Working memory is the brain’s short-term scratchpad, the system that lets a person hold a few digits of a security code in mind for the seconds needed to type them. It is a remarkable balancing act. The memory must be stable enough to survive distractions and persist over a useful interval, yet it must also be rapidly deployable and just as rapidly discarded, because a brain that permanently memorized every number or smell it encountered would waste precious energy on useless information. Understanding how neural tissue achieves this dual demand of stability and flexibility has been a central puzzle in neuroscience, and the new fly study offers a concrete, experimentally verified answer in a system small enough to watch cell by cell.

The fruit fly has become one of neuroscience’s most powerful model organisms for precisely this kind of question. Although the human brain contains tens of billions of neurons and the fly brain fewer than 200,000, the two are organized in strikingly similar ways, with comparable principles governing how neurons communicate and how circuits are wired together. More importantly for researchers, every connection between every neuron in the fly brain, the complete wiring diagram known as the connectome, has been fully mapped. That means scientists can directly trace how individual neuron types interact to produce specific behaviors, rather than inferring those interactions indirectly as is often necessary in larger brains.

In the new experiments, the researchers exposed flies to a heady whiff of apple cider vinegar and watched what happened in the insects’ brains. The flies traveled toward the odor and continued moving toward its remembered location for a few seconds even after the smell itself disappeared, a behavior that depends on some internal record of where the odor had been. By monitoring brain activity during this process, the team found that two different types of neurons responded to the smell with similar patterns of electrical activity, suggesting that the cells were working together to control the flies’ movement in response to the odor. Those two cell types, known as PFG and hΔK neurons, turned out to be the key players in the memory mechanism.

What the researchers discovered is that PFG and hΔK neurons form what is called an attractor network, a type of neural circuit in which a set of neurons effectively talks to itself, with activity circulating among the cells until a stable signal emerges. Attractor networks have long been proposed as the theoretical substrate for working memory, because a self-sustaining loop of activity could hold information online after the original stimulus is gone. Scientists have assumed that an arrangement like this powers working memory, but the new experiments confirm that such an arrangement exists in a real brain and show exactly how it operates in a specific behavioral context, namely a fly navigating toward a remembered smell.

The twist is that this attractor network is not continuously active. Most of the time, communication between the PFG and hΔK neurons is blocked, with hΔK activity suppressed. During those blocked periods, the PFG neurons perform a different job: they track the fly’s orientation in space by receiving information from the fly’s internal compass system, continuously updating the animal’s heading. The circuit is therefore not a simple memory device but a versatile one whose function changes depending on whether the gate between its two components is open or closed.

When the block is lifted and the hΔK and PFG neurons are allowed to communicate, the fly becomes able to lock in on a particular origin, such as the source of an odor, and then move toward it. In that moment, the content of the memory, carried by the PFG neurons, is combined with a timing signal controlled by the hΔK neurons, and the communication block acts as a gate deciding when the memory is written. The researchers call this arrangement a split attractor network, a design in which the jobs of storing content, controlling timing, and gating access are divided among distinct elements of the same circuit. That division of labor, the study suggests, is what provides both the stability and the flexibility that working memory requires.

Senior investigator Katherine Nagel, PhD, an associate professor in the Department of Neuroscience at NYU Grossman School of Medicine, framed the finding as a confirmation of a long-standing hypothesis. According to Dr. Nagel, the study shows how a neuronal circuit forms a short-term memory in response to a fly sensing an odor, enabling the animal to remember a direction and travel toward a smell it wants to remember. She noted that scientists have long assumed an arrangement like this powers working memory, and that the experiments confirm this arrangement exists and reveal how it works in a specific context. Her hope, she said, is that the fly, which has an amazing track record for revealing how human biology works in a clear and simple way, will provide insight into processes like working memory that researchers have not yet had the tools to study in depth.

The study arrives at a moment when mapping specific network functions has become one of the frontiers of neuroscience. With complete connectomes now available for the fly and increasingly detailed wiring data emerging for other animals, researchers can move beyond asking whether particular brain regions are active during a task and begin asking what specific networks are actually doing, cell by cell and connection by connection. Dr. Nagel’s laboratory plans to pursue several questions raised by the current work, including how the circuit is controlled across different time frames, what kinds of information other types of neurons in the system are tracking, and how and why different regions of the brain are capable of controlling similar functions simultaneously. Answers to those questions could clarify whether split attractor designs are a general strategy that brains use to manage short-term memories of many kinds.

