Gamma and beta power and the 1/f slope vary across a spectrum of depression severity

Abstract

Depression is a highly prevalent mental disorder that impacts an individual’s functioning, societal productivity, and quality of life. It is associated with disrupted neural activity (e.g., balance of excitation-inhibition) across networks implicated in emotional processing, such as between prefrontal and limbic regions. Gamma and beta activity measured with electroencephalography (EEG) differ between healthy individuals and those with depression, and predict treatment outcomes following pharmacological intervention. However, to date research has focussed on binary comparisons between individuals with a clinical depression diagnosis relative to healthy control populations, providing limited insight into how these measures may shift as a function of illness severity. To establish the utility of EEG measures as potential biomarkers for depression, an improved understanding across the spectrum of symptom profiles is required. Here, we aimed to bridge this gap and investigate changes in beta and gamma power and the 1/f slope in resting-state EEG across a spectrum of mild to severe depression symptom presentations. In line with expectations, we demonstrate graded alterations to gamma and beta power and the 1/f slope across a spectrum of depression severity. Our findings provide critical new insights into the neurophysiological signature of depression symptoms, and highlight the utility of EEG markers to inform future precision psychiatry approaches to more effectively assess and treat depression.

Introduction

Depression is a highly prevalent mental disorder characterized by low mood and/or anhedonia, which can have devastating impacts on daily functioning, social engagement, and occupational productivity [1, 2]. Although depression symptoms are notoriously heterogeneous [3], neuroimaging studies indicate they are commonly associated with disruptions across networks implicated in emotional processing (e.g., frontolimbic regions; for review, see: [4]). Identification of neural markers that reflect communication within these neural circuits may assist in tailoring treatment approaches to individual clinical presentations.

Electroencephalography (EEG) is an accessible way to measure neural communication. The EEG signal at a given electrode is comprised of both periodic, oscillatory activity, and aperiodic activity. The overall signal can be decomposed into the relative contribution of each frequency band in the form of a power spectrum. Different compositions of spectral power underpin neural communication between spatially distinct brain regions ([5, 6], see Fig. 1A (left)). For example, synchronization of high frequency oscillations facilitates emotional expression recognition by integrating information from several spatially distinct brain regions [7]. Individuals with depression have exhibited disruption in connectivity between brain regions involved in processes such as emotional regulation, emotional perception, and social cognition [8], likely related to aberrant oscillatory communication. In particular, gamma and beta activity have garnered interest as possible markers of depression severity [9, 10].

Fig. 1. Theoretical framework for mapping EEG features to depression symptomology.

A (left) Disruptions in neural circuits can be characterized via analysis of oscillatory activity measured by non-invasive EEG. Past studies have identified elevated gamma and beta power in individuals with depression. A (right) The 1/f slope provides an electrophysiological measure of excitation-inhibition balance, with steeper slopes reflecting greater inhibition and flatter slopes reflecting greater excitation. B Schematic representing the current study’s methodology and hypotheses. EEG data is transformed into the frequency domain, providing a power spectrum outlining the contributions of relative frequencies. This data also undergoes “spectral parameterization”, in which the aperiodic component of the signal (1/f slope) is extracted. Gamma and beta activity and the 1/f slope are expected to be sensitive to increasing depression severity. C The current study contributes to efforts to better tailor treatment to individuals by providing potential markers that, in conjunction with other established measures, can stratify individuals into subgroups based on their neural characteristics. Created with BioRender.com.

Gamma oscillations (30 Hz and greater) index the rhythmic firing of inhibitory interneurons [11], which occur as part of a negative feedback loop in the communication between excitatory and inhibitory neurons [12]. Thus, gamma oscillations are thought to reflect a region being “active” [13], and have also been used as a measure of the ratio of excitation to inhibition across the cortex [14]. Gamma oscillations are involved in numerous cognitive and affective processes relevant to depression such as perceptual awareness, attention, memory, [15], and emotional processing [16, 17].

