Abstract
γ oscillations (20–80 Hz) are associated with sensory processing, cognition, and memory, and focused attention in animals and humans. γ activity can arise from several neural mechanisms in the cortex and hippocampus and can vary across circuits, behavioral states, and developmental stages. γ oscillations are nonstationary, typically occurring in short bouts, and the peak frequency of this rhythm is modulated by stimulus parameters. In addition, the participation of excitatory and inhibitory neurons in the γ rhythm varies across local circuits and conditions, particularly in the cortex. Although these dynamics present a challenge to interpreting the functional role of γ oscillations, these patterns of activity emerge from synaptic interactions among excitatory and inhibitory neurons and thus provide important insight into local circuit operations.
Introduction
Rhythms in the brain have provided insights into local circuit interactions (Csicsvari et al., 2003; Montgomery and Buzsáki, 2007; Cardin et al., 2009; Sohal et al., 2009) and network architecture (Berger, 1929; Contreras and Steriade, 1995; Haider et al., 2006; Steriade, 2006) and remain intriguing for their potential roles in information encoding and transmission (Fries, 2005; Fries et al., 2007; Womelsdorf et al., 2007). Considerable attention has been given to high-frequency oscillations in the 20–80 Hz, or γ, range (Jasper, 1938; Das, 1955; Bressler and Freeman, 1980) and more recently in the high γ (90–140 Hz) range (Colgin et al., 2009; Sullivan et al., 2011). γ oscillations are associated with sensory processing (Gray and Singer, 1989; Gray et al., 1990), cognition and memory (Engel et al., 2001; Harris et al., 2003; Fries, 2005; Montgomery and Buzsáki, 2007; Cabral et al., 2014), and focused attention (Tiitinen et al., 1993; Fries et al., 2001; Vidal et al., 2006) in animal models. γ activity is likewise associated with cognition and perception in humans and is compromised in human neurodevelopmental disorders, including autism (Milne et al., 2009; Peiker et al., 2015) and schizophrenia (Spencer et al., 2003, 2008a, b; Gallinat et al., 2004; Light et al., 2006).
Activity in the γ range can represent the outcome of more than one neural mechanism, either independently or in coordination. Although the observed frequency bands are approximately similar, γ activity varies across an array of species from insects to humans (Ribary et al., 1991; Eckhorn et al., 1993; Laurent et al., 1996; Nase et al., 2003) and across brain regions, including the olfactory bulb, thalamus, hippocampus, and cortex (Ghose and Freeman, 1992; Pinault and Deschênes, 1992; Steriade et al., 1993, 1996; Chrobak et al., 2000; Kay, 2015). γ oscillations are also significantly modulated in peak frequency and power across behavioral and developmental states. Further complicating interpretation, observed increases in γ-band power may sometimes represent enhanced power in a broad frequency range, more analogous to filtered noise than a true oscillation. It is therefore challenging to assign a single, definitive role for γ oscillations in brain function. However, these oscillations emerge from dynamic interactions among excitatory and inhibitory neurons and thus provide a crucial and informative real-time window into local circuit operations.
Multiple network mechanisms
γ oscillations are observed in the local field potential (LFP), which emerges as a result of spatially organized, summed current flow largely due to synaptic activity within a neural network (Mitzdorf, 1985; Barth et al., 1989; Einevoll et al., 2013). γ is the result of rhythmic synaptic output from cohorts of neurons with precise temporal relationships. However, there are varied mechanisms by which these temporal patterns occur within local neural circuits.
