Modulatory effects of ketamine on EEG source-based resting state connectivity in treatment resistant depression

Abstract

Treatment-resistant depression (TRD) accounts for approximately 30% of major depressive disorder cases and has been characterized by altered functional connectivity within and between the Default Mode (DMN) and Frontoparietal networks (FPN). Ketamine can be an effective treatment for TRD, and its antidepressant response has been associated with alterations in resting state functional connectivity (rsFC). Here, we evaluated the effect of a single subanesthetic dose of racemic ketamine (0.5 mg/kg) on electroencephalogram (EEG) derived source-based measures of rsFC from 24 participants with TRD (16 women; aged 44.35 ± 15.86 years). Ninety-six channel resting state EEG data were collected 24 h before and after ketamine infusion. Exact low-resolution electromagnetic tomography (eLORETA) was used to estimate theta and beta-band rsFC within and between the DMN and FPN. Ruminative symptoms were assessed using the Ruminative Response Scale. Analogous data were collected from 34 healthy control participants (25 women, aged 32.49 ± 14.07 years) who did not receive any intervention. Twenty-four hours post-infusion, depressive, anhedonic, and ruminative symptoms for the TRD sample were significantly reduced. Interestingly, symptom reduction was not correlated with any changes in rsFC but was associated with initial pre-ketamine rsFC. Moreover, individuals with TRD displayed broad increases in rsFC within the DMN and FPN as well as between these two networks. Based on preclinical findings, we posit that ketamine’s synaptogenic effects may be driving this general increase in connectivity. However, these synaptogenic effects can be short lived, and future work probing the full time-course of rsFC via EEG pre- and post-ketamine administration is warranted.

Introduction

Major depressive disorder (MDD) is characterized by cognitive, emotional, and physical symptoms [1, 2]. Substantial clinical heterogeneity renders treating MDD challenging, evidenced by the fact that up to 50% of individuals with MDD do not benefit from first-line and subsequent antidepressant therapies [3]. Indeed, those not improving across at least two antidepressant trials (appropriate in dose, duration, and adherence) in their current episode are typically considered to suffer from treatment resistant depression (TRD), also known as difficult to treat depression [4]; these individuals account for approximately 30% of MDD cases [5]. Relative to individuals who benefit from antidepressant treatment, individuals with TRD experience negative outcomes including high relapse and hospitalization rates [6–8], highlighting urgent needs for different treatments.

Ketamine, a N-methyl-D-aspartate (NMDA) receptor antagonist, has emerged as an effective treatment for TRD [9, 10]. Two overlapping models posit ketamine’s mechanism of action, the disinhibition and direct inhibition hypotheses [11–14]. The former suggests that ketamine preferentially blocks NMDA receptors on gamma amino butyric acid inhibitory (GABA) interneurons. These interneurons disinhibit the prefrontal cortex (PFC), increasing glutamate, leading to a sustained activation of α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptors that triggers synaptic release of brain-derived neurotrophic factor (BDNF). BDNF binding then results in mammalian target of rapamycin (mTOR) mediated synaptogenesis [11, 15, 16]. The direct inhibition hypothesis, by contrast, suggests that ketamine directly antagonizes PFC NMDA receptors which directly cascade into similar synaptogenic pathways [11, 13, 17]. Notably, both hypotheses describe synaptogenesis as the final common pathway of ketamine’s mechanism of action, and these alterations in connectivity might underlie its antidepressant effect [18].

MDD has been increasingly characterized as involving disrupted large-scale brain network connectivity, and research has generally centered on Menon’s tripartite model of the Default Mode (DMN), Central Executive (i.e., Frontoparietal Network; FPN), and Salience Networks (SN) as a key pathophysiological framework [19–22]. TRD is no different, and recent reviews have suggested that altered connectivity within the DMN, and between the DMN and the FPN, as well as subcortical structures, may characterize TRD [23, 24]. More specifically, individuals with TRD exhibit within-DMN hyperconnectivity relative to those with MDD who benefit from antidepressant treatment and non-depressed controls [25, 26]. Several subcortical structures including the amygdala, hippocampus, and habenula and their connectivity with cortical DMN loci have also been implicated in TRD [27–29]. Importantly, Ma et al. [30] noted that patterns of connectivity differentiate individuals with TRD from treatment-responsive individuals and posited that node-specific approaches may be important for understanding treatment resistance.

Recent research has probed these assumptions by using measures of neural connectivity to understand ketamine’s antidepressant effects (see [31–33] for recent reviews), and treatment-related changes in rsFC have emerged [34]. Indeed, several studies have found that ketamine alters global brain connectivity of the PFC (which includes key DMN, FPN, and SN structures) [35, 36], with node-based analyses suggesting that ketamine reduces within-PFC connectivity and enhances PFC connectivity to other brain regions [35, 37]. Notably, one study demonstrated increased global brain connectivity in the DMN and FPN [38], echoing other fMRI-based connectivity studies that implicate similar regions and networks [25, 39, 40]. That said, it is worth noting that some global connectivity work has failed to find any relationship between PFC connectivity and antidepressant response in MDD participants, despite observing reduced baseline connectivity in these participants [41]. The authors postulate that study- (e.g., scan interval) or subject-related variables (e.g., sex distribution) may account for these results, however, it is also possible that changes in PFC connectivity may not index ketamine’s antidepressant response.

