Brain entropy and neurotrophic molecular markers accompanying clinical improvement after ketamine: Preliminary evidence in adolescents with treatment-resistant depression

Abstract

Background:

Current theory suggests that treatment-resistant depression (TRD) involves impaired neuroplasticity resulting in cognitive and neural rigidity, and that clinical improvement may require increasing brain flexibility and adaptability.

Aims:

In this hypothesis-generating study, we sought to identify preliminary evidence of brain flexibility correlates of clinical change within the context of an open-label ketamine trial in adolescents with TRD, focusing on two promising candidate markers of neural flexibility: (a) entropy of resting-state functional magnetic resonance imaging (fMRI) signals; and (b) insulin-stimulated phosphorylation of mammalian target of rapamycin (mTOR) and glycogen synthase-3-beta (GSK3β) in peripheral blood mononuclear cells.

Methods:

We collected resting-state functional magnetic resonance imaging data and blood samples from 13 adolescents with TRD before and after a series of six ketamine infusions over 2 weeks. Usable pre/post ketamine data were available from 11 adolescents for imaging and from 10 adolescents for molecular signaling. We examined correlations between treatment response and changes in the central and peripheral flexibility markers.

Results:

Depression reduction correlated with increased nucleus accumbens entropy. Follow-up analyses suggested that physiological changes were associated with treatment response. In contrast to treatment non-responders (n=6), responders (n=5) showed greater increase in nucleus accumbens entropy after ketamine, together with greater post-treatment insulin/mTOR/GSK3β signaling.

Conclusions:

These data provide preliminary evidence that changes in neural flexibility may underlie symptom relief in adolescents with TRD following ketamine. Future research with adequately powered samples is needed to confirm resting-state entropy and insulin-stimulated mTOR and GSK3β as brain flexibility markers and candidate targets for future clinical trials.

Clinical trial name:

Ketamine in adolescents with treatment-resistant depression

URL:

https://clinicaltrials.gov/ct2/show/NCT02078817

Registration number:

NCT02078817

Introduction

Depression is a leading cause of global disability and a key risk factor for suicide (Friedrich, 2017; Whiteford et al., 2013). Depression often emerges for the first time during adolescence (Kessler et al., 2005), a time notable for rapid changes in brain development and synaptic remodeling (Giedd et al., 1999; Paus et al., 2008; Raznahan et al., 2011). Although effective treatments are available, a significant number of adolescents with depression do not respond to standard treatments (March et al., 2006), allowing the disease course to progress (Amital et al., 2008; Maalouf et al., 2011). Such treatment-resistant depression (TRD) is associated with poor outcomes, including academic failure, loss of relationships, and exacerbation of depression symptoms such as poor self-esteem, hopelessness and suicide attempts (Asarnow et al., 2011; Crown et al., 2002; Greden, 2001). TRD negatively impacts development and impedes the successful transition to adulthood (Aalto-Setälä et al., 2002; Fergusson and Woodward, 2002; McLeod et al., 2016). Hence, novel treatments are urgently needed for adolescent TRD, with the goal of restoring healthy neurobehavioral development early in the disease course and preventing long-term negative outcomes. The process of developing such treatments will require a deeper knowledge of the key neurobiological changes that are associated with achieving therapeutic clinical outcomes.

Accordingly, ketamine has emerged as a promising intervention for adults with TRD (Newport et al., 2015). The rapid onset and large effect sizes observed in trials of ketamine for TRD (Bobo et al., 2016; Coyle and Laws, 2015; Wilkinson et al., 2017) underscore its potential value as a research probe for illuminating biological mechanisms underlying clinical improvement in this difficult to treat patient population. A core feature of depression in general, and TRD in particular, is the tendency to get “stuck” in negative moods (Holtzheimer and Mayberg, 2011). Cognitively, patients with severe depression perseverate, or ruminate, on negative thoughts, experience a rigidly negative perspective on themselves and the world, and perceive a sense of narrowed options (Cleary, 2012; Lennings, 1994). This lack of flexibility may be reflected in cellular functioning (e.g. limited molecular responsivity to environmental stimuli) and in brain signaling (e.g. low variability and adaptability). Examination of central and peripheral indices of brain flexibility may be a promising approach for understanding neural correlates of clinical response to rapid-acting treatments such as ketamine.

