Anti-memories through compartmentalised activity in a single neuron in a Drosophila memory centre

Most neurons communicate through self-regenerating signals. They take in and sum up signals in the input part of the neuron, and if these inputs surpass a certain threshold, they send a self-regenerating electrical impulse to the output part of the neuron, which releases chemicals to signal to other neurons. But in many neurons, activity is 'compartmentalised': activating the neuron in one part makes it release output signals only in that one part, not other parts, because electrical activity does not spread readily to other parts of the neuron. Why do they do this? In most cases, we don't know: there are few examples where the behavioural function of compartmentalised neuronal activity is clearly understood.

We address this gap by studying olfactory memory in fruit flies. Flies learn to avoid punished odours and approach rewarded odours. These memories are suppressed by a neuron in the fly brain called 'APL', which inhibits memory-storing neurons called Kenyon cells (KCs). The explanation isn't as simple as 'APL inhibits memories because APL inhibits KCs.' Rather, we propose that APL suppresses learning because its activity is compartmentalised.

This is because behavioural responses to odours are controlled by the balance between two opposing types of neurons, which make the fly either approach or avoid the odour. These 'approach' and 'avoidance' neurons, called 'mushroom body output neurons' (MBONs), are activated by KCs, which are activated by odours. Odour+punishment training weakens connections from KCs onto approach (but not avoidance) MBONs, so that avoidance dominates and flies avoid the punished odour. Similarly, odour+reward weakens KC->avoidance connections. Thus, learned behaviour is determined by the balance in MBON signalling, not the total output of KCs.

Therefore, APL can only suppress memory if it lessens the imbalance between odour-evoked activity in approach and avoidance MBONs. To do this, it must inhibit some KC->MBON connections more strongly than others: e.g., suppress punishment memories by inhibiting KC->avoidance connections more than KC->approach connections. To achieve this:

(1) APL activity (and thus its inhibitory output) must be compartmentalised. This would allow the single neuron APL to differentially inhibit KC->approach and KC->avoidance connections because they are in different spatial 'zones'.

(2) APL must have different activity in approach vs. avoidance zones. APL's activity is controlled by KCs, so this would occur if KC->APL connections are modified by learning in the same way as KC->MBON connections in the same zone: e.g., odour+punishment selectively weakens KC->APL connections in the 'approach' zone. In this scenario, because APL inhibits KCs locally, after learning APL would inhibit KCs (and thereby MBONs) more strongly in the avoidance zone than in the approach zone. This would lessen the imbalance in MBON activity induced by odour+punishment (avoidance greater than approach).

Thus, if APL activity is compartmentalised, learning could simultaneously induce synaptic modifications that support memory (KC->MBON connections) and modifications that oppose memory (local KC->APL connections). We call the latter 'anti-memories' because they are an active change that acts as a 'mirror opposite' to memory, rather than passive decay. Such anti-memories might gate memory formation or expression, and would provide the first clear cognitive function for compartmentalised activity.

Our preliminary data has already confirmed some of the predictions above. We will test the rest using brain imaging: we will take an engineered protein that lights up when neurons are active, put the protein in KCs, APL or MBONs, and image the whole volume of the neurons of interest while the fly smells odours. We will do this before and after training, or while manipulating activity in small areas of the neural circuit, and we will analyse how activity differs in different parts of APL.

Many neurons do not fire action potentials, meaning that electrical activity is spatially localised in the neuron: synaptic inputs lead directly to transmitter release locally where the inputs arrive, not in other parts of the neuron. Why? Few examples exist where this neuronal compartmentalisation has a clear behavioural function.

We propose a novel hypothesis to link compartmentalisation to behaviour: that in the Drosophila brain, compartmentalised activity in an inhibitory interneuron, 'APL', allows localised synaptic modifications to oppose memory formation. Learning modifies (usually weakens) synapses between odour-responsive Kenyon cells (KCs) and behaviour-controlling mushroom body output neurons (MBONs). MBONs that control opposing behaviours (approach vs. avoid the odour) occupy segregated 'MBON zones' along KC axons. Together with these segregated 'MBON zones', the fact that APL provides local feedback inhibition to KCs could explain why APL suppresses learning.

APL would suppress learning if APL locally inhibits MBONs (via KCs) and KC->APL plasticity in each 'MBON zone' mirrors KC->MBON plasticity in that zone. The latter means that learning causes both APL odour responses and APL's inhibitory output strength in each zone to change in the same direction as MBON odour responses in that zone (but not other zones). Because APL inhibits KCs, this KC->APL plasticity would oppose KC->MBON plasticity, and if this opposition is local to each zone, it would lessen the learning-induced imbalance in MBON activity that underlies olfactory memory. (This logic is explained in more detail in the main summary.) We will test these predictions using two-photon volume imaging of APL and MBONs, using genetically encoded calcium indicators. We will record neural activity while presenting odours to the fly, and/or locally activating or inhibiting APL or dopaminergic neurons by ectopically expressing receptors for chemicals that we locally apply to specific MBON zones.

Training
The postdoc employed on this project will benefit by learning both technical skills (two-photon microscopy, scientific programming) and transferable skills (project management, presentation, writing). Students (undergraduate, MSc and PhD) in the lab will also learn similar skills. Having gained valuable skills, these individuals will go on to contribute to the UK's skilled workforce and knowledge economy.

Wider public
We will actively engage with the public to generate interest in our research and neuroscience in general. For example, we will publicise our results to general audiences through the media and the lab and university websites, and organise public engagement events where visitors can experience the excitement of science through hands-on exhibits (e.g., through university events like Festival of the Mind, Researchers' Night, Discovery Night, etc.). In recent years we have run very successful events about sensory neuroscience and about the use of Drosophila as a model organism in biology, and we will continue to run and further improve these events. In this way, members of the public will enhance their appreciation of the excitement, beauty and elegance of scientific research.

Artificial intelligence
AI engineers often seek to build intelligent systems that work with imperfect inputs and minimal or unreliable hardware (e.g., miniature autonomous flying robots). The simple nervous system of the fruit fly has solved this problem. Thus, understanding how the fly brain uses subcellular dynamics within a neuron for network computations may help engineers improve the capabilities of minimalistic artificial brains. We will explore these potential benefits via collaborations with computer scientists and technology companies.

Clinical impact
Understanding basic principles of how memory works will help us understand what goes wrong when memory is impaired. Thus, this research may eventually lead to improved treatment, diagnosis, or prevention for dementia or cognitive disabilities, potentially benefitting patients and their families.