Beta frequency shifts in decision making: Spectral fingerprints or communication channels?

The gist: Beta frequency shifts in frontal cortex signal categorical decision outcomes, arising from changes in connectivity between weakly coupled oscillators, which reflect active mechanisms to (re)-activate behaviorally relevant communication channels.

Evidence: Beta frequency shifts in decision making

Beta peak frequency is highly variable between and within individuals and tasks A coarse classification of this variability usually differentiates between low (20 Hz; Roopun et al., 2006; Kopell et al., 2011; Stanley et al., 2018; Oswal et al., 2021), often conceptualized as distinct sub-bands within the beta frequency range, possibly originating from different (sub-)cortical sources and serving different motor and cognitive functions Moment-to-moment modulations of frequency within a particular band have received far less attention (Rassi et al., 2023a) and are the subject of this review In a recent study in NHP, we show that beta frequency, rather than power, is in fact the key feature: in a series of duration- and distance-categorization tasks in which the boundary between categories changed from one block of trials to the next, beta peak frequency consistently reflected the context-dependent categorical decision, regardless of objective stimulus properties (Rassi et al., 2023b) Analysis of local field potential (LFP) recordings in the dorsolateral prefrontal cortex (dlPFC) and presupplementary motor area (pre-SMA) during that delay showed that beta frequency predicted the monkey’s decision, independently of the subsequent movement, and independently of the accuracy of the response (Figure 1A) Importantly, the stimuli and categorical boundaries changed from one block to the next (e.g., an interval of 500 ms could be considered short in one block, but long in another), but the same two distinct beta frequencies consistently reflected the two context-dependent categories, regardless of objective stimulus properties Even when stimuli had identical magnitudes but belonged to different relative categories across task conditions (i.e., depending on the context-defined boundary), beta frequency predicted the animal’s response We conceptualized these two beta frequencies as “channels” of communication, each having distinct spectrotemporal and connectivity profiles We showed that dlPFC and pre-SMA were connected via these frequency channels, and that these beta dynamics could be characterized as transient bursts (see also Box 1) rather than sustained oscillations Finally, we showed that the frequency shift was driven by dlPFC, and that category-selective neurons in dlPFC (Mendoza et al., 2018) synchronized with the beta rhythm at the respective category-selective frequency: short-selective cells synchronized with the frequency reflecting the short category, and long-selective cells synchronized with the frequency reflecting the long category (Rassi et al., 2023b) Strikingly, we observed a similar pattern of results in human M/EEG recordings across a range of tasks: beta frequency shifts in frontal cortex consistently allowed readout of the subjective decision outcome, independent of physical stimulus properties (Rassi et al., 2025) In sum, we find that beta frequency shift in frontal cortex is a decision-related signal that is robust across task designs, decision types, stimuli, analysis approaches, and recording techniques