Because the fly brain and the human brain share deep organizational similarities, the mechanics uncovered in this study may illuminate principles that apply far beyond a vinegar-seeking insect. If the same logic of gated, split attractor circuits governs how mammalian brains write and erase short-term memories, the finding could eventually inform research into conditions in which working memory falters. For now, the study stands as a vivid demonstration of how a small, fully mapped brain can resolve a question that has hovered over neuroscience for decades: the memory that guides a fly toward a vanished smell is not a vague impression but a precisely engineered circuit state, written at the moment a gate opens, held stable by self-reinforcing activity, and ready to steer the animal until it is no longer needed. The work was supported by National Institutes of Health grants R01NS127129 and R01DC017979 and by National Science Foundation grant 2014217, with additional contributions from researchers including Aaron J. Lanz, Nicholas D. Kathman, and Emily Hao at NYU Langone and Bard Ermentrout at the University of Pittsburgh.

Subject of Research: Neural circuit mechanisms of working memory in the fruit fly brain

Article Title: ​​​​​​​Fruit fly study reveals mechanics of working memory

Article References: ​​​​​​​Fruit fly study reveals mechanics of working memory. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: working memory, fruit fly, neuroscience, attractor network, connectome, PFG neurons, hΔK neurons, odor navigation, NYU Langone, Nature, neural circuits, short-term memory

Most insects, including flies, see the world through compound eyes composed of hundreds of tiny lenses. Behind each lens sits a cluster of eight light-detecting cells, called photoreceptors, arranged in a pattern so fundamental that scientists believe it has been conserved since before the diversification of the major insect groups. Butterflies are the rare exception to this rule. They carry nine photoreceptors per eye unit, an addition that fundamentally changes how much color information their visual systems can gather. According to Michael Perry, an associate professor in the Department of Cell and Developmental Biology at UC San Diego’s School of Biological Sciences, butterflies see far more color than flies because at some point in their evolution they added a photoreceptor to every unit of their compound eye, a rare break from an eye design otherwise conserved across insects for hundreds of millions of years. His team, he said, found the genetic switch that did it.

The study focused on the painted lady butterfly, one of the most widespread butterfly species on Earth and a familiar visitor to gardens across multiple continents. In flies, the light-sensing cells are known as photoreceptors R1 through R8. The researchers discovered that butterflies expanded their color vision by adding a second R7 photoreceptor to each unit of the eye. This duplication is not merely a matter of having more cells; it means that every single sampling unit of the butterfly’s compound eye carries an additional channel of color information, effectively upgrading the resolution of the insect’s perception of the visual world at every point where light is captured.

To determine whether this genetic change was sufficient on its own to produce the butterfly eye plan, the team performed a striking experiment. They recreated the modification in a fruit fly, an insect that normally follows the standard eight-cell design. By switching on a gene in cells where it is normally kept silent, and timing that activation to the brief developmental window during which the eye is being built, they produced what the researchers describe as a butterfly fly. This engineered insect grows its eyes according to the butterfly blueprint, with nine cells per eye unit instead of the usual eight. The result demonstrates that a single, targeted genetic change can redirect the construction of an entire sensory organ toward a fundamentally different architecture.

An additional light detector, however, would be useless without a brain capable of interpreting its signals. The prevailing assumption in sensory biology has been that when a new sensory input evolves, the brain must gradually evolve a matching neural partner to receive the information on the other end. Evolution, in this view, would need to coordinate changes on both sides of the connection: the eye producing new signals and the brain developing new circuitry to process them. The UC San Diego team expected to find evidence of such a slow, coupled adaptation in butterflies. What they found instead surprised them. No adaptive change in the brain was needed at all.

The explanation lies in a remarkable feature of how insect brains develop. The fly brain routinely overproduces neurons during development, and those surplus cells die off if they fail to find a connection. This built-in redundancy creates a population of standby neurons that normally have no function and are pruned away. In the butterfly fly, the researchers found that the brain knew precisely how to handle the new photoreceptor. It simply recruited its extra standby neurons and put them to work. Perry explained that when the team gave those spare neurons something to connect to, they survived and wired up correctly, immediately, with no further genetic change. In his words, the brain was ready before the eye asked, a rare and concrete case of evolution making use of neurons that were otherwise going to die.

This finding carries significant implications for understanding how major sensory innovations arise. Rather than requiring perfectly synchronized mutations in both the eye and the brain, the evolution of a new photoreceptor could proceed because the developing nervous system already possessed the flexibility to accommodate novel inputs. The brain’s habit of generating excess neurons and keeping them available until connections are established provided a pre-existing substrate for the new visual channel. In effect, the neural infrastructure for enhanced color vision was already in place, waiting for the eye to supply it with something to do. This mechanism allowed insects with a flexible brain framework to adapt to an emerging ecological need: the elevated ability to visualize flowering plants as they became prominent features of the environment.

