Andrew C. Lin

Neural networks maintain stable activity levels by compensating for perturbations through homeostatic plasticity. However, homeostatic mechanisms operating at different levels may conflict with each other. For example, in inhibitory feedback circuits, if inhibitory neurons receive excess excitation, compensation at a ‘local’ level (e.g. reducing inhibitory neurons’ activity) could conflict with ‘network‐level’ compensation (e.g. suppressing the excitatory neurons responsible for overexciting the inhibitory neurons). We studied this problem in the Drosophila mushroom body, where excitatory Kenyon cells (KCs) receive feedback inhibition from the anterior paired lateral (APL) neuron. Dual‐colour calcium imaging revealed that prolonged (24 h) artificial activation of KCs causes APL to become less sensitive to KC activity. Meanwhile, KCs compensate for their excess activity by reducing excitation, yet this change is opposed by reduced inhibition from APL. This conflict meant that KCs did not consistently show the expected homeostatic reduction in odour responses. Our findings show that neurons sometimes adapt their activity locally in a way that counteracts broader adaptations in the network. image Key points: Neural networks maintain stable activity levels through homeostatic plasticity – but what physiological variables are stabilised? In inhibitory feedback circuits, local and network‐level compensation might conflict. For example, if excitatory neurons are overactive, they might compensate by becoming less excitable. But if inhibitory neurons compensate for the excess excitation by also becoming less excitable, this would decrease inhibition onto the excitatory neurons and increase their activity. We tested this idea in the fruit fly brain, where excitatory Kenyon cells (KCs) get negative feedback from an inhibitory neuron called anterior paired lateral (APL). After overactivation of KCs, APL becomes less sensitive to KCs. The resulting loss of inhibition onto KCs counteracts KCs’ attempts to reduce their activity. These results show that adaptation at local and network levels can conflict with each other.

Neurons stabilize their physiological properties in part by homeostatic compensation between different ion channel conductances. However, little is known about this process in the central brain in vivo. We studied this problem in Kenyon cells, the third‐order olfactory neurons in the fruit fly Drosophila that store olfactory associative memories. We investigated whether Kenyon cells regulate their excitability homeostatically by testing how their activity is affected by ectopic expression of the bacterial voltage‐gated sodium channel NaChBac, a manipulation previously reported to increase neuronal excitability in other systems. Surprisingly, NaChBac expression decreases Kenyon cell excitability. Whether expressed constitutively (throughout development) or only in the adult, NaChBac expression in Kenyon cells suppresses Kenyon cell spiking, reduces odour‐evoked calcium influx, and prevents olfactory aversive conditioning. However, odour‐evoked calcium influx in Kenyon cell axons and dendrites (but not somata) is normal after 4‐day adult‐only NaChBac expression (but not 2‐day adult‐only or constitutive expression), suggesting limited homeostatic regulation of calcium influx that is prevented by developmental NaChBac expression. NaChBac expression also decreases expression of endogenous voltage‐gated sodium channels (Para) in the spike initiation zone, suggesting homeostatic regulation of sodium influx. Indeed, a compartmental model best fits the data when the exogenous NaChBac conductance is accompanied by a decrease in endogenous sodium conductance. These results suggest that manipulating neuronal activity with ion channels can have unexpected effects depending on compensatory plasticity. image Key points: Neurons stabilize their physiological properties through compensation between key ion channels, but little is known about this process in the central brain in vivo. Here we test homeostatic compensation in vivo in Drosophila Kenyon cells, the neurons that store olfactory associative memories, by ectopically expressing NaChBac, a bacterial voltage‐gated sodium channel commonly used to increase neuronal excitability. Surprisingly, NaChBac expression in Kenyon cells decreases their excitability: it reduces expression of endogenous sodium channels, prevents spiking, reduces odour‐evoked calcium influx, and impairs learning. The electrophysiological phenotype is reproduced in a compartmental model. However, odour‐evoked dendritic/axonal calcium influx (but not spiking or learning) returns to normal if NaChBac is expressed for 4 days only in adults (not during development), suggesting limited homeostatic regulation of voltage‐gated calcium influx. These results show that manipulating ion channels can have unexpected effects depending on homeostatic compensation.

Romantic engagement can bias sensory perception. This ‘love blindness’ reflects a common behavioural principle across organisms: favouring pursuit of a coveted reward over potential risks¹. In the case of animal courtship, such sensory biases may support reproductive success but can also expose individuals to danger, such as predation^2,3. However, how neural networks balance the trade-off between risk and reward is unknown. Here we discover a dopamine-governed filter mechanism in male Drosophila that reduces threat perception as courtship progresses. We show that during early courtship stages, threat-activated visual neurons inhibit central courtship nodes via specific serotonergic neurons. This serotonergic inhibition prompts flies to abort courtship when they see imminent danger. However, as flies advance in the courtship process, the dopaminergic filter system reduces visual threat responses, shifting the balance from survival to mating. By recording neural activity from males as they approach mating, we demonstrate that progress in courtship is registered as dopaminergic activity levels ramping up. This dopamine signalling inhibits the visual threat detection pathway via Dop2R receptors, allowing male flies to focus on courtship when they are close to copulation. Thus, dopamine signalling biases sensory perception based on perceived goal proximity, to prioritize between competing behaviours.

