The number of spike patterns expressible by even a modest number of neurons
is astronomical, particularly if precise spike times, as well as firing
rates, are taken into account. In auditory cortex, neural activity can be
triggered by sound stimuli, but may also occur spontaneously during silence.
We found that both sensory-evoked and spontaneous activity patterns are
intricately structured but highly and similarly constrained. Individual
neurons responded to stimuli with consistent temporal dynamics, revealing a
stereotyped spread of activity across the recorded population. The temporal
sequence of neurons firing was quantitatively similar for both spontaneous
and evoked events, as well as across different acoustic stimuli, suggesting
that the primary difference between responses consisted of the combination
of neurons participating in an event, rather than their sequential order.
The set of possible neural combinations activated by presentations of any
single stimulus was confined to a subset of the set of all possible
combinations; the subspaces corresponding to individual stimuli lay within a
larger but still constrained realm outlined by the set of spontaneous
events. To investigate how such constraints could arise in neural circuits,
we constructed a recurrently connected network simulation of diverse
excitatory and inhibitory units, which produced similar behavior. Although
constraints on spike patterns are inefficient from the point of view of
energy efficient coding, we suggest they are an inevitable consequence of
information processing in recurrent neuronal networks.