Neurophysiological evidence that frontoparietal connectivity and changes in the GABA-A receptor underlie the antidepressant response to ketamine

Abstract. Identification of acute cortical pharmacodynamics of the dose of the antidepressant ketamine in people with depression is key to identification of specific mechanism(s) of action to alleviate symptoms. Although the following effects are characterized by increased plasticity and reduced symptoms of depression, this cascade of events triggers an acute response in the brain. Computational modeling of cortical interlaminar and cortico-cortical communication and receptor dynamics makes it possible to investigate this question using human electroencephalography (EEG) data recorded during infusion ketamine. Here, a resting EEG was recorded in a group of 30 patients with major depressive disorder (MDD) at baseline and at a ketamine dose of 0.44 mg/kg, including bolus and infusion. Frontoparietal connectivity was assessed using dynamic causal modeling to match the thalamocortical model to hierarchically related nodes in the medial prefrontal cortex and superior parietal lobe. We found a significant increase in AMPA-mediated parietofrontal connectivity and a significant decrease in the frontal time constant of GABA. Changes in both parameters correlated between participants with the response of antidepressants to ketamine. Changes in NMDA receptor time constant and inhibitory intraneuronal injection into superficial pyramidal cells did not stand up to correction for multiple comparisons and did not correlate with response to antidepressants. These results suggest that the antidepressant effect of ketamine may be mediated by acute frontoparietal connectivity and GABA receptor dynamics. In addition, it supports a large number of literature sources suggesting that the acute mechanism underlying the antidepressant properties of ketamine is related to GABA-A and AMPA receptors rather than NMDA receptor antagonism.

Entry. It is argued that the discovery of the fast-acting antidepressant effect of ketamine is the most important event in the study of depression in the last 50 years [1]. The first empirical evidence was provided by Berman and Cappiello [2], who demonstrated the efficacy of a single subanesthetic intravenous (IV) infusion of ketamine (0.5 mg/kg) over 40 minutes in the treatment of major depressive disorder (MDD). The antidepressant properties of ketamine have so far been repeated many times, and it has been empirically shown that that ketamine demonstrates efficacy among patients, including those considered resistant to treatment with conventional antidepressants [3]. The identification of the acute pharmacodynamic effect of ketamine, which precedes its subsequent antidepressant effects, remains an open question of the study.

Ketamine is often described simply as an N-methyl-d-aspartate (NMDA) receptor antagonist, however, the failure of other NDMA antagonists, such as memantine [4], to exert antidepressant effects suggests that the complex and possibly synergistic pharmacological effects of ketamine are necessary to induce an antidepressant response. Possible direct effects include ketamine, which preferentially binds to NMDA receptors, which densely populate γ-aminobutyric acid (GABA)-mediated interneurons [5]. Possible indirect effects include subsequent disinhibition of α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptors on the postsynaptic neuron with increased flow excitation and triggering proplasticity cascades [5].

These pharmacodynamic effects of ketamine are likely to interact at the meso- and macroscale through alteration of cortical and cortico-cortical coupling. MDD is often described as a disorder characterized by abnormal communication of brain networks at rest [6, 7], and hypotheses of depression in the brain network suggest that treatment with antidepressants may have therapeutic benefits due to the return of resting network connections to optimal levels [8, 9]. Although it has been argued that the acute effects of ketamine on frontoparietal connectivity may play a role in its antidepressant properties, this has been done based on healthy samples and has never been tested in patients with IDD [10,11,12,13].

Here, we recorded a resting electroencephalography (EEG) during anesthesia, an antidepressant infusion of ketamine in patients with IDD. We set out to identify acute changes in the cortical chip and GABA, AMPA, and NMDA receptors that generate spectral effects on EEG using generative computational modeling. To date, simulations have only been used to assess long-term (3–9 hours post-infusion) effects in people with depression [14,15,16,17] or effects in people without depression [10, 18]. The model used here was more complex [18], more carefully parameterizing the mutual connection of the thalamus with the cortex, as well as a separate superficial and deep interneuronal input. GABA-A, AMRA, and NMDA receptors are modeled as in Moran, Symmonds [19]. In addition, we investigated whether acute disruption of frontoparietal connectivity with ketamine found in healthy participants [10,11,12,13] is also found in patients with IDD. In addition, whether changes in connectivity in the acute phase of ketamine infusion are associated with response to antidepressants. Finally, a dataset of healthy young men [13] who underwent an identical EEG recording and ketamine infusion protocol is also modeled for comparison.

