Anticipatory human subthalamic area beta-band power responses to dissociable tastes correlate with weight gain.

The availability of enticing sweet, fatty tastes is prevalent in the modern diet and contribute to overeating and obesity. In animal models, the subthalamic area plays a role in mediating appetitive and consummatory feeding behaviors, however, its role in human feeding is unknown. We used intraoperative, subthalamic field potential recordings while participants (n = 5) engaged in a task designed to provoke responses of taste anticipation and receipt. Decreased subthalamic beta-band (15-30 Hz) power responses were observed for both sweet-fat and neutral tastes. Anticipatory responses to taste-neutral cues started with an immediate decrease in beta-band power from baseline followed by an early beta-band rebound above baseline. On the contrary, anticipatory responses to sweet-fat were characterized by a greater and sustained decrease in beta-band power. These activity patterns were topographically specific to the subthalamic nucleus and substantia nigra. Further, a neural network trained on this beta-band power signal accurately predicted (AUC ≥ 74%) single trials corresponding to either taste. Finally, the magnitude of the beta-band rebound for a neutral taste was associated with increased body mass index after starting deep brain stimulation therapy. We provide preliminary evidence of discriminatory taste encoding within the subthalamic area associated with control mechanisms that mediate appetitive and consummatory behaviors.

Introduction

The subthalamic area, including the subthalamic nucleus (STN), substantia nigra (SN), and zona incerta (ZI), is classically considered part of the basal ganglia involved in motor processing including locomotion, reaching, and orofacial movement (Nambu et al., 1996; Quinn et al., 2015; Baunez et al., 2002). Moreover, the subthalamic area is well-positioned to integrate and regulate not only motor but also motivation and cognitive processing needed for coordinated responses to incentives, and even food to maintain energy homeostasis (Lardeux et al., 2009; Espinosa-Parrilla et al., 2013; Weintraub and Zaghloul, 2013; Stice et al., 2008a). How this region mediates appetitive and consummatory feeding behavior in humans remains unknown, but animal studies have suggested topographically heterogenous roles in feeding within the subthalamic area (Baunez et al., 2002; Lardeux et al., 2009; Zhang and van den Pol, 2017). Specifically, STN neuronal firing rates responded to food-related cues and corresponded to food preference (Lardeux et al., 2009; Espinosa-Parrilla et al., 2013). Further, ZI GABAergic neurons were reported to exhibit increased activity with food deprivation and administration of ghrelin, and driving these neurons increased sweet, high-fat (sweet-fat) food interaction and consumption and increased weight gain (Zhang and van den Pol, 2017).

While electrophysiological data from humans are lacking, there have been variable reports of increased body weight and consummatory feeding behaviors in a subset of patients following subthalamic area deep brain stimulation (DBS) (Serranova et al., 2011; Serranova et al., 2013; Aiello et al., 2017). While these effects are likely to be multifactorial, they suggest, in line with preclinical findings, that a physiologic control signal exists in the subthalamic area that is disrupted by DBS (and lesions in rodents) and may explain the observed variability in weight gain (Baunez et al., 2002; Baunez et al., 2005; Uslaner et al., 2008). Examining this focal area in awake feeding human subjects is made possible by collecting electrophysiological recording data during DBS surgery (Zaghloul et al., 2009).

The sweet-fat macronutrient combination has been reported to provoke feeding beyond homeostatic needs across species (Zhang and van den Pol, 2017; Drewnowski, 1997). We hypothesized that responses to a sweet-fat solution exist in topographically-specific regions of the subthalamic area and used a validated taste-incentive task paradigm to isolate anticipation and receipt phases of feeding (Stice et al., 2008b). Further, we hypothesized that the characteristics of these responses may correlate with subject-specific behavioral outcomes represented by postoperative body mass index (BMI) (Balestrino et al., 2017; Foubert- Samier et al., 2012). To our knowledge, this is the first study to identify discriminatory and topographically organized taste encoding within the human subthalamic area.

Methods and materials

Intraoperative cohort

We recruited patients with Parkinson’s disease (PD) undergoing clinically indicated DBS surgery. The study was approved by the Stanford University Institutional Review Board (IRB#33146) and all participants provided written informed consent.