Complementing hemodynamic neuroimaging, electrophysiological metrics can also be used to index functional connectivity [42, 43]; however, only a limited number of studies have used such approaches to probe the effects of ketamine. Studies using magnetoencephalography (MEG) to estimate source connectivity have found that ketamine exposure is associated with significant reductions in connectivity between specific nodes (e.g., the insulo-temporal nodes and the amygdala, and a precentral cortical network and the subgenual anterior cingulate cortex; [44]), and more widely across the brain [45]. Notably, these more widespread reductions in connectivity did not correlate with antidepressant response, even though other connectivity parameters did distinguish between those who were and were not responsive to ketamine [45]. Indeed, several studies using electroencephalography (EEG) provided similar results and have shown reduced connectivity in frontal structures for those benefitting from ketamine [46], as well as increased connectivity in visual cortices [47] and AMPA-related frontoparietal signaling [48].

Notably, while EEG-based measures of connectivity have been used to examine depression-related treatment outcomes (e.g., [49–51]), to the best of our knowledge, these examinations have not included ketamine. McMillan and colleagues’ [52] analysis of simultaneously captured EEG and fMRI data provides a possible starting point, whereby widespread changes in source-based spectral power estimates during ketamine infusion were identified, though these changes did not predict antidepressant response. These changes reflect alterations to the underlying neural oscillations but notably, do not index functional connectivity between neural networks. Thus, the primary goal of the present study was to evaluate the effect/s of a single subanesthetic dose of racemic ketamine (0.5 mg/kg) on EEG-derived source-based measures of rsFC. Based on previous literature, we hypothesized that, 24 h post-infusion, ketamine would: 1) reduce depressive, anhedonic, and ruminative symptoms; 2) decrease within-DMN connectivity; 3) increase within-FPN connectivity; and 4) decrease connectivity between the DMN and FPN.

Materials and methods

The broader study of which the present data were part was registered on ClinicalTrials.gov, NCT04239963, and was carried out at McLean Hospital, Belmont, Massachusetts.

Participants

Prior to the start of the study, we estimated the necessary sample size based on the results of relevant neuroimaging literature. An integrative analysis of 9 ketamine trials found that ketamine elicits medium-to-large reductions in depressive symptoms when compared to midazolam (Cohen’s d = 0.70) and saline (d = 1.60), and that the pooled 24-hour response rate (i.e., percentage of individuals with a more than 50% reduction in symptoms) was 46% [53]. Using this information, we found that at a two-tailed α of 0.05 and a mean effect size of 1.15, n = 30 TRD participants will be associated with a power > 0.85 to detect changes after ketamine injection, differences between responders and non-responders, and identify baseline EEG and/or behavioral markers predicting treatment response.

Thirty-six people with TRD scheduled for ketamine therapy were recruited from McLean Hospital’s Ketamine Service. Prior to enrolment, participants completed a screening session where their diagnosis was confirmed using the Mini International Neuropsychiatric Interview (MINI; [54]) and symptoms assessed using several questionnaires (see Supplement) including the Hamilton Depression Rating Scale (HAMD; [55]), the Quick Inventory of Depression Scale (QIDS; [56]), the Mood and Anxiety Symptom Questionnaire (MASQ; [57]), and the Snaith-Hamilton Pleasure Scale (SHAPS; [58]). Forty-three additional participants were screened via the MINI to be psychologically healthy and serve as a control group (HC), which led to 2 HC participants being excluded (see Supplement for Inclusion/Exclusion criteria). Furthermore, 11 participants (2 HC and 9 TRD) withdrew prior to EEG data collection, 6 participants (3 HC and 3 TRD) were lost to follow up, and data from 2 HC participants were excluded due to excessive EEG artifacts.

Thus, the final EEG sample consisted of 24 participants with TRD (16 women, aged 44.35 ± 15.86 years) and 34 HC participants (25 women, aged 32.49 ± 14.07 years). The two groups were similar in their education and sex distribution (p’s ≥ 0.704). However, TRD participants were, on average, older than HC participants (t(1,45.79) = −2.94, p = 0.005), and more likely to self-identify as white (χ² = 12.17, p = 0.002). A detailed demographic description of both groups is presented in Supplementary Table 1; notably, all TRD subjects were medicated (see Supplementary Table 2) and did not change medications across the study sessions.

Study procedure

Study procedures were approved by the Mass General Brigham Institutional Review Board, and participants provided written informed consent. All study procedures were performed in accordance with the relevant guidelines and regulations. Participants were compensated for their time, and those in the TRD group had their first ketamine dose paid for by the study.

