Siegel et al. PNAS 2009
Gamma Band Synchronization and Information Transmission – Chapter from Principles of Neural Coding (Summary / Abridged Version)
Vinck, Womelsdorf, and Fries in a chapter entitled “Gamma Band Synchronization and Information Transmission” from the book Principles of Neural Coding question old neuroscientific biases regarding the role of firing rate in coding information, the sensitivity of neuronal membranes to gamma frequency oscillations, the mechanisms underlying gamma oscillations and gamma synchronization, and the place of gamma phase coding in providing reliable, informative signals within the brain. To begin, the gamma band is classically referred to as 40-80 Hz oscillatory signal identified within a local field potential, or in other words, a continuous electrical recording from an extracellular space. Further, gamma band synchronization refers to the organization of other signals – be they spikes from other neurons or other field potentials from independent electrode channels – according to properties of the aforementioned gamma band signal (e.g. the phase). Over the last decades, the role for this gamma band signal in information processing has expanded as such phenomena have been found somewhat universally across cortical regions, and repeatedly involved in attentional processes.
Mechanism – ING and PING: Gamma band oscillations are believed to result from the interaction between pyramidal neurons and fast-spiking, inhibitory basket cells. These fast-spiking spiking cells are parvalbumin+ (a type of calcium binding protein), hyperpolarize post-synaptic neurons mostly in peri-somatic regions, and are capable of firing more than one action potential per gamma cycle. These specific cells play a key causal role in the generation of gamma, for cortical and subcortical slices bathed in biculline – a GABAergic antagonist – are incapable of generating gamma-rhythm oscillations in their local field potentials. Moreover, the optogenetic stimulation of PV+ cells infected with channelrhodopsin2 produced an increase in power in the gamma band. Two neural models have been developed to explain the role of the PV cell and the generation of gamma – the ING model (inhibitory network gamma) and the PING model (pyramidal inhibitory network gamma). According to the ING model, the generation of gamma synchronization arises from basket cell mutual inhibition. The timing of the pyramidal cell firing is secondary to the rhythmic inhibition, and the parameter determining the timing of these gamma frequency is the rate of decay of GABA currents. Alternatively, the PING model proposes that as pyramidal-basket cell interactions are not secondary, but entirely causal in the generation of gamma. As pyramidal cells recover from inhibition, they cause an increase in feedback inhibition, which in turn leads to a decrease inhibition, and subsequently a new volley of inhibition, producing what is observed as this gamma rhythmicity.
Overall, Vinck, Womelsdorf, and Fries believe that there exists a preponderance of evidence in favor of the PING model. One,the PING model predicts a delay of several milliseconds, constituting a characteristic gamma phase lead, between pyramidal neurons and fast spiking basket cells. In contrast, the ING model due to its entrainment of pyramidal neurons should fire in phase with fast spiking basket cells. Two, although the ING model demonstrates robustness against heterogeneity in excitatory drive, the large excitatory drive on these pyramidal neurons may be somewhat onerous and energy inefficient. Three, labs have shown that PV activation with AMPA (receptors on postsynaptic inhibitory interneurons) and NMDA receptor blockade (receptors on postsynaptic pyramidal neurons) fails to produce gamma rhythms.
Gamma – A Common Cortical Phenomenon: Gamma band synchronization has been a somewhat common finding across cortical and subcortical regions.
1. Within the primary visual cortex, there is an increase in synchronization with visual stimulation. This increase in synchronization depends on the salience, contrast, and size of the visual stimulus. Moreover, it is substantially larger in unaesthetized, active awake animal. This synchronization occurs in superficial layers of cortex, whereas the beta band is strong in infragranular layers.
2. This increase in synchronizations occurs commonly – in mouse hippocampus, auditory cortex, somatosensory cortex, barrel cortex, parietal cortex, frontal cortex, and ventral striatum.
