Volitional Regulation and Transferable Patterns of Midbrain Oscillations

Abstract

Dopaminergic brain areas are crucial for cognition and their dysregulation is linked to neuropsychiatric disorders typically treated with pharmacological interventions. These treatments often have side effects and variable effectiveness, underscoring the need for alternatives. We introduce the first demonstration of neurofeedback using local field potentials (LFP) from the ventral tegmental area (VTA). This approach leverages the real-time temporal resolution of LFP and ability to target deep brain. In our study, two male rhesus macaque monkeys (Macaca mulatta) learned to regulate VTA beta power using a customized normalized metric to stably quantify VTA LFP signal modulation. The subjects demonstrated flexible and specific control with different strategies for specific frequency bands, revealing new insights into the plasticity of VTA neurons contributing to oscillatory activity that is functionally relevant to many aspects of cognition. Excitingly, the subjects showed transferable patterns, a key criterion for clinical applications beyond training settings. This work provides a foundation for neurofeedback-based treatments, which may be a promising alternative to conventional approaches and open new avenues for understanding and managing neuropsychiatric disorders.

Keywords: midbrain; neurofeedback; ventral tegmental area.

Figure 1.

Experimental setup. a, Schematic of the neurofeedback paradigm. VTA LFP signals were used to calculate the instantaneous NBP, which were linearly mapped onto the computer screen as the vertical position of the cursor. b, Experimental blocks. During the Control and Washout blocks, the computer screens and the juice reward system were turned off. The cursor provided visual feedback in the Main blocks. By contrast, in the Transfer blocks, only the target was visible. The blocks were color-coded throughout all figures.

Figure 2.

Normalized beta power is superior to raw or z-scored beta power. a, b, The distributions of raw, z-scored, and normalized beta power in the Control blocks in each subject. Each box plot included the 5th, 50th, and 95th percentiles of the distribution. Raw beta power suffered from day-to-day variance, while the z-scored beta power remained skewed (c, d). Absolute skewness considered left- and right-skew similarly. Normalized beta power was significantly less skewed than the other two metrics in both Monkey A (c) and Monkey B (d), as tested by one-way ANOVA. Monkey A: F_(2,42) = 4.274, p = 0.020, partial eta-squared = 0.170; post hoc Tukey’s test: NBP and raw p = 0.039 (Hedge's g = −1.038), NBP and z-scored p = 0.039 (Hedge's g = −1.038), and raw and z-scored p = 1.000. Monkey B: F_(2,42) = 64.96, p = 1.40 × 10⁻¹³, partial eta-squared = 0.756; post hoc Tukey’s test: NBP and raw p = 4.93 × 10⁻¹² (Hedge's g = −4.018), NBP and z-scored p = 4.93 × 10⁻¹² (Hedge's g = −4.018), and raw and z-scored p = 1.000. Error bars represent standard errors of the mean.

Figure 3.

Optimizing NBP using rank correlations between NBP and beta power. a, The Spearman rank correlation coefficients relating frequencies at start of trial (x-axis) and end of trial (y axis). The frequencies were selected so that the denominating frequency band for normalization was broader than the beta band. Thereby, we could ensure that the NBP was strictly bound between zero and one. b–d, 2D histograms of the beta power (in log scale) and NBP using different representative denominating frequency bands, as indicated by the grids with corresponding letters in (a). e, 2D histogram of the gamma power (in log scale) and the NBP using the optimal frequency range (d). In (b–e), the Spearman rank correlation coefficients are provided. The gradient colors indicate the number of samples accumulated in each pixel of the 2D histograms.

Figure 4.

Normalized beta power (NBP) and target locations. a, Target locations. The probability density plot shows the NBP distribution from a Control block. The target sizes were the same and covered around 5–6 percentiles of data in High and Low targets. Target locations were also linearly mapped to the computer screen. The targets were color-coded throughout all figures. b, Stability of NBP within a Control block. The five-minute recording of a Control block was segmented into five one-minute snippets. The NBP distributions and the three percentiles (inset) were estimated for each snippet. A Friedman test was used to test whether there are differences in percentiles across time (Q(4) = 4.53, p = 0.339, n = 5).

Figure 5.

Successful volitional regulation and improvement of VTA NBP with training. a, Trial durations in all Main blocks (Monkey A: n = 15; Monkey B: n = 16) were separated for each target and smoothed by a 20-sample boxcar filter. Insets show average trial durations without separating the targets, indicating general improvement over the Main blocks. b, Maximal improvements for different targets across Main blocks. Bar plots represent the average. One-tailed one-sample t test was used to verify if maximal improvements were greater than zero. Monkey A: High: p = 9.20 × 10⁻⁵ (Cohen's D = 1.298), Center: p = 1.39 × 10⁻⁴ (Cohen's D = 1.242), Low: p = 2.80 × 10⁻³ (Cohen's D = 0.844); Monkey B: High: p = 1.81 × 10⁻⁸ (Cohen's D = 2.562), Center: p = 8.50 × 10⁻⁵ (Cohen's D = 1.241), Low: p = 3.88 × 10⁻⁴ (Cohen's D = 1.050). c, Example trajectories of early, middle, and late trials for hitting High targets from a Main block, color-coded with NBP as shown by the color bar. Dashed lines represent borders of the target (here NBP = 0.800–0.884). Arrows above the trajectories mark the periods when the cursor was in the target region and shade reflects duration within the target region. d, NBP aligned to reward (end of trials) and averaged across trials for individual targets. Each interval in the x axis represents 0.5 s. Shaded areas and error bars are standard error of the mean.