Several studies have identified significant differences in resting-state gamma power when comparing depression and healthy control groups ([18–20]; for review, see [9]), and also demonstrated that gamma power differentiates individuals with depression from those with other psychiatric diagnoses, such as bipolar disorder [17, 21]. Further, gamma power is altered following antidepressant treatments, such as ketamine and repetitive transcranial magnetic stimulation [22–24]. However, there is also evidence that oscillatory changes in depression are not specific to gamma. Differences in the beta band (~14 Hz to 30 Hz) have also been observed between depression and healthy controls [25–27], and following ketamine treatment [28]. Beta power is thought to relate to working memory and attention, which are negatively impacted in depression [29], and increased beta power is correlated with the number of depressive episodes and refractory depression [25–27, 30, 31]. Taken together, these findings suggest that gamma and/or beta power could be a viable neural marker for identifying participants with depression. However, it remains unclear how these markers may map onto the spectrum of depression severities.

In addition to oscillatory patterns, there is evidence that the aperiodic 1/f slope of the EEG signal may reflect excitation-inhibition (E/I) balance across the cortex ([32, 33], see Fig. 1A (right)). The slope is extracted by applying a model to the frequency transformed EEG data and can be described by its two parameters, the exponent (curvature of the slope) and offset (y-intercept), where the slope reflects characteristic differences between excitatory and inhibitory current decay and the offset reflects broadband shifts in power ([32], see Fig. 1A (right)). Differences in E/I balance have been reported between individuals with a history of depression relative to healthy controls [34–37]. However, measurements to date have been taken from either post-mortem tissue cell counts [34], or via magnetic resonance spectroscopy [35–37] with the latter providing an overall estimate of neurometabolite concentration that is thought to predominantly index metabolic and extrasynaptic processes [38]. Thus, measuring the 1/f slope with EEG may provide an alternative non-invasive way to index synaptic E/I balance, and examine corresponding changes in E/I balance across depression severities.

Recent studies have investigated differences in 1/f slope parameters between individuals with depression relative to healthy controls. Studies have observed lower exponent values (i.e., a flatter slope) in individuals with depression [39–41], reflecting increased excitation, and also decreased 1/f offset values [39, 41, 42], indicating shifts in spectral power. However, other studies have observed no differences [42]. Further, only two studies to our knowledge have examined how these parameters change across a spectrum of depression severity, using invasive intracranial electrode recordings in a small sample of individuals with refractory depression [43, 44]. Of these, one study reported lower exponent values with increasing depression severity in the ventromedial prefrontal cortex [43], and another reported higher exponent values with increasing depression severity within the subcallosal cingulate [44]. Assessing differences in the 1/f slope across a spectrum of depression severity in a larger sample using non-invasive EEG may provide a more accessible and scalable aggregate marker of E/I balance across the cortex (see Fig. 1B).

An improved understanding of how neural transmission is altered as a function of depression severity may advance screening and assessment of depression and facilitate more effective intervention (see Fig. 1C). To this end, the present study aimed to assess changes in gamma and beta power and the 1/f slope across a spectrum of depression severities (see Fig. 1B) using a large open-source dataset (the Brainclinics Foundation [45]). We hypothesized that both beta and gamma power would differ across individuals experiencing mild, moderate, and severe depression symptoms. We also hypothesized that the 1/f slope, a marker of E-I balance, would significantly differ across depression severities. To assess the reproducibility and robustness of observed effects, we examined these same associations in an independent dataset [46].

Methods

Data acquisition & participants

This study utilized open-source data from the Brainclinics Foundation (https://www.synapse.org/TDBRAIN), titled ‘Two Decades Brainclinics Research Archive for Insights in Neurophysiology’ [45], referred to here as TDBRAIN (2022). The dataset contains resting-state data recorded from a 26 channel EEG montage and self-reported depression symptom data from 1274 participants with mental health diagnoses collected between 2001 to 2021 [45] (age range 6–83 years, M = 37.48 ± 19.1 SD). The EEG sampling rate was 500 Hz and recorded relative to a virtual ground. The data was low-pass filtered at 100 Hz prior to digitization. BDI scores and eyes-open resting-state recordings (two minutes duration) were obtained from N = 181 for analysis in the current study. We restricted our analyses to the 181 participants for whom Beck Depression Inventory II (BDI) scores were recorded. Appropriateness of sample size was supported by previous studies reporting associations between gamma power and depression [18, 20, 47].

