Link to paper

The full paper is available here.

You can also find the paper on PapersWithCode here.

Abstract

• Neurons can emit action potentials in different patterns.
• Electrophysiologists labeled action potentials emitted at a high frequency as “bursts”.
• The burst coding hypothesis suggests that the neural code has three syllables: silences, spikes and bursts.
• Evidence is reviewed to support the ternary code in terms of mechanisms for burst generation, synaptic transmission and synaptic plasticity.
• Learning and attention theories are reviewed that benefit from the triad.

Paper Content

Introduction

• Problem of neural coding is to interpret neuronal signals
• Neurons represent input features with number of action potentials
• High impulse frequencies imply presence of feature
• Inputs are represented by computing averages
• Not all spikes are equal, code is ternary
• Evidence for burst coding hypothesis in 3 variations
• Alignment between generation and synaptic mechanisms
• Nervous system can process two streams of information simultaneously

What is a burst?

• Brain may use a ternary code
• Neurons may display bimodal or unimodal interspike interval distributions
• A unimodal distribution of intervals is not necessarily associated with a drop of information transmitted using a burst code
• Burst coding hypothesis suggests that bursts are a family of firing patterns that trigger physiological mechanisms not engaged by the same number of spikes in relative isolation
• Features associated with bursting are high intra-burst frequencies, sustained depolarization, decreasing action potential amplitude and widening of action potential shape
• Defining bursting based on the number of spikes at a high frequency
• Synapses can show frequency-dependent short-term plasticity
• Firing patterns can trigger long-term synaptic plasticity
• Events of more than 3 spikes at frequencies higher than 100 Hz can be labeled a burst in the hippocampus and cortex
• A higher frequency may be required in the thalamus and the cerebellum
• Bursts can have a variable duration and can have variable intra-burst intervals
• Defining burst based on a nonlinear readout of calcium means that frequency and number of spikes are both required to establish the boundary between burst and non-burst

Disjunctive and conjunctive burst generation

• Pathway 1 and Pathway 2 are two fictitious synaptic pathways
• Pathway 1 drives action potential firing and Pathway 2 is required for burst firing
• Pathway 2 can act autonomously or drive bursting only when Pathway 1 has generated an action potential
• These two possibilities are referred to as disjunctive and conjunctive burst generation
• The rate of either types of events (spikes or bursts) is dependent on input 1 and input 2
• The firing rate is highly correlated with burst rate, which explains why many studies have rejected burst coding in favor of rate coding

Disjunctive dendritic calcium spikes -cerebellum.

• Purkinje cells emit two types of potentials: simple spikes and complex spikes
• Simple spikes are short action potentials mediated by sodium and potassium ion channels
• Complex spikes are bursts mediated mainly by calcium currents in the dendrites
• Complex spikes are triggered by input from climbing fibers, not parallel fibers
• Dendritic organization may regulate calcium currents underlying complex spikes

Conjunctive bursting -hippocampus.

• Pyramidal cells of the CA1 region have a prominent apical dendrite and a number of basal and radial oblique dendrites
• Input onto basal and radial oblique can trigger action potentials, but input onto apical tuft undergoes electrotonic attenuation
• Inputs from CA3 can trigger single action potentials but rarely bursts
• In vitro studies suggest a conjunctive code, which has seen further experimental support in vivo
• Backpropagating action potentials, distal NMDA-spikes and large amplitudes of distal post-synaptic potentials shape input integration in these neurons
• Granule cells from dentate gyrus and cells from CA3 can produce bursts of action potentials

Conjunctive bursting -cortex.

• Cortex has many types of principal cells
• Deep layer neurons can generate bursts of action potentials
• Inputs to basal and radial oblique dendrites receive recurrent and thalamic connections
• Bursting arises when one or both input pathways are active
• Correlation between singlet rate, burst rate, event rate, burst fraction, and firing rate with either inputs
• Cortex uses a conjunctive code where feedforward and feedback inputs are required to generate bursts
• Potent feedforward inhibition counteracts ability to generate action potentials
• T-type voltage-gated calcium channels control burstiness in thalamic relay neurons
• Controlling hyperpolarization controls burst fraction without affecting firing rate

Burst-dependent synaptic plasticity

• LTP is a cellular basis for learning and memory
• LTP requires NMDARs, elevated post-synaptic calcium and local glutamate release
• Calcium elevation at the synapse can come from NMDAR-mediated currents, local release from calcium stores or VGCC
• Post-synaptic bursting can control LTP expression
• Hebb’s theory needs to be revised when bursting is not triggered by the same pathway undergoing the potentiation
• In vitro recordings have confirmed a central role of post-synaptic bursting in the expression of LTP
• Plasticity in the cerebellum also shows burst dependence
• Burst-dependent learning rules in the electric sense of electric fish

Differential transmission of firing patterns

• Transmission must be frequency dependent to separate bursts from single spikes post-synaptically.
• Francis Crick hypothesized properties of synapses called von der Marlsburg synapses.
• STP is well established and can be triggered by burst-like events with a frequency closer to 100 Hz.
• STP can demix a mixture of spikes and bursts and can communicate independent information to different post-synaptic targets.