3. This synchronization occurs across long-range pathways, including between spinal cord and motor cortex, visual cortex and parietal cortex, V4 and FEF, LIP and FEF, and hippocampus and prefrontal. Mechanistically, this synchronization across areas may result from entrainment of pyramidal cells, as they reset the phase of fast spiking activity in their local assemblies. Such a mechanism, however, would induce a phase delay due to the times associated with synaptic conduction. Long-range zero lag situations would necessarily results from an alternative mechanism – e.g. according to the authors, it may result from long-range excitatory connections causing doublet spikes in FS basket.
Consequences of Rhythmic Neuronal Synchronization - Feedfoward Coincidence Detection: An important functional role for gamma synchronization may be to organize spiking patterns into discrete time bins so that spikes from particular neuronal ensembles can impact post-synaptic neurons nearly simultaneously and increasingly influence post-synaptic spiking dynamics. Although coincident spikes do not require oscillations for they may result secondary to common inputs or co-variation of “stimulus locked rate changes,” a consequence of coherent activity of neuronal ensembles is the grouping of spikes into narrow temporal windows. However, does the timing of these spikes within particular phases of the gamma rhythm (40-80 Hz), leading to expected interspike intervals of no more than 6-10 ms, actually differentially influence postsynaptic neurons?
According to Vinck et al., the effective temporal integration may not be as prolonged as had been previously demonstrated. First, in vivo, contrary to slice, there is a much higher level of background synaptic activity causing the membrane to be consistently nearer to threshold and thus leak conductance to be larger. The rise in the leak conductance naturally leads to prolonged decrease in the membrane time constant, for any increase in membrane potential can be compensated through current passing through such leak channels. Moreover, inhibitory neurons with their shorter time constants and rapidly decaying EPSPs may be the more appropriate target population for identifying coincidence detectors.
Secondly, and surprisingly for me, the level of depolarization of the membrane potential is not the best predictor of the membrane potential. Rather, the first derivative of the membrane potential, e.g. the changes in the membrane potential, are strong determinants of spiking activity. A fast depolarization will lower the threshold of the neuron, and interestingly the orientation tuning of cells in V1 are mainly driven by fast fluctuations of the membrane potential [After a bit of thought, this reported finding is not entirely surprising – a static interpretation of a neuron’s threshold is certainly inappropriate. Conductances are highly dynamic, and with changes in the conductance across the membrane the thresholds will vary correspondingly. Moreover, since numerous channels are sensitive to voltage changes with different time constants, it is expected that thresholds would vary as a function of the changes in membrane potential.] If instead of different levels of membrane depolarization but rather the rate of membrane voltage shift is relevant to driving spiking activity, then it is important to examine how different lagged spikes can drive changes in membrane potential deviations. Expectedly, the temporal integration times would be far shorter, near to the time course of the capacitive membrane current – a few milliseconds. Any synaptic inputs separated by any longer time period would unlikely influence the probability of spiking. Moreover, given a refractory period of several milliseconds, any burst of spikes from a single neuron is unlikely to influence the post-synaptic neuron more than it would in a single spike. The influence of bursts would result only in a weakly additive effect on the membrane potential of the post-synaptic neuron.
Thirdly, neurons are not only sensitive to synchronous or near-synchronous inputs, but also to the specific temporal sequence of dendritic inputs. The membrane potential of the post-synaptic neurons may be tuned, as been previously demonstrated in the retina, by the velocity and direction by which synaptic inputs from dendritic to soma are activated. Within this context, dendrites could not be just simple linear integrators of synaptic inputs. This sensitivity to sequence may, for Vinck et al. allows for the possibility that neurons may in fact detect regular sequences of synaptic inputs organized by different phases of a gamma cycle.
Consequences of Rhythmic Neuronal Synchronization - Balanced excitation and feedback inhibition shape synaptic integration: As was made clear at the beginning of the chapter, gamma oscillations rely on a careful balance between excitation and inhibition between pools of pyramidal neurons and PV+ interneurons. This balance producing the gamma oscillations may also
1. prevent run-away excitation of the network,
2. allow strong recurrent connections that facilitate fast responses to external network input,
3. remove any output noise correction (via global inhibitory feedback)
4. sharpen neuronal selectivity by canceling out noisy fluctuations, ( in other words, the inhibition period of the gamma cycle may strongly limit the temporal window of integration and quench any slow fluctuations in excitatory inputs).