Lead implantation

Participants were instructed to stop long-acting and short-acting dopaminergic medications over 24 and 12 h prior to surgery, respectively. On the day of surgery, we carried out computer-assisted, image-guided stereotactic placement of the DBS electrodes into the subthalamic nucleus via a frameless approach described previously (Quinn et al., 2015). Pre-operatively, stereotactic coordinates for STN were used (12 mm lateral, 3 mm posterior and 4 mm inferior to the midpoint of the AC–PC line) and optimized with direct targeting using a T2-sequence (Benabid et al., 1996; Hamid et al., 2005). Conscious sedation during operative opening (including trephination) and closing was achieved using dexmedetomidine and propofol to minimize neurocognitive effects. All agents were turned off at least 20 min prior to microelectrode recordings to ensure the patient was appropriately conscious during the single pass of microelectrode recordings (Neuroprobe Sonus-Shielded Tungsten STR-009080; Alpha Omega, Inc.) and sensorimotor testing that preceded DBS lead implantation and task-based recordings. Once identified, the microelectrode was replaced by an 8-contact DBS lead (Boston Scientific DB-2201, N = 4, or DB-2202, N = 1, Marlborough, MA; USA). After implantation of the left hemispheric lead, field potentials were recorded from the left hemisphere as the participant completed the task described below. Following task completion, intra-operative test stimulation was used to confirm therapeutic response at target, and the leads was permanently fixed.

Sweet-fat incentive task paradigm

Participants completed a task that elicits neural responses during anticipatory and receipt of a sweet-fat solution (Stice et al., 2008b; Bohon and Stice, 2011; Bohon, 2017; Stice et al., 2010). Prior to each session, a sweet-fat solution (McDonalds chocolate milkshake) and taste-neutral solution were prepared and stored in a refrigerator as described previously (Bohon and Stice, 2011).

This task lasted approximately 13 min and was presented on an LCD monitor. The task consisted of 80 trials divided, with each divided into 5 screens (Fig. 1A top). For each trial, participants were instructed to fixate to a point on the screen (2 s), cued (1 s) with an image of a sweet-fat (glass of chocolate milkshake; 50% of trials) or a taste-neutral (glass of water), and instructed to again fixate to a point on the screen as they anticipated receipt of the cued solution (2 s). The order of stimulus presentation was randomized across participants. Solutions were loaded into respective, 60 ml syringes and attached to a MATLAB-programmable dual syringe pump (Braintree Scientific BS-8000 Dual) to ensure consistent volume, rate, and timing of taste delivery. Syringes were connected via Tygon tubing to a custom manifold fitted into the participants’ mouths to provide solution delivery to a consistent segment of the tongue (Stice et al., 2008b). To maintain attention, participants were directed to press a keypad button using their dominant hand, immediately after cue presentation (Stice et al., 2008b). Next, a “Delivering taste” message (2 s) followed by a fixation screen (1 s) appeared, during which a 0.5 cc solution of the cued taste was administered to the participant. Finally, a “Swallow” message (2 s + 0.5–1 s jitter) directed the participant to swallow and prepare for the next trial. Subjective ratings of solution palatability using a 10-point Likert scale and solution preference (sweet-fat vs taste-neutral) were collected following a practice session immediately prior to surgery.

Fig. 1.

Baseline beta-band power and beta-band responses to the sweet-fat incentive task predominate over power changes in other frequency-band ranges (A top) Schematic of a single trial in the sweet-fat incentive task. After fixation (baseline), participants were cued with an image of a sweet-fat solution (chocolate milkshake; 50% of trials) or a taste-neutral solution (water; anticipatory phase), and subsequently received a taste of the cued solution (receipt phase) for a total of 80 trials. (A bottom) Time-frequency (TF) spectrogram from the all 40 recorded channels showing a focus of time-locked activity changes in the beta-band (15–30 Hz) frequency range. Onset times for the cue and solution delivery (soln delivery) marked by vertical dashed lines. (B) Frequency-band power during anticipation (top) and receipt (bottom) of sweet-fat and taste-neutral solutions and during the pre-stimulus baseline. Frequency bands analyzed along x-axis include delta (δ), theta (θ), alpha (α), beta (β), low gamma (γlow), and high gamma (γhigh). (A inset) Comparison of baseline power at each frequency band. At baseline, beta-band power was prominent in the subthalamic area (FDR-adjusted p < 0.05). Further, during both anticipation and receipt, changes in frequency-band power are significant only in the beta-band frequency range in response to both sweet-fat and taste-neutral stimuli. *FDR-adjusted p < 0.05 on one-way ANOVA.

DBS lead localization and subcortical atlas

Pre-operative T1-weighted MRI and post-operative CT images obtained after DBS lead placement were linearly registered using Advanced Normalization Tools (ANTs). DBS contacts were then reconstructed in each participant’s native space according to electrode artifacts from the co-registered post-operative CT images. ANTs was used to perform a two-step linear and non-linear registration between each participant’s native T1 and the standard MNI152 space. The non-linear registration was optimized for the subcortical regions for improved registration accuracy. The optimized local registration was achieved by using a dilated subcortical mask to constrain the cost function. Regions of interest (ROIs) were extracted from the MNI-defined DBS Intrinsic Template Atlas (DISTAL), which contains 101 thalamic and extra-thalamic nuclei including the STN, the SN, and the ZI (Chakravarty et al., 2006; Ewert et al., 2017). The ROIs were registered onto the participant’s native brain T1-space and overlaid to the reconstructed DBS contacts.