Treatment-related changes in neural flexibility can be measured through quantification of entropy. Entropy, a concept originally born in physics and more recently adapted to information theory, is a univariate measure of signal variability and unpredictability (Coifman and Wickerhauser, 1992; Shannon, 1997). Although the entropy of any brain signal can be estimated from its time course using a range of methods, the focus here is on resting-state functional magnetic resonance imaging (rs-fMRI) entropy. Large scale studies of rs-fMRI entropy in healthy human subjects have revealed hierarchical brain entropy networks consistent with conventional functional and anatomical brain parcellations (Wang et al., 2014). Multi-scale rs-fMRI entropy studies in healthy subjects have further shown that regional entropy is correlated with network functional connectivity in a frequency specific manner (at low frequencies, rs-fMRI entropy is positively associated with functional connectivity, while at high frequencies, entropy is inversely associated with connectivity) (Wang et al., 2018).

The cognitive inflexibility associated with depression may be characterized by rigidity in neural signals, whereas increased signal complexity (less rigidity) may reflect clinical improvement. Increased rs-fMRI entropy has been associated with increased intelligence in healthy individuals, suggesting that intellectual capacity may critically depend on the brain’s flexibility and ability to access highly variable and complex neural states (Saxe et al., 2018). Further compelling evidence for utilizing rs-fMRI entropy as a potential biomarker of neural flexibility comes from a large study of healthy subjects (n=386) in which increased rs-fMRI entropy was positively correlated with different aspects of divergent thinking—fluency, flexibility and originality (Shi et al., 2020). Importantly, this study included validation of these rs-fMRI entropy findings in two additional independent samples of healthy human subjects (n=431 and n=132). Entropy of rs-fMRI data has also been used to characterize signal rigidity in mental health conditions such as schizophrenia (Bassett et al.,2012) and attention deficit hyperactivity disorder (Sokunbi et al., 2013). Finally, rs-fMRI entropy of subcortical regions such as the amygdala and putamen has been associated with better performance in flexibility-related cognitive tasks such as verbal fluency and abstract thinking (Yang et al., 2013).

There is preliminary evidence that ketamine treatment may lead to increased entropy regardless of conscious state: one electroencephalography (EEG) study in humans reported significantly increased entropy following the administration of ketamine versus saline in anesthetized adult patients (Hans et al., 2005) and another reported increased spontaneous magnetoencephalography (MEG) signal complexity following administration of ketamine in awake adult patients (Schartner et al., 2017). However, to date, few studies have explicitly investigated the utility of rs-fMRI entropy in the study of depression or its treatment. Although rs-fMRI has a lower temporal resolution than EEG/MEG, it provides excellent spatial resolution of individual brain regions and functional networks. Additionally, whereas EEG/MEG data generally reflect rapid changes in whole brain electrical activity, rs-fMRI provides information about more sustained, gradual changes in hemodynamic activity. Thus, measures of rs-fMRI entropy provide complementary information to measures of EEG/MEG entropy. Although EEG entropy measures can be useful for characterizing instantaneous brain states, we suggest that rs-fMRI entropy measures can be used to extrapolate sustained changes in regional signal complexity which may correspond with behavioral changes in the context of TRD.

Additionally, a lack of adaptability at the cellular level may contribute to poor treatment outcomes in depression, due to failed upregulation of neurotrophic responses to treatment (Price et al., 2018). While such molecular changes in brain cells are difficult to capture in human studies, recent work suggests that using insulin probes to study molecular changes in peripheral blood mononuclear cells (PBMCs) may shed light on treatment-associated changes in pro-trophic events that promote neural flexibility. Animal studies have shown that activation of neurotrophic signaling cascades (i.e. brain derived neurotrophic factor and its receptor TrkB), together with the promotion of mammalian target of rapamycin (mTOR) and glycogen synthase-3-beta (GSK3β) phosphorylation, mediates the rapid antidepressant effects of ketamine, including behavioral changes and dendritic spine growth (Li et al., 2010; Liu et al., 2013). These neurotrophic responses are critical for synapse formation, synaptic plasticity and neural network remodeling, which are in turn thought to underlie improved mood regulation (Bessa et al., 2009; Price et al., 2018) and cognitive flexibility (Xu et al., 2019) in depression following effective treatment. At the molecular level, insulin serves as a critical moderator of such neurotrophic growth responses within the brain. Using a rodent model of TRD, we have established that these central pro-trophic insulin signaling responses can be derived in PBMCs upon ex vivo stimulation (Walker et al., 2019). These data suggest that assessing PBMC mTOR and GSK3β responses to insulin may serve as a minimally invasive and complementary approach for delineating treatment-induced pro-trophic molecular events supporting neural flexibility.