**Analysis of local field potential (LFP) recordings in the dorsolateral prefrontal cortex (dlPFC) and presupplementary motor area (pre-SMA) during that delay showed that beta frequency predicted the monkey’s decision, independently of the subsequent movement, and independently of the accuracy of the response The stimuli and categorical boundaries changed from one block to the next (e.g., an interval of 500 ms could be considered short in one block, but long in another), but the same two distinct beta frequencies consistently reflected the two context-dependent categories, regardless of objective stimulus properties We conceptualized these two beta frequencies as “channels” of communication, each having distinct spectrotemporal and connectivity profiles We showed that dlPFC and pre-SMA were connected via these frequency channels, and that these beta dynamics could be characterized as transient bursts (see also Box 1) rather than sustained oscillations The frequency shift was driven by dlPFC, and that category-selective neurons in dlPFC (Mendoza et al., 2018) synchronized with the beta rhythm at the respective category-selective frequency: short-selective cells synchronized with the frequency reflecting the short category, and long-selective cells synchronized with the frequency reflecting the long category (Rassi et al., 2023b) Generalization to other paradigms shows that beta frequency shifts in frontal cortex consistently allowed readout of the subjective decision outcome, independent of physical stimulus properties In sum, we find that beta frequency shift in frontal cortex is a decision-related signal that is robust across task designs, decision types, stimuli, analysis approaches, and recording techniques One caveat is that while our LFP data showed that both monkeys’ beta frequency shifted in the same direction to signal long vs. short, our EEG data showed that beta frequency shifted in one direction for two thirds of participants, and in the other direction for the other third It therefore seems that the direction of the frequency shift is not critical, rather, it is the frequency differential which allows consistent and significant readout of the decision outcome on a trial-by-trial basis within a given participant Next, we asked whether this finding would translate to different tasks, sets of stimuli, and recording techniques We analyzed data from two MEG datasets in which participants performed decision tasks (i.e., visual delayed match-to-sample and audio-tactile discrimination) that used response mapping to dissociate decisions from motor outputs Again, we found that the decision outcome was significantly mapped onto frontal beta frequency—meaning a slower frequency was associated with one decision and a faster frequency associated with another decision—with the direction of the effect again differing across participants In sum, we find that beta frequency shift in frontal cortex is a decision-related signal that is robust across task designs, decision types, stimuli, analysis approaches, and recording techniques The evidence discussed here focused primarily on (categorical) decision-making tasks with binary decisions An obvious question is whether our observations hold for cases with more than two possible decision outcomes, that is, does the framework generalize beyond binary categorical decisions Broadly speaking, we suggest two main approaches to tackle these questions in the future: i) expand to cases with more than two categories, and ii) observe how frequency shifts come about during learning and in response to rule updating We hypothesize that adding more categories will result in additional distinct beta-frequency channels that can be flexibly shifted, i.e., adding more categories will likely also shift the frequencies of the original categories (which is to say, we predict that adding a “middle” category in our original temporal categorization paradigm would not lead to that decision outcome being represented by a frequency in between the two we originally observed, but rather lead to a remapping of all categories onto a new set of frequencies) Further, we predict that during learning, specific decision outcomes become “mapped” onto particular frequencies by virtue of the cell ensemble that dynamically forms to represent that particular decision (Antzoulatos and Miller, 2014; Stanley et al.

1. Bold header: Behavioral Decision Readout via Frequency Channels

The improved AI system can decode categorical decision outcomes by analyzing beta frequency shifts in frontal cortex, as this signal is proposed to be a proxy for information content. This allows the AI to read out decisions independent of physical stimulus properties and robust across different task designs, suggesting the system can perform high-level categorization tasks based on these spectral fingerprints.

2. Bold header: Dynamic Ensemble Recruitment Modeling

The system can model cognitive states by tracking the activation of one ensemble vs. another, as beta frequency shifts are proposed to reflect the recruitment of distinct neural ensembles when different (categorical) decisions are made. This enables the AI to infer the underlying population activity that codes for decision outcomes, even without access to raw spike data.

3. Bold header: Communication Channel Multiplexing

The system can selectively transmit information downstream by leveraging distinct frequency channels, which is conceptualized as frequencydivision multiplexing where oscillations at a particular frequency serve as a channel to selectively transmit the code downstream. This allows the AI to filter and interpret different types of cognitive signals based on their specific resonant frequencies.

4. Bold header: Causal Neuromodulation Simulation

The system can simulate experimental manipulations by testing how artificially creating oscillatory electrical fields biases decision-making performance, allowing it to test whether this biases decision-making performance by stimulating at the frequency associated with decision A vs. B. This provides a framework for understanding the causal role of these beta dynamics in behavior.

5. Bold header: Context-Dependent Frequency Mapping

The system can learn to map specific categories to particular frequencies, as the paper suggests that specific decision outcomes become 'mapped' onto particular frequencies by virtue of the cell ensemble that dynamically forms to represent that particular decision. This allows for flexible remapping of categories during learning and rule updating.

Recent evidence suggests that beta-band activity plays a key role in decision-making. Here we review our recent work in humans and non-human primates showing that beta-band frequency shifts in frontal cortex signal categorical decision outcomes. We revisit our previous proposal suggesting that content-specific beta reflects the flexible recruiting of transient neural ensembles and update it to emphasize frequency as the relevant parameter. We argue that beta frequency shifts arise from changes in connectivity between weakly coupled oscillators and that, more than a spectral fingerprint, they reflect an active mechanism to (re)-activate behaviorally relevant communication channels in the brain.