Successful simulation of the activity of the acute effects of ketamine, which are thought to mediate its antidepressant response, will not only support and enable the translation of preclinical work that predicts the antidepressant properties of ketamine, but will also provide a benchmark for testing the pharmacodynamic and therapeutic properties of ketamine. Effects of new drugs developed on or related to ketamine. Such methods can be used to monitor predicted effect in the brain and are developed in conjunction with pharmacokinetic modelling to support dose determination.

Methods. Participants. The main dataset used comes from a study that was a randomized, double-blind, active, placebo-controlled, cross-planning trial involving 30 participants who met the DSM-IV criteria for major depressive disorder (MDD). Data from 27/30 participants were included (mean age = 30.2 years; SD = 7,9; range = 18–48; 15 women). Data from one participant was excluded due to an excessive movement artifact that affected data quality, and two due to infusion interruptions. Some data from participants in the current study have been published previously [15, 16, 20], and full information about the cohort can be found in these publications (a summary and full inclusion and exclusion criteria are provided in the Supplementary Material).

Participants received racemic ketamine at one study visit and active placebo remifentanil hydrochloride (Ultiva®, GlaxoSmithKline, Auckland, New Zealand) at the other, the order of which was randomized and balanced with a minimum three-week withdrawal period. Participants provided informed written consent. This study has been approved by the New Zealand Health and Disability Ethics Committee (15/NTB/53) and the study has been registered at: https://www.anzctr.org.au/Trial/Registration/TrialReview.aspx?id= 368052&isReview=true, registration number: ACTRN12615000573550.

Administration of the drug. All interventions were performed by an anesthesiologist. The medication was delivered through an intravenous cannula into the left pre-ulnar fossa in a magnetic resonance imaging (MRI) environment. Ketamine was administered as a 0.25 mg/kg bolus followed by an infusion of 0.25 mg/kg/hour over 45 minutes. Active placebo remifentanil was administered as a 9-minute targeted controlled infusion to achieve a predicted plasma concentration of 1.7 ng/mL using the Minto pharmacokinetic model [21, 22]. If a participant’s body mass index (BMI) exceeded 30 kg/m2, they received a dose according to their calculated ideal body weight (IBW). See Sumner, McMillan [15] for a summary of side effects and blindness.

Response to antidepressants for correlation analysis with neuroimaging measurements was defined as the percentage change in the Montgomery Asberg Depression Rating Scale (MADRS) one day after ketamine ingestion from baseline, and the effect of ketamine compared to active placebo on the MADRS score for all time points was assessed using linear mixed modeling with limited maximum likelihood score.

EEG data collection. EEG data was continuously recorded from 64 channels using standard BrainCap MR caps and BrainAmp MR Plus amplifiers (Brain Products, Munich, Germany). For more information on collection options, see the Supplementary Material. Resting EEG data was collected during a functional MRI (fMRI) resting state scan, in which participants kept their eyes open while lying down for 7 minutes prior to ketamine, for two minutes in between, including and immediately after the ketamine bolus, and for seven minutes during infusion.

Analysis of EEG data. Pre-treatment. The initial stages of pre-processing EEG data are presented in the Supplementary Material and described earlier [20]. The data were erased into 2200 ms segments corresponding to the volumes collected by fMRI and then compared to the overall mean reference score.

Spectral analysis. Global covariance matrices were created and obtained after filtering into seven frequency bands using a fourth-order bidirectional Butterworth filter: delta (1–4 Hz), theta (4–8 Hz), alpha (8–13 Hz), low beta ( 15–26 Hz), high beta (28–40 Hz), low gamma (42–53 Hz), and high gamma (55–67 Hz). These frequencies were chosen to avoid slice artifact frequency and its harmonics contaminating EEG data collected simultaneously with fMRI (see data with scanner harmonics in Supplementary Fig. S1). Linearly constrained minimum variance beamforming [23] was applied using 6 mm resolution grids warped to the template head model provided with FieldTrip [24] to generate spatial filters for each voxel (i.e., 5604 locations) for each frequency band. To calculate volumetric images of the effect of ketamine on source amplitude, a Hilbert transform was used to obtain the amplitude envelope of the oscillations at each site, with the mean amplitude calculated for the pre-infusion and intra-infusion time periods. Differences at each site were calculated at the group level using paired samples t-tests.