The coevolution of flowering plants and their insect pollinators is one of the most celebrated stories in evolutionary biology, and the new study adds a mechanistic chapter to it. Butterflies depend on flowers for nectar, and flowers depend on butterflies and other pollinators for reproduction. The ability to distinguish subtle differences in color would confer a direct foraging advantage, guiding butterflies toward rewarding blossoms and helping them avoid flowers that have already been depleted. Enhanced color vision may also play a role in mate recognition, as butterflies often use visual signals to identify conspecifics. The addition of a ninth photoreceptor to every eye unit would have amplified all of these capabilities at once, making it a powerful adaptation with consequences across multiple aspects of the butterfly’s life.

The researchers also found evidence that this evolutionary transition is not confined to the distant past. They identified a hawkmoth species that appears to be partway through the same change. The lower half of its eye is butterfly-like, containing two of these R7-type cells per unit, while the upper half remains fly-like, with only one. This mosaic pattern is roughly what would be expected if the genetic change began in one region of the eye and is spreading across the organ over evolutionary time. The hawkmoth thus offers a living snapshot of an intermediate stage, a rare opportunity to observe a major sensory transition in progress rather than inferring it solely from comparisons between finished endpoints.

Questions remain, and the team is actively pursuing them. The most difficult to study is whether the engineered butterfly fly can actually see more vivid color with its additional photoreceptor, as natural butterflies do. Behavioral and physiological tests will be needed to confirm that the new cell contributes functional color discrimination rather than simply existing as an anatomical novelty. Still, the study stands as a compelling demonstration of how a single genetic switch, acting at the right moment in development, can rebuild a sensory organ, and how a brain pre-adapted with spare neural capacity can absorb the change without further modification. It offers a concrete, experimentally verified account of one of the most consequential upgrades in the visual history of insects, and a vivid illustration of evolution’s capacity to repurpose existing biological material in service of new ways of seeing the world.

Subject of Research: Genetic basis of photoreceptor expansion and neural accommodation in butterfly color vision

Article Title: The story behind the butterfly’s enhanced visual world of color

Article References: The story behind the butterfly’s enhanced visual world of color. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: butterflies, color vision, photoreceptors, compound eye, genetics, evolution, neuroscience, Science Advances, painted lady butterfly, fruit fly, hawkmoth, sensory biology

The research team, led by Yingqiong Peng and Yuling Jin, built their system around a deceptively simple observation. Convolutional neural networks, the workhorses of modern image segmentation, progressively downsample images as they analyze them. This downsampling is essential for capturing the broad context of a scene, but it systematically destroys high-frequency information, the fine-grained signals that encode sharp edges, delicate textures, and intricate contours. For large objects, this loss is tolerable. For a fruit fly whose body measures only a few dozen pixels in a field photograph, it is devastating, because the insect’s identity rests almost entirely on those fine details: the veins of its wings, the outline of its mesonotum, the crisp boundary between body and background.

The researchers’ answer is a segmentation architecture they call HFIC-DroSeg, which stands for high-frequency information compensation. The framework adopts the familiar encoder-decoder structure of U-Net, a network originally designed for biomedical imaging, but augments it with two purpose-built modules. The first, the High-Frequency Information Extraction module, or HFIE, operates in the frequency domain. Rather than looking at the image in its raw spatial form, the module applies a two-dimensional discrete Fourier transform, which decomposes the image into a spectrum of frequency components. Low frequencies carry the broad shapes and smooth color gradients of a scene; high frequencies carry the edges and textures that make an object visually distinct.

What makes the HFIE module unusual is how it decides which high-frequency components to keep. Field photographs are full of high-frequency noise: the serrated edges of leaves, the speckle of soil, the dappled patterns of sunlight. A naive frequency filter would preserve all of this clutter along with the fly. Instead, the team employs a genetic algorithm, an optimization technique inspired by evolutionary search, to adaptively select an optimal threshold for each input image. Components whose spectral magnitude exceeds the threshold are retained; the rest are discarded. The selected components are then transformed back into the spatial domain through an inverse Fourier transform, producing a high-frequency information map that is aligned in scale with the original features and fused back into the network through a lightweight convolutional layer. Unlike classical edge operators such as Sobel or Laplacian filters, which use fixed kernels and capture only local, fixed-direction gradients, this pipeline performs a global spectral analysis and dynamically suppresses noisy components that do not correspond to the pest.