The current study examined neuroimaging markers of brain flexibility and associated peripheral pro-trophic molecular indicators of cellular adaptability to identify potential neurobiological correlates of clinical response to ketamine in adolescents with TRD. Markers were obtained during our recently conducted open-label pilot study testing intravenous ketamine as an intervention for adolescents with TRD (Cullen et al., 2018). Clinical results (high tolerability of the ketamine intervention and a significant reduction in depression symptoms for five of the participants) have already been published (Cullen et al., 2018). Here we examined rs-fMRI entropy and insulin-stimulated mTOR and GSK3β activation before and after treatment to identify neurobiological correlates of clinical improvement in the context of ketamine treatment in this sample. We predicted that clinical improvement following ketamine treatment would be accompanied by evidence for increased brain flexibility as measured by increased rs-fMRI entropy in specific depression-related regions of interest (ROIs) and by upregulation of insulin-stimulated mTOR and GSK3β phosphorylation in PBMCs. We also postulated that these brain and blood markers would correlate with each other, providing converging evidence for brain flexibility as a correlate of clinical improvement.

Methods

Overview

A detailed description of this sample has been published previously (Cullen et al., 2018). Briefly, this study was approved by the Institutional Review Board of the University of Minnesota (UMN). Participants were recruited via community postings and clinic referrals. Inclusion criteria were age 12–18 years, current diagnosis of major depressive disorder, Children’s Depression Rating Scale–Revised (CDRS-R) (Poznanski et al., 1984) raw score >40, and treatment resistance defined as a failure to exhibit a satisfactory response to at least two antidepressant medications. Rigor of antidepressant trials was assessed using the Antidepressant Treatment History Form (Sackeim, 2001); past trials were considered sufficient if they scored at least a 3 (on a scale of 1 to 4; as an example, a rating of 3 for fluoxetine is 4 weeks or more and dosage 20–39 mg/day, and a rating of 4 is 4 weeks or more and dosage ⩾40 mg/day), or if the trial was truncated due to intolerance (as opposed to an early decision regarding inefficacy). Current psychotropic medications had to be dose stable for 2 months. If participants opted to discontinue any psychotropic medications before the study, we required a washout period of 2 weeks for mood stabilizers and antipsychotic medications, 4 weeks for antidepressants, and 1 week for stimulants. Exclusion criteria were the presence of a current substance use disorder, a primary psychotic disorder, bipolar disorder, autism spectrum disorder, a history of intellectual disability, a neurological disorder, or a significant medical illness.

Participants completed baseline clinical assessments and baseline rs-fMRI one day before the first ketamine infusion; post-treatment clinical and rs-fMRI assessments were completed one day after the last infusion. Blood was drawn directly before the first infusion and 2 h after the last infusion.

Clinical assessments and ketamine infusions

After completing the informed consent and assent (where applicable) process, participants were evaluated using the Kiddie Schedule for Affective Disorders and Schizophrenia, Present and Lifetime Version (K-SADS-PL) (Kaufman et al., 1997). Adolescents and parents were interviewed separately by trained clinicians. Clinicians assessed depression using the CDRS-R (based on both adolescent and parent report) (Poznanski et al., 1984). A consensus meeting following the interviews integrated all available clinical information for diagnostic and inclusion finalization. At post-treatment, CDRS-R was repeated. The primary clinical outcome measure was percentage change in depression symptoms as measured by the CDRS-R. Additionally, participants were classified into groups of treatment responders and non-responders based on whether they had or did not have at least a 50% reduction in depression symptoms. For all results shown, a CDRS-R score change of 100% would be equivalent to complete remission of depression symptoms.

