Sensory encoding and memory in the mushroom body: signals, noise, and variability

Abstract

To survive in changing environments, animals need to learn to associate specific sensory stimuli with positive or negative valence. How do they form stimulus-specific memories to distinguish between positively/negatively associated stimuli and other irrelevant stimuli? Solving this task is one of the functions of the mushroom body, the associative memory center in insect brains. Here we summarize recent work on sensory encoding and memory in the Drosophila mushroom body, highlighting general principles such as pattern separation, sparse coding, noise and variability, coincidence detection, and spatially localized neuromodulation, and placing the mushroom body in comparative perspective with mammalian memory systems.

Figure 1.

Diagram of the Drosophila olfactory system. Odors activate olfactory receptor neurons (ORNs). ORNs signal to matching projection neurons (PNs; indicated by matching colors: brown, purple, and green) via glomeruli in the antennal lobe; these signals are transformed by local neurons (LNs). PNs project to the mushroom body for flexible behavior and to the lateral horn for innate behavior. In the mushroom body, PNs activate Kenyon cells (KCs), which respond sparsely to odors. Each KC sends axons through different compartments. In compartments for appetitive memory, KCs get local neuromodulatory input from reward-encoding dopaminergic neurons (DANs) and send output to mushroom body output neurons (MBONs) that trigger avoidance behavior (–). Conversely, in compartments for aversive memory, KCs get input from punishment-encoding DANs and send output to MBONs that trigger approach behavior (+). Learning occurs by depressing outputs to the “wrong” action.

Figure 2.

Expansion recoding aids pattern separation. (A) Diagram of expansion recoding circuitry. In expansion layer circuits, a small number of input layer neurons (purple) synapse onto a much larger number of expansion layer neurons (gray), which in turn converge onto a small number of output layer neurons (green). This basic architecture is used by both the insect mushroom body and the vertebrate cerebellum. (B) Why expansion recoding aids discrimination. Each dot represents the neuronal response to a single exposure to an odor in “activity space,” where each dimension is one neuron's activity. Due to noise, different trials elicit slightly different responses, represented by the cluster of dots of the same color. Here, in a two-dimensional space, the clusters for odor 1 and odor 2 overlap, making it difficult to cleanly discriminate between the two odors by drawing a line between the two clusters. In contrast, in three-dimensional space, the two clusters can be separated by a plane. The same principle applies if the red dots are rewarded odors and the blue dots are punished odors. Adapted from Cayco-Gajic and Silver (2019).

Figure 3.

Local neuromodulation in the mushroom body. Illustration of examples of local neuromodulation. Dopaminergic PPL2 neurons (purple) innervate KC dendrites and enhance KC responses to the odor that was paired with the unconditioned stimulus (the CS⁺) but not the unpaired odor (the CS⁻). The GABAergic APL neuron (brown) innervates the whole mushroom body, but because it shows localized activity, it may have different functions on KC dendrites (sparsening KC odor responses) versus KC axons (perhaps gating synaptic plasticity). KC–KC synapses (black) implement axonal lateral inhibition via mAChR-B and thereby prevent flies from erroneously showing learned responses to the unpaired odor (CS⁻). The DPM neuron (turquoise) releases serotonin (5-HT) locally on KC axons and regulates the coincidence time window; i.e., the largest interstimulus interval (ISI; gap in time between the CS and US) that still allows flies to learn.