In the hippocampus, extensive exploration of computational and experimental models has identified multiple network mechanisms by which γ oscillations can be generated. Experimental work using pharmacological manipulations of the in vitro hippocampal slice preparation found that the 40 Hz-range field potential oscillations evoked under different circumstances exhibited different reliance on synaptic excitation and inhibition. γ oscillations induced by activating mGluR (Whittington et al., 1995) or kainate (Fisahn et al., 2004) receptors require GABAergic, but not glutamatergic, synaptic transmission. In contrast, oscillations induced by the cholinergic receptor agonist carbachol (Fisahn et al., 1998; Fellous and Sejnowski, 2000; Gulyás et al., 2010) or by elevated potassium levels (LeBeau et al., 2002) rely on both GABAergic and glutamatergic components. γ-band activity can thus be evoked in the hippocampal circuit by at least two network mechanisms: (1) by strongly activating interneuron-interneuron networks (I-I; sometimes referred to as interneuron network γ) (Van Vreeswijk et al., 1994; Ermentrout, 1996; Wang and Buzsáki, 1996; Bartos et al., 2002) or (2) by coactivating reciprocally connected groups of interneurons and pyramidal neurons (E-I; sometimes referred to as pyramidal interneuron network γ) (Wilson and Cowan, 1972; Ermentrout and Kopell, 1998; Whittington et al., 2000; Brunel and Wang, 2003; Traub et al., 2005; Tiesinga and Sejnowski, 2009; Whittington et al., 2011). These two key mechanisms may occur independently or in concert, but both rely on fast inhibitory synaptic transmission by GABAergic interneurons.
Although pyramidal-interneuron and interneuron-interneuron network mechanisms for γ generation have been clearly demonstrated in hippocampal circuits, pyramidal-interneuron mechanisms appear to predominate in cortex. Optogenetic activation of fast-spiking basket interneurons (Cardin et al., 2009; Sohal et al., 2009) or pyramidal neurons (Adesnik and Scanziani, 2010) in sensory cortex evokes robust γ oscillations that depend on both GABAergic and glutamatergic synaptic transmission. Indeed, spontaneous γ oscillations in vivo are eliminated by optogenetic suppression of interneurons (Sohal et al., 2009) and both spontaneous and optogenetically evoked cortical oscillations are abolished by application of AMPAR and NMDAR blockers (Cardin et al., 2009). Together, these data strongly suggest that coordinated activity of excitatory and inhibitory neurons is necessary for expression of cortical γ rhythms.
The canonical E-I model of γ generation requires that reciprocally connected excitatory and inhibitory cells fire at different times during the γ cycle. Consistent with this model, some work in vitro and in anesthetized animals has found precisely offset phase relationships between excitatory and inhibitory cortical cell firing during γ oscillations (Hasenstaub et al., 2005). In contrast, several recent studies in awake rodents found little or no average phase difference between excitatory and inhibitory cortical cells (Vinck et al., 2015b; Perrenoud et al., 2016). However, the existence of multiple populations of cortical fast-spiking interneurons with entrainment to early and late phases of γ (Vinck et al., 2015b) may indicate multiple ongoing circuit mechanisms for cortical γ generation. These might include I-I γ generation or more complex relationships among interneurons during bouts of cortical activity. The direct contribution of I-I γ oscillations to cortical rhythms remains to be experimentally demonstrated. However, interneuron-interneuron interactions may contribute to γ rhythms even under circumstances where E-I γ is generated. Indeed, fast-spiking interneurons receive rhythmic IPSPs that likely enhance their entrainment to γ rhythms by modulating spike probability (Tamás et al., 2000; Hasenstaub et al., 2005; Sohal and Huguenard, 2005; Oren et al., 2006; Salkoff et al., 2015).
Hippocampal and neocortical GABAergic interneurons are highly diverse, but several lines of evidence point to fast-spiking basket cells that coexpress the calcium binding protein parvalbumin (PV+) as a key source of synaptic inhibition in the generation of γ oscillations. PV+ basket cells are the most abundant type of interneuron, and they are extensively connected to each other via both chemical and electrical synapses (Cobb et al., 1997; Tamás et al., 1998, 2000; Galarreta and Hestrin, 1999; Gibson et al., 1999; Bartos et al., 2001, 2002). These interneurons also exhibit a highly promiscuous reciprocal synaptic connectivity with nearby pyramidal neurons, potentially allowing them to powerfully synchronize and respond to excitatory spiking (Hofer et al., 2011; Packer and Yuste, 2011). Basket cells are capable of firing rapidly, have intrinsic resonance in the γ range, and are highly entrained to endogenous γ oscillations (Tukker et al., 2007; Klausberger and Somogyi, 2008). Extensive theoretical and computational work has suggested that the specialized synaptic and firing properties of fast-spiking interneurons promote γ oscillations (Wang and Buzsáki, 1996; Rinzel et al., 1998; Bartos et al., 2001, 2002, 2007; Börgers et al., 2008). The time course of GABAA receptor-mediated IPSPs is optimal for generating a 40 Hz oscillation cycle (Whittington et al., 1995; Wang and Buzsáki, 1996). IPSPs with rapid temporal kinetics also enforce a narrow window for postsynaptic spiking (Pouille and Scanziani, 2001; Wehr and Zador, 2003; Wilent and Contreras, 2004; Higley and Contreras, 2006; Cardin et al., 2010), enhancing the entrainment of excitatory neurons.