For visualization purposes, images were loaded into MATLAB 2016b (MathWorks, Inc., Natick, MA; USA). Using the Lead-DBS toolbox (v2), these imaging protocols were normalized into standard MNI152 space (point X = 0, Y = 0, Z = 0 at the anterior commissure), co-registered, and the results of which were confirmed manually (Horn et al., 2019). DBS contacts were automatically pre-reconstructed using the phantom-validated and fully-automated PaCER method and results were manually confirmed in 3DSlicer (v4) (Husch et al., 2018; Fedorov et al., 2012). The reconstructed electrode trajectory was merged with ROIs as defined in DISTAL.

DBS channel data acquisition and field potential processing

DBS recordings (1375 Hz sampling rate) were amplified and stored for offline analysis using a NeuroOmega recording system (AlphaOmega, Alpharetta, GA; USA). Data were analyzed in MATLAB using the FieldTrip toolbox (v2.3) (Oostenveld et al., 2011). Line noise was removed using a zero-phase, band-stop filter at 60 Hz harmonics. The field potential signal was extracted using a 300 Hz, 6th order, two-pass, zero-phase, Butterworth IIR low-pass filter. Re-referencing was carried out using a Laplacian scheme on sequential channels to maximize the local population-level signal and to keep all contacts as sources (Li et al., 2018; Hindriks et al., 2016). In the participant with a segmented lead (“1–3–3–1” scheme), both segmented rings were referenced against the adjacent unsegmented ring and the corresponding segment of the adjacent segmented ring. Both unsegmented rings were referenced against the average of all segments in the adjacent segmented ring. The data were then epoched to task cue onset from − 2 to 6 s (s) with onsets of cue and solution delivery at 0 s and 3 s; respectively. Time-frequency (TF) analysis was carried out with a multi-taper convolution method using discrete prolate spheroidal sequences as tapers with power extracted by squaring the magnitude of the complex Fourier-spectra. The TF spectrogram was log transformed, and z-scored against the pre-cue baseline (− 0.6 to − 0.1 s) within each channel. For frequency-band-specific power analyses, the TF spectrograms were averaged in the frequency domain by division into 6 discrete frequency-bands: delta (1–4 Hz), theta (5–7 Hz), alpha (8–14 Hz), beta (15–30 Hz), low gamma (35–70 Hz), and high gamma (75–150 Hz). To determine frequency bands of interest, the power signal in each frequency-band was averaged during baseline, anticipatory, and receipt time periods for both tastes. Responsive frequency bands were defined as those in which the magnitude of power during anticipation or receipt differed from baseline power. Significance was determined using one-way ANOVA with p-values corrected using the false discovery rate method (FDR-adjusted) at an alpha threshold of 0.05.

Definition of electrode responsivity as an electrode inclusion criteria

A channel was classified as cue- or receipt- responsive if band power significantly differed for either taste during anticipation (0 to 3 s) or receipt periods (3 to 6 s), respectively, compared to baseline (−0.1 to −0.6 s). A channel was classified as cue- or receipt-specific if band power significantly differed between tastes during anticipation or receipt, respectively. All 6 frequency-band ranges were examined. Statistical significance was calculated using a cluster-based permutation approach (Maris and Oostenveld, 2007). Briefly, channel-time cluster significance was quantified by the maximum sum of an independent-sample t-test value with a cluster alpha-threshold of 0.05. The significance probability was then calculated using a Monte-Carlo estimate (resampling statistic) with 500 random permutations using the cluster-based statistic and a critical alpha threshold of 0.05. Observations for unpaired permutation testing were taken from trials at each channel of interest per participant, independently.