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 theDrosophilamushroom 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.

Romantic engagement can bias sensory perception. This ’love blindness’ reflects a common behavioral principle across organisms: favoring pursuit of a coveted reward over potential risks. In the case of animal courtship, such sensory biases may support reproductive success but can also expose individuals to danger, such as predation. How do neural networks balance the trade-off between risk and reward? Here, we discover a dopamine-governed filter mechanism in male Drosophila that reduces threat perception as courtship progresses. We show that during early courtship stages, threat-activated visual neurons inhibit central courtship nodes via specific serotonergic neurons. This serotonergic inhibition prompts flies to abort courtship when they see imminent danger. However, as flies advance in the courtship process, the dopaminergic filter system reduces visual threat responses, shifting the balance from survival to mating. By recording neural activity from males as they approach mating, we demonstrate that progress in courtship is registered as dopaminergic activity levels ramping up. This dopamine signaling inhibits the visual threat detection pathway via Dop2R receptors, allowing male flies to focus on courtship when they are close to copulation. Thus, dopamine signaling biases sensory perception based on perceived goal proximity, in order to prioritize between competing behaviors.

Significance: How does variability between neurons affect neural circuit function? How might neurons behave similarly despite having different underlying features? We addressed these questions in neurons called Kenyon cells, which store olfactory memories in flies. Kenyon cells differ among themselves in key features that affect how active they are, and in a model of the fly’s memory circuit, adding this interneuronal variability made the model fly worse at learning the values of multiple odors. However, memory performance was rescued if compensation between the variable underlying features allowed Kenyon cells to be equally active on average, and we found the hypothesized compensatory variability in real Kenyon cells’ anatomy. This work reveals the existence and computational benefits of compensatory variability in neural networks.

``Sparse'' neural networks, in which relatively few neurons or connections are active, are common in both machine learning and neuroscience. While, in machine learning, ``sparsity'' is related to a penalty term that leads to some connecting weights becoming small or zero, in biological brains, sparsity is often created when high spiking thresholds prevent neuronal activity. Here, we introduce sparsity into a reservoir computing network via neuron-specific learnable thresholds of activity, allowing neurons with low thresholds to contribute to decision-making but suppressing information from neurons with high thresholds. This approach, which we term ``SpaRCe,'' optimizes the sparsity level of the reservoir without affecting the reservoir dynamics. The read-out weights and the thresholds are learned by an online gradient rule that minimizes an error function on the outputs of the network. Threshold learning occurs by the balance of two opposing forces: reducing interneuronal correlations in the reservoir by deactivating redundant neurons, while increasing the activity of neurons participating in correct decisions. We test SpaRCe on classification problems and find that threshold learning improves performance compared to standard reservoir computing. SpaRCe alleviates the problem of catastrophic forgetting, a problem most evident in standard echo state networks (ESNs) and recurrent neural networks in general, due to increasing the number of task-specialized neurons that are included in the network decisions.

Scientific studies often require assessment of similarity between ordered sets of values. Each set, containing one value for every dimension or class of data, can be conveniently represented as a vector. The commonly used metrics for vector similarity include angle-based metrics, such as cosine similarity or Pearson correlation, which compare the relative patterns of values, and distance-based metrics, such as the Euclidean distance, which compare the magnitudes of values. Here we evaluate a newly proposed metric, pairwise relative distance (PRED), which considers both relative patterns and magnitudes to provide a single measure of vector similarity. PRED essentially reveals whether the vectors are so similar that their values across the classes are separable. By comparing PRED to other common metrics in a variety of applications, we show that PRED provides a stable chance level irrespective of the number of classes, is invariant to global translation and scaling operations on data, has high dynamic range and low variability in handling noisy data, and can handle multi-dimensional data, as in the case of vectors containing temporal or population responses for each class. We also found that PRED can be adapted to function as a reliable metric of class separability even for datasets that lack the vector structure and simply contain multiple values for each class.

Echo state networks (ESNs) are a powerful form of reservoir computing that only require training of linear output weights whilst the internal reservoir is formed of fixed randomly connected neurons. With a correctly scaled connectivity matrix, the neurons' activity exhibits the echo-state property and responds to the input dynamics with certain timescales. Tuning the timescales of the network can be necessary for treating certain tasks, and some environments require multiple timescales for an efficient representation. Here we explore the timescales in hierarchical ESNs, where the reservoir is partitioned into two smaller linked reservoirs with distinct properties. Over three different tasks (NARMA10, a reconstruction task in a volatile environment, and psMNIST), we show that by selecting the hyper-parameters of each partition such that they focus on different timescales, we achieve a significant performance improvement over a single ESN. Through a linear analysis, and under the assumption that the timescales of the first partition are much shorter than the second's (typically corresponding to optimal operating conditions), we interpret the feedforward coupling of the partitions in terms of an effective representation of the input signal, provided by the first partition to the second, whereby the instantaneous input signal is expanded into a weighted combination of its time derivatives. Furthermore, we propose a data-driven approach to optimise the hyper-parameters through a gradient descent optimisation method that is an online approximation of backpropagation through time. We demonstrate the application of the online learning rule across all the tasks considered.