5. act as an important gating device to flexibly modulate the gain of excitation
Consequences of Rhythmic Neuronal Synchronization - Rhythmic gain modulation: Due to the rhythmic rise and fall in excitation of various neuronal ensembles, their interactions will depend on the phase relationships between their respective gamma cycles. This idea that “selective coherence between sender’s and receiver’s gamma band activity” is entitled ……… and represents “ a potentially very powerful mechanism for the flexible routing of signals in the nervous system.” During gamma band activity, pyramidal cell entrained by gamma rhythmic inhibition, characterized by a rise in threshold due to a membrane potential near the Cl- reversal and effective perisomatic shunting due to such proximal targeting of PV inhibitory interneurons. The prediction of this model is that a good gamma phase relationship between two neuronal ensembles will improve their interactions. There is some evidence for this prediction – visual activation of primary visual cortices resulted in an increase in synchronization (with only minor variation around a mean phase relationship) and good gamma phase relationships “preceded strong interactions by about 5 ms.” In the motor cortex and spinal cord, the amplitude of post TMS motor end plate potentials depended on the pre-TMS phase of the spinal beta rhythm.
Consequences of Rhythmic Neuronal Synchronization - Attentional Selection by Selective Gamma Band Synchronization: One of the major roles of the central nervous system is to make decisions that result in particular actions. In order to do so, the nervous system must encode sensory stimuli, select those dimensions of the stimuli that are found to be behaviorally relevant, analyze these dimensions of the sensory space, make a decision, and produce an action. In order to select the relevant information in a rapid way, neuronal ensembles encoding the sensory information must be able to flexibly adjust their relative importance in a timescale which is much ‘faster than the timescale at which synaptic potentiation occurs.” Viinck and others propose that gamma synchronization may play an important in regulating the flow of information between different ensembles. Fries et al. found that in V4 “synaptic inputs that are strongly gamma rhythmic and coherent with the gamma rhythm have an advantage over competing synaptics inputs” add that “enhancement of gamma band synchronization with attention modulation in V4” resulted in reduced reaction times. Moreover, they found that the “induced gamma band oscillation in V4 emerge before top-down modulations on firing rate arise” and occur as soon as an attentional cue is presented to the animal.
Coding and Gamma Band Synchronization: Finally, in addition to investigating the role of gamma in modulating interactions between neuronal groups, Vinck, Womelsdorf, and Fries examine its role in potentially serving as internal clock on which a temporal code could rely upon. In other words, “endogenously generated oscillations can serve as reliable, internal clocks to define the timing of spikes, constituting a phase coding for sensory data and assembly information.” Phase coding has been previously reported – phase precession occurs in the hippocampus, prefrontal cortex, and ventral striatum. Mechanistically, this phase precession may arise from multiple possible mechanisms:
1. Theta phase advance would arise with higher firing rates due to larger dendritic excitation overcoming perisomatic inhibition
2. Theta phase advance may arise from input which is already organized by a slightly fast theta rhythm, resulting in an interference pattern
3. Theta phase advance may arise from sort of network effect
If information is encoded by phase, decoding of phase “requires a waiting or updating time on the order of the cycle duration.” The short duration of the gamma cycle consequently is well suited to organize the encoding of sensory information if its phase coded, for alternative rhythms are slower (faster than alpha, theta, delta). Fries and colleagues found that when an isolated single unit was stimulated by its preferred orientation, spikes were on average advanced in the gamma cycle. Differences between preferred and non-preferred orientations (high vs low spike densities) resulted in phase differences on the order of 2-3 ms, which should be detectable by feedforward coincidence detection mechanisms. Importantly, the absolute degree of the phase shift may be somewhat less important. If the gamma rhythms have high levels of inhibition, the phase shift may be smaller but it also may be more reliable and equally informative. It is not so simple, however, that stronger excitation simply allows it a neuron to fire early in the gamma, for it should then also overcome inhibition for longer periods of time during the gamma cycle and lead to a loss in phase locking. To deal with this, there still needs to be some local negative feedback system over and above the more global gamma rhythmic inhibition.