Each participant underwent six sessions of ketamine (0.5 mg/kg) infusions over the course of 2 weeks following the baseline MRI session. (As noted previously (Cullen et al., 2018), dosing was based on ideal body weight for the first five participants and then based on actual body weight for the remaining participants.) We elected to include a series of six infusions, rather than a single infusion, based on evidence emerging around the time of our study that repeated infusions led to higher responses (Shiroma et al., 2014).

Neuroimaging data acquisition

Scanning took place before the first ketamine infusion and one day after the 6th ketamine infusion. All neuroimaging data for the study were acquired using a 3T Siemens Prisma scanner at the Center for Magnetic Resonance Research at UMN. We utilized a multiband echo planar imaging sequence to improve the spatial and temporal resolution of the acquired fMRI data over conventional methods (Feinberg et al., 2010). Individual rs-fMRI data (eyes open, fixation cross, multiband factor of 8, time repeat of 710 ms, echo time of 30 ms, 2 mm isotropic voxel size, 680 volumes (~8 min)), along with a B0 field map and high resolution T1-weighted magnetization-prepared rapid acquisition with gradient echo anatomical scan (time repeat of 2530 ms, echo time of 3.65 ms, inversion time of 1100 ms, 7 degree flip angle, 1 mm isotropic voxel size, 4 min), were collected before and after the ketamine intervention for all 13 study participants.

rs-fMRI preprocessing

rs-fMRI data were preprocessed using FEAT (FMRI Expert Analysis Tool) Version 6.00, part of FSL (FMRIB’s Software Library, www.fmrib.ox.ac.uk/fsl). The following preprocessing pipeline was applied: motion correction using MCFLIRT (Jenkinson et al., 2002), B0 field map unwarping and distortion correction using FUGUE, non-brain removal using BET2 (Smith, 2002), spatial smoothing using a Gaussian kernel of full-width half-maximum 3 mm, grand mean intensity normalization of the entire 4D dataset by a single multiplicative factor and high-pass temporal filtering (Gaussian-weighted least-squares straight line fitting, with sigma=50.0 s). Independent component analysis–based exploratory data analysis was carried out using MELODIC (Beckmann and Smith, 2004) prior to automated identification and removal of artifactual components (Kelly et al., 2010) using FSL FIX. Registration to standard Montreal Neurological Institute space was carried out using FLIRT (Jenkinson et al., 2002; Jenkinson and Smith, 2001). Image quality of each resting-state scan was evaluated using the method of Power et al. (2012). Volumes with framewise displacement values of more than 0.5 mm and/or temporal derivative of time courses (DVARS) which exceeded 8 after motion correction only were flagged as having excessive motion, along with the previous volume and next two volumes. If a scan had more than 30% of volumes with “excessive motion,” that scan was excluded from analysis.

rs-fMRI entropy analysis

Time courses from 132 ROIs (cortical and subcortical regions from the FSL Harvard-Oxford atlas and cerebellar regions from the automated anatomical labeling atlas) were extracted using SPM’s functional connectivity toolbox (CONN) (Whitfield-Gabrieli and Nieto-Castanon, 2012). Entropy analysis was carried out using custom MATLAB scripts, including functions from the Wavelet Toolbox. rs-fMRI time courses for all 132 ROIs were first bandpass filtered (0.08–0.12 Hz) using a 6th order Butterworth filter. We chose to assess this high frequency range instead of the conventional 0.01–0.1 Hz range based on several studies that have shown higher mean frequency resting-state blood oxygen level dependent activity in limbic regions in depression (Ries et al., 2018; Wu et al., 2008). We focused on the Shannon entropy since it is a well-established concept in information theory and straightforward to compute (Bassett et al., 2012; Shannon, 1997). For this calculation, the Shannon entropy E of a signal s is defined by

