Neural Codes
A spike train can carry information in both the average firing rate and the pattern of spikes in the train. But can such a spike-pattern code be supported by cortical circuits? Neurons in vitro produce a spike pattern in response to the injection of a fluctuating current. However, cortical neurons in vivo are modulated by local oscillatory neuronal activity and by top-down inputs. In a cortical circuit, precise spike patterns thus reflect the interaction between internally generated activity and sensory information encoded by input spike trains.
For a single neuron, the potential information content of precise and reliable spike times is many times larger than that which is contained in the firing rate, which is averaged across a typical interval of 100 milliseconds. The information contained in spike timing is available immediately, rather than after an averaging period. Furthermore, the timing of patterns of spikes can potentially transmit even more information than the timing of the individual constituent spikes. The relevance of spike patterns is apparent when considering populations of neurons: When a group of similar neurons (a "pool") produces precise and reliable spike trains, the neurons to which they project receive volleys of synchronous spikes. This opens up the possibility of communicating between different cortical areas through synchronous spike volleys.
The imprecision in spike timing is mainly due to variability in the membrane voltage just before the spike, and that this variability is inversely proportional to the rate of change of the voltage. Thus, a precisely timed spike follows a rapidly depolarizing current. There are other sources of imprecision: the spike threshold can change with the rate of voltage change, or membrane currents can be activated by neuromodulators. As the rate of change of the membrane voltage generally increases with the amplitude of the stimulus, the precision also improves as the stimulus amplitude increases.
Many neurons have a preferred frequency for stimulus waveforms. For a subthreshold sinusoidal current, the amplitude of the voltage deflection will be maximal when the stimulus frequency matches the preferred frequency. When the stimulus amplitude is increased above the spike threshold, the firing rate, reliability, and precision will be optimal for stimulus waveforms at the neuron's preferred frequency. Experiments show that interneurons have intrinsic frequencies in a higher range (20–70 Hz) than in pyramidal cells (4–12 Hz). Models predict that the preferred frequency arises from the dynamics of voltage-gated channels.
In recordings from the lateral geniculate nucleus (LGN) of the cat in response to an irregularly fluctuating whole-field stimulus, neurons fired spikes at highly reliable spike times, with submillisecond precision, across trials, across neurons, and across cats. In the middle temporal area (area MT) of the monkey visual cortex, where neurons respond selectively to the direction of motion of a visual stimulus, the spike-timing precision in responses to moving stimuli that fluctuated between the preferred and nonpreferred directions was less than 5 milliseconds. This is a surprising result given that the visual inputs to neurons in area MT must go first through the primary visual cortex and are at least eight synapses removed from photoreceptors. Overall, the degree of precision of spiking in response to repeated presentations of the same stimulus appears to decrease along the visual pathway, whereas spike-count variability increases. This could be due to the presence of background cortical activity.
Spike-Time Patterns
A single-neuron spike pattern is a sequence of spike times that either occur together in a trial or do not occur at all. We developed an automatic clustering procedure for finding spike-time patterns in repeated trials. The method is capable of finding patterns that are not apparent to the human eye, but after the spike trains have been clustered and reordered the multiple patterns are obvious and highly statistically significant. When applied to recordings from cortical slices, we identified specific windows of time, a few hundred milliseconds long, during which multiple spike-time patterns were possible, but at other times only unique patterns were found.
We also applied the new clustering methods to in vivo data from cats and monkeys from other laboratories and were surprised to find that these too contained multiple spike patterns. In the monkey area MT, the time windows were a few hundred milliseconds long, but in the cat LGN the patterns lasted for many seconds. One implication of these findings is that cortical neurons are even more reliable than previously thought.
A cortical neuron receives on the order of 10,000 synaptic inputs, most of which are from other cortical neurons and only a small fraction of which are active at any one time. Although a stimulus waveform is locked to the stimulus onset, the phase of oscillations is set internally and is therefore typically not connected to the stimulus onset. The precision of firing therefore reflects a balance between intrinsic reverberations (including oscillations) and stimulus properties. In general there will be interaction between these two types of input. Depending on the nature of the interaction, stimulus locking can still be obtained. For instance, stimulus locking persists when the oscillatory inputs change the number of spikes that a neuron produces in response to the stimulus-related inputs.
A pyramidal cell has access to inputs from several cortical layers and must integrate these and produce one output spike train. Each dendritic branch integrates its input independently through a local nonlinearity. This suggests a two step process: First, synchrony decoding occurs in the dendritic branches, and then global integration with the inputs from other dendritic branches at the soma follows. There are multiple ways to generate action potentials. Some methods lead to precise spikes that are conducive to generating synchronous volleys; others are more appropriate for the propagation of firing rate modulations. The state of neuromodulation may shift the mode of spiking between spike-rate coding and spike-time coding.
Inhibitory Regulation of Spike Timing
Precise inhibition generated by fast cortical oscillations can gate and modulate spike trains. Cortical basket cells make synapses near the soma of pyramidal cells. In the hippocampus, the spike of one basket cell can synchronize the activity of a large number of pyramidal cells. Inhibitory cells are involved in the generation of fast oscillations, especially those in the gamma frequency range. Consistent with this role, for in vivo recordings the power spectrum of currents generated by inhibitory synapses has more power in the gamma frequency range than the power spectrum of currents generated by excitatory synapses. Interneurons can be transiently synchronized through "synchrony by competition," in which a top-down projection depolarizes a subset of interneurons, increasing their firing rate and synchrony. This reduces the firing rate of the remaining interneurons and synchronizes the "winners." In this way, selective attention could increase the firing rate of a subset of inhibitory neurons.
When an interneuron network is in a synchronized oscillation, a postsynaptic neuron receives volleys of synchronized inhibitory inputs from the network, which could act as a gate, preventing spiking when the network is asynchronous and allowing spiking when the network is synchronous. In either case, the neuron produces spikes that are precisely timed with respect to the inhibitory rhythm. Only excitatory inputs that arrive during the period when the inhibitory conductance is low can be transmitted into an output spike, which is locked to the local inhibitory rhythm. Using this mechanism, the specific path that feedforward information follows along multiple groups could be rapidly altered to achieve behavioral goals. Thus, inhibitory interneurons may have a dynamic role for regulating the flow of information in the cortex by controlling spike timing and synchrony in cortical circuits.
One major conclusion is that we need to reconsider current views on the reliability of cortical neurons, since we have demonstrated that they are capable of highly reliable spike responses with millisecond precision. The spike timing could be used as an internal variable in the cortex to represent the importance of a stimulus. Spike times could also have consequences for how memories are stored in the cortex, since the relative spike timing of cortical neurons also controls synaptic plasticity.