Cortex.

• Study surveyed over 20,000 synaptic pairs in animal slices and over 2500 in human tissue
• Most connections showed some level of STD over a wide range of stimulation frequencies
• Connections to and from PV-positive cells tend to show depression
• Some synapses show STD with a cutoff at 50 or 100 Hz
• Connections onto and from SST- and VIP-positive cells display STF
• Combining synaptic properties in cellular connectivity motifs can select bursts
• Polarization of local connections in terms of STD and STF is reflected in a polarization of impinging pathways

Hippocampus.

• CA1 pyramidal cells receive inputs from two main pathways: the perforant path and the schaffer collateral
• Perforant path inputs target mainly apical dendrites and NPY-positive cells
• Schaffer collateral inputs target mainly apical oblique dendrites, PV-, SST-as well as NPY-positive cells
• All inputs show short-term facilitation with frequency dependence
• CA1 pyramidal neurons project to multiple targets, with many connections going to the subiculum, which also display short-term facilitation

Cerebellum.

• Purkinje cells are the most numerous cells in the cerebellum.
• They receive inputs from granule cells and the inferior olive.
• PF inputs show STF, while CF inputs show an absence of short-term plasticity.

Thalamus.

• Sensory thalamus receives feedforward information from the senses.
• Thalamocortical afferents from sensory thalamus to sensory cortices show STD.
• Cortex sends feedback information to sensory thalamus with STF projections.
• Cortex sends feedforward information to higher-order thalamus with STD.

What logic.

• Inputs to principal cells are both STD and STF in the thalamus, more STD in the cortex and strictly STF in the hippocampus.
• Outputs from principal cells are strictly STD in the cerebellum and thalamus, but STF in the hippocampus.
• STD is used to communicate feedforward information, supported by the thalamus and cortex.
• STF inputs are used to control LTP in the cortex, hippocampus and cerebellum.

Algorithmic requirements for burst coding

• Representing information using two distinct types of activity can be beneficial for the nervous system
• Algorithms that require two streams of information can be implemented using a ternary code
• Supervised learning algorithms require two types of information
• Reinforcement learning algorithms benefit from burst coding
• Unsupervised learning algorithms require two types of information
• Attention signals take the form of gain modulation and plasticity gating, which can be facilitated by burst coding

Bursting in vivo

• Bursts occur spontaneously in awake and anesthetized animals
• Bursting is correlated with firing rate
• Different methods of measuring burst fractions lead to different results
• Anesthesia and sleep are associated with higher burstiness
• Relationship between arousal and bursting may extend to other areas
• Burst fraction in awake states is typically between 15-40%
• Variations in burstiness across areas indicate higher propensity in visual and prefrontal areas

Attention states.

• Visual attention enhances or suppresses saliency of locations and features in the visual field.
• Modulations of local field potential and noise correlations are observed in the visual cortex.
• Individual neurons show increases in firing rate and decreases in trial-to-trial variability.
• Direct measures of bursting modulation with attention have been observed in the visual cortex.
• When attention is directed onto a neuron’s receptive field, there is an increase in firing rate and a reduction in burstiness.
• In contrast, attention-dependent increases in bursting have been observed in anterior cingulate and prefrontal cortex.
• In the primary somatosensory cortex of rodents, changes in bursting correlate with a shift in perceptual performance.

Learning.

• Place fields are more likely to form in cells that produce bursts
• Correlation between spontaneous plateau potentials and place field formation
• Artificial induction of plateau potentials can give rise to a new place field
• Inputs from entorhinal cortex, CA3 and disinhibition from SST-positive are required for in vivo bursting
• EC bursting acts as an instructive signal
• High-frequency bursting of long duration occurs in cortex during associative learning tasks
• Blocking feedforward or feedback connections blocks learning
• Dendrite targeting inhibition also blocks learning
• Burstiness encodes errors with a delay of 1 second
• Silencing pyramidal cell activity during reward period alters learning of sensory association task

Discussion

• Brain uses a ternary code for representing, transmitting and processing information
• Presence of burst-generating calcium spikes depend on neuron size, cortical layer and species
• Neurons engage in bursting in vivo and have burst-dependent transmission and synaptic plasticity
• GABAergic interneurons of the cortex may show burst-dependent transmission
• Putative GABAergic cells modulate bursting in an attention task in vivo

Conclusion

• Ternary neural code is a way for cells to communicate both elevated membrane potential and elevated levels of intracellular calcium
• Synaptic plasticity is engaged by elevated calcium rather than elevated membrane potential
• Burst coding hypothesis suggests that synaptic pathways engage either spikes or bursts
• Burst coding also arises from the observation that STD and STF synapses transmit bursts differently
• Burst coding surfaces from long-term potentiation, pairing pre-post activity with bursts leads to LTP
• Features associated with bursting are high intra-burst frequencies, sustained depolarization, decreasing action potential amplitude and widening of action potential shape