Neural network modeling

TF spectrograms were used to train and test an artificial neural network (ANN) model with a single-hidden layer designed in JMP Pro v14 (SAS Institute Inc., Cary, NC; USA). The model was designed with nodes using both hyperbolic tangent (TanH) and Gaussian activation functions. The number of nodes under each function was selected from preliminary analysis evaluating the minimum root-mean-square error of models with 1 to 20 TanH and 1 to 10 Gaussian function nodes (Spuler et al., 2015). Model observations were generated from the trials of cue- and receipt-responsive channels across participants. The features fed into the model consisted of power in responsive frequency-bands and of time, with anticipatory and receipt periods divided into 6, 500 ms time windows. ANN classification was performed separately for anticipation and for receipt. Observations were obtained from trials at each channel per participant and semi-randomly divided into independent training (70%), validation (20%), and testing (10%) subsets. Semi-random partitioning of observations was performed to ensure that trials within each participant were evenly divided between the 3 subsets. To minimize overfitting, the validation set was used to apply an optimal penalty on model parameters using the absolute penalty method (Tibshirani, 1996). In addition, after training was complete, we performed a final, independent assessment of the model’s performance with observations not used in training or validation (i.e. the testing subset). Model performance was assessed using receiver operant curve analysis to calculate the area under the curve (AUC) as an outcome measure in addition to the overall accuracy, and the true and false positive rates. Statistical significance was determined by permutation testing with a critical alpha threshold of 0.05. Features were assessed for their contribution to the model’s total effect using JMPs ‘Assess Variable Importance’ function within the prediction profiler toolbox (Saltelli, 2002). Briefly, feature importance is measured by varying feature value sampling with the Monte Carlo method to estimate impact on response variability of the feature alone and in combination with other features (total effect). Features with the largest contribution to the effect size were defined as having a total effect above the 95% confidence interval (95% CI) of the mean distribution of all sampled effects across features and permutations.

Body-mass index and band power association

In all 5 participants, available BMI measures were collected from pre- and post-operative clinic visits. Using BMI measures, participants were stratified into a BMI gain (vs no BMI gain) group if the average post-operative BMI was significantly greater than pre-operative BMI. Statistical significance was determined by paired t-testing. Then using band power measures sampled from all 40 trials per condition of all 8 channels per participant, we examined if participants could independently be stratified into the same two groups. To quantify the normalized band power distributions during the time segments of interest, mean values and 95% CIs were derived utilizing bootstrapping (random sampling with replacement, using 1000 repetitions). Statistical significance was classified as average power above the 95% CI.

Availability of materials and data

The data that support the findings of this study are available from the corresponding author, upon reasonable request.

Results

Five participants (average age of 67.6 years, 2 females, all right-handed) undergoing STN DBS surgery performed the sweet-fat incentive task (Table 1). All participants underwent DBS implantation with task-related electrophysiological recordings from the subthalamic area. All participants preferred the sweet-fat over the taste-neutral solution with the average rating for the sweet-fat solution being 7.4 (range 5 to 10) on a 10-point Likert rating scale. The median preoperative BMI was 29.9 kg/m2 (range: 15.3 to 36.8 kg/m2).

Table 1

Intraoperative cohort participant characteristics.

ID	Age (years)	Sex	Disease duration (years)	Clinical phenotype	Pre-operative UPDRS	Sweet-fat rating	Preferred sweet-fat	Preoperative BMI
P1	61	F	8	AR	31	7/10	Yes	28.2
P2	71	M	5	TR	35	6/10	Yes	29.3
P3	66	M	14	AR	40	10/10	Yes	36.8
P4	75	M	13	TR	66	9/10	Yes	34.9
P5	65	F	8	TR	26	5/10	Yes	15.5

All participants were right-hand dominant. F = female, M = male, AR = akinetic-rigid, TR = tremor-dominant.

Beta-band power characterizes subthalamic area responses to sweet-fat vs taste-neutral tastes

Compared to other measured frequency bands, beta-band (15–30 Hz) power predominated in the subthalamic area (N = 40 channels in 5 participants; Fig. 1A, FDR-adjusted p < 0.05; Fig. 1B top inset; one-way ANOVA, FDR-adjusted p < 0.05) and on average, decreased significantly in response to both tastes during anticipation and receipt (Fig. 1B–C, one-way ANOVA, FDR-adjusted p < 0.05). Of the 40 channels, 17 (42.5%) demonstrated cue-responsive and 14 (35%) demonstrated cue-specific beta-band power changes in4 (80%) participants during anticipation (Fig. S1A; Fig. 2A). Of the 40 channels, 7 (17.5%) demonstrated receipt-responsive and 4 (10%) demonstrated receipt-specific beta-band changes in 3 (60%) participants. Thus, further analysis was focused on the beta-band range.

Fig. 2.