The beck depression inventory II

The Beck Depression Inventory II (BDI-II) is a well-validated 21-item self-report measure of depression symptoms in adults and adolescents [48, 49]. For each item, respondents are instructed to select the option that best describes how they have been feeling during the past two weeks. Each item is scored from 0 to 3 (e.g., 0 = I do not feel sad; 1 = I feel sad much of the time; 2 = I am sad all the time; 3 = I am so sad or unhappy that I can’t stand it). Final scores range from 0 to 63, with 0–13 classed as “minimal”, 14–19 as “mild”, 20–28 as “moderate”, and 29 and above as “severe” [48]. The scale is regarded as a valid and reliable measure of depression symptoms [50, 51], with an internal consistency of 0.9 and test-retest reliability from 0.73 to 0.96 [52].

Data analysis

EEG preprocessing

All EEG files were pre-processed using the Reduction of Electroencephalographic Artifacts (RELAX) automated pre-processing pipeline utilizing EEGLAB functions within MATLAB [53–55]. Cerebellar, ocular, and mastoid electrodes were excluded and line noise was removed using ZaplinePlus [56, 57]. The data was downsampled to 500 Hz, and bandpass filtered at (high-cutoff 0.5 Hz, 120 Hz low-cutoff) with the pop_eegfiltnew function [54]. Multi-Weiner Filtering (MWF) and Independent Component Analysis (ICA) (Picard algorithm) were employed to remove muscle and eye movement artifacts, with subsequent data interpolation. Default parameters were used for all other RELAX pre-processing steps. N = 8 were excluded on the basis of excessive noise remaining in resting-state EEG data following pre-processing.

Spectral power extraction

Pre-processed EEG files were imported into Brainstorm (2023), and the Welch Power spectrum density function was utilized to transform the data from the time domain to the frequency domain (window length = 4 s, overlap ratio = 50%, units = physical μV²/Hz). Frequency bands were specified as theta: 4–7 Hz, alpha: 8–12 Hz, low beta: 13–20 Hz, high beta: 21–29 Hz, low gamma: 30–45 Hz, and high gamma: 55–90 Hz (avoiding 50 Hz line noise).

Fitting the 1/f slope (Spectral Parameterization)

1/f slopes were calculated from frequency transformed data using the ‘specparam: Fitting oscillations and 1/f’ package in Brainstorm using the default parameters [32]. The frequency range for analysis was set at 1–40 Hz, with a gaussian peak model (peak width limits = 0.5–12 Hz, maximum number of peaks = 3, minimum peak height = 3 dB, proximity threshold = 2 standard deviations (SD) of peak model, fixed aperiodic model i.e., without a ‘knee’), yielding individual estimates of the aperiodic exponent and offset. Goodness of model fit metrics can be found in Fig. 2.

Fig. 2. Goodness of 1/f model fit.

Explained variance (R²) of the model fit and mean squared error, averaged across participants and electrodes, shown for each dataset.

Cluster-based permutation statistical analyses

For group-level analyses, outliers (+−3SD) were identified independently per dependent measure (beta and gamma frequency bands, 1/f exponent and offset values). This decision stemmed from concern that outlier removal on parameters calculated across all frequency bands (i.e., low beta up to high gamma) may result in values in the middle frequency bands (i.e., high beta and low gamma) being erroneously retained despite being abnormally low or high for their own band. Descriptive statistics following outlier removal can be found in Table 1.

Table 1.

Descriptive Statistics.