Eligible participants completed two study sessions separated by 48 h; individuals with TRD were tested 24-hours before and after their first ketamine infusion, and healthy participants waited the 48-hour period without any intervention. At both sessions, participants completed the HAMD, QIDS, SHAPS, and the Ruminative Response Scale (RRS; [59]), as well as an EEG protocol described below.

EEG protocol and recording

Continuous EEG data were recorded using a 96-channel equidistant spherical actiCAP and actiChamp amplifier (Brain Products GmbH, Gilching, Germany) digitized at 500 Hz using BrainVision Recorder Software (v1.25.0201; Brain Products GmbH). Data were referenced online to the vertex channel with a ground electrode located approximately at AFz, and impedances were maintained below 25 KΩ.

Both sessions began with recording 8-min of resting EEG data in 1-min segments. Four minutes were recorded with eyes open and four minutes with eyes closed; the order of which was randomized and counterbalanced across participants. This resting baseline was followed by two tasks (not reported here). Consistent with our previous studies investigating EEG-based functional connectivity [50, 60, 61], only eyes-closed resting EEG data were analyzed.

EEG preprocessing

EEG data were processed using BrainVision Analyzer (v2.2; Brain Products GmbH). Raw data were first visually inspected to remove gross muscle artifacts and identify artifactual channels. Data were then filtered between 1–100 Hz using a second-order Butterworth zero-phase IIR filter. Next, independent component analysis was used to remove components containing artifact sources (e.g., eyeblinks and movements, cardiac and muscle signals). Artifactual channels were interpolated using spherical splines [62], and data were re-referenced to the common average. Non-overlapping 2048 ms segments were then extracted for functional connectivity analyses. The lowest number of segments exported was 105 corresponding to at least 215 sec of data, a value beyond the recommended minimum 40 sec [63]. Moreover, the average number of segments did not differ between HCs and TRDs at session 1 (116.38 ± 1.56 vs. 118.04 ± 7.29; t = −1.098, p = 0.283) or session 2 (117.06 ± 3.13 vs. 118.25 ± 5.14; t = −1.010, p = 0.319). Finally, the number of segments did not differ between sessions for both groups (p’s ≥ 0.27).

eLORETA measures and quantification

Functional connectivity estimates were computed using eLORETA implemented via the LORETA software (v20221219; [63]). eLORETA is a linear inverse problem solution capable of correctly reconstructing localized cortical activity from scalp-recorded data. It uses a solution space of 6239 cortical gray matter voxels embedded in the realistic head model [64] from the Montreal Neurological Institute 152 template [65]. Several studies validated the original LORETA algorithm [66–68], and the eLORETA algorithm has been validated using MEG [69, 70] and EEG combined with MRI [71].

Functional connectivity was estimated using lagged phase synchronization (LPS; see Supplement). LPS quantifies the nonlinear relationship between two regions of interest (ROI) after removing instantaneous EEG connectivity, which likely reflects volume conduction effects [63]. Based on prior hypotheses and findings in independent MDD samples [50, 60], LPS was estimated across several ROIs within the DMN and FPN. We elected to focus on these two networks due to their inclusion in Menon’s tripartite model of psychopathology [21], and their possible roles in the pathophysiology of depression generally [20, 22] and treatment-resistant depression [23, 24]. ROIs were defined using seed points from our previous MDD studies [50, 60] and included all gray matter voxels within a 10 mm radius; details for all ROI seeds are listed in Supplementary Table 3.

Normalized Fourier transforms applied to all artifact-free segments of data were used to calculate the LPS between ROIs in the following frequency bands: theta (4.5–7 Hz), beta1 (12.5–18 Hz), beta2 (18.5–21 Hz), and beta3 (21.5–30 Hz). Analyses were confined to these frequency bands as: 1) recent reviews of relevant oscillatory and connectivity EEG analyses highlight the theta, beta, and gamma frequency bands as relevant to depression and ketamine treatment of depression [33, 72, 73] and; 2) previous EEG-based rsFC studies from our laboratory have implicated theta and beta band connectivity in depression and its treatment [50, 60, 61]. Finally, at the suggestion of an anonymous reviewer, we also examined global lagged connectivity to further contextualise our ROI specific approach; these results are presented in the Supplement.

Data analysis

Analytically, we evaluated Group, Session, and Group × Session differences in within- and between-network connectivity by simultaneously comparing LPS between all pairs of ROIs at each frequency. Within-group between-Session change was evaluated using dependent sample t-tests, while the between-Group within-session and the Group × Session differences were evaluated using independent sample t-tests assuming unequal variances. All tests were submitted to non-parametric permutations based on the maximal statistic [74, 75] with 5000 randomizations to correct for multiple comparisons. Additionally, to further probe group-level differences in LPS, we also evaluated these tests using a t-value threshold that corresponded to an uncorrected p < 0.005; The respective t-values that corresponded to an uncorrected p < 0.005 were 3.009 and 3.105 for the HC and TRD within-group comparisons, and 2.923 for the between-group comparison.