$$E\left(s\right)=-\sum _i{s}_i^2\log \left(s_i^2\right)$$

with the convention 0log(0) = 0 (Coifman and Wickerhauser, 1992). Entropy values for each rs-fMRI scan were calculated for each ROI’s bandpass filtered time series data by using Matlab’s wentropy function and taking the negative of the output, to obtain positive Shannon entropy values per the equation above. We transformed the resulting values by log(E(s)) to help visualize changes in entropy. We examined local entropy changes in a subset of 14 ROIs that have been previously implicated in TRD, including the subcallosal cingulate cortex (Mayberg et al., 2005), regions within the default mode network which have been implicated in depression and rumination (posterior cingulate, precuneus, and anterior cingulate) (Greicius et al., 2007; Hamilton et al., 2015; Zhu et al., 2017) and limbic regions known to be implicated in emotion regulation and reward processing (bilateral hippocampus, amygdala, nucleus accumbens (NAc), insula and thalamus) (Phillips et al., 2003a, 2003b).

Individual changes in local entropy were also plotted against CDRS-R score changes and mTOR measures for correlation analysis. Additionally, Student’s t-tests (alpha=0.05) were used to compare mean local entropy changes in each of the 14 ROIs between the responder (n=5) and non-responder (n=6) groups. We examined uncorrected results and also applied a Bonferroni correction (p < 0.05/14=0.0036) to the ROI results to adjust for multiple tests.

PBMC assays

Whole blood was collected immediately prior to the first ketamine infusion and following the last ketamine infusion. Samples were then shipped overnight for PBMC isolation and testing. Upon receipt of the samples, PBMCs were isolated via ficoll gradient and slow frozen as previously described (Walker et al., 2019). Following completion of sample collection from all subjects, cells from each subject were counted and 1 × 10⁷ cells were placed into one of four 35 mm dishes with 5 mL of media. Then 10 μg insulin was added to two of these, allowing the others to serve as baseline. The cells were incubated for 5 min and then immediately centrifuged, and washed once with 10 mL of phosphate-buffered saline. The cells were then lysed with RIPA lysis buffer to prepare protein whole cell lysates. Relative change in total and phosphorylated levels of mTOR, pmTOR, GSK3β, pGSK3β were determined post-insulin exposure using ELISA (Cell Signaling Technology, Danvers, MA), in accordance with the manufacturer’s instructions. We measured the extent to which stimulating the cells with insulin increased the ratio of phosphorylated to non-phosphorylated mTOR (pmTOR/mTOR) and GSK3β (pGSK3β/GSK3β), before and after ketamine treatment. We considered insulin-mediated upregulation of pmTOR/mTOR and pGSK3β/GSK3β protein levels, relative to no insulin baseline levels, as variables for correlation analyses with clinical improvement and entropy measures, respectively.

Results

Participants

As previously reported, 13 adolescents completed the 2-week ketamine protocol (mean age 16.9 years, range 14.5–18.8 years, 8 biologically male). Five were considered responders (at least 50% decrease on CDRS-R scores). All 13 adolescents also completed the neuroimaging protocol before and after treatment, but data from two of these participants were excluded due to excessive motion (greater than 30% of the volumes in the pre and/or post scan exceeded our motion threshold described above.) There were 11 participants with usable scans for the entropy analysis and 10 of these participants had viable blood for the PBMC insulin assays.

Entropy analysis

After Bonferroni correction for multiple comparisons, of the 14 ROIs examined, only entropy changes in the right NAc significantly correlated with individual changes in CDRS-R scores (r=0.86, p=0.00066) (see Figure 1). All responders showed an increase in right NAc log(entropy) after treatment (n=5, mean=0.35, SEM=0.13, change range: 0.046 to 0.77) while all non-responders showed a decrease in right NAc log(entropy) after treatment (n=6, mean=−0.42, SEM=0.10, change range: −0.17 to −0.77). Entropy changes in the right NAc were significantly different between responder and non-responder groups (t=4.73, p=0.0011), with a Cohen’s effect size value (d=2.84), suggesting a very high practical significance of this difference (see Figure 2).

Figure 1.

