Interactions among neurons in brain circuits underlie sensory perception and information storage. Work in locusts shows how the timing of different neuronal signals is synchronized to ensure effective communication.
Most biological systems can adapt to different conditions and environments. The nervous system has elaborated on this ability and developed mechanisms that use prior experience to predict future events. Many of these mechanisms could potentially support behavioural prediction. However, little is known about which specific mechanisms are used during common tasks, such as learning how to hit a baseball or remembering to avoid poison ivy. In a seminal study, Cassenaer and Laurent1 (page 709 of this issue) demonstrate a specific predictive mechanism that operates during olfactory learning in locusts.
In both mammals and insects, olfactory stimuli trigger diffuse, but reproducible, patterns of neural activity in many interconnected brain regions2. At the initial processing stage, odorants in the environment evoke all-or-none electrical discharges, which are recorded in neurons as spikes (action potentials). As the cells involved in the odorant-to-spiking conversion have only broad selectivity3, the activity of any one neuron is a poor predictor of odorant identity. Instead, odorant identity seems to be encoded by populations of neurons whose activity becomes transiently synchronized in response to sensory stimulation. Individual neurons often respond to several odorants and probably participate in many transient 'cell assemblies'2. The insect brain affords excellent accessibility for electrical recordings from several neurons, making it useful for determining how odorant-evoked activity patterns develop.
Network oscillations also have an important role in the processing of olfactory information by linking together the neurons that collectively represent a specific odorant. The presence or absence of a single spike on a specific oscillation cycle defines cell assemblies that are activated by an odorant. In honeybees4 the disruption of network oscillations impairs olfactory discrimination, highlighting the oscillations' relevance to information processing.
Olfactory information is processed sequentially by different brain regions that are linked by network oscillations. In insects, simple olfactory stimuli activate large subsets of projection neurons in the antennal lobe — a region analogous to the olfactory bulb in mammals. The neural representation of sensory information becomes significantly sparser in the second5- and third1-order stages of olfactory processing (Fig. 1a). Sparse coding is advantageous because it facilitates the recall of memories from partial cues and allows for denser, more reliable storage of biological information6.
As several stages of the insect olfactory system represent sensory information as sparse cell assemblies that are tightly linked by network oscillations2, small perturbations in the timing of single spikes in constituent neurons can potentially disrupt sensory coding. Thus, the olfactory system has the difficult task of maintaining the temporal precision required to generate odorant-linked cell assemblies. Propagation of information through neuronal circuits typically results in a loss of temporal precision due to randomness associated with the mechanisms by which neurons communicate. The decrease in timing precision (increased jitter) is especially large when target cells are activated by relatively few, but powerful, inputs. Downstream brain regions must, therefore, maintain temporal precision without the advantage of reducing jitter by averaging over many inputs.
Cassenaer and Laurent1 found that the insect olfactory system uses information channels comprised of relatively few, but powerful, connections and, surprisingly, that it avoids the temporal-precision penalty normally associated with sparse-coding pathways. In contrast to the weak connections between projection neurons and Kenyon cells7, the authors found that Kenyon-cell spiking causes strong activation of β-lobe neurons. The amplitude of the connections from Kenyon cells to β-lobe neurons was almost 20 times larger than that from projection neurons to Kenyon cells. However, inputs from Kenyon cells to β-lobe neurons were relatively rare, occurring 25 times less frequently than weak connections between projection neurons and Kenyon cells. Both the large amplitude and low incidence of these connections are characteristic of sparse-coding pathways8. Despite the sparse nature of their main inputs, β-lobe neurons still fired with high temporal precision.
The authors went on to identify a mechanism for circuit modification that explains how infrequent, large inputs from Kenyon cells generate precise timing in the downstream β-lobe neurons (Fig. 1b). In most brain circuits, the strength of connections between neurons is relatively constant. However, dramatic changes in strength can occur after a 'coincident' event in which both the initiating and target cells are active at roughly the same time. The ability of some neural circuits to modify themselves after detecting coincident activity — termed spike-timing-dependent plasticity (STDP) — is one of the best-understood mechanisms for regulating connection strength9.
Cassenaer and Laurent1 report the first example of STDP in the olfactory system, showing for the first time that STDP facilitates oscillation-linked spike synchronization. Studying the sparse connections between Kenyon cells and β-lobe neurons, the authors find — as shown previously in mammals9 — that STDP can bidirectionally regulate connection strength. The strengthening or weakening of connections depends on the sequence in which initiating and target cells are active. If an inappropriately timed input causes a target cell to spike late, this circuit-modification mechanism will strengthen that specific connection so that future spikes occur earlier, thereby preserving spike synchrony across the cell assembly. A parallel mechanism weakens the subset of connections that are active after a premature target-cell spike, thereby altering the aggregate drive to the target cell so that it fires at the appropriate time on subsequent oscillation cycles. The bidirectional nature of STDP provides an elegant means to avoid the expected loss of temporal precision as information propagates through oscillating sparse networks.
Besides providing transient representations of sensory information, cell assemblies may also serve as the initial substrates for long-term memory. An outstanding question is whether the same mechanisms that mushroom-body neurons use to generate precise timing also trigger long-term circuit modifications that underlie sensory memory — extending the neuroscience axiom that “neurons that fire together, wire together” to transient cell assemblies. This study1 will probably also spark a search for STDP in 'secondary' olfactory regions of mammals — including many of the relatively unexplored olfactory areas, such as the olfactory tubercle and agranular insula cortex — that may be analogous to assemblies of β-lobe neurons.
As synchronous oscillations can be recorded in many diverse mammalian brain regions that exhibit STDP, the model presented by Cassenaer and Laurent1 may be broadly applicable. Subsequent studies will probably test whether neuronal assemblies with self-correcting timing underlie predictive memory in other behavioural contexts, as described by theoretical models9,10.
About this article
IEEE Transactions on Autonomous Mental Development (2009)