The peaks allocated for the simulation were chosen, as in Muthukumaraswamy, Shaw [10]. The Montreal Neurological Institute (MNI) identified a voxel representing a peak increase in frontal theta near the left medial prefrontal cortex [MNI: −12 36 60] and a voxel representing a peak alpha decrease in the left parietal cortex [MNI: −24 −66 66] were selected from a large mean localized source volume.

Computational modeling. Basic research. The thalamocortical model described in Shaw, Muthukumaraswamy [18] was fitted to resting EEG data (see Fig. 1 for the model diagram). Data were filtered from 3 to 100 Hz, and epochs of 2200 ms were averaged according to whether they were recorded 7 minutes before ketamine administration, during 2 minutes of bolus administration, and within 5 minutes of ketamine infusion immediately after bolus.

The model itself is an 8-population model that builds on the previous 4-population spm_fx_cmm_NMDA.m model available in the SMP12 software (http://www.fil.ion.ucl.ac.uk/spm/) [10, 19]. The extended model additionally contains the thalamic relay and deep pyramidal projection populations. Inhibitory interneurons act separately on the superficial and deep layers, as well as the inhibitory reticular input on the thalamus. AMPA, NMDA, and GABA-A receptors are present in each cell populations.

The model equations and neural and observational model parameterizations for intrinsic connections and receptors are fully detailed in Ref. [18] and available on GitHub (https://github.com/alexandershaw4/TransPsych_KetDep_GABA_SumnerShaw ). From here, the model looks like the spm_fx_cmm_NMDA model; external connections are mediated by AMPA and NMDA receptors, as in Muthukumaraswamy et al. [10]. Concretely modeled connections and receptors are shown in fig. 1 and are informed by the canonical microcircuits proposed by Douglas and Martin [25] based on Hilbert and Wiesel [26]. Our choice of model is based on the observation that cortical-only (e.g., canonical IC) models exhibit a limited repertoire of simultaneous frequency responses insufficient to reproduce the broadband spectral changes induced by our data. Accordingly, we have shown that our extended model overcomes this bandwidth limitation and outperforms canonical cortical models in Bayesian model selection [18] even when adjusted for the increased number of parameters.

The model was established using standard dynamic causal modeling (DCM) for spectral densities, as in Moran et al. [27] which uses free energy as an objective function, so that the objective function is not just a term of plausibility (i.e., correspondence), but rather a term of accuracy minus complexity. Thus, the published result represents not only the best match of the model’s spectrum to the data, but also the “simplest best match” in terms of parameter updates.

Repeated measurement variance (RM) analysis was used to assess the effect of ketamine on pre-infusion, during bolus, and during infusion. The parameters analyzed include external AMPA and NMDA, feedback and forward connections between the parietal and frontal nodes. Internal connections analyzed included self-reinforcement on superficial pyramidal (SP) cells, administration of superficial inhibitory interneurons (SIs) on SPs, relays (RLs) on spiny stellate (SS) and thalamic pyramidal (TP) cells on RLs in both parietal and frontal nodes. . The time constants analyzed included the frontal and parietal populations of AMPA, GABA-A, and NMDA receptors separately.

Connection parameters were analyzed by connection and time constant as separate ANOVA. A correction for multiple comparisons was applied to the result of a univariate comparison in each ANOVA analysis using the False Discovery Rate (FDR) of Benjaminy and Hochberg [28].

The Pearson correlation was performed on the absolute change in the MADRS score at 24 hours and 7 days after ketamine and parameters that were significantly altered by ketamine if not adjusted for multiple comparisons. Correlations have been adjusted for many comparisons, as noted above, using FDR.