Relatively little is known about the roles of nonbasket interneuron populations in directly generating γ-band activity. Recent work identified two separate hippocampal γ rhythms: one at the pyramidal cell soma and one at the apical dendritic tuft. The firing of basket cells was coupled to somatic, but not dendritic γ, suggesting a possible role for other GABAergic neurons in at least one local form of γ activity (Lasztóczi and Klausberger, 2014). Intriguingly, under some circumstances, hippocampal basket cells are less strongly entrained to γ than other GABAergic populations, including bistratified cells (Tukker et al., 2007; Klausberger and Somogyi, 2008). In the cortex, PV+ cells receive innervation from other interneuron populations, such as the somatostatin- and vasoactive intestinal peptide-expressing interneurons, and their firing is strongly regulated by these inputs (Cottam et al., 2013; Pfeffer et al., 2013; Pi et al., 2013). γ activity could thus also be strongly modulated by synaptic inhibition of PV+ cells from multiple sources. Computational modeling of the emergence of γ oscillations from neural networks suggests that synaptic inhibition of PV+ basket cells may promote the flexible expression of γ oscillations with varying frequencies (Börgers et al., 2008).
γ generators: hippocampus versus cortex
Although γ oscillations have been observed in many circuits, their properties and underlying synaptic mechanisms vary. As described above, findings from in vitro, in vivo, and computational work suggest several network mechanisms for γ generation in the hippocampus. In addition, the hippocampal-entorhinal cortex circuit comprises multiple anatomically distinct oscillation generators that produce activity in both fast and slow γ bands. Evidence from in vivo recordings suggests at least two independent γ current sources within the hippocampus: one in the CA3-CA1 regions and one in the dentate gyrus (Bragin et al., 1995; Csicsvari et al., 2003). CA3 pyramidal neurons entrained to the γ rhythm activate local CA3 basket cells as well as cells in CA1 (Csicsvari et al., 2003), and γ activity in CA1 depends on input from CA3 (Fisahn et al., 1998). In contrast, γ activity in the dentate gyrus depends on input from the entorhinal cortex (Bragin et al., 1995).
The hippocampal formation generates not only the canonical “40 Hz” γ rhythm but also a distinct fast γ oscillation (defined as 90–140 Hz or 65–140 Hz) (Csicsvari et al., 1999; Colgin et al., 2009; Sullivan et al., 2011; Schomburg et al., 2014; Zheng et al., 2016). γ oscillations in the medial entorhinal cortex are most prominent in the superficial layer, entrain both inhibitory and excitatory neurons, and are strongly modulated by local θ oscillations (Chrobak and Buzsáki, 1998; Cunningham et al., 2003; Quilichini et al., 2010). Entorhinal fast γ oscillations are synchronized with fast γ in the hippocampal CA1, whereas slow γ activity is coherent between CA1 and CA3, suggesting a possible function for these distinct frequency bands in routing information through different networks (Colgin et al., 2009). Synaptic mechanisms underlying fast γ rhythms have been examined in much less detail than those for slow γ. Indeed, much of the mechanistic information inferred from pharmacological work in vitro and anesthetized recordings in vivo may pertain more to CA3-generated slow (30–80 Hz) γ than to fast (90–140 Hz) γ, which depends on entorhinal inputs that are removed or attenuated by slicing for in vitro preparations and by anesthesia in vivo.