Moreover, as age significantly differed between the two groups, we used ordinary least squares regression models to evaluate any identified significant Group or Group × Session differences when also controlling for age. These models estimated the main effect of age at session 1 (mean centered to the full population mean), and either the Group or Group × Session effect. To account for possible extreme values, LPS parameters were winsorized at the 5^th and 95^th percentile prior to analyses.

Lastly, we conducted a bivariate correlation analysis to examine the associations between change in LPS and change in rumination scores in TRD participants, as well as between initial LPS and change in rumination.

Results

First, we evaluated if depressive symptoms changed between sessions (Table 1). As hypothesized, depressive symptoms of individuals with TRD significantly decreased between sessions when assessed by the HAMD (15.58 ± 4.96 vs. 11.48 ± 5.70; t = −2.754, p = 0.012) and the QIDS (13.86 ± 4.29 vs. 9.68 ± 4.76; t = −2.799, p = 0.011). The HAMD and QIDS scores of HC participants did not differ between testing sessions (p’s ≥ 0.161) Similar results were observed when examining anhedonic symptoms (SHAPS scores) where TRD participants reported a small but significant reduction (37.46 ± 5.79 vs 35.29 ± 5.88; t = −2.126, p = 0.044), while HC participants reported no change (p = 0.129). Moreover, we observed that the total rumination scores of TRD participants significantly decreased between the two sessions (64.78 ± 9.09 vs. 60.42 ± 9.84, t = −3.161, p = 0.005), as did their scores on the brooding (t = −2.206, p = 0.038) and depression-related rumination (t = −2.953, p = 0.007) subscales. Interestingly, the HC group also showed a significant reduction in their total (t = −2.332, p = 0.027) and depression-related rumination (t = −2.487, p = 0.018); however, these decreases were notably smaller than for the TRD participants (3.98% vs. 6.73% and 5.93% vs. 8.40%). Finally, neither group significantly changed in their reflective thinking (p’s ≥ 0.086).

Table 1.

Clinical symptoms for healthy controls and TRD participants at both testing sessions.

Scale	Group	T1 Mean (SD)	T2 Mean (SD)	t	df	p
Depression
HAMD	HC	0.34 (0.79)	0.15 (0.57)	1.44	30	0.161
	TRD	15.58 (4.96)	11.41 (5.7)	2.75	21	0.012a
QIDS	HC	0.16 (0.45)	0.18 (0.58)	−1.00	30	0.325
	TRD	13.86 (4.29)	9.68 (4.76)	2.80	19	0.011a
Anhedonia
SHAPS	HC	18.26 (4.99)	19.06 (5.65)	−1.56	33	0.129
	TRD	37.46 (5.79)	35.29 (5.88)	2.13	23	0.044a
Rumination
RRS – Total	HC	28.83 (5.02)	27.68 (4.71)	2.33	29	0.027a
	TRD	64.78 (9.09)	60.42 (9.84)	3.16	22	0.005a
RRS – Brooding	HC	6.58 (1.54)	6.38 (1.44)	1.76	30	0.088
	TRD	14.43 (2.86)	13.33 (2.70)	2.21	22	0.038a
RRS – Depression	HC	15.19 (2.55)	14.29 (2.37)	2.49	31	0.018a
	TRD	38.67 (5.02)	35.42 (5.59)	2.95	23	0.007a
RRS – Reflection	HC	7.36 (2.64)	7.00 (1.91)	1.36	32	0.184
	TRD	12.12 (3.66)	11.67 (3.69)	1.80	23	0.086

HAMD Hamilton Depression Rating Scale, HC Healthy Control, QIDS Quick-Inventory of Depression, RRS Ruminative Response Scale, SHAPS Snaith-Hamilton Pleasure Scale, T1 Testing Session 1, T2 Testing Session 2, TRD Person with Treatment Resistant Depression.

^a = Statistical Significance.

Group differences in functional connectivity

Next, we compared LPS between the two groups at session 1 (Fig. 1), i.e., prior to TRD participants receiving treatment. Significant group differences emerged in beta3 within-DMN connectivity. Specifically, relative to HC, participants with TRD had significantly higher LPS between the left parahippocampal gyrus and left angular gyrus (t = 3.009) and left mid-temporal gyrus and right angular gyrus (t = 3.313); both effects were retained after controlling for age (t = 3.167, p = 0.003; and t = 2.771, p = 0.008).

Fig. 1. Beta3 lagged phase synchronization for healthy controls and TRD participants at testing session 1.

Note: Plot A presents the beta 3 LPS values for the ROI pairs/connections that significantly differed between groups (i.e., healthy controls and TRD participants) at session 1. LPS values were winsorized before plotting and are plotted as the Mean ± SD (error bars); all other connections did not significantly differ between the two groups and have not been plotted for clarity. Plot B localizes these connections within the Colin-27 MNI brain volume. AG Angular Gyrus, DMN Default Mode Network, HC Healthy Control, l Left, LPS Lagged Phase Synchronization, MTG Mid-temporal Gyrus, PHG Parahippocampal Gyrus, r Right, SMG Supramarginal Gyrus, TRD Treatment Resistant Depression, * Statistical significance.