Figure 6.

Rewards per minute in each block. The rewards per minute (RPM) in the Control and Washout blocks correspond to chance and were computed post hoc because the subjects did not actually receive any rewards during these blocks. We first calculated the target locations using the three percentiles in the Control block, and then examined through the NBP time series to count the number of times the cursor stayed within the defined target regions for more than 200 ms. The RPMs in the Main and Transfer blocks were calculated by the actual number of rewards received divided by the duration of the block. a, Monkey A: F_(3,47) = 35.891, p = 3.26 × 10⁻¹², partial eta-squared = 0.696, one-way ANOVA; post hoc Tukey tests: Control vs Main p = 2.46 × 10⁻⁹ (Hedge's g = −2.641), Control vs Transfer p = 5.62 × 10⁻⁷ (Hedge's g = −2.830), Control vs Washout p = 0.988 (Hedge's g = 0.162), and Main vs Transfer p = 1.000 (Hedge's g = −9.71 × 10⁻³). b, Monkey B: F_(3,53) = 156.49, p = 2.59 × 10⁻²⁶, partial eta-squared = 0.899, one-way ANOVA; post hoc Tukey tests: Control vs Main p = 1.78 × 10⁻¹⁵ (Hedge's g = −4.97), Control vs Transfer p = 1.78 × 10⁻¹⁵ (Hedge's g = −7.44), Control vs Washout p = 0.89 (Hedge's g = 0.42), and Main vs Transfer p = 0.305 (Hedge's g = 0.514). Error bars represent standard errors of the mean.

Figure 7.

Differential contribution of frequency bands to volitional VTA regulation. a, Spectrograms were aligned to reward for each target and normalized by the power spectral density in the Control block (b) to show relative changes in power for each frequency bin. Dashed lines represent the boundaries of the beta power band (20 and 35 Hz). c, Changes in beta and gamma power were summed across the beta band (20–35 Hz) and averaged across trials (n = 100 trials). Error bars are standard error of the mean. One-way ANOVA was used to test whether changes in power differed across targets. Beta power: F_(2,298) = 296.52, p = 1.32 × 10⁻⁷¹, partial eta-squared = 0.666, one-way ANOVA; post hoc Tukey’s test p < 0.001 for all combinations. Gamma power: F_(2,298) = 43.60, p = 2.45 × 10⁻¹⁷, one-way ANOVA; partial eta-squared = 0.226, post hoc Tukey’s test p < 0.001 for Low, but not for High and Center (n.s.). d, The raw beta and gamma powers were aligned and averaged to start of trial and reward (end of trial). Shaded areas are standard error of the mean.

Figure 8.

Stability of powers other than beta and gamma frequency bands during regulation. a, Similar to Figure 7c, changes in each band power in the Main block were summed across each band (left: delta 1–4 Hz, center: theta: 4–8 Hz, and right: α: 8–12 Hz) and averaged across trials (n = 100 trials). Error bars are standard error of the mean. One-way ANOVA was used to test whether changes in power differed across targets. Delta power: F_(2,298) = 1.11, p = 0.331, partial eta-squared = 7.4 × 10⁻³. Theta power: F_(2,298) = 1.200, p = 0.303, partial eta-squared = 7.99 × 10⁻³. Alpha power: F_(2,298) = 11.660, p = 1.30 × 10⁻⁵, partial eta-squared = 0.073, post hoc Tukey’s test p < 0.001 for High, but not for Low and Center (n.s.). b, Similar to Figure 9e, changes in power in the Transfer block were summed and averaged across trials. Error bars are standard error of the mean. One-way ANOVA was used to test whether changes in power differed across targets. Delta power: F_(2,297) = 2.119, p = 0.122, partial eta-squared = 0.014. Theta power: F_(2,297) = 2.539, p = 0.081, partial eta-squared = 0.017. Alpha power: F_(2,297) = 1.626, p = 0.198, partial eta-squared = 0.011.

Figure 9.