Cavanagh (2021) dataset
Outcome measure	All n	Control group n	Depression group n	Age M(SD)	Female %
Low beta	95	61	24	18.9(1.1)	42.1
High beta	94	61	23	18.8(1.1)	46.8
Low gamma	83	55	20	19.0(1.3)	51.8
High gamma	89	60	20	19.0(1.2)	49.4
Exponent	101	66	23	18.9(1.2)	42.6
Offset	103	65	25	18.9(1.2)	38.8

TDBRAIN (2022) dataset
Outcome measure	All n	Mild n	Moderate n	Severe n	Age M(SD)	Female %
Low beta	152	19	49	73	40.6(13.6)	49.3
High beta	150	20	47	72	39.9(13.3)	50.0
Low gamma	140	19	41	70	39.5(13.4)	48.6
High gamma	143	21	40	70	39.6(13.1)	48.3
Exponent	161	20	47	83	39.6(13.3)	47.2
Offset	165	20	50	84	40.2(13.7)	49.7

Few participants scored in the “minimal” symptom severity range in the TDBRAIN dataset (n = 12). As such, this range was removed from the analysis. The Cavanagh dataset was divided into controls (those scoring in the “minimal” range) and individuals experiencing depression (those scoring in the “moderate” and “severe” range).

To identify differences in spectral power and the 1/f slope as a function of depression symptomology, cluster-based permutation analyses were conducted using the Monte Carlo method in the MATLAB-based toolbox FieldTrip [58]. Cluster-based permutation analyses are non-parametric tests free from distributional assumptions [59]. Cluster-based Spearman correlations (two-tailed) were calculated between BDI-II scores and EEG outcomes of interest to assess associations between EEG measures (gamma and beta power, and 1/f estimates) and depression symptoms (cluster statistic = “maxsum”, number of randomizations = 10000, alpha threshold = 0.05), while accounting for skew in BDI symptom presentations. Cluster-based independent-samples t-tests were also conducted to identify differences between diagnostic categories (e.g., mild vs severe symptom presentations).

Inspection of the TDBRAIN (2022) dataset indicated ‘mild’ (n = 22), ‘moderate’ (n = 52), and ‘severe’ (n = 87) symptom presentations. Therefore, separate cluster-based independent t-tests were performed between these groups (i.e., mild vs moderate, mild vs severe, and moderate vs severe).

Dataset 2

While the TDBRAIN (2022) dataset encompassed a variety of scores and a large number of participants, it has some limitations. 52 participants were listed as having taken medication between 0 and 15 h prior to data collection, and EEG data was collected using a relatively low-density 26 electrode array. To test robustness of our findings and account for these limitations, we sourced a second dataset with a carefully characterized sample and a higher density electrode array [46]. This dataset (herein referred to as “Cavanagh (2021)”; [46]) was sourced from the OpenNeuro platform and contains resting-state data recorded from a 60 channel 10–20 EEG montage and self-reported depression symptom data from 122 young adults aged 18–24 (M = 18.57 ± 1.19 SD). The EEG data was recording at a sampling rate of 500 Hz, using a channel between Cz and CPz as a reference. A 0.5–100 Hz band-pass filter was applied. Participants had no history of head trauma or seizures and no current psychoactive medication use. “Control” participants had stable low BDI scores (remaining <7 for 2–14 weeks), no self-reported history of Major Depressive Disorder, and no self-reported symptoms associated with mental illness (previously referred to as Axis I disorders). Participants in the “depression” group had stable high BDI scores (remaining ≥ 13 for 2–14 weeks), and many were administered the Structured Clinical Interview for DSM-5, revealing almost all primary diagnoses were depression or anxiety-related.

Each resting-state EEG recording was six minutes in duration, with alternating epochs of 1-minute of rest with eyes open, and one minute with eyes closed. A custom MATLAB (2023b) script was used to exclude data during the eyes-closed condition to ensure consistency between datasets. Data from 6 participants were excluded on the basis of missing resting-state data (n = 5) or BDI-II scores (n = 1), leaving a total of N = 116.

Methodology for pre-processing and data analysis for this dataset was largely the same as for the TDBRAIN (2022) dataset, with the following exceptions:

• N = 3 participants were excluded due to excessive noise remaining in resting-state EEG data following pre-processing.

• In spectral power extraction, low gamma power was defined as 30–55 Hz and high gamma as 65–90 Hz in order to avoid 60 Hz line noise for this North American dataset.