In addition, we also observed a difference in beta3 DMN-FPN connectivity, where people with TRD exhibited greater right supramarginal gyrus and left mid-temporal gyrus LPS when compared to HC (t = 3.239; Fig. 1). This effect, again, survived when controlling for age (t = 2.577, p = 0.013).

Change in functional connectivity between groups

The next step was to probe the Group × Session interaction term, i.e., group differences in the change in LPS (ΔLPS) between sessions.

Within-Network

Within the FPN, participants with TRD showed significantly greater ΔLPS relative to HCs. More specifically, TRD participants were characterized by greater change in theta precuneus to left supramarginal gyrus connectivity (t = 2.974; Fig. 2A/B), and beta2 right to left supramarginal gyri connectivity (t = 3.181; Fig. 2C/D). Both effects remained when controlling for age (t = 3.691, p < 0.001; and t = 3.032, p = 0.004).

Fig. 2. Differences in the between-session change in within-network connectivity between the healthy control and TRD groups.

Note: Plot A presents the change in theta LPS values (computed as the session 1 LPS value subtracted from the session 2 LPS value) for the within-FPN connection that significantly differed in the degree of between session change between the HC and TRD groups. Plots C and E present analogous change in beta2 LPS data for connections within the FPN and DMN, respectively. LPS values were winsorized before plotting and are plotted as the Mean ± SD (error bars); all other connections did not significantly differ and are not plotted for clarity. Plots B, D, and F localizes these connections within the Colin-27 MNI brain volume. AG Angular Gyrus, DMN Default-Mode Network, FPN Frontoparietal Network, HC Healthy Controls, l Left, LPS Lagged Phase Synchronization, PC Precuneus, PHG Parahippocampal Gyrus, r Right, SFG Superior Frontal Gyrus, SMG Supramarginal Gyrus, TRD Treatment Resistant Depression, * Statistical significance.

Turning to the DMN, participants with TRD showed a significantly larger change in beta2 connectivity between the right parahippocampal gyrus and left superior frontal gyrus (t = 3.109), and the right and left angular gyri (t = 2.991; Fig. 2E/F). These group differences were confirmed after controlling for age (t = 2.870, p = 0.006; and t = 2.841, p = 0.006).

Between-Network

With respect to DMN-FPN connectivity, we observed several significant group differences in the degree of theta, beta1, and beta2 ΔLPS, such that TRD participants showed significantly greater changes relative to HCs. More specifically, TRD participants had greater change in: theta band precuneus and left angular gyrus connectivity (t = 3.483; Fig. 3A/B); beta1 band paracingulate gyrus and right superior frontal gyrus connectivity (t = 3.017; Fig. 3C/D); and beta2 band connectivity between the right supramarginal gyrus and left angular gyrus (t = 3.595; Fig. 3E/F), and the right mid-temporal gyrus and left parahippcampal gyrus (t = 3.036). Each of these effects remained after controlling for age (t’s = ≥ 2.846, p’s ≤ 0.006).

Fig. 3. Differences in the between-session change in between-network connectivity between the healthy control and TRD groups.

Note: Plot A presents the change in theta LPS values (computed as the session 1 LPS value subtracted from the session 2 LPS value) for the FPN-DMN connection that significantly differed in the degree of between session change between the HC and TRD groups. Plots C and E present analogous change in LPS data for beta1 and beta2 frequency bands, respectively. LPS values were winsorized before plotting and are plotted as the Mean ± SD (error bars); all other connections did not significantly differ and are not plotted for clarity. Plots B, D, and F localize these connections within the Colin-27 MNI brain volume. AG Angular Gyrus, DMN Default Mode Network, FPN Frontoparietal Network, HC Healthy Control, l Left, LPS Lagged Phase Synchronization, MTG Mid-temporal gyrus, PC Precuneus, PCG Paracingulate Gyrus, PHG Parahippocampal Gyrus, SMG Supramarginal Gyrus, TRD Treatment Resistant Depression, * Statistical significance.

Session-related differences in functional connectivity

After examining the Group × Session effect, we then evaluated the between-session differences in rsFC of the TRD group. We observed that DMN-FPN beta2 connectivity between the right supramarginal gyrus and the left angular gyrus (t = 3.215) and between the left frontoparietal mid-temporal gyrus and the right angular gyrus (t = 3.148) significantly increased (Fig. 4A/B). Probing the connections implicated in the Group × Session analysis provided further evidence for increases in DMN-FPN connectivity, identifying a significant rise in precuneus and left angular gyrus theta band connectivity (t = 2.591, p = 0.016), and in right frontoparietal mid-temporal gyrus and right angular gyrus beta2 band connectivity (t = 2.831, p = 0.009).

Fig. 4. Between testing session change in lagged phase synchronization of the TRD participants.