Rather than occurring in sustained patterns of activity, both slow and fast γ oscillations in the hippocampus are generated in short bouts, suggesting brief entrainment of subnetworks of inhibitory and excitatory neurons. Slow and fast γ activity patterns occur on different θ oscillation phases and during different behavioral epochs related to spatial exploration and memory (Colgin et al., 2009; Carr et al., 2012; Kemere et al., 2013; Bieri et al., 2014; Cabral et al., 2014; Takahashi et al., 2014; Yamamoto et al., 2014; Zheng et al., 2016). However, the frequency and amplitude of γ oscillations are not stationary, even within these short bouts. Instead, instantaneous frequency and amplitude change rapidly both in vitro and in vivo (Bragin et al., 1995; Womelsdorf et al., 2007) and are coupled to underlying fluctuations in synaptic excitation and inhibition (Atallah and Scanziani, 2009). This variation suggests that the temporally patterned interactions between excitatory and inhibitory neurons that underlie γ oscillations are transiently recruited or enhanced by ongoing events in the surrounding network. It remains unclear whether this transient recruitment is due to fluctuations in local synchrony at lower frequencies, changes in overall excitatory drive to the circuit, or another mechanism. Overall, these findings highlight not only the diversity of hippocampal γ activity patterns, but also how a detailed understanding of these oscillations has provided insight into the architecture and function of the hippocampal circuit.
γ oscillations in cortical circuits exhibit a pronounced laminar profile. γ power (30–80 Hz) in the LFP and γ-band synchrony among neurons are highest in supragranular layers, suggesting that densely recurrent networks in these layers may contribute to amplification of activity in this frequency band (Lund, 1988; Malach et al., 1993; Buffalo et al., 2011; Xing et al., 2012; Smith et al., 2013). In agreement with this idea, optogenetic activation of layer 2/3 pyramidal cells with a ramp stimulus generates robust γ oscillations in mouse primary visual cortex in vivo and in vitro (Adesnik and Scanziani, 2010). Evidence from laminar recordings of mouse visual cortex suggests potentially independent sites of local γ generation in superficial and deep cortical layers (Welle and Contreras, 2016). Recordings from layers 4 and 6 in primate visual cortex further indicate amplification of γ power from input to output layers (Briggs and Usrey, 2009). In turn, γ oscillations may exert different levels of control over spike timing in specific subnetworks of cortical pyramidal neurons, mediating more precise spike control in supragranular pyramidal neurons than those in deeper layers (Otte et al., 2010; Hasenstaub et al., 2016). Although the retina and lateral geniculate nucleus generate some activity in the γ range, recordings in primate and cat LGN and primary visual cortex suggest that the thalamus is not a major contributor to γ power in cortex (Castelo-Branco et al., 1998; Koepsell et al., 2009; Ito et al., 2010; Bastos et al., 2014). In contrast, recent evidence suggests that narrowband, but not broadband, γ in mouse primary visual cortex is inherited from the thalamus (Saleem et al., 2016). γ oscillations in the 20–80 Hz range in cortex are thus largely cortical in origin and may result from the activity of multiple laminarly distinct generators, but γ sources may vary across species. γ power in the LFP can also be inherited from long-range corticocortical synaptic connections and can entrain local neurons (Engel et al., 1991; Roelfsema et al., 1997; Tallon-Baudry et al., 2001; Gregoriou et al., 2009). γ activity recorded in the LFP at any particular cortical site may thus emerge from a mix of local and long-range circuit operations, presenting experimental and analytical challenges to assigning underlying sources and mechanisms.
Spontaneous and evoked oscillations
Spontaneously generated and stimulus-evoked γ oscillations differ in their engagement of cortical networks. Recordings in cats and rodents highlight the different subnetworks engaged by these two forms of γ activity. Spontaneous γ is generated throughout the cortical layers and largely entrains fast-spiking, putative inhibitory interneurons (Steriade and Amzica, 1996; Steriade et al., 1996; Sukov and Barth, 1998; Welle and Contreras, 2016). In contrast, stimulus-evoked γ is primarily generated in layer 4 and supragranular layers and also entrains excitatory neurons, suggesting distinct circuit involvement (Welle and Contreras, 2016). Intracellular recordings of neurons in superficial visual cortex in awake mice reveal that PV+ interneurons show strong but brief spike entrainment to γ activity in response to short epochs of correlated background synaptic input during both baseline and visual stimulus periods. In contrast, excitatory pyramidal neurons show little entrainment in the absence of a visual stimulus (Perrenoud et al., 2016). Work from primate visual cortex area V4 similarly suggests that phase-locking to spontaneous γ during baseline epochs is largely restricted to putative interneurons, whereas entrainment of putative excitatory neurons is observed upon visual stimulation (Vinck et al., 2013). In both V1 and S1 cortex of rodents, fast-spiking putative interneurons always show entrainment to γ activity, whereas regular spiking putative excitatory neurons become entrained during periods of arousal and locomotion (Vinck et al., 2015a, b). Together, these findings suggest that resonant circuits in different layers may be differentially recruited by internally and externally generated drive to the cortex.