Individual changes in right nucleus accumbens (rNAc) log(entropy) versus individual percentage changes in Children’s Depression Rating Scale–Revised (CDRS-R) ratings. Entropy changes were assessed for the 0.08–0.12 Hz range. With respect to baseline, all responders (n=5, blue) showed an increase in rNAc log(entropy) after ketamine, while all non-responders (n=6, red) showed a decrease in rNAc log(entropy) after ketamine. Increased rNAc log(entropy) strongly correlated with depression response.

Figure 2.

Group-level entropy changes (0.08–0.12 Hz) for depression-related regions of interest. Entropy changes with respect to baseline were assessed in the following regions of interest: left and right hippocampus (lHipp and rHipp), left and right amygdala (lAmyg and rAmyg), left and right nucleus accumbens (lNAc and rNAc), subcallosal cortex (SubCal), left and right insula (lInsula and rInsula), anterior cingulate cortex (ACC), posterior cingulate cortex (PCC), precuneus (PreC), and left and right thalamus (lThal and rThal). A significant group difference in rNAc log(entropy) change was observed between responders (n=5, blue) and non-responders (n=6, red).

Error bars show the standard error of the mean.

Insulin-stimulated change in mTOR and GSK3β

Figure 3 summarizes the results of the mTOR analyses. Post-ketamine treatment, responders showed significantly higher (t=2.41, p=0.042) insulin-induced upregulation of pmTOR/mTOR (n=5, mean=26.12%, SEM=4.37%) compared with non-responders (n=5, mean=1.05%, SEM=9.43%). In contrast, pre-ketamine treatment, no significant group difference in insulin-induced upregulation of pmTOR/mTOR was observed. Percentage change in CDRS-R score over the course of the study was significantly correlated (r=0.80, p=0.0053) with post-ketamine insulin-induced upregulation of pmTOR/mTOR. Further, post-ketamine insulin-induced upregulation of pmTOR/mTOR correlated with baseline-to-post-ketamine changes in right NAc entropy (r=0.72, p=0.019).

Figure 3.

Post-ketamine percentage changes in pmTOR/mTOR versus baseline-to-post-ketamine changes in Children’s Depression Rating Scale–Revised (CDRS-R) ratings and right nucleus accumbens (rNAc) log(entropy). Responders (n=5, blue) had a significantly greater percentage change in post-ketamine pmTOR/mTOR compared with non-responders (n=5, red) (a). Post-ketamine percentage changes in pmTOR/mTOR strongly correlated with baseline-to-post-ketamine changes in CDRS-R ratings (b) and baseline-to-post-ketamine changes in rNAc log(entropy) (c), respectively.

Error bars show the standard error of the mean.

Similar findings were observed for the GSK3β analyses (Figure 4). Post-ketamine treatment, responders showed significantly higher (t=3.73, p=0.0058) insulin-induced upregulation of pGSK3β/GSK3β (n=5, mean=16.20%, SEM=3.14%) compared with non-responders (n=5, mean=−3.99%, SEM=4.40%). No significant group difference in insulin-induced upregulation of pGSK3β/GSK3β was observed pre-ketamine exposure. Percentage change in CDRS-R score over the course of the study was also significantly correlated (r=0.87, p=0.0012) with post-ketamine insulin-induced upregulation of pGSK3β/GSK3β. Finally, post-ketamine insulin-induced upregulation of pGSK3β/GSK3β correlated with baseline-to-post-ketamine changes in right NAc entropy (r=0.72, p=0.019).

Figure 4.

Post-ketamine percentage changes in pGSK3β/GSK3β versus baseline-to-post-ketamine changes in Children’s Depression Rating Scale–Revised (CDRS-R) ratings and right nucleus accumbens (rNAc) log(entropy). Responders (n=5, blue) had a significantly greater percentage change in post-ketamine pGSK3β/GSK3β compared with non-responders (n=5, red) (a). Post-ketamine percentage changes in pGSK3β/GSK3β strongly correlated with baseline-to-post-ketamine changes in CDRS-R ratings (b) and baseline-to-post-ketamine changes in rNAc log(entropy) (c), respectively.

Error bars show the standard error of the mean.