Note: Plots A presents the beta2 LPS values for connections that differed for the TRD group. LPS values were winsorized before plotting and are plotted as the Mean ± SD (error bars); all other connections did not significantly differ and are not plotted for clarity. Plot B localizes these connections within the Colin-27 MNI brain volume. AG Angular gyrus, FPN Frontoparietal Network, l Left LPS Lagged Phase Synchronization, MTG Midtemporal Gyrus, r Right, SMG Supramarginal Gyrus, T1 Testing Session 1, T2 Testing Session 2, * Statistical significance.

The Group × Session follow-up analysis also revealed between-session increases in within-network connectivity. Within the DMN, beta2 connectivity increased between the right parahippocampal gyrus and left superior frontal gyrus (t = 2.506, p = 0.020), and between the left and right angular gyri (t = 2.499, p = 0.020). Within the FPN, precuneus and left supramarginal gyrus theta connectivity increased (t = 2.257, p = 0.034) as did beta2 connectivity between the right and left supramarginal (t = 2.752, p = 0.011).

Associations between changes in LPS and ruminative thinking

Finally, we examined if any implicated ΔLPS parameter was significantly associated with changes in HAMD and RRS scores among TRD participants. The correlation analysis revealed no significant association between ΔLPS of any implicated connections and change in HAMD score (r’s ≤ |0.16|, p’s ≥ 0.48), total rumination score (r’s ≤ |0.27|, p’s ≥ 0.24), brooding score (r’s ≤ |0.32|; p’s ≥ 0.13), and depression-related rumination score (r’s ≤ |0.24|, p’s ≥ 0.28). Interestingly, change in right parahippocampal gyrus and the left superior frontal gyrus beta2 connectivity (i.e., within-DMN connectivity) was positively associated with change in reflective thinking (r = 0.41, p = 0.04), such that a greater increase in connectivity predicted a greater increase in reflective thinking; all other associations were not significant (r’s ≤ |0.39|, p’s ≥ 0.06).

In addition, we also conducted an exploratory correlation analysis that used session 1 LPS values rather than ΔLPS; these results are presented in Table 2. In contrast to the ΔLPS correlations, we observed significant positive associations between changes in total ruminative thinking and connectivity within the DMN (r = 0.520, p = 0.016) and within the FPN (r = 0.478, p = 0.029), such that greater reductions in rumination were associated with reduced rsFC at session 1. Notably, these associations appear to be largely driven by depression-related ruminative thinking which showed a similar pattern of significance (right column of Table 2; r’s ≥ 0.434, p’s ≤ 0.044), rather than brooding and reflecting thinking where no significant associations with any LPS value were observed (p’s ≥ 0.06).

Table 2.

Zero-order associations between change in ruminative thinking scores and initial within- and between-network lagged phase synchronization of TRD participants.

	Δ Total Rumination Score		Δ Depression-Related Rumination
	r	p	r	p
θ PC – l AG	0.080	0.731	−0.198	0.378
θ PC – l SMG	0.140	0.544	−0.124	0.583
β2 l FPN MTG – r AG	0.469	0.032	0.479	0.024a
β2 r SMG – l AG	0.427	0.053	0.434	0.044a
β2 r PHG – l SFG	0.148	0.521	0.225	0.315
β2 r AG – l AG	0.520	0.016a	0.587	0.004a
β2 r SMG – l SMG	0.478	0.029a	0.468	0.028a
β2 r FPN MTG – l PHG	0.335	0.137	0.586	0.004a

Change in Hamilton Depression Rating Scale Score as well as Ruminative Brooding and Reflective thinking were not significantly associated with any LPS value and are not presented for clarity.

AG Angular Gyrus, FPN Frontoparietal Network, l Left, LPS Lagged Phase Synchronization, MTG Mid-temporal gyrus, PC Precuneus, PHG Parahippocampal Gyrus, r Right, SFG Superior Frontal Gyrus, SMG Supramarginal Gyrus, TRD Treatment Resistant Depression, θ theta band, β2 beta2 band.

^a = Statistical significance.

Discussion

The present study evaluated rsFC changes associated with the first dose of ketamine among individuals with TRD. In line with our hypothesis and existing literature, within 24-hours, ketamine significantly reduced depressive, anhedonic, and ruminative symptoms. Interestingly, while demonstrating the expected pre-treatment elevation in DMN connectivity relative to their HC counterparts, we observed that the rsFC of patients with TRD demonstrated a broad increase following ketamine infusion. More specifically, in contrast to our hypotheses, we observed an increase in within-DMN beta2 connectivity, as well as increased theta and beta DMN-FPN connectivity. We did, however, observe the expected increase in within-FPN connectivity. Notably, these ROI-specific effects contrasted with the observed reduction in theta, beta2, and beta3 global lagged connectivity (see Supplementary Results). Together, these results suggest that ketamine acutely and broadly modifies neural connectivity but may have unique effects within depression-relevant brain regions, aligning with existing imaging and preclinical literature.