During spontaneous UP states in prefrontal cortex in vitro and in vivo, inhibitory synaptic and spiking activity in layer 5 is locked to the γ rhythm, indicating that interneurons are entrained during spontaneous γ as observed in sensory cortex (Hasenstaub et al., 2005). However, γ activity associated with UP states also entrains the firing of layer 5 putative excitatory neurons (Hasenstaub et al., 2005; Ushimaru and Kawaguchi, 2015). Similar phase relationships between γ oscillations and both excitatory and inhibitory events have been observed during spontaneous UP states in superficial entorhinal cortex in vitro (Salkoff et al., 2015). This variation in observed entrainment of pyramidal cells may result from true distinctions in circuit architecture across different cortical areas, potential differences in the type of γ oscillations generated by UP states and other regimens of spontaneous activity, or differences in analytical techniques between groups.
Stimulus-induced γ in visual cortex has a broad peak in both primates (Ray and Maunsell, 2010) and rodents (Vinck et al., 2015a), consistent with observations that γ oscillations occur in transient bouts and with varying frequencies. Rather than being stationary, evoked cortical γ oscillations depend on the ongoing stimulus regimen. Repeated visual stimulation results in increased γ-band synchronization among putative cortical pyramidal neurons and interneurons, even when overall firing rates decrease (Brunet et al., 2014). γ power likewise varies with a number of other stimulus features, as reviewed extensively by Ray and Maunsell (2015). γ frequency rises with increasing stimulus contrast and varies over cortical distance in response to stimuli with spatially variable contrast values (Ray and Maunsell, 2010). Similarly, the frequency of visually induced γ oscillations in humans changes with the velocity of stimulus motion and the eccentricity of the stimulus (Swettenham et al., 2009; van Pelt and Fries, 2013; Orekhova et al., 2015). Because the frequency of oscillations is tied to the timing of the underlying synaptic interactions between neurons, these observations suggest evolving recruitment of local circuit elements over time during periods of high external drive to the cortex.
In sum, these findings suggest that distinct subnetworks in different cortical layers not only generate γ oscillations independently but may also be recruited differentially by internally generated spontaneous activity and sensory-evoked drive to the cortex. Excitatory drive to cortical networks engages the resonant properties of local circuits, facilitating the transient engagement of local synchrony among excitatory and inhibitory neurons. The varying frequencies exhibited during these bouts of γ oscillations suggest a range of temporal dynamics in the underlying synaptic interactions, but these temporal variations still remain to be fully explored in vivo.
State dependence of oscillation frequencies and entrainment
γ oscillations vary significantly with behavioral state in both cortex and hippocampus, suggesting state-dependent modulation of the underlying synaptic interactions. Hippocampal γ activity increases during locomotor behavior (Buzsáki et al., 1983) and peak γ frequency increases with running speed (Ahmed and Mehta, 2012; Kemere et al., 2013). In contrast, in both visual and somatosensory cortex, the power and peak frequency of the γ oscillation increase with arousal and locomotion but appear uncorrelated with locomotion speed (Niell and Stryker, 2010; Vinck et al., 2015a, 2015b). In mouse V1, locomotion- and visual stimulus-induced γ bands demonstrate separable peak frequencies (Vinck et al., 2015a), suggesting that they may be generated through different local and long-range circuits. Locomotion-related γ oscillations have been reported to be narrow-band in frequency (Niell and Stryker, 2010; Hoy and Niell, 2015), highlighting the possible engagement of a single γ generator circuit that may rely on thalamic input (Saleem et al., 2016). Intriguingly, several recent studies found that the state-dependent γ oscillations associated with locomotion emerge over the course of development. Cortical γ oscillation power and spike-field coherence increase in a laminar pattern during the critical period in rodents (Chen et al., 2015; Hoy and Niell, 2015). This trajectory likely reflects developmental changes in the intrinsic properties of inhibitory interneurons (Doischer et al., 2008; Okaty et al., 2009; Goldberg et al., 2011) as well as critical-period changes in network connectivity (Fagiolini and Hensch, 2000; Chattopadhyaya et al., 2004; Lazarus and Huang, 2011).