• The sampling strategy for this dataset resulted in the presence of a ‘healthy’ control group (BDI mean = 1.78, SD = 1.70) and another group with moderate-severe depression symptoms (BDI mean = 22.44, SD = 4.70). To accommodate the sampling differences between datasets, cluster-based permutation independent t-tests were performed between participants with scores between 0 and 13 (‘minimal’ category, n = 73), and participants with scores 20 and above (‘moderate’ and ‘severe’, n = 41). Correlations were performed across all participants, that is, with the inclusion of participants in the ‘mild’ category, n = 13.

Results

Dataset 1 – TDBRAIN

Demographics

N = 173 were included in the final analysis. Participants had a mean age of 40.36 + −13.59, range 14–78 years. There were no significant differences in age (M = 40, SD = 13–14) or biological sex (47–50% female) between sub-samples (i.e., between frequency bands). Control analysis indicated that there was no significant relationship between age and BDI scores (rho = −0.05, p = 0.53).

Associations between gamma power and depression symptomology

Topographical representations of gamma power as a function of depression symptom severity are represented in Fig. 3a, b. Increased gamma power was observed in individuals experiencing severe symptoms, relative to individuals with moderate symptoms across a central-posterior electrode cluster (low gamma: t = −2.74, p = 0.002, Fig. 3c; high gamma: t = −2.64, p = 0.003, Fig. 3d). Furthermore, a correlation analysis across the entire sample revealed that low gamma power was positively correlated with BDI scores (Fig. 3e), particularly across a left anterior-central cluster (rho = 0.20, p = 0.01) and right posterior-central cluster (rho = 0.19, p = 0.04).

Fig. 3. Topographical representation of gamma power across depression symptoms.

Top row: low (a) and high (b) gamma power in mild, moderate, and severe Beck Depression Inventory (BDI) categories. Second row: t-statistics for comparisons of low (c) and high (d) gamma power between mild and moderate scorers, mild and severe scorers, and moderate and severe scorers. Third row: Cluster-based permutation Spearman correlations for low (e) and high (f) gamma power. Scatterplot depicts the correlation between BDI scores and average power across significant electrodes. For both t-test and correlation topography maps, red denotes an increase in power with greater depression severity, while blue denotes a decrease in power with greater depression severity. White circles: p < 0.05, corrected for multiple comparisons. White diamonds: p < 0.05, uncorrected.

Associations between beta power and depression symptomology

There was some evidence of increased beta power (both high and low) in individuals experiencing severe depression symptoms, relative to those with moderate symptoms (see Fig. 4a, b). This difference was most pronounced in a central-anterior electrode cluster for high beta power (t = −2.22, p <0.05 uncorrected) but did not remain statistically significant after adjusting for multiple comparisons (p = 0.07, Fig. 4d). No other significant differences or associations were observed (Fig. 4e, f).

Fig. 4. Topographical representation of beta power across depression symptoms.

Topographical representation of beta power across depression symptoms. Top row: low (a) and high (b) beta power in mild, moderate, and severe Beck Depression Inventory (BDI) categories. Second row: t-statistics for comparisons of low (c) and high (d) beta power between mild and moderate scorers, mild and severe scorers, and moderate and severe scorers. Third row: Cluster-based permutation Spearman correlations for low (e) and high (f) beta power. Scatterplot depicts the correlation between BDI scores and average power across significant electrodes. For both t-test and correlation topography maps, red denotes an increase in power with greater depression severity, while blue denotes a decrease in power with greater depression severity. White circles: p < 0.05, corrected for multiple comparisons. White diamonds: p < 0.05, uncorrected.

Changes in the 1/f slope across differing depression severities

Topological representations of the spectral parameterization (1/f exponent and offset) values for each sub-group are shown in Fig. 5a, b. For the 1/f exponent, no significant group differences were observed (Fig. 5c, Fig. 6). There was some evidence of a positive correlation between depression symptoms and exponent values across a frontocentral topography (rho = 0.17, p < 0.03 uncorrected), though these results did not remain significant after correcting for multiple comparisons (p = 0.09, Fig. 5e).

Fig. 5. Topographical representation of 1/f parameters across depression symptoms.

