A Role of Phase-Resetting in Coordinating Large Scale Neural Networks During Attention and Goal-Directed Behavior

Abstract

Short periods of oscillatory activation are ubiquitous signatures of neural circuits. A broad range of studies documents not only their circuit origins, but also a fundamental role for oscillatory activity in coordinating information transfer during goal directed behavior. Recent studies suggest that resetting the phase of ongoing oscillatory activity to endogenous or exogenous cues facilitates coordinated information transfer within circuits and between distributed brain areas. Here, we review evidence that pinpoints phase resetting as a critical marker of dynamic state changes of functional networks. Phase resets: (1) set a "neural context" in terms of narrow band frequencies that uniquely characterizes the activated circuits; (2) impose coherent low frequency phases to which high frequency activations can synchronize, identifiable as cross-frequency correlations across large anatomical distances; (3) are critical for neural coding models that depend on phase, increasing the informational content of neural representations; and (4) likely originate from the dynamics of canonical E-I circuits that are anatomically ubiquitous. These multiple signatures of phase resets are directly linked to enhanced information transfer and behavioral success. We survey how phase resets re-organize oscillations in diverse task contexts, including sensory perception, attentional stimulus selection, cross-modal integration, Pavlovian conditioning, and spatial navigation. The evidence we consider suggests that phase-resets can drive changes in neural excitability, ensemble organization, functional networks, and ultimately, overt behavior.

Keywords: alpha; coding; cross frequency coupling; gamma; inter-areal coordination; oscillations; phase reset; theta.

Figure 1

Coordination of information flow is maintained by oscillatory dynamic circuit motifs. (A) Canonical circuit motifs give rise to frequency-specific oscillations, such as theta (blue), alpha and/or beta (green), and gamma (red). (B) These motifs are embedded within cortical micro-columns and support information processing between layers and cortical areas. Summarized here are the results of six studies that suggest specific relationships of oscillatory activation signature and anatomical circuit structure. The left most inset depicts the cortical layers S, G, and I, corresponding to the supra-granular, granular, and infragranular layers, respectively. (i,ii) In visual cortex, gamma activity follows feedforward connections (red; van Kerkoerle et al., ; Bastos et al., 2015) and co-occurs with low frequency theta (Bastos et al., 2015). On the other hand, alpha/low beta activity correlates with the feedback direction (green; Bastos et al., 2015). (iii,iv) The entrainment of cortical oscillations may depend on thalamic input at theta, alpha and/or beta band frequencies (Saalmann et al., ; Fuentemilla et al., 2014). (v) Long-distance oscillatory coordination can then affect processing within a micro-column; for example, theta activity and alpha generated in deep layers modulated gamma activity in superficial layers (Spaak et al., ; Florez et al., 2013). (C) The six studies outlined in (B) have been combined to highlight a putative oscillatory motif coordinating distant sites via interactions between different frequencies. The putative combined motif suggests that a mixture of oscillatory dynamic circuit motifs coordinates information processing between cortical laminae and across brain areas. Different circuit motifs are responsible for generating/propagating specific oscillations, and are recruited to fulfill specific functions. Note that this conceptual model is dominated by studies of visual processing. We stress that the purpose of combining these studies is to set out a framework to understand how local and distant circuits may functionally interact. Thus, it is likely that brain areas with different laminar structures have a different coordination profile, and future studies across multiple brain areas are necessary to test the predictions of such a framework (see Box 1). (D) Not only are oscillations constrained by anatomy, but oscillatory networks also emerge in relation to specific task contexts. Tracking changes in the phase of interacting oscillatory sources can thus be used to make predictions about anatomical and neurophysiological mechanisms underlying functional changes. Abbrevations: TEO, temporo-occipital; MT, medial temporal; mPFC, medial prefrontal cortex.

Figure 2

Phase alignment permits integration of information. (A) Stimuli can be encoded even before the phase of maximum excitation (i.e., if a strong depolarization occurs when the local circuit is resistant to perturbation), whereas stimuli with a weaker activation are encoded closer to the high excitability phase (when the circuit is sensitive to perturbations). (Bi) Neural codes that rely on temporal precision either pool the response of spatially distributed neurons (pooled response code) or maintain the spatial distribution responses (joint response code; left). However, taking the phase of a population oscillation can be leveraged to provide another dimension to coding (right). (ii) In the auditory cortex of macaques listening to natural sounds, accounting for the phase using either a pooled or joint response increases the informational content of the neural representation (illustration adapted from Kayser et al., 2009). (C) Phase-dependent rate coding is evident in area V1, where the firing rate increases associated with a preferred orientation occur during a specific phase of the gamma cycle. In other words, orientation selectivity is dependent on the phase of ongoing gamma (adapted from Womelsdorf et al., 2012). (D) Near threshold somatosensory stimuli are detected when the stimulus occurs during the rising phase (but not the falling phase) of an ongoing infra-slow oscillation (~0.1 Hz). This further suggests that encoding of somatosensory stimuli depends on the phase of the oscillation (figure adapted from Monto et al., 2008).