γ oscillations vary not only with global arousal and motor activity, but also with attention and cognitive demand. Attention is associated with increased γ power in the LFP (Fries et al., 2001, 2008; Hauck et al., 2007), and this increased power is observed independently of ongoing cortical firing rates (Fries et al., 2001, 2008; Vinck et al., 2013). Computational work has identified potential roles for pyramidal-interneuron γ oscillations in attentional selection between stimuli (Börgers et al., 2008). In agreement with this general model, attention in the absence of visual stimulation increases the spike-field locking of putative fast-spiking interneurons, but not putative excitatory neurons, to γ rhythms in primate V4 cortex (Vinck et al., 2013). Similarly, the allocation of attention is linked to enhanced PV+ interneuron entrainment to γ oscillations in prefrontal cortex in mice (Kim et al., 2016). Enhanced synchronization at γ frequencies in primate visual cortex is correlated with faster task reaction times, suggesting a potential role for this patterned activity in optimizing encoding and transmission of behaviorally relevant information (Womelsdorf et al., 2006; Fries et al., 2007; Fries, 2015). γ power in visual cortex is likewise modulated by attention in humans (Gruber et al., 1999; Siegel et al., 2008; Self et al., 2016).
Functional consequences
Current evidence suggests a functional role for the rhythmic synaptic interactions underlying γ oscillations in directly modulating sensory processing by individual neurons. The rhythmic engagement of inhibitory, or inhibitory and excitatory, synaptic activity can enhance the precision and efficacy of sensory encoding. Indeed, encoding of visual stimulus orientation varies over the γ cycle (Womelsdorf et al., 2012). Hyperpolarizing IPSPs can increase subsequent spike probability, potentially raising response amplitude (Mainen and Sejnowski, 1995; Tamás et al., 2004). Inhibition may further regulate the signal-to-noise ratio and gain of sensory responses by several mechanisms (Fellous et al., 2003; Börgers et al., 2008; Atallah et al., 2012; Lee et al., 2012; Wilson et al., 2012; Khubieh et al., 2016). Phasic inhibition from PV+ interneurons at γ frequencies restricts the spiking of excitatory neurons to specific phases of the γ cycle and increases the temporal precision of sensory-evoked spiking (Cardin et al., 2009; Siegle et al., 2014). Increased precision of spike timing across a local population of neurons may contribute to synchrony, potentially enhancing the synaptic efficacy of these spikes at downstream targets. The inhibition-dependent regulation of evoked responses at specific γ phases is associated with improved behavioral detection of low-amplitude sensory stimuli (Siegle et al., 2014). γ oscillations may also play more complex roles in the encoding and transmission of information (Fries et al., 2007; Fries, 2015), but the circuit-level underpinnings and functional consequences of these roles remain to be fully explored.
Open questions
Despite decades of exploration, several aspects of the mechanisms underlying γ oscillations remain unknown. γ is often observed as a nonstationary activity pattern, typically occurring in short bouts, whose peak frequency may vary systematically with a number of developmental, behavioral, and stimulus-dependent variables. Rather than confounds arising from specific experimental paradigms, these variations are hints of underlying network and synaptic dynamics. The current model for γ generation is based on recurrent E-I interactions in local circuits, with rhythmic engagement of PV+ interneurons and pyramidal neurons. It is still unclear exactly how short epochs of rhythmic synaptic activity and spike synchrony with varying frequency components are repeatedly recruited and quenched in continuously active cortical or hippocampal circuits in vivo.
Extensive in vitro, in vivo, and computational work has provided a nuanced view of the sources of γ activity in the hippocampal-entorhinal circuit. However, the local mechanisms of γ generation in cortical circuits are less well understood. In particular, the participation of inhibitory and excitatory neurons across different cortical layers and the relative contributions of E-I and I-I interactions to cortical γ rhythms in vivo remain to be explored in more detail.