Pre-treatment hyperconnectivity within the DMN and between the DMN and FPN in individuals with TRD, relative to their HC counterparts, is consistent with meta-analytic evidence from fMRI studies [20, 23, 24]. Interestingly, the implicated within-DMN ROI pairs involve, among other nodes, the angular gyri. The cortical regions of these gyri are thought to be involved in integrating multimodal information for subsequent processing [78] and have key roles in self-referential cognition and autobiographical memory [79, 80]. These functions might that explain why initial rsFC (i.e., LPS at session 1) in several ROI pairs that included the angular gyri were significantly associated with change in depression-related rumination. The nature of these correlations suggests that those with the greatest ketamine-related antidepressant response were characterized by reduced connectivity within the DMN and FPN, as well as between the two networks. It is possible that such changes could guide future research that proactively uses EEG-based measures of rsFC in guiding antidepressant therapy. However, capturing 96-channel EEG data and subsequent eLORETA analysis requires more technical expertise, and thus refinements to the data collection procedure are warranted.

Beyond these initial differences in connectivity, given ketamine’s antidepressant effects, it might be surmised that ketamine would also reduce the within-DMN and DMN-FPN hyperconnectivity and increase the within-FPN hypoconnectivity often observed in MDD and TRD [20, 23, 24]; accordingly, the observed broad rsFC increase among individuals with TRD following their first ketamine infusion is also largely unexpected. That said, prior EEG studies have reported increased connectivity in and between frontoparietal and visual cortices following ketamine [47, 48], suggesting that the specific neural nodes implicated may define the direction of change. Nonetheless, following the mechanisms of action described by the disinhibition and direct inhibition hypotheses, we speculate that the observed increases in connectivity represent an initial spike in synaptogenesis [11]. Preclinical work using a mouse model of depression supports this notion and has shown that ketamine restores functional connectivity in the cortex and does so by both recovering lost dendritic spines and triggering de novo spineogenesis (i.e., the formation of spines in new dendritic locations) that initializes somewhere between 6–12 h post-exposure and peaks around 24-hours post-exposure [81].

Additionally, Moda-Sava et al. [81] note that a majority (~55%) of newly formed dendritic spines are lost within four days of ketamine exposure which suggests some form of pruning process. It may be that following this pruning we would observe the expected patterns of change in functional connectivity, i.e., reduced DMN and DMN-FPN connectivity accompanied by increased FPN connectivity. With this in mind, we would encourage any future research to record EEG at multiple timepoints following ketamine administration to evaluate this hypothesis. Of note, for many, without additional doses, depressive symptoms often re-emerge within approximately one week of ketamine treatment [82, 83], and maintenance of antidepressant effects typically requires repeated, sometimes long-term, dosing. Indeed, most existing studies of the brain-based correlates of ketamine’s antidepressant response have only examined the effect of a single infusion [33]. Thus, conducting a study that evaluates changes in rsFC across the full time-course of ketamine treatment (e.g., pre-infusion, during infusion, 4–24 h post infusion, and 2–7 days post-infusion for all infusions in a treatment program), may identify alterations in connectivity across different timescales (i.e., immediate vs. long-term) and of different types (i.e., transient vs. stable) and provide greater ability to understand ketamine-related treatment outcomes and responsivity. Similarly, while our specific goal was to evaluate putative changes in rsFC 24-hours post-infusion, ketamine’s antidepressant effects emerge rapidly within a few hours post-infusion [84] and significant synaptic changes occur within the first 24 h [81]. Moreover, EEG informed fMRI analysis has demonstrated the existence of different time-courses of ketamine-induced neural changes [52], and so, it is possible that a denser sampling around/during ketamine infusions would provide more insight into the dynamics of ketamine’s effects on rsFC.

Finally, we examined if any of the implicated changes in rsFC were correlated with change in depressive symptoms. In contrast to our hypothesis, there were no significant associations between these parameters, although this is perhaps not too unexpected when considered from the synaptogenic framework discussed above. Moda-Sava et al., [81] observed a timing disconnect between ketamine’s behavioral and synaptogenic effects (i.e., 3-hours vs. 12–24 h post exposure), as well as no association between spine formation and immobility behavior (i.e., a depressive symptom homologue for rodents). These results align with some prior human research that also observed no such associations (e.g., [45, 52]) and suggest that synaptogenesis may only be important for sustaining ketamine’s antidepressant effect not initiating it. Perhaps then the present lack of correlations is more the result of where in time rsFC parameters were assessed rather than the true absence of said association suggesting the need for future verification studies. Related to depressive symptoms, beyond the expected changes in TRD participants, the ruminative scores of HC participants also significantly decreased between session. This effect could be related to measurement variability, however, given that the RRS was a self-report scale, we would suggest that it is more likely a result of the confined variance in rumination scores for HCs, i.e., minimized variance renders differences of smaller magnitudes (e.g., the observed mean change in depressive rumination for HCs was less than 1 point) more likely to be statistically significant. Nonetheless, this is something for future research to consider when evaluating change in depressive symptoms.