Top row: exponent (a) and offset (b) parameters in mild, moderate, and severe Beck Depression Inventory (BDI) categories. Second row: exponent (c) and offset (d) t-statistics for comparisons between mild and moderate scorers, mild and severe scorers, and moderate and severe scorers. Third row: Cluster-based permutation Spearman correlations for exponent (e) and offset (f) parameters. Scatterplot depicts the correlation between BDI scores and average parameter value across significant electrodes. For both t-test and correlation topography maps, red denotes an increase in parameter magnitude with greater depression severity, while blue denotes a decrease in parameter magnitude with greater depression severity. White circles: p < 0.05, corrected for multiple comparisons. White diamonds: p < 0.05, uncorrected.

Fig. 6. Topographical representation and scatter plot of correlation between low gamma power and 1/f offset.

Left: Spearman’s rho values for the correlation between the two variables at each electrode. White circles: p < 0.017, Bonferroni-corrected. Right: Scatterplot depicting 1/f offset values against low gamma power values for each participant.

A greater 1/f offset was observed in individuals with severe relative to moderate depression symptoms, across a centroparietal topography (t = −2.57, p = 0.007, Fig. 5d). A significant positive correlation was also observed between offset and depression symptoms (rho = 0.21, p = 0.004, Fig. 5f). Interestingly, the topography for this cluster was qualitatively similar to low gamma. As such, a correlation analysis was performed between low gamma power and 1/f offset for each electrode. This revealed significant moderate positive correlations for each electrode, with rho values spanning from 0.29 to 0.65 (see Fig. 7). No other significant differences were observed.

Fig. 7. Topographical representation and scatter plot of correlation between low gamma power and 1/f offset.

Left: Spearman’s rho values for the correlation between the two variables at each electrode. White circles: p < 0.017, Bonferroni-corrected. Right: Scatterplot depicting 1/f offset values against low gamma power values for each participant.

Control analyses

Given recent drug and alcohol use has the potential to alter neural activity, additional analyses were conducted on a subset of the sample, including only participants with no history of drug or alcohol use. Similar trends were observed across both samples and many relationships observed in the full sample were strengthened when drug and alcohol use was removed (see Supplementary Materials). To account for potential confounding of aperiodic signal within the spectral data, additional analyses were also conducted with the aperiodic component removed. Results remained largely the same (see Supplementary Materials).

Overall, significant increases in gamma power and the offset parameter of the 1/f slope model were observed across depression severities, particularly between individuals with “moderate” and “severe” BDI scores. These effects had a distributed topography. We did not observe significant changes in the beta band, although there was a general trend towards increased beta power with increased depression severity. To assess the reproducibility of these findings, the analyses were also conducted on the Cavanagh (2021) dataset.

Dataset 2 - Cavanagh

N = 113 were included in the final analysis. Participants had a mean age of 18.90 + −1.22, range 18–24 years. There were no significant differences in age (M = 19, SD = 1) or biological sex (38–53% female) between sub-samples (i.e., between frequency bands). Control analysis indicated that there was no significant relationship between age and BDI scores (rho = −0.11, p = 0.25).

Associations between gamma power and depression symptomology

No significant differences in gamma power were observed between individuals with depression and controls for the Cavanagh dataset. There was evidence of a positive correlation between gamma power and BDI score in a left frontocentral and right posterior cluster, but this was a weak relationship (rho = 0.30, p < 0.02 uncorrected) that did not survive correction for multiple comparisons (p = 0.07), see Fig. 8D).

Fig. 8. Topographical representation of comparisons and correlations of spectral power and 1/f slope parameters across depression symptoms.

Rows, from top to bottom: t-statistics for comparisons between control group and depression group for spectral power and 1/f parameters, cluster-based permutation Spearman correlations for spectral power and 1/f parameters, and scatterplots depicting the correlation between Beck Depression Inventory (BDI) scores and average spectral power across significant electrodes. For both t-test and correlation topography maps, red denotes an increase in parameter magnitude with greater depression severity, while blue denotes a decrease in parameter magnitude with greater depression severity. White circles: p < 0.05, corrected for multiple comparisons. White diamonds: p < 0.05, uncorrected.