Figure 3

Nested oscillatory interactions route local and long-range activity to optimize stimulus representation. (A) The phase of slower (e.g., theta) oscillations affect processing facilitated by faster oscillations. Theta phase can organize gamma bursts, evident as a modulation of gamma amplitude. (B,C) Each data point is obtained by (1) extracting unique CFC patterns based on anatomical location and frequency content; and (2) from these patterns of location and frequency content, selecting LFPs with statistically significant coupling (for details, see van der Meij et al., 2012). (B) The frequency peak was calculated from each unique CFC pattern as the peak of the spectral profile of coupled phase or amplitude-providing LFPs. Phase providing LFPs had a spectral peak that was characteristically lower than amplitude providing LFPs. (C) In the spatial map, the mean Euclidian distance between LFPs is generally greater for phase providing rather than amplitude providing LFPs. This implies that during cross frequency coordination, low frequency activity has an influence over a greater space than high frequency activity. (D) In macaques, coherence in the gamma band between V1–V4 increased for attended but not unattended stimuli (Bosman et al., 2012). Gamma coherence between distant sites was modulated as a function of theta phase (adapted from Bosman et al., 2012). (E) (Top panel) In this schematic, phase resetting in low or high frequency sources (or both) can facilitate stimulus encoding. (Bottom panel) While low frequency oscillations may emerge to support a specific function, high-frequency activity can significantly contribute to internal stimulus representation. For example, during comprehensible but not incomprehensible speech, theta phase is informative of the speech envelope. However, theta-modulated gamma amplitude significantly reduces uncertainty about the stimulus (see Gross et al., 2013). Phase-modulated gamma activity also contributes to attention switching in macaques (Voloh et al., 2015).

Figure 4

Pyramidal-interneuron-gamma model (PING) motif with differential drives and targeted neuron populations results in differential cross frequency coupled profiles. The PING model gives rise to cross frequency coordination (see Onslow et al., 2014). (A) The PING motif has been widely studied as a generator of fast gamma oscillations. Arrowheads correspond to excitatory connections, while circles correspond to inhibitory ones. The excitatory and/or inhibitory population can receive inputs. The response of the population to an input is governed by a sigmoidal response curve. (B) The excitatory or inhibitory populations can be targeted with a constant and/or periodic input. This panel contains a selection of response profiles, and abstracts over specific input values responsible for the observed output profile (for a full description, see Onslow et al., , in particular Figures 3C, 4B,E, 5A, 6D). (From left to right) (i) A constant input to the excitatory population generates gamma oscillations. (ii) An 8 Hz periodic stimulation of the excitatory population generates a signal with the same period for pyramidal cells and interneurons, but with gamma activity locked to the peaks. (iii) Increasing the input past a critical point results in gamma activity locked to the ascending and descending slopes. (iv) Targeting both the inhibitory and excitatory populations with a constant input results in similar gamma activity as in (i). (v) However, when the input to the inhibitory population is periodic, gamma activity locked to the descending phase of the local population.

Figure 5

Theta phase alignment to salient cues coordinates networks and correlates with behavioral success. Cue-aligned theta oscillations in the medial prefrontal cortices (i) organizes neural activity that drives behavior when (ii) the low frequency phases reset in response to a salient cue, and (iii) routes information across long-distances, all of which (iv) can be understood in terms of a Dynamic Circuit Motif, visualized near the bottom of the figure (for a review, see Womelsdorf et al., 2014b). (A) In a Pavlovian conditioning task, rodents learn to associate a neutral cue (conditioned stimulus, CS) with an aversive stimulus. Recordings were made in the rodent mPFC (adapted from Courtin et al., 2014). (i) Theta oscillations were often apparent when the animals exhibited freezing behavior. Neural spikes were locked to theta peaks, suggesting that when present, theta oscillations drove local circuit modulation that resulted in a fear response. (ii) Cue-aligned theta oscillations emerged in response to cue onset (left), but only when the animal was in a high (bottom) but not low (top) fear state. Optogenetically inducing theta oscillations also led to freezing (right). (iii) Neurons reorganized by theta projected to targets responsible for instantiating the fear response. Thus, theta phase reset in upstream areas is critical for affecting fear processing in downstream areas. (iv) A proposed dynamic circuit motif relating the reactivation of a conditioned rule (the function) to spike-theta phase locking (the neural signature). The long distance synaptic integration is achieved by inhibiting PV+ interneurons, leading to the disinhibition of projecting excitatory principal neurons. (B) In a cued selective attention task, macaques had to covertly shift the focus of attention to a target in order to correctly identify a change in the target (adapted from Voloh et al., 2015). (i) Following onset of the cue that triggered the attention shift, high frequency gamma amplitude was locked to the peaks of theta oscillations in prefrontal and anterior cingulate cortex. Importantly, gamma-amplitude to theta phase locking on correct trials was systemically different than on error trials. (ii) Correct, but not erroneous, covert attention shifts were also accompanied by an increase in theta phase consistency in those LFPs that modulated gamma activity. (iii) During theta-gamma correlation, LFPs with theta information predominated in the ACC, while those with gamma information were more likely in the LPFC. This study suggests that theta modulation of local activity may extend to distant sites. In line with the study in (A), theta phase reset drives neural reorganization, leading to observable behavioral changes. (iv) In the framework of a dynamic circuit motif, where a function (attention switching) is correlated with a neural signature (theta-gamma correlation). Causal connection between them could be subserved by neural elements known to generate gamma nested in theta oscillations. Abbreviations: BLA, basolateral amygdala; PGA, periaqueductal gray area; PV+, parvalbumin+.