The second innovation, the Multi-Scale High-Frequency Aggregation module, or MS-HFA, addresses a different problem: the high-frequency features extracted at different depths of the network carry different kinds of information. Shallow layers preserve fine edge details, while deeper layers encode semantic context about what the object actually is. The MS-HFA module gathers high-frequency feature maps from four encoder stages, resizes them to a common resolution, and weighs them through a dedicated attention mechanism. Each feature map is processed with average and max pooling, passed through a small fully connected network, and converted into a weight vector that determines how much that scale contributes to the final aggregated representation. The experiments showed that this aggregation matters enormously: using any single feature layer yielded mean intersection-over-union scores between 0.69 and 0.79, while combining all four layers pushed the score to 0.94.

To sharpen boundaries even further, the team added a Fine-grained Edge Guidance branch, an auxiliary task that forces the network to pay explicit attention to edges during training. The branch applies traditional edge detection filters to both the predicted segmentation mask and the ground truth, then computes a loss based on the Euclidean distance between the resulting edge maps. This distance metric outperformed both cosine similarity and Kullback-Leibler divergence in controlled comparisons, achieving a Dice coefficient of 0.96. The edge loss is computed on small 28-by-28 single-channel masks, which keeps the additional computation modest while still steering the network toward crisp, well-localized contours.

The empirical results are striking. On a dataset of 2,400 field images covering four Drosophila species, including Bactrocera cucurbitae, Bactrocera dorsalis Hendel, Bactrocera scutellata, and Bactrocera tau, HFIC-DroSeg achieved a precision of 0.93, a recall of 0.94, an accuracy of 0.95, a mean intersection-over-union of 0.94, and a Dice coefficient of 0.96, outperforming established architectures including FCN, U-Net, PSP-Net, U-Net++, DeepLabV3+, SegFormer, and UDTransNet. Ablation experiments confirmed that each module contributes measurably: adding HFIE alone lifted the baseline U-Net’s mean intersection-over-union from 77.32 to 86.83 percent, and the full system reached 93.91 percent, with paired t-tests confirming the improvements were statistically significant across five random seeds.

Crucially, the team also ruled out the possibility that the gains came simply from a bigger model. An enlarged U-Net with a nearly identical parameter count of 28.3 million, but without the frequency-compensation modules, managed only a mean intersection-over-union of 0.79, compared with 0.94 for HFIC-DroSeg at 28.1 million parameters. The architecture also generalized reasonably well beyond its training domain: on the public IP102 pest benchmark it achieved an accuracy of 0.92 and a mean intersection-over-union of 0.90, and on Pascal VOC 2012 it reached 0.82, competitive with specialized methods though not the top score. When applied directly to real sticky-trap photographs containing three other mulberry field pests, the model segmented the fruit flies cleanly and captured the main body regions of the non-target insects, though with some edge inaccuracies attributable to domain differences.

The authors are candid about the system’s limits. Under heavy leaf occlusion, extreme uneven illumination, or scenes crowded with multiple similarly textured small insects, performance drops, with mean intersection-over-union falling to between 0.81 and 0.85 on challenging subsets. The frequency-domain explanation is intuitive: occlusion breaks the closed contour of the fly, harsh lighting attenuates the amplitude of valid high-frequency signals, and overlapping spectral distributions confuse the adaptive threshold selection. Edge-pixel errors on non-target pest species rose by 14.3 percent relative to fruit-fly targets, underscoring a generalization bottleneck that the team hopes to address with unsupervised domain adaptation in future work.

Beyond the technical achievements, the study raises practical questions about deploying such systems in the field. The authors caution that models trained on data from specific orchards, cameras, and lighting conditions may underperform in unfamiliar environments, and they advocate a human-machine collaborative workflow in which the AI provides segmentation cues while final pest-management decisions rest with agronomists or growers. They also point toward future directions including lightweight designs for mobile deployment and extension to other agricultural pests. If those steps succeed, the frequency-compensation strategy demonstrated here could become a standard ingredient in the computer vision toolbox of precision agriculture, helping farmers spot the smallest threats before they become the largest losses.

Subject of Research: Deep learning image segmentation for Drosophila pest detection using high-frequency information compensation

Article Title: An improved image segmentation method for pest with high-frequency information compensation

Article References: Peng, Y., Jin, Y., Li, Y., Liao, M., Xu, Y., & Yin, C. (2026). An improved image segmentation method for pest with high-frequency information compensation. Smart Agricultural Technology, 15, Article 102566. https://doi.org/10.1016/j.atech.2026.102566

Image Credits: AI Generated

DOI: 10.1016/j.atech.2026.102566

Keywords: image segmentation, Drosophila, fruit fly, deep learning, high-frequency information, Fourier transform, genetic algorithm, U-Net, precision agriculture, pest monitoring, computer vision, attention mechanism