The sweet-fat incentive task elicits beta-band power activity that characterizes topographically heterogenous responses to sweet-fat and taste-neutral stimuli in the subthalamic area. (A) Reconstructed DBS lead trajectories of all participants (colored lines) within the subthalamic area – substantia nigra (SN), subthalamic nucleus (STN), zona incerta (ZI) – showing anticipation-(top) and receipt- (bottom) specific (black outline), responsive (green filled spheres), and nonresponsive (grey spheres) contacts. (Inset top) Representative image of localization and medial view orientation of depicted reconstruction within the left hemisphere. Onset times for the cue and solution delivery (soln delivery) marked by vertical dashed lines. (B) Beta-band power tracing (top) from all responsive channels demonstrating decreased beta-power to both solutions in 0 to 1 s after cue, with greater decreased magnitude for sweet-fat over taste-neutral cues (cluster-based permutation testing, p < 0.05). During the subsequent 1 to 3 s after cue, beta-band is discriminatory as it remains decreased for sweet-fat, but not taste-neutral cues (cluster-based permutation testing, p < 0.05). During receipt, significant discriminatory signal changes are observed between 5 and 6 s after cue (cluster-based permutation testing, p < 0.05). (B bottom) Beta-band responses to sweet-fat (left) and taste-neutral (right) stimuli averaged over 500 ms time windows. Anticipation of sweet-fat stimuli is characterized by a sustained decreased power response while the taste-neutral anticipatory response is shortened by a beta-band rebound (i.e. switch from decreased to increased power). (C) Beta-band power tracings averaged across responsive channels within 3 subregions of the subthalamic area. During anticipation, the STN (top) and the SN (middle) were characterized by an early decrease in beta-band power to both solution cues that then became discriminatory (cluster-based permutation testing, p < 0.05). In the ZI (bottom), both decreased (Type 1, left) and increased (Type 2, right) power responses were observed during anticipation. However, decreased power responses were specific for sweet-fat and not taste-neutral cues. During receipt, the STN had early responses followed by those in the SN. No receipt responses were captured in the ZI. *p < 0.05 on cluster-based permutation testing.

Cluster-based permutation testing among cue-responsive channels (N = 17 channels in 4 participants) revealed decreased beta-band power0.5 to1s after both taste cues, however, sweet-fat cues elicited a greater magnitude response (Fig. 2B top; cluster-based permutation, p < 0.05). To further characterize the signal’s temporal dynamics, beta-band power was divided and averaged into 500 ms time windows (from 0.5 to 1 s – i.e. T1 – up to 5.5 to 6 s – i.e. T12 after cue; Fig. 2B bottom). Following the initial decrease, beta-band power increased earlier for taste-neutral cues (beta-band rebound) 1.5 s after cue (T4) while remaining decreased for sweet-fat cues throughout anticipation (cluster-based permutation testing, p < 0.05). During receipt, decreased beta-band power was taste-neutral specific and occurred within 1 safter solution delivery (cluster-based permutation testing, p < 0.05). Beta-band rebounds were observed 1 s after delivery of both tastes.

Participant button-press reactions times to cue onset did not significantly differ between sweet-fat (median: 0.73 s, interquartile range: 0.37 s) and taste-neutral conditions (0.71 s, interquartile range: 0.58 s; pooled t-test, p = 0.55). With trials time-locked to the button-press (instead of taste cue) onset, similar patterns of beta-band power decrease and rebound were observed (Fig. S2A–B; cluster-based permutation testing, p < 0.05). However, the overall magnitude of the beta-band power decrease during the first second was greater when trials were time-locked to the taste cue vs the button press events. Further, during the beta-band rebound period, rebound magnitude was significant greater for the neutral taste, and beta-band power was significantly more suppressed for the sweet-fat taste when trials were time-locked to the taste cue vs the button press event (Fig. S2C–D; FDR-adjusted p < 0.05).

The STN, SN, and ZI exhibit topographically heterogenous beta-band responses

During both anticipation and receipt, most responsive channels observed were situated in the ventroposterior aspect of the recorded subthalamic area corresponding to the STN (Fig. S1B–C; one-way ANOVA, FDR-adjusted p < 0.05). During anticipation in the STN (N = 10 channels), an early, non-discriminatory, beta-band power decrease was observed (Fig. 2C top; T2: 0.5 to 1 s; cluster-based permutation, p < 0.05), and followed by a taste-neutral specific beta-band rebound (Fig. 2C top; T4: 1.5–2 s; cluster-based permutation, p < 0.05). A similar pattern of responses was observed in the SN (N = 3 channels; Fig. 2C middle; cluster-based permutation, p < 0.05). During receipt, the STN and SN were characterized an early (T8: 3.5 to 4 s) and late (T11-T12: 5 to 6 s) taste-neutral-specific beta-band power decreases (cluster-based permutation, p < 0.05).

The changes in beta-band power in the ZI (N = 4 channels) differed from those in the STN and SN (Fig. 2C bottom). First, while the STN and SN exhibited exclusively decreased beta-band power during anticipation, the ZI displayed both increased (N = 2) and decreased (N = 2) power responses. The decreased power responses were specific to sweet- fat cues (cluster-based permutation, p < 0.05) and situated more dorso-laterally in the ZI (Fig. S1D, one-way ANOVA, FDR-adjusted p < 0.05). The increased beta-band power cue-response changes were non-discriminatory. No receipt-responsive or –specific changes were measured in either region of the recorded ZI.

