Cells exhibit a mysterious form of selective heritable short-term memory, influencing outcomes as diverse as cell fate decisions in embryos and environmental responses in cancer cells and bacteria. Here, we present a simple theoretical framework explaining how this selective memory can arise from the reactions regulating molecular levels in cells. Our key insight is that related cells retain more similar molecular concentrations relative to random cells when a greater variance of possible concentration states is created during a single cell generation than is created by cell division across a population. This persistence of molecular similarity down a lineage constitutes a form of heritable short-term memory. We identify the biochemical networks that produce, modify, and degrade molecules as an underexplored source of these additional molecular concentration states. Using experimentally informed simulations, we find that the strength and duration of molecular similarity down a lineage depend on tunable network properties, explaining why some cellular traits persist only briefly while others last generations. These contributions to molecular concentration variance from biochemical reaction networks act in concert with gene expression and other regulatory processes to shape the protein composition of cells. Our framework yields clear, testable predictions for determining how biochemical network architectures drive non-genetic cellular inheritance.

The endoplasmic reticulum (ER) is a highly interconnected membrane network that serves as a central site for protein synthesis and maturation. A crucial subset of ER-associated transcripts, termed secretome mRNAs, encode secretory, lumenal and integral membrane proteins, representing nearly one-third of human protein-coding genes. Unlike cytosolic mRNAs, secretome mRNAs undergo co-translational translocation, and thus require precise coordination between translation and protein insertion. Disruption of this process, such as through altered elongation rates, activates stress response pathways that impede cellular growth, raising the question of whether secretome translation is spatially organized to ensure fidelity. Here, using live-cell single-molecule imaging, we demonstrate that secretome mRNA translation is preferentially localized to ER junctions that are enriched with the structural protein lunapark and in close proximity to lysosomes. Lunapark depletion reduced ribosome density and translation efficiency of secretome mRNAs near lysosomes, an effect that was dependent on eIF2-mediated initiation and was reversed by the integrated stress response inhibitor ISRIB. Lysosome-associated translation was further modulated by nutrient status: amino acid deprivation enhanced lysosome-proximal translation, whereas lysosomal pH neutralization suppressed it. These findings identify a mechanism by which ER junctional proteins and lysosomal activity cooperatively pattern secretome mRNA translation, linking ER architecture and nutrient sensing to the production of secretory and membrane proteins.

Visual systems across species transform photoreceptor inputs into diverse perceptual representations through hierarchical networks that extract features via parallel pathways. In Drosophila, the optic lobes are layered, retinotopic visual processing centers that contain two-thirds of the brain’s neurons and support diverse visually guided behaviors. Although this architecture has long suggested hierarchical and parallel organization, a system-wide account of how behaviorally relevant visual features are routed and integrated across a complete visual system—in any animal—has remained elusive. The new male fly connectome now provides the synapse-level wiring needed to trace visual information from photoreceptors through the optic lobes and across the central brain. Applying a network-based analysis of information flow, we reveal a multi-layered architecture organized into distinct, functionally interpretable pathways. Using this framework to propagate signals through these pathways predicts receptive-field structure and feature selectivity consistent with physiological data, enabling large-scale functional annotation of thousands of neuron types. We find that distinct visual input channels are broadly distributed throughout the brain, yet converge in focal regions of feature specificity and acute spatial vision. Together, these analyses provide a neuron-level, connectome-based view of how a brain organizes and transforms visual input.

Understanding how neurons integrate signals from thousands of input synapses requires methods to monitor neurotransmission across many sites simultaneously. The fluorescent protein glutamate indicator iGluSnFR enables visualization of synaptic signaling, but the sensitivity, scale and speed of such measurements are limited by existing variants. Here we developed two highly sensitive fourth-generation iGluSnFR variants with fast activation and tailored deactivation rates: iGluSnFR4f for tracking rapid dynamics, and iGluSnFR4s for recording from large populations of synapses. These indicators detect glutamate with high spatial specificity and single-vesicle sensitivity in vivo. We used them to record natural patterns of synaptic transmission across multiple experimental contexts in mice, including two-photon imaging in cortical layers 1–4 and hippocampal CA1, and photometry in the midbrain. The iGluSnFR4 variants extend the speed, sensitivity and scalability of glutamate imaging, enabling direct observation of information flow through neural networks in the intact brain.