Discussion

This is the first study to report on neurobiological correlates of treatment response to ketamine in adolescents with TRD. In these adolescents, clinical improvements after the ketamine infusions were accompanied by baseline-to-post increases in NAc rs-fMRI entropy and greater post-treatment insulin signaling (pmTOR/mTOR and pGSK3β/GSK3β). Baseline-to-post NAc entropy changes were also associated with post-ketamine upregulation of insulin-stimulated pmTOR/mTOR and pGSK3β/GSK3β levels in PBMCs. This suggests that cellular neurotrophic responses to insulin post-ketamine may have a mechanistic link with the associated effects on NAc entropy. Importantly, we did not observe generalized ketamine-induced changes in rs-fMRI entropy and insulin-stimulated pmTOR/mTOR and pGSK3β/GSK3β expression; these findings were specific to responders. Collectively, these data provide preliminary evidence suggesting that enhancing neural flexibility may be critical for clinical improvement in the context of ketamine treatment in adolescents.

Entropy has emerged as a promising measure for quantifying different brain states based on the variability of neural activity (Carhart-Harris, 2018; Carhart-Harris et al., 2014) and as a metric of neuroplasticity (Foz et al., 2002; Tecchio et al., 2006), across a number of methods. It has been proposed that entropy can reflect the flexibility of a state (Shi et al., 2020), and that while depression is associated with lower entropy, certain psychoactive drugs can increase brain entropy (Carhart-Harris et al., 2014). A recent MEG study found that ketamine, LSD and psilocybin all increased resting-state entropy measures at doses conferring psychoactive effects, suggesting that these dissociative drugs can generally increase the complexity of neural activity (Schartner et al., 2017). These previous studies of entropy focused on real-time changes in brain activity. In contrast, our work investigated sustained changes in entropy, given that the post-treatment assessment was one day after the final ketamine infusion of the six infusions delivered over the course of two weeks. We found that adolescent responders to ketamine showed increased signal complexity in the NAc following the intervention. This could signify enhanced neural flexibility, allowing the adolescent to break out of rigid and pathological signaling pathways. Although depression research has previously identified neural networks that become entrenched in pathological feedback loops that contribute to ongoing problems with mood and behavior (Hamilton et al., 2015), our results suggest that in adolescents with TRD, a clinical improvement in the context of ketamine treatment may result from enhanced neural flexibility in key limbic regions, allowing for a therapeutic shift in the behavior of these entrenched neural networks.

In our study, entropy change associated with clinical improvement was particularly noted in the NAc, a subcortical brain region which serves as a nexus point for pathways mediating emotion, cognition and motor function (Nauczyciel et al., 2013). Reduced activation in this region has been associated with severe depression and has been thought to underlie dysfunction in the reward system (Pizzagalli et al., 2009). Furthermore, dendritic atrophy in medium spiny neurons of the NAc is known to mediate stress-induced depression behaviors in animal models (Francis et al., 2017). Thus, recent efforts to treat severe depression in adults using deep brain stimulation have targeted the NAc as well as the ventral striatum (Bewernick et al., 2010, 2012). Numerous studies have also associated the antidepressive effects of ketamine with modulation of NAc activity (Abdallah et al., 2017). Studies in rats have found that ketamine enhances high frequency oscillations (130–180 Hz) and reduces gamma band activity (30–90 Hz) within the NAc (Hunt et al., 2006); these changes could potentially relate to the increased local rs-fMRI entropy that we observed in our group of adolescent responders. Collectively, our findings suggest that ketamine’s effects on NAc entropy could be a critical step for facilitating neural flexibility and treatment response in adolescents with TRD.

Finally, based on prior research suggesting that signal transduction along the mTOR pathway mediates ketamine’s antidepressant effects (Welberg, 2010), we examined ketamine-associated changes in insulin-stimulated pmTOR/mTOR and pGSK3β/GSK3β expression. We found evidence for increased pmTOR/mTOR and pGSK3β/GSK3β as markers of clinical response in this small sample. Furthermore, greater post-treatment pmTOR/mTOR and pGSK3β/GSK3β levels were associated with increased NAc entropy. These findings add to previous work suggesting that mTOR/GSK3β signaling is critical for ketamine’s effects on neurotrophic morphology changes (Li et al., 2010). Taken together, our results suggest that upregulation of insulin-stimulated mTOR/GSK3β signaling may be an important mechanistic marker of neural flexibility that occurs in concert with increased NAc entropy in the process of adolescents with TRD responding to ketamine. Moreover, it underscores the established role for cellular responses to insulin that could be engaged through adjunctive treatment options (Nguyen et al., 2018; Rasgon and McEwen, 2016; Watson et al., 2018). Indeed, there is a growing literature highlighting the important role of insulin signaling, and the enhancement thereof in antidepressant treatment response (Nguyen et al., 2018; Rasgon and McEwen, 2016; Watson et al., 2018). While it is conceivable that ketamine’s promotion of the neurotrophic effects of insulin within the brain may contribute to the increased entropy observed in responders, further mechanistic studies are required to demonstrate a causal functional relationship.