Regulation transfers to occluded cursor condition and improves over time. a, Trial durations (Monkey A: n = 7; Monkey B: n = 10). b, Maximal improvements for different targets. One-tailed t tests assessed the difference from zero. Monkey A: High: p = 2.82 × 10⁻³, Cohen's D = 1.590, Center: p = 3.74 × 10⁻³, Cohen's D = 1.495, Low: p = 0.020, Cohen's D = 0.983. Monkey B: High: p = 5.37 × 10⁻⁴, Cohen's D = 1.496, Center: p = 9.8 × 10⁻⁵, Cohen's D = 1.905, Low: p = 6.09 × 10^−4, Cohen's D = 1.468. c, NBP aligned to reward and trial-averaged for individual targets. d–f, Spectral analyses from representative Transfer block. e, One-way ANOVA tested differences in changes in power across targets. Beta power: F_(2,297) = 373.65, p = 8.11 × 10⁻⁸², partial eta-squared = 0.716, post hoc Tukey’s test p < 0.001 for all combinations. Gamma power: F_(2,297) = 29.12, p = 2.83 × 10⁻¹², partial eta-squared = 0.164, post hoc Tukey’s test p < 0.001 except High and Center (n.s.). g, Co-modulation patterns are preserved between Main and Transfer blocks (data from representative session). Changes in power are relative to Control block. Triangles represent centroids of each group. High target: slope is 2.76 (R² = 0.645, p = 8.67 × 10⁻²⁴) and 2.62 (R² = 0.666, p = 4.68 × 10⁻²⁵) for Main and Transfer blocks. Low target: slope is 0.30 (R² = 0.895, p = 2.92 × 10⁻⁵⁰) and 0.24 (R² = 0.828, p = 3.24 × 10⁻³⁹) for Main and Transfer blocks. h, Correlation between changes in beta and gamma power. Paired t tests reveal no difference between Main and Transfer. Monkey A: High p = 0.434 (Cohen's D = 0.316), Low p = 0.424 (Cohen's D = 0.234) (n = 7). Monkey B: High p = 0.741 (Cohen's D = 0.104), Low p = 0.693 (Cohen's D = 0.176) (n = 10). Error bars represent standard error of the mean.

Figure 10.

Theta power in the Transfer blocks shows positive RPE after reward. a, The raw theta power was aligned and averaged to start of trial and reward (end of trial) in both Main (top) and Transfer (bottom) blocks, similar to the plotting scheme in Figures 7d and 9f. Shaded areas are standard error of the mean. b, Theta power accumulated between t = 0.2 to t = 0.5 s from Reward for each target condition in the Main and Transfer blocks of a representative session and smoothed using a 20-sample boxcar filter. This interval was selected due to its relevance to the RPE signal (Kim et al., 2012). c, Average ratio of theta power between Transfer and Main blocks. One-tailed one-sample t tests assessed whether the ratios were greater than one. Monkey A: High target p = 0.006 (Cohen's D = 1.550), Center target p = 0.100 (Cohen's D = 0.603), and Low target p = 0.041 (Cohen's D = 0.890). Monkey B: High target p = 4.39 × 10⁻⁴ (Cohen's D = 1.716), Center target p = 7.24 × 10⁻⁴ (Cohen's D = 1.583), and Low target p = 4.69 × 10⁻⁴ (Cohen's D = 1.697). Error bars represent standard errors of the mean. d, Theta power was aligned to Hold and averaged based on whether the holds are successful. If successful, rewards were delivered at t = 0.2 s. Theta power between successful and failed holds from t = 0.4 to t = 0.7 s (identical to t = 0.2 to t = 0.5 s from Reward, as shown in b) were statistically different in both Main and Transfer block based on one-sample t tests. Main p = 3.00 × 10⁻⁶ (Cohen's D = 0.350); Transfer p = 1.34 × 10⁻¹⁶ (Cohen's D = 0.568).

Figure 11.

Changes in beta and gamma power and NBP after neurofeedback training. a, For each session, the average beta and gamma power in the Washout block was divided by the average power in the same frequency band in the Control block. b, The mean NBP in the Washout block was divided by the mean NBP in the Control block for each session. One-tailed one-sample t test served to test whether the changes were greater than one: Monkey A: beta power p = 0.097 (Cohen's D = 0.365), gamma power p = 0.180 (Cohen's D = 0.254), and NBP p = 0.642 (Cohen's D = 0.100), n = 14; Monkey B: beta power p = 0.013 (Cohen's D = 0.640), gamma power p = 1.28 × 10⁻⁴ (Cohen's D = 1.253), and NBP p = 0.999 (Cohen's D = 1.335), n = 15. Error bars represent standard errors of the mean.

Figure 12.

Distribution and variance of NBP in the four blocks. a, NBP distributions in a representative session. Inset shows the 5th, 50th, and 95th percentiles of each distribution. b, Variance of each block in all sessions. One-way ANOVA was used to test for differences across blocks. The code above each bar indicates significance levels using post hoc Tukey tests. Monkey A: F_(3,47) = 1.568, p = 0.210, partial eta-squared = 0.091. Monkey B: F_(3,53) = 15.91, p = 1.68 × 10⁻⁷, partial eta-squared = 0.474; post hoc Tukey tests: Control vs Main p = 3.40 × 10⁻⁴ (Hedge's g = −1.382), Control vs Transfer p = 1.81 × 10⁻⁴ (Hedge's g = −2.069), Control vs Washout p = 0.92 (Hedge's g = 0.311), Main vs Transfer p = 0.804 (Hedge's g = −0.288), Main vs Washout p = 2.00 × 10⁻⁵ (Hedge's g = 1.647), and Transfer vs Washout p = 1.40 × 10⁻⁵ (Hedge's g = 2.394). Error bars represented standard errors of the mean.