Associations between beta power and depression symptomology

Significantly higher low beta power was observed in individuals with depression relative to controls, across anterior, central, and temporal regions (t = −2.64, p = 0.005; see Fig. 8E). Positive correlations between beta power and BDI scores were also observed. In the low beta band, after adjusting for multiple comparisons, this effect was most pronounced across a left temporal cluster (rho = 0.24, p = 0.047 corrected; see Fig. 8F). In the high beta band, this effect was most prominent across an anterior cluster (rho = 0.24, p = 0.045 corrected; see Fig. 8G).

Changes in the 1/f slope across differing depression severities

For the 1/f exponent, no significant group differences were observed (Fig. 8I and J; Fig. 6). There was some evidence of a negative correlation between depression symptoms and exponent values across an anterior (rho = −0.22, p < 0.02, uncorrected) and central-posterior (rho = −0.22, p < 0.03., uncorrected) topography, though these results did not remain significant after correcting for multiple comparisons (anterior: p = 0.10, central-posterior: p = 0.13). No significant group difference or correlation was observed for the 1/f offset (Fig. 8K, L).

Discussion

In this study, we demonstrate changes in gamma and beta activity across a spectrum of depression symptom severities. We show modulation of gamma and beta power as a function of depression severity, and alterations in the aperiodic slope. We assessed the reproducibility of these effects using an independent dataset, and observed evidence of altered high and low beta power, albeit with a smaller effect size. To our knowledge this study is the first to demonstrate graded changes in oscillatory power across depression symptoms. We discuss the potential utility of these EEG features as possible markers of depression symptomology to aid in improved assessment and treatment.

Gamma power is altered across depression severities

In line with our expectations, we observed altered gamma power across depression severities. Specifically, we demonstrate a general pattern of increased gamma power across increasing severity. Our findings align with previous work reporting differences in gamma power between individuals with depression and healthy controls when assessed at rest [19, 60, 61], and during presentation of emotional stimuli ([17, 21, 62]; for review, see [9]). Extending on these findings, our results indicate that differences in gamma power are not only apparent between healthy individuals and those with a depression diagnosis, but show a relationship with depression symptom severity, with the most pronounced differences for moderate to severe symptom presentations.

We note certain inconsistencies between our findings of increasing gamma power and previous studies with regard to the direction of observed effects. For instance, Liu and colleagues reported decreased gamma power in drug-naïve individuals with first-episode depression (Liu, Liu [18]. Chronicity (and by extension, history of medication/treatment) may influence the relationship between gamma activity and depression. Indeed, certain antidepressant treatments (such as rTMS and SSRIs) have been associated with increased gamma power [22, 24], and in certain cases change in gamma power has been associated with symptom remission [24]. We observed some differences in the strength of associations between datasets in the current study, which may also plausibly reflect differences in medication usage between samples. Overall, we show evidence that gamma power increases with increasing depression severity. Future work may explore the role of chronicity and treatment history in mediating the direction and magnitude of these associations.

Greater depression severity is associated with greater beta power

We also observed evidence of increased beta power as a function of depression severity. Analysis of the primary dataset revealed preliminary evidence of increased high beta power in individuals with severe depression symptoms, relative to those with moderate symptom presentations. This finding was corroborated by analysis in an independent dataset, which similarly showed elevated beta power across a widespread topography in individuals with depression relative to healthy controls. Broadly, our findings align with several previous studies reporting altered beta power in individuals with depression across anterior [26, 63], parieto-occipital [25], and distributed [27] topographies. Interestingly, in line with the current findings indicating a relationship between symptom severity and beta power, previous studies have also identified links between beta modulation and chronicity (e.g., relapse and number of depressive episodes) [30, 31]. Collectively these results indicate potential utility of beta oscillations as a sensitive marker of depression severity and duration.

