The discovery that certain neurons' odour responses differ between individual fruit flies, but are consistent across the hemispheres of each fly's brain, indicates that sensory processing depends on an individual's experience. See Letter p.258
Organisms process sensory information to learn about their environment and to inform future behaviours. The first layers of sensory processing transform information from the sense organs into sparse and stimulus-specific activity patterns across large populations of neurons in the brain1. But the subsequent processing steps are less well understood. In a paper on Nature's website, Hige et al.2 (page 258 of this issue) address this gap in knowledge by studying the olfactory system of the fruit fly Drosophila melanogaster.
In insects, olfactory sensory neurons project to a processing centre in the brain called the antennal lobe. At this stage, information about different odours is encoded by overlapping but specific patterns of activity across a relatively small population of projection neurons (about 150 in fruit flies). These activity patterns are transmitted to two higher brain regions, one of which is the mushroom body. Here, odour representations are transformed into sparse activity patterns across a much larger number of neurons called Kenyon cells (flies have around 2,000 Kenyon cells; Fig. 1).
The activity patterns evoked in Kenyon cells by different odours show little overlap. Patterns seem to be generated by random connections to projection neurons, so they vary between individuals3,4. It is therefore thought that this expansion of odour representations in the mushroom body supports their subsequent classification, for example during learning1. Indeed, the mushroom body is crucial for learning associations between arbitrary odours and specific values or behaviours. Learning is probably enabled by adjusting the strength of synaptic connections between Kenyon cells and a population of 35 mushroom-body output neurons (MBONs)5,6,7. The low number of MBONs implies that odour representations are recompressed as they are transferred to higher brain regions, but it remains unclear what information these neurons encode, and how they do this.
Fruit-fly MBONs can be genetically separated into 21 types8. Hige et al. expressed a fluorescent calcium-sensor protein in most of these cell types, usually targeting one type per fly. Odour-evoked activity of MBONs was measured using a high-resolution multiphoton microscope to detect changes in fluorescence. The authors averaged calcium signals over time and pooled the results across flies to construct mean tuning curves (responses to a set of odours) for each MBON type. They also constructed 'virtual activity patterns' for the whole MBON population by combining the responses of different MBON types.
As observed in other insects7,9, MBONs were broadly tuned, each responding to multiple odours. Moreover, the mean tuning curves of different MBON types were often similar. Unlike activity patterns across Kenyon cells, virtual activity patterns across MBONs overlapped and were not well suited for fine odour discrimination. These results suggest that MBONs do not encode odour identity with high efficiency or accuracy. Nevertheless, Hige and colleagues demonstrated that representations of innately attractive odours could be reliably discriminated from repulsive ones. This finding is consistent with the hypothesis10 that MBONs can transmit information about certain derived properties of the stimulus, such as its valence (positive or negative value).
When interpreting these experiments, two issues should be considered. First, the authors pooled data from several individuals. This approach provides a representative picture when the response of a neuron is stereotyped between individual flies, as in antennal-lobe projection neurons, or when it varies randomly, as in Kenyon cells3. However, pooling is not ideal when responses vary systematically between individuals. Consider the same word written by different people — despite the idiosyncratic handwriting, each word is legible, but the average is a blur. Second, the calcium-imaging method used by Hige et al. cannot resolve the fine temporal structure of odour-evoked activity. But in locusts, for example, the precise temporal patterning of MBON activity provides a great deal of information about odour identity9.
Hige et al. found that the tuning curves of some MBONs were stereotyped across flies, whereas others were more variable. To explore the source of this variation, they took simultaneous electrical recordings from a Kenyon cell and an α2sc neuron, the MBON type that showed the greatest variation between individuals. The probability of an excitatory synaptic connection between these cells was only around 30%, suggesting that variable tuning of MBONs is a consequence of variation in the connections they form with Kenyon cells.
Although the tuning curves of α2sc neurons differed between individuals, the authors showed that the tuning of α2sc neurons in the two hemispheres of individual flies was nearly identical. Moreover, variability between individuals was abolished in rutabaga mutant flies, which have severe learning deficits. The mutation in the rutabaga gene disrupts the plasticity process that modifies the strength of synaptic connections. These results suggest that the variation in MBON tuning, unlike that of Kenyon-cell tuning, is not simply caused by randomness in the wiring between Kenyon cells and MBONs, but instead reflects a coordinated process that shapes, or 'individualizes', the responses of MBONs in each fly (Fig. 1). Information transfer from Kenyon cells to MBONs might therefore be shaped by experience. Because this process depends on rutabaga, the underlying synaptic-plasticity mechanisms might overlap with those involved in associative learning.
Hige and colleagues' results indicate that MBONs map high-dimensional odour representations in the Kenyon-cell population onto a low-dimensional output. This mapping differs between individuals, perhaps because the precise connectivity between Kenyon cells and MBONs is shaped by experience. Last year, a behavioural study10 of fruit flies showed that stimulating different MBONs had specific appetitive or aversive effects that interacted additively, suggesting that the population-wide activity pattern across MBONs represents valence. Hige and colleagues' results thus indicate that MBONs map representations of odours onto a representation of valence in an experience-dependent fashion. The MBONs then transmit this valence representation to higher brain areas to modulate specific behaviours.
This conclusion is consistent with prevailing views of mushroom-body function1,4 and can now be tested further. For example, it may be predicted that spontaneous behavioural reactions to odours are less variable in rutabaga flies than in wild-type flies because their odour-to-valence mapping is more stereotyped.
As Marcel Proust highlighted11, the emotions associated with an odour depend on an individual's experience. Hige and colleagues have revealed how this may work in fruit flies. Does higher olfactory processing in vertebrates follow a similar scheme? A major target of the olfactory bulb (the vertebrate equivalent of the antennal lobe) is the piriform cortex, in which odours are represented by distributed activity patterns across many neurons. Unlike in the mushroom body, however, the output of the piriform cortex is not funnelled through a small neuronal population, suggesting that higher olfactory processing in vertebrates is more complex than in fruit flies. This notion is consistent with the richness of our olfactory perception and its powerful links to specific memories that we experience every day.
About this article
Pesticides Drive Stochastic Changes in the Chemoreception and Neurotransmission System of Marine Ectoparasites
International Journal of Molecular Sciences (2016)