Looking ahead, while the present study focused on intravenous racemic ketamine, future EEG studies might consider similar assessment before and after intranasal delivery of esketamine which has similar antidepressant effects [85]. This could also be combined with adding a placebo-controlled group of participants with TRD and a sham treatment group for HC as well as adjusting to a cross-over design, all of which would better enable causal analyses regarding ketamine’s effects on neural connectivity and overcome an important limitation of the present study. Furthermore, the availability of FDA-approved esketamine for psychiatric indications may also allow recruitment of larger samples, increasing statistical power, and thus navigating a challenge faced by ROI-based connectivity studies (i.e., the number of statistical tests conducted). That said, while modest, our sample size of n = 24 TRD participants is well aligned with existing literature, whereby across 34 relevant electrophysiological studies of functional connectivity in depression and its treatment included in recent reviews the average number of participants with depression included in the analysis was approximately 22 people [33, 72].

When speaking to our sample, it is worth considering other possible sources of group differences. As previously noted, TRD subjects were, on average, older than their HC counterparts and all self-identified as White. While our analysis controlled for effects of age, it did not factor in racial identity. This was, in part, because it would lead to subgroups too small for meaningful analyses, and because we had no a priori hypothesis of how LPS and change in LPS would be affected by race. Nonetheless, we know that race-related socioenvironmental factors can have neurobiological effects [86] and larger follow-up studies should factor the racial identity of their subjects into their analytical approach. Additionally, TRD participants were all medicated (Supplementary Table 2) and it is possible that this may also affect neural connectivity. As such, the inclusion of a group of unmedicated TRD participants, while challenging for several reasons (e.g., washout procedures, ethical requirements of care) would help future research to further specify ketamine’s effects.

Finally, to focus specifically on future directions for EEG-based functional connectivity, while we constrained our analysis to the theta and beta frequencies based on a synthesis of several reviews and empirical studies, Miljevic et al., [72] note several studies have observed alpha-band connectivity effects relevant to depression. Alpha-based EEG parameters, most often alpha asymmetry, have been regularly used in depression-research [87] and so, future studies should consider evaluating if connectivity at the alpha frequencies are meaningful targets. Moreover, such studies may also consider adding task-based measures of functional connectivity to supplement any resting-state findings. The design structure and standardized nature of typical EEG tasks could potentially be used to constrain the neural processes and associated systems activated and subsequently directly target networks of interest, i.e., the DMN, FPN, and/or SN.

In sum, we evaluated the effect of ketamine on EEG-derived source-based measures of DMN and FPN rsFC in individuals with TRD. Within 24-hours of their first infusion, individuals with TRD showed significant reductions in depressive, anhedonic, and ruminative symptoms. Moreover, they were characterized by broad increases in functional connectivity within the DMN and FPN as well as between the two networks. Based on preclinical evidence, we speculate such general connectivity increases might be driven by ketamine’s synaptogenic effects. Because preclinical models suggest that these synaptogenic effects can decline rapidly, an important next step will be to use EEG to track the full time-course of ketamine therapy, starting prior to the first infusion and following through all subsequent doses.

Author contributions

Conceptualization: D.A.P., R.C.M. Methodology: D.A.P. Formal Analysis: T.L. Investigation: J.N.S. Jr., S.M.E., M.L., S.E.W. Resources: D.A.P. Data Curation: T.L., S.R.L., R.D. Writing – Original Draft: T.L. Writing – Review & Editing: T.L., J.N.S.Jr., S.M.E., M.L., S.E.W., S.R.L., R.D., M.B., B.W.B., C.M., P.B., S.L., R.C.M., D.A.P. Visualisation: T.L. Supervision: D.A.P. Project Administration: D.A.P. Funding Acquisition: D.A.P.

Data availability

The data used in the present work are available upon reasonable request to the corresponding author.

Code availability

We used the following software in preprocessing the raw EEG data and the subsequent analysis: BrainVision Analyzer (v2.2), LORETA (v20221219) via its graphical interface; R (v4.3.0) [76] using R-Studio (v2023.06.0.421) [77], and the stats (v4.3.0) [76] R package. The R code and EEG preprocessing templates are available from the corresponding author on reasonable request.

Competing interests

Over the past 3 years, Dr. Pizzagalli has received consulting fees from Abbvie, Arrowhead Pharmaceuticals, Boehringer Ingelheim, Circular Genomics, Compass Pathways, Engrail Therapeutics, N1 Biocorp Inc, Neumora Therapeutics, Neurocrine Biosciences, Neuroscience Software, Takeda, TP Sciences, and Xenon Pharmaceuticals; he has received honoraria from the American Psychological Association, Psychonomic Society and Springer (for editorial work) and Alkermes; he has received research funding from the Bird Foundation, Brain and Behavior Research Foundation, Circular Genomics, Dana Foundation, Millennium Pharmaceuticals, NIMH, and Wellcome Leap; he has received stock options from Ceretype Neuromedicine, Compass Pathways, Engrail Therapeutics, Neumora Therapeutics, and Neuroscience Software. All other authors have no conflicts of interest or relevant disclosures. All views expressed are solely those of the authors.

Supplementary information

The online version contains supplementary material available at 10.1038/s41398-026-03928-4.