1/f parameters exponent and offset are associated with depression severity

To our knowledge, we provide the first evidence of altered aperiodic activity across a spectrum of depression severities using non-invasive EEG. Across datasets, we observed preliminary evidence of altered 1/f slopes as a function of increased symptom severity at midline electrodes. The 1/f slope is a putative marker of E/I balance, with flatter slopes (i.e., lower exponent values) thought to reflect a shift towards excitation, and steeper slopes (i.e., higher exponent values) indicating a shift towards inhibition [32]. Hence, our findings may reflect changes in excitation:inhibition balance as a function of depression severity. Indeed, post-mortem tissue studies of individuals with depression indicate widespread changes in GABAergic and glutamatergic cell counts across regions/networks implicated in emotional regulation, such as prefrontal cortex and limbic regions [34]. Our findings also align with previous longitudinal evidence demonstrating associations between symptom remission and 1/f slope, as measured via intracranial EEG in prefrontal and limbic regions [43, 44]. Collectively, our findings add to previous work indicating that altered excitation-inhibition balance underlies depression symptomology. We demonstrate the utility of the 1/f slope as a non-invasive measure of E-I balance to differentiate depression severities. Future work may expand on this finding to explore how the 1/f slope may be utilized as a physiological marker of treatment and recovery.

We also report an association between depression and the offset parameter of the 1/f slope, which indexes broadband shifts in power [32], and has been associated with neuronal spiking and fMRI measures of blood-oxygen-level-dependent activity [64, 65]. It is plausible that our results reflect increases in excitability as a function of increased symptom severity. Broadly, this finding aligns with previous work reporting altered offset parameter values in individuals with depression relative to healthy controls, and individuals in remission [39, 41, 42]. A significant moderate correlation was also observed between the 1/f offset and low gamma power across electrodes. It is methodologically challenging to dissociate the aperiodic component of the EEG signal from gamma activity due to the absence of clear high-frequency gamma peaks. As such, it is plausible that to some degree gamma power and the offset parameter reflect a shared/common neurophysiological signal. Indeed, there is evidence that the 1/f offset reflects neural population spiking [32], which is positively correlated with gamma power [64, 66, 67]. The strength of the association does suggest that either of these measures may provide a useful correlate of depressive symptomology, and raises the question of whether multiple EEG signatures are indeed required as a neurophysiological marker of depression severity. It will be important to determine the unique contribution of these measures to depression severity in future work.”

Limitations & future directions

This study has limitations which should be acknowledged. For instance, the validity of self-report measures can be impacted by insight, which can be impaired in depression [68, 69] possibly leading to inaccurate scores. However, the BDI-II is a widely used self-report measure that has been validated in numerous populations and, in some instances, has demonstrated greater sensitivity than clinician-administered scales [48, 52, 70].

There were also sampling differences across the two datasets analyzed in this study. Participants in the TDBRAIN (2022) dataset were not required to have depression as their primary diagnosis, and some participants presented with other diagnoses such as obsessive-compulsive disorder and attention-deficit hyperactivity disorder, and/or recent history of medication use. Similarly, the datasets utilised in this study did not provide information on chronicity factors such as stages of disease, number of depressive episodes, or responsiveness to treatment. However, depression is a mental health condition with a high degree of co-morbidity and heterogeneity. A strength of this study is that our findings emerge despite the variability in individuals with a depression diagnosis. We observed similar trends across datasets (i.e., altered high-frequency oscillations) indicating that these EEG markers are robust to these sampling differences and may indeed be specific to depression symptomology. It should also be noted that the cross-sectional nature of this study limits inferences regarding disease progression and remission. While our findings provide a foundational evidence base associating periodic and aperiodic EEG measures with depression severity, longitudinal data is required to establish how these measures change with symptom progression, and treatment response. Our results offer novel insights into the utility of non-invasive EEG markers of depression severity, but future work is required to delineate the extent to which gamma power and the spectral parameterisation model offer complementary markers of symptomology.

Conclusion

In this study, we report evidence of altered gamma and beta power and aperiodic activity across a spectrum of depression severity. We show similar trends across two independent datasets, indicating that these non-invasive EEG measures may provide reliable markers of symptom progression. Future research should investigate how these markers shift in response to intervention, which may inform their utility in precision psychiatry approaches aimed and improved assessment and treatment.

Author contributions

J.G.H, J.C., J.J.H designed and conducted data analysis for this study. J.G.H wrote the initial draft. All authors edited the manuscript. J.C and J.J.H supervised the research.

Code availability

All analyses were completed in MATLAB R2023b. Code is available upon request.

Data availability

This study utilised data from two existing open source datasets available online [45, 46].

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41398-026-04268-z.