Trial-level subthalamic activity predicts anticipated and received tastes

To further characterize the observed patterns of beta-band activation, we utilized ANN for non-linear predictive modeling (Fig. 3A). Observations were selected from trials of the 15 cue-responsive channels characterized by decreased beta-power in the STN, SN, and ZI, excluding the 2 channels with increased beta-power responses. The ANN classified sweet-fat vs taste-neutral trials with an AUC = 0.75 (confidence interval–CI: 0.74–0.75) during anticipatory (Fig. 3B left) and an AUC = 0.73 (95% CI: 0.72–0.74) during receipt periods (Fig. 4C left). During anticipation, the overall accuracy was 69% with false-positive rates of 28% and 34% for sweet-fat and taste-neutral solutions, respectively. During receipt, the overall accuracy was 67% with false-positive rates of 32% and 33% for sweet-fat and taste-neutral solutions, respectively. Of the features used during anticipation, the beta-band rebound period (T4: 1.5 to 2 s after cue-onset) had the largest contribution to the observed total effect (p < 0.05; Fig. 3B right). During receipt, the early (T8: 3.5 to 4 s) and late (T12: 4.5 to 5 s) taste-neutral -specific response periods had the largest contributions to the observed total effect (p <0.05; Fig. 4C right).

Fig. 3.

Subthalamic area beta-band response is sufficient in classifying trials belonging to sweet-fat vs taste-neutral conditions. (A) Schematic of the artificial neural network (ANN) design. Observations were formed from each trial of the 15 and 7 channels across participants (P1, P4, P5) with decreased beta-band responses with decreased beta-band power responses during anticipation and receipt, respectively. The 2 channels in P2 with increased beta-band anticipatory responses were excluded. The 6 model features consisted of beta-band power divided into 6, 500 ms time segments (T1-T6) for both anticipation and receipt. (B left) Performance of the trained ANN in classifying sweet-fat vs taste-neutral trials (in testing subset independent from data used to train the classifier) during the anticipatory phase. Classifier performance had an overall accuracy of 69% (95% CI: 68–70%) with overall true-positive rates (TPR) of 66% and 72% and false-positive rates 28% and 34% for sweet-fat and taste-neutral stimuli. (B right) Feature importance in model performance during anticipation as measured by main total effect (i.e. relative contribution of each factor alone and in combination). Beta-band responses during 1.5 to 2 s (T4) after cue had the largest impact on model performance (p < 0.05). (C left) Performance of the trained ANN in classifying sweet-fat vs taste-neutral trials (in testing subset independent from data used to train the classifier) during the receipt phase. Classifier performance had an overall accuracy of 67% (95% CI: 66–78%) with overall TPR of 67% and 68% and false-positive rates 32% and 33% for sweet-fat and taste-neutral stimuli. (C right) Feature importance in model performance during receipt as measured by main total effect (i.e. relative contribution of each factor alone and in combination). Beta-band responses during soln delivery at 3.5 to 4 s (T8) and 6.5 to 6 s (T12) had the largest impact on model performance (p < 0.05). *p < 0.05 using the 95% confidence interval (CI) of calculated means across time-segments with 100 permutations.

Fig. 4.

Subthalamic area beta-band rebound to taste-neutral food is associated with weight gain. (A) Averaged pre- and post-operative BMIs for each subject. Measures were taken from 3 distinct pre-operative clinic visits within 6 months of implantation, and variable number of at least 2 clinic visits to 15-months post-operatively. P1 and P5 had significant BMI gain post-operatively (BMI gain group; FDR-adjusted p-value <0.05). (B) Overall BMI change from day of DBS implantation. In P1 and P5, there was a significant trend toward increased BMI (R2 = 0.94, F(1,7) = 114, p < 0.0001, BMI Gain group). The remaining participants did not show a consistent post-operative BMI trend (R2 = 0.01, F(1,10) = 0.11, p = 0.75, No BMI Gain group). (C) Beta-band responses to taste-neutral stimuli during anticipatory (T4: 1.5 to 2 s) and receipt periods (T2: 3.5 to 4 s and T6: 5.5 to 6 s) that discriminated solutions averaged across all 8 contacts in each participant (P1- P5). Beta-band response to taste-neutral stimuli, in each period, was significantly different between participants that had an increase vs no increase in BMI (p < 0.05). (D) Beta-band responses to sweet-fat stimuli during anticipatory and receipt periods that discriminated solutions average across contacts in each participant (P1-P5). No beta-band response threshold to sweet-fat stimuli discriminated between participants that had an increase or no increase in body-mass index (BMI). *p < 0.05 using the 95% confidence interval (CI) of calculated means across time-segments with 100 permutations.