Repeatability of ketamine effects. To test whether the model would produce the same pharmacodynamic effects of ketamine, a second set of data collected under equivalent conditions was used. Ketamine was administered as in the “Medication Administration” section. EEG data was collected, as in the “EEG Data Collection” section, and the EEG was pre-processed and prepared for simulation, as in the “Computer Simulation” section. These data are published in Forsyth, McMillan [29]. This dataset includes 30 healthy male participants without IDD (mean age = 27.3 years; SD = 6,2; range = 19–37).

Results. Clinical effects of mood. The significant antidepressant response of these participants has been extensively described elsewhere [15, 16, 20, 30]. The absolute reduction in MADRS from baseline to 24 h after ketamine, which is used in correlation, can be found in the Supplementary Material.

Ketamine-induced volumetric spectral EEG changes. The ketamine-induced volumetric spectral EEG changes were computed in baseline space by comparing baseline to ketamine with ketamine infusion data and presented in the Supplementary Material. According to similar analyses in healthy volunteers [10, 31] and results in a subgroup of these participants [20], ketamine significantly reduced spectral amplitude in the delta, alpha and beta bands, while an increase in spectral amplitude was observed in the frequency bands of high beta radiation, low levels of gamma radiation and high range of gamma radiation. The theta frequency band showed a relatively small increase in frontal spectral amplitude and a posterior decrease.

Modelling of pharmacodynamic effects of ketamine. The models provided a good match for all individual spectral densities (variability >90% explained), the model’s correspondence compared to the data is plotted in the Supplementary Material.

 

A summary of the RM ANOVA results is shown in Table 1. Ketamine significantly enhanced the direct connection of AMPA (parietal to frontal) in bolus and infusion in an apparently linear manner (Fig. 2, panel 2.1A), but did not alter AMPA feedback or NMDA receptor-mediated coupling (Fig. 2, panel 2.1B–D). Figure 3 depicts these results in a thalamocortical model.

Ketamine significantly reduced the frontal time constant of the GABA-A population, with most of this reduction occurring in the bolus (Table 1, Figure 2, Panel 2.2A). Without correction, the time constant of the parietal NMDA population increased linearly, and the internal superficial inhibitory intraneuronal input to the frontal superficial pyramidal populations decreased linearly. These two results did not stand up to FDR correction.

An increase in AMPA-mediated direct association was significantly correlated with a smaller decrease in MADRS 24 hours after ketamine (r = 0.43, p = 0.02) (Fig. 2, panel 2.1E). The time constant of the GABA-A receptor was also significantly correlated with the change, a smaller decrease in the time constant of the GABA-A receptor was also associated with a greater decrease in MADRS (r = -0.38, p = 0.04) (Fig. 2, panel 2.2B). There was no significant correlation between the NMDA receptor time constant and the change in MADRS. 7 days post-infusion, only a decrease in the GABA-A time constant was still significantly correlated with a decrease in MADRS (see supplementary material).

Repeatability of ketamine effects in a healthy volunteer dataset. The thalamocortical model used provided a good match for all individual spectral densities with an explained variability of >90%.

A summary of the RM ANOVA results is shown in Table 2. The greatly increased direct AMPA connection is consistent with the core dataset (Fig. 4, Panel 4.2A). The reduction in superficial intraneuronal insertion into the superficial pyramidal cells was consistent with the original dataset, although in this sample the result was very significant and withstood correction. The reduction in frontal self-reinforcement on superficial pyramidal cells was also significant. All other intrinsic parameters were also significantly altered by ketamine. Parietal superficial interneuronal input on superficial pyramidal cells and self-amplification on pyramidal cells were significant. All thalamocortical connections were significantly enhanced by ketamine, including frontal and parietal relay to spinous stellate cells and thalamic pyramid to relay parameters.

Interestingly, the time constant of the GABA receptor did not change significantly in Table 2. Instead, the time constant of the AMPA receptor in the parietal and frontal nodes was significantly increased (Fig. 4, panel 4.2A, D). The time constant of the parietal NMDA receptor also increased (Fig. 4, panel 4.2F), although it did not survive after FDR correction.