Finally, although rhythmic interactions between pyramidal cells and PV+ interneurons have been deeply examined, the roles of non-PV+ interneurons in γ oscillations remain largely unknown. The diverse populations of GABAergic interneurons have distinct intrinsic properties and synaptic targets. They may thus contribute in varying ways to generating and regulating the expression, amplitude, and spread of γ oscillations. Non-PV+ interneurons may either directly generate rhythmic activity via novel mechanisms or indirectly influence the pyramidal-PV+ circuit by inhibiting excitatory neurons or other inhibitory interneurons. Understanding their role in the broad suite of neural circuit behaviors that contribute to γ activity will provide further insight into the complex circuit architecture underlying network computations.
Footnotes
Dual Perspectives Companion Paper: How Close Are We to Understanding What (if Anything) γ Oscillations Do in Cortical Circuits?, by Vikaas S. Sohal
This work was supported by the Alfred P. Sloan Foundation, the Brain and Behavior Foundation, McKnight Foundation, National Eye Institute EY022951, and National Institute of Mental Health MH102365. I thank Vikaas Sohal for discussions and Michael J. Higley for comments on the manuscript.
The authors declare no competing financial interests.
- Correspondence should be addressed to Dr. Jessica A. Cardin, Yale University, 333 Cedar Street, PO Box 208001, New Haven, CT 06520. jess.cardin{at}yale.edu
References
Response from Dual Perspective Companion Author–Vikaas Sohal
Jess' review highlights the extraordinary diversity of γ oscillations which may be generated by excitatory-inhibitory or interneuron-interneuron networks, occur within pyramidal cell somata or their dendritic tufts, oscillate at slow (30–80 Hz) or fast (90–140 Hz) frequencies, arise from sources in the superficial versus deep layers of cortex, reflect thalamic input versus local cortical circuit interactions, and appear spontaneously or in response to a stimulus. Furthermore, Jess notes that “γ oscillations occur in transient bouts and with varying frequencies.” These important observations are fully compatible with our proposal that γ oscillations “flexibly reroute the flow of information between groups of neurons” and/or “modulate patterns of activity within local circuits.” The specific mechanisms we describe, through which γ oscillations may contribute to information processing, do not require that γ oscillations have a constant frequency or amplitude, or that they occur in a stereotyped way. What they do require are that γ oscillations “enforce a narrow window for postsynaptic spiking” and entrain the firing of at least some excitatory neurons. As Jess notes, many forms of γ oscillations appear to satisfy these conditions. When γ-frequency fluctuations in inhibition inhibit the firing of excitatory neurons at certain times, then there are invariably consequences for how information is processed at a local level and/or transmitted between structures. The mechanisms we outlined simply describe these consequences. In this context, the heterogeneous and nonstationary nature of γ oscillations noted by Jess does not complicate potential function(s) of γ oscillations but rather enhances them. Specifically, even within a single microcircuit, one type of γ oscillation may appear at one moment in time, transiently enhancing a particular circuit operation or the flow of information along a particular pathway, only to be supplanted a short time later by another type of γ oscillation performing an alternate function. This highlights the potential utility of γ oscillations for endowing neuronal circuits with profound flexibility. As Jess notes, fast-spiking basket cells have intrinsic resonance in the γ range, as well as synaptic and firing properties that promote γ oscillations, suggesting (to us) that they are optimized to exploit these potential functions, rather than to produce γ oscillations simply by happenstance.
In conclusion, there is now a compelling body of theoretical work showing how γ oscillations ought to impact information processing, in vivo observations that are consistent with these mechanisms, and studies that have observed behavioral consequences after optogenetically manipulating γ oscillations. All that being said, the field is still missing the “smoking gun” that brings together all of these elements: by describing the behavioral effects of optogenetic manipulations, which create or disrupt the detailed properties of γ oscillations that are required by specific mechanisms through which γ oscillations are hypothesized to impact information processing. As Jess notes, the dynamic and heterogeneous nature of γ oscillations makes establishing this sort of link extremely challenging. Nevertheless, this represents the crucial next step that the field needs to take to definitely establish how γ oscillations impact neural circuit function.
In this issue