Increased post-operative BMI is associated with beta-band responses

Two participants (P1 and P5) had significantly increased BMI in the months post-operatively compared to a pre-operative BMI baseline (BMI gain group; Fig. 4A–B: FDR-adjusted p-value<0.05 on one-way ANOVA; Fig. 4B: R2 = 0.94, F(1,7) = 114, p < 0.0001). The remaining 3 participants (P2–4) had no statistically significant change in BMI several months post-operatively (no BMI gain; R2 = 0.01, F(1,10) = 0.11, p = 0.75). Averaged beta-band power from all channels (N = 40 channels in 5 participants) revealed significantly larger response magnitudes for neutral tastes during all the 3 discriminatory time periods (T4, T8, and T12) in the BMI gain vs no BMI gain group (Fig. 4 C, p < 0.05 bootstrap test). Beta-band responses for sweet-fat during these time periods did not stratify participants into these groups (Fig. 4D). Both participants that gained weight had chronically-active contacts which were task responsive (Fig. 2A).

Discussion

Our everyday environments are filled with sensory cues that modulate neural activity and influence behavior. The subthalamic area is a major node in this process that is thought to mediate inhibitory control, incentive salience, and goal-oriented behavior (Serranova et al., 2011; Frank, 2006). To our knowledge, we demonstrate, for the first time using intracranial electrophysiological recordings, that the human subthalamic area responds to taste incentives during periods of anticipation and receipt. Specifically, beta-band power changes in the STN, SN, and ZI, respond to and can accurately discriminate (AUC ≥ 74%) tastes during both anticipatory and receipt. Finally, subthalamic beta-band power responses to neutral taste can stratify individuals that gain vs do not gain BMI following DBS. Together, these preliminary findings provide experimental and clinical evidence that the subthalamic area is not only involved in motor planning but also in taste anticipation and receipt.

In this study, beta-band power was prominent, compared to a wide spectrum of frequency-band ranges within the subthalamic area (Kuhn et al., 2004; Levy et al., 2000). This aligns with prior literature associating subthalamic beta-band power with motor processing and control (Jenkinson and Brown, 2011; Williams et al., 2005; Wingeier et al., 2006). Further, these studies demonstrate that beta-band rebounds (i.e. switch from decreased to increased power) may function to dynamically adapt motor performance and behavior in the context of imagined, observed, or performed movements (Kuhn et al., 2004; Marceglia et al., 2009; Kuhn et al., 2006; Torrecillos et al., 2018). In go/no-go task paradigms, movement during ‘go’ trials (i.e. favored movement initiation) was characterized by a greater decrease in beta-band power compared to ‘no-go’ trials (i.e. suppressed movement initiation) where beta-band rebounds interrupted the decreased response (Kuhn et al., 2004; Marmor et al., 2020; Alegre et al., 2013). In the few studies describing subthalamic beta-band responses to reward, greater beta-band power decreases have been associated with increased size of monetary rewards, and with increased magnitude of received reinforcements (Duprez et al., 2019; Schroll et al., 2018). With regard to taste-motivated activity in the human subthalamic area, prior studies have been limited by functional MRI which offers poor temporal resolution that does not match up to behavioral timescales, provides indirect measures of local neuronal population activity, and has a low specificity for small subcortical nuclei g (Branch et al., 2013; Thanarajah et al., 2019; de Hollander et al., 2017). Therefore, our findings, in conjunction with the previously described reward studies, implicate decreased beta-band power with motivational reinforcement for sweet-fat tastes and the beta-band rebound response with inhibitory control toward the less preferred neutral taste for the first time. However, the results of this study are preliminary and further research with a broader range of foods and dietary tastes is needed to explore the inhibitory effects of these responses across the subthalamic area.

The differences in anticipatory responses to sweet-fat and neutral taste were related to taste-preference for the sweet-fat solution and not to the button-press action requested upon cue presentation. The button press in this task was used to maintain participant attention, it was used for both sweet-fat and taste-neutral solutions, was not coded to change receipt outcome, and there were no reaction time differences between the two stimuli. This simple motor action has been shown to drive robust topographically-dependent beta-band power fluctuations in the subthalamic area (Torrecillos et al., 2018; Priori et al., 2013). However, if the observed beta-band response reported here was due solely to motor activity, we would expect similar changes between sweet-fat and taste-neutral conditions. Nonetheless, the influence of the button press on beta-band activity cannot be excluded. Reassuringly, Buot et al. (2012) found that they could elicit higher magnitude, event-related potentials in the STN to unpleasant vs neutral whether a motor or passive response was required following the image, but the magnitude was larger when paired with a motor response (Baunez et al., 2005). Further, subthalamic beta-band power responses to rewards have been demonstrated in contexts where no motor action was required (Oswal et al., 2012; Al-Ozzi et al., 2020). Here, we demonstrate that when trials are time-locked to the taste cue vs the button press, the magnitude of the beta-band power decrease and beta-band rebound are significantly greater for the former, further suggesting role of anticipatory taste processing.