 

Discussion. The current study examined the resting EEG before ketamine infusion for depression, during bolus, and during infusion. During infusion, ketamine significantly modulated the spectral amplitude in all seven frequency bands analyzed and was largely typical of those documented in healthy volunteers [10, 11, 31, 32, 33].

A thalamocortical model of laminar microcircuitry in the medial prefrontal cortex and superior parietal lobe was fitted to the spectra using DCM based on previous work on the modulatory effects of ketamine on the brain [10,11,12,13] (Fig. 1). The study revealed a ketamine bolus and infusion-mediated increase in AMPA fronto-parietal anterior connectivity (from parietal to frontal) and a decrease in the GABA-A receptor time constant (Fig. 3). The increased time constant of the parietal NMDA receptor and the decrease in the inhibitory intraneuronal input to the surface pyramidal cells did not withstand multiple correction of comparisons (Fig. 3). Together, these results are consistent with the known effects of ketamine from preclinical work (as reviewed in references [5, 34]). Significant changes in AMPA anteroparietal connectivity and the GABA-A receptor time constant were significantly correlated with antidepressant response to ketamine at 24 hours. The correlation with the GABA-A receptor time constant was still significant 7 days after ketamine.

A benchmark data set was used to test the consistency of the model results with ketamine. The dataset used was collected and preprocessed under equivalent conditions. A key difference was that the cohort consisted of healthy young males. Many of the parameter changes obtained in the model output were in the same direction as in the ADHD cohort, but overall more parameters were statistically significant in this data set than in the ADHD cohort, including increased thalamocortical connectivity. The reduction in the number of significant effects in the depression data set may be due to the greater heterogeneity of this cohort, including factors such as gender, age, co-medications, co-morbidities, and depression. This heterogeneity may both reduce the power to detect significant changes and account for inherent physiological differences in response to ketamine between cohorts.

Contextualization of the finding with data previously analyzed from the main cohort with VDR. According to the current results, the more the time constant of the GABA-A receptor was reduced and the more frontoparietal AMPA was increased by ketamine, the less likely an individual was to experience an antidepressant response (Fig. 2, panels 2.1E and 2.2B). Paradoxical is the idea that ketamine has a stronger effect on the brain in those who have a lower response to antidepressants. Previous publications report a more intense experience in the same data set; in particular, experiences of unity, spirituality, and insight, as measured by the 11D-ASC, were correlated with greater antidepressant response [30], indicating that these nonresponders subjectively had less intense experiences. At 3 h post-ketamine, a greater increase in uplinking in response to unexpected sensory input (auditory mismatch response) was associated with a greater response to antidepressants, indicating improved sensitivity to prediction errors and short-term plasticity mechanisms [16]. This suggests that responders will have greater improvements in plasticity than non-responders.

However, there are also preliminary results from this dataset that are consistent with this finding of greater acute changes associated with poor response to antidepressants. The Blood Oxygen-Dependent Signal (BOLD) was significantly increased in participants with lower improvements in MADRS on concomitantly recorded resting fMRI [20]. This has occurred in many spatially distributed areas of the brain. Therefore, even though the result seems paradoxical at first, it is in the broader context of this cohort’s data.

AMPA-mediated correlation of direct communication with antidepressant response. In the broader literature, greater change in direct AMPA coupling has previously been associated with antidepressant response. Specifically, in depression 6–9 h after ketamine administration during a tactile task and using a four-population model (cmm_NMDA), it was shown that greater changes in direct AMPA communication from the somatosensory (parietal) to the frontal cortex were associated with the change of MADRS [14]. This suggests that our finding of acute changes may have persisted until the time of infusion, and that changes in AMPA-mediated feedforward communication appear to be a reliable correlate of the antidepressant response to ketamine. Furthermore, the current study adds to growing evidence that ketamine consistently alters frontal and frontoparietal connections, and this effect is robust to a number of robust and functional connectivity analysis methods [10,11,12,13].

