including MRI, fMRI, DTI, and EEG, across different sample groups, and which have been published elsewhere (Pedraz-Petrozzi et al., 2020). Between 8:00 a.m. and 12:00 p.m., venous blood samples of participants who had undergone fasting before the assessment were collected, totaling 3 mL. EDTA-K blood sample tubes (SARSTEDT AG & Co. KG, Nümbrecht, Germany) were utilized for the sample collection. Subsequently, the tubes were centrifuged at 4°C, with a force of 1100 x g, for 15 minutes. The plasma was carefully extracted and divided into two 0.5 mL aliquots after centrifugation. These aliquots were transported at -20°C to a university research facility, where they were stored at -80°C for further analysis of inflammatory parameters. The levels of interleukin (IL)-6, IL-1β, and TNF-α were determined using ELISA (Quantikine ELISA kits, R&D Systems Inc., Minneapolis, Minnesota, United States of America) with detection limits as follows: IL-6 = 3.13 pg/mL, IL-1β = 3.9 pg/mL, and TNF-α = 15.6 pg/mL. The intra- and inter-precision values were both < 10%. Concentrations of the pro-inflammatory markers were calculated using the Tecan Reader and Magellan Reader Software (Tecan Group Ltd., Männedorf, Switzerland) employing Marquardt's 4-parameter estimation method for parameter calculation.
Sustained attention task - paradigm and measurement procedure
A sustained attention test was carried out with a corresponding subtest of the Test Battery on Attention (TAP) version 2.3.1 (Zimmermann and Fimm, 2002). The objective of this 15-minute task was to assess cognitive resource utilization through sustained attention and investigate fatigue or mental exhaustion processes over time. The stimulus sequence on a computer screen contained stimuli of different colors, shapes, sizes, and fillings. Subjects were asked to respond to target stimuli as quickly and accurately as possible. The targets matched one of the previous stimuli in one of the characteristics (i.e., same color, shape, size, or filling). More details about the task can be found at https://www.psytest.net/. The test application in the current study consisted of a sequence of 450 images, each shown for 500 milliseconds. The interstimulus interval between stimulus onsets was 2000 milliseconds. Response times and accuracy (number of correct answers) were recorded.
Evaluation of perceived fatigue – Fatigue Impact Scale
This study evaluated the perceived fatigue psychometrically using the Fatigue Impact Scale - German Version (FIS-D) (Häuser et al., 2003). The FIS-D assesses the impact of fatigue on Health-Related Quality of Life using a three-dimensional structure consisting of three sub-scales: psychosocial (PSY-F; 20 questions and a maximum score of 80 points), somatic (SOM-F; 10 questions and a maximum score of 40 points), and cognitive (COG-F; 10 questions and a maximum score of 40 points). The maximum score achievable is 160 points (4 points for each item). According to the FIS-D manual, the cut-off values for increased fatigue are > 20 points for the psychosocial dimension, > 10 points for the somatic or cognitive dimension, and > 40 points for the overall test (Pedraz-Petrozzi et al., 2020).
Electroencephalography (EEG) – Procedures and analysis of electrophysiological data
During the sustained attention task, a 32-lead EEG was recorded using BrainVision EEG recording hardware and software (Amplifier DC; Brain Vision Recorder; Brain Products GmbH, Germany). The electrodes were mounted with EEG caps (extended 10–20 System; https://www.easycap.de) using electrode gel Abralyt 2000 (EASYCAP GmbH, Germany). Electrode resistance was kept below 5 kΩ. Data sampling was performed at 500 Hz. EEG recordings were conducted in an electrically shielded and soundproofed room with each subject alone and free from distractions.
For this study, the channel Pz was considered for ERP analysis, as recommended in the literature (Howe et al., 2014; Käthner et al., 2014; Trongnetrpunya et al., 2019). EEG data analysis was conducted using BrainVision Analyzer, version 2.2.0 (Brain Products GmbH, Germany).
The EEG data preprocessing involved several steps. First, bandpass filtering was applied between 2 Hz (slope 4 dB/oct) and 15 Hz (slope 8 dB/oct). Automatic eye artifact reduction was performed using Independent Component Analysis (ICA), with the vertical electrooculogram (vEOG) used for blink detection. Further automatic artifact reduction was carried out with the following criteria: maximum gradient of 50 µV/ms, maximum difference of 200 µV within a 200 ms interval, minimum amplitude of -200 µV, maximum amplitude of 200 µV, and a minimum activity of 0.3 µV within a 100 ms interval. Re-referencing was performed for all EEG channels (except TP9 and TP10) to an average reference. Next, the entire data set was divided into segments for all correctly answered stimuli. Each segment lasted 800 ms and started 50 ms before the corresponding stimulus occurrence. To remove baseline offsets, the segments were baseline-corrected using the mean amplitudes of 10 ms before and after the stimulus. After preprocessing, the artifact-free segments were averaged, and the P300 components (search window 270 ms – 370 ms) were semi-automatically identified at the Pz electrode, with a visual check for potential misidentifications. The resulting peak amplitudes (in µV) and latencies (in ms) were used for the subsequent statistical analysis.
Statistical analysis
Quantitative variables were presented in tables as either mean (standard deviation) or median (interquartile ranges), while categorical variables and count data were expressed as frequencies or fractions. Decimal values were rounded to the nearest decimal point, and values smaller than 0.01 were denoted as < 0.01. Furthermore, the ERP components (P300 amplitude and latency) were subjected to z-transformation, following recommendations (John et al., 1993; Wang et al., 2000). Likewise, the cytokine concentrations underwent a log transformation due to the skewed distribution in the original dataset of this sample, aiming to approximate a normal distribution, as recommended elsewhere (Fjell et al., 2013; Janelidze et al., 2011; Keene, 1995).
P300 data were analyzed using linear mixed models (LMM), estimated using the restricted maximum likelihood (REML) method. Initially, we examined the effects of group status (DE vs. HC), sex, stimulus (target vs. non-target), participants' age, perceived fatigue, logIL-6, logIL-1β, and logTNF-α, as well as the two-way interactions (group status × sex; group status × stimulus; and sex × stimulus), and three-way interaction (group status × sex × stimulus). The ID variable was utilized as a cluster variable (intercept) for the random effects. In the LMM, fixed effect omnibus tests were employed to determine the main effects of the factors in the model and to compare the model against the null model (supplementary material, tables S4-S5). Results of the LMM were presented in tables and graphically, and two-tailed p-values and 95% confidence intervals (95CI) were used to describe the results in the text. For this study, significance was defined as p ≤ 0.05.
Two separate generalized linear models (GLZM) were conducted to assess the effects of sex, group status (DE vs. HC), participants' age, perceived fatigue, logIL-6, logIL-1β, and logTNF-α, as well as the two-way interaction (group status × sex), on the total number of correct answers (i.e., recognized targets) and the mean response time (RT). For the total number of correct reactions during the trial, a Poisson regression was computed due to the count nature of the dependent variable, which followed a Poisson distribution (Kolmogorov-Smirnov-Test for goodness-of-fit for Poisson distribution: Z = 0.90, p = 0.39). In cases of significant two-way or three-way interactions, post hoc tests were performed. Statistical significance was defined as a two-tailed p ≤ 0.05 for these effects. Loglikelihood tests were computed for the GLZM to determine the goodness of fit of the main effects of the variables in the model (supplementary material, tables S6-S7).
All statistical analyses were performed using the R-based software jamovi 2.0.0 (Love et al., 2020) and the toolbox GAMLj (Gallucci, 2019).