Compared to prior electrophysiologic studies, the task paradigm in this study allowed for, the food-taste component of receipt to be better separated from the motor (i.e. swallowing) phase (Espinosa-Parrilla et al., 2013). After solutions were delivered onto the participant’s tongue, participants were instructed to refrain from swallowing until prompted 3 s later. During receipt, decreased power responses specific to the neutral taste were observed early in the STN and later in the SN. Prior PET imaging studies have similarly reported on a temporal dissociation between these regions with dopamine release within 5 min, and 15–20 min after sweet-fat (vs “tasteless solution”) intake in the STN and SN, respectively… (Thanarajah et al., 2019) Unfortunately, BOLD changes were not reported or characterized in these regions likely due to the previously mentioned limitations of this modality. The temporally shifted responses between these two physiologically inter-dependent regions may relate to their connectivity within the basal ganglia system circuitry, however, further investigation teasing out these components is needed.

Finally, subthalamic beta-band activity was associated with having increased (vs not increased) BMI after starting DBS therapy. Here, increased BMI was observed only among participants with greater beta-band power magnitude during important discriminatory time periods, namely an exaggerated anticipatory beta-band rebound response to the neutral taste. In prior literature, weight gain following subthalamic DBS for PD has been extensively reported (Foubert-Samier et al., 2012; Mills et al., 2012; Locke et al., 2011). One proposed mechanism is that participants have both decreased energy demands and increased ability to feed due to a reduction in resting tremor and rigidity (Balestrino et al., 2017; Foubert-Samier et al., 2012; Sauleau et al., 2016). However, these findings are inconsistent across studies (Locke et al., 2011). Other studies suggest a primarily metabolic mechanism for weight gain, with decreased lipid and increased glucose oxidation, measured by indirect calorimetry, following STN-DBS-vs levodopa-treated PD patients (Perlemoine et al., 2005).While the present preliminary work is unable to substantiate a causal relationship, we propose two possible mechanisms. First, the beta-band rebound shown to underlie inhibitory control may be disrupted by DBS, leading to increased likelihood of weight gain. Prior literature has highlighted the multimodal effects of DBS both as an inhibitor (i.e. functional lesion) and exciter of local and distant neural activity (Xiao et al., 2018; McIntyre et al., 2004). Participants who require heightened beta-band rebound responses to avoid less-preferred foods may be more susceptible to weight gain following disruptive DBS. A second possibility is that patients gained weight through a pre-existing process, not captured in our study (lifestyle, eating habits, etc.) that was perhaps facilitated by DBS. Prior animal studies have reported that STN lesions differentially modulate appetitive behavior in mice with preexisting binge-drinking (i.e. high preference for alcohol over water) vs those without this preexisting behavior (i.e. no high preference for alcohol over water) (Lardeux and Baunez, 2008). Following STN lesioning, mice with pre-existing alcohol preference had further increased place preference for alcohol-paired environments. However, mice without pre-existing alcohol preferences had increased place preference for water-paired environments following STN lesions. Here, the differential beta-band power patterns may serve as a marker for patients with a pre-existing preference for increased feeding that was facilitated by STN DBS, as a functional lesion, and resulted in increased weight gain. Of note, the two patients with increased BMI after DBS were both female and had the lowest pre-operative baseline BMIs. However, a review of 38 studies examining predictors of BMI change following STN DBS for PD reported no effect of gender or pre-operative BMI (Steinhardt et al., 2020). Our preliminary results pave the way for additional studies to further evaluate these mechanisms as predictors of post-operative weight gain.

Our study provides preliminary evidence that the subthalamic area beta-band power plays a role in taste processing. Nonetheless, it should be noted that the sample size was limited in large part due to the unique study design of delivering actual liquid per oral intraoperatively. However, given multi-channel recording, each participant provided multiple data points for analysis. Moreover, given the inherent limitation of the recording apparatus used for the study, field potentials were only collected unilaterally in the participant’s dominant hemisphere to standardize across the extended case series and to optimize evaluation of any possible reaction time differences (Kerr et al., 1963). We cannot exclude the influence of expected and actual putative differences between the two solutions or ocular saccades on reported electrophysiologic difference. However, these differences are not expected to contribute heavily during the anticipatory phase prior to taste delivery or receipt phase prior to explicit swallow instructions. It is also unclear if our findings are generalizable to subjects without PD. Nevertheless, the finding of a potential control signal within the subthalamic area that when disrupted by DBS leads to weight gain is a conceptual advance that demands further translational study.