Conclusions of this study should be considered in light of several study limitations. First, the sample is small, raising the risk for spurious or non-generalizable results, and limiting our power to detect all of the links examined here (Button et al., 2013). The results of this study should be considered hypothesis-generating rather than hypothesis-confirming. Second, this was an open-label design with no placebo control. While our study focused on correlates related to clinical improvement versus non-improvement, the changes observed in these adolescents with TRD cannot be said to be specific to ketamine. The ketamine-associated changes described could be attributable more generally to clinical improvement. These changes might also have been observed in response to placebo or possibly associated with spontaneous remission of depression. Third, the non-uniformity in dose strategy in this study (where those that received the actual body weight dosing strategy had a greater rate of responders (5 of 8) versus those that received the ideal body weight dosing strategy (0 of 5)) (Cullen et al., 2018) introduced an additional source of variability. Future studies are needed to more systematically examine dose–response relationships with respect to ketamine’s (versus placebo’s) effects on both clinical and biological outcomes. Fourth, while in this study we opted to examine neural changes 24 h after the last infusion in an attempt to coincide biological assessment with peak clinical response, additional data collection in real time would help to paint a more complete picture of ketamine’s short- and long-term effects in adolescents. Fifth, although the current study assessed changes in rs-fMRI entropy, the relatively low temporal resolution of fMRI does not fully capture the high frequency of neural signaling that can be captured by EEG. Additionally, low frequency signals have inherently less entropy than high frequency signals. Future studies should consider a multimodal imaging approach, potentially adding EEG to the intervention protocol to leverage its high temporal resolution for obtaining real-time entropy measures during each administration of ketamine. Sixth, rs-fMRI entropy can be affected by other pharmacological or behavioral factors, such as caffeine (Chang et al., 2018) and fatigue (Shan et al., 2018), so future work will be needed to assess its potential as a robust trait marker of treatment response in the context of TRD. Seventh, given the relatively high noise of the CDRS measure, it is possible that the observed associations between depression score changes and mTOR measures were confounded by individual changes in sleep and activity patterns. To address this, we conducted correlation analyses which revealed a trend-level relationship between sleep/activity levels and mTOR measures. However, this trend did not fully account for the observed changes in depression scores. Regardless, due to the low statistical power of this study, this null result cannot definitively reject the hypothesis that sleep/activity measures may be confounding the conclusions drawn from our mTOR analyses. Finally, while the insulin signaling pathway may provide a link to cellular energetics and neurotrophic responses limiting functional adaptability, as supported by our group’s recent study showing that peripheral and central markers of insulin signaling (mTOR and GSK3) are directly correlated with antidepressant response in a rodent model of treatment resistance (Walker et al., 2019), future mechanistic studies are needed to demonstrate this functional link.

Conclusion

There is an urgent need for understanding mechanisms of TRD and developing novel treatments to address it in adolescents. Our preliminary findings suggest that measures of central and peripheral neural flexibility (including increased rs-fMRI entropy, particularly in the NAc which was associated with increased mTOR and GSK3β signaling) represent neural correlates of clinical response in adolescents with TRD. These preliminary findings suggest that rs-fMRI entropy and insulin-mediated mTOR/GSK3β signaling could represent candidate neurobiological targets in future research investigating optimization strategies designed to enhance neural flexibility and plasticity in adolescents with TRD. It is possible that these physiological changes are due to ketamine treatment but this conclusion would need to be confirmed with subsequent studies.