Correlation of the GABA-A receptor time constant with antidepressant response. The role of GABAergic inhibitory interneuron neurotransmission in the medial prefrontal cortex and depression has been established [35], and is widely described as important for the antidepressant response to ketamine [34]. GABA-A-mediated disinhibition of interneurons is most specifically accounted for by two model parameters: surface inhibitory input of interneurons to superficial pyramidal cells and the time constant of the GABA-A receptor. A reduction in inhibitory interneuron input to superficial pyramidal cells was previously shown using a four-population model (spm_fx_cmc.m) after ketamine infusion [36]. In the current study, both surface inhibitory interneuron input to superficial pyramidal cells and the GABA-A receptor time constant were reduced by bolus and infusion of ketamine, although only the GABA receptor time constant survived correction. Interestingly, while the comparative data set of healthy young men tended to show more significant changes in the same parameters as well as additional significant parameters during infusion, the decrease in the GABA time constant was not significant.

The addition of GABA-A receptors is unique to the current model, and it has been found that their alteration by ketamine is associated with the magnitude of the antidepressant response. In the type of simulation used for this study, the time constants can be considered a concentrated parameter of receptor kinetics, however, the changes that can be detected at the mesoscale (i.e., local field potential spectra) probably contain information including changes in the abundance and density of the receptor, (a) synchronous discovery, as well as the time of opening of the channel and the decay constant itself. Dysfunction of the GABA-A receptor is well documented in depression, where GABA deficiency has been found [37]. Increased expression of GABA receptor mRNA in people who have died by suicide may be a compensator for low GABA levels [35, 38].

We hypothesize that greater reactivity of the time constant of the GABA-A receptor in people who responded less to ketamine indicates greater GABA dysfunction. For example, because ketamine increases the sensitivity of GABA-A receptors to low GABA concentrations [39], participants with lower GABA levels [37] may have experienced the greatest change from baseline.

Strengths and limitations. One of the strengths of this study was the availability of a control group to compare with the main cohort with VPR. However, the ability to compare groups was limited by the fact that they did not match by age, medication, or gender. For this reason, we decided not to make quantitative comparisons between the two cohorts. In general, the similarity in changes in synaptic parameters between groups suggests the reliability of restoring the parameters and effects of the drug, this includes the direction of the changes, even if they are minor (Fig. 2, Panel 2.1A–D compared to Fig. 4, Panel 4.1). A–D and Fig. 2, panel 2.2A compared to Fig. 4, panel 4.2A–F). In considering these similarities, it is important to note that, unlike many spectral studies of DCM (e.g., [10]), we did not use the overall linear contrast of the model to encourage this direction of change.

Computational models of the generated electrophysiological data are limited in that they do not provide an estimate of every potential parameter in the brain (i.e., every channel and every cell type). However, they represent a mesoscopic and biologically plausible representation of the main participants in the electrophysiologically generated signal, and in the current model and similar DCM-equipped models, parameter estimates are limited to biologically sound prior and expected posteriori estimates distributions. One of the biggest sources of evidence that models complement invasive electrophysiology and pharmacodynamic studies is in cases where models cause plausible and realistic changes. TCM’s current data and results are a prime example of this.

Future directions and conclusion. Current evidence suggests that the model used, with subsequent validation and testing, may play a future role in improving treatment. For example, people with smaller changes in MADRS and a greater reduction in GABA receptor time constant and an increase in AMPA coupling may benefit from different infusion regimens, such as less and slower administration of medications to regulate the cortical response. In addition, These data may indicate a marker of people who will not respond to ketamine, in which case the model can be used to prevent unnecessary subsequent doses of ketamine and accelerate the transition to another potentially more beneficial treatment. Such hypotheses can be tested and confirmed using the methods and model used in the current study.

Overall, this study supports a large body of literature suggesting that the acute mechanism underlying the antidepressant effects of ketamine is related to the properties of the GABA-A and AMPA receptors, rather than NMDA receptor antagonism. While generative computational modeling is limited to describing the data in terms of their parameters in the event of further validation, the relationship between key parameters mediated by GABA and AMPA receptors indicates that simulations can also be used as a precision medicine tool during a patient’s first infusion, informing their individual regimen, including dose, infusion rate (if appropriate), as well as reinforcing and complementary therapies. Such testing may include simulations of preclinical work or further studies of drugs in humans that disrupt GABA, AMPA, and other systems.

Source https://www.nature.com/articles/s41398-024-02738-w