The early formation of sensorimotor circuits is essential for survival. While the development and function of exteroceptive circuits and their associated motor pathways are well characterized, far less is known about the circuits that convey viscerosensory inputs to the brain and transmit visceromotor commands from the central nervous system to internal organs. Technical limitations, such as the in utero development of viscerosensory and visceromotor circuits and the invasiveness of procedures required to access them, have hindered studies of their functional development in mammals. Using larval zebrafish—which are genetically accessible and optically transparent—we tracked, in vivo, how cardiosensory and cardiomotor neural circuits assemble and begin to function. We uncovered a staged program. First, a minimal efferent circuit suffices for heart-rate control: direct brain-to-heart vagal motor innervation is required, intracardiac neurons are not, and heart rate is governed exclusively by the motor vagus nerve. Within the hindbrain, we functionally localize a vagal premotor population that drives this early efferent control. Second, sympathetic innervation arrives and enhances the dynamics and amplitude of cardiac responses, as neurons in the most anterior sympathetic ganglia acquire the ability to drive cardiac acceleration. These neurons exhibit proportional, integral, and derivative–like relationships to heart rate, consistent with controller motifs that shape gain and dynamics. Third, vagal sensory neurons innervate the heart. Distinct subsets increase activity when heart rate falls or rises, and across spontaneous fluctuations, responses to aversive stimuli, and optogenetically evoked cardiac perturbations, their dynamics are captured by a single canonical temporal kernel with neuron-specific phase offsets, supporting a population code for heart rate. This temporally segregated maturation isolates three experimentally tractable regimes—unidirectional brain-to-heart communication, dual efferent control, and closed-loop control after sensory feedback engages—providing a framework for mechanistic dissection of organism-wide heart–brain circuits.

Unveiling the genetic profiles of spatially distinguished cells is an important aspect in many areas of brain research, as the genetic identity contains information about a cell’s physiological properties and internal state. On top of this, knowledge of the genetic details of each cell can reveal structural organization within tissue. As image-based spatial transcriptomics moves toward applications in tissues with dense cellular packing, accurate assignment of detected mRNA transcripts ("spots") to correct segmented cells becomes increasingly difficult, rendering simple methods insufficient with many incorrect assignments to neighboring cells. Here we introduce SpotDMix, a statistical model for assigning spots to cells by modeling spots as coming from a mixture model of distributions matching segmented cell shapes, with assignment probabilities and shape parameters optimized using the Expectation Maximization algorithm. Performance is assessed and compared against several simple methods in various scenarios on both surrogate data and larval zebrafish data. In all tested scenarios SpotDMix outperforms the simple methods on all evaluated metrics, including individual transcript assignment accuracy, total assigned number of spots per cell error and cell type classification. Further, SpotDMix produces a higher degree of exclusivity between genes which are known to not or rarely co-express.

To navigate their environments effectively, animals frequently track time elapsed or distance traveled while seeking food and avoiding threats. The hippocampus is implicated in this process, but the neural mechanisms remain unclear. Using virtual reality tasks that require mice to integrate time or distance to collect a reward, we identified two previously unknown functional subpopulations of CA1 pyramidal neurons. Both subpopulations encode time or distance via distinct ramping dynamics. The first subpopulation exhibits a rapid, synchronous rise in activity upon movement-initiated integration. Subsequently, individual neurons ramp down at heterogeneous rates, creating progressively diverging firing rates that encode elapsed time or distance. Closed-loop optogenetic inactivation of somatostatin-positive (SST) interneurons counterintuitively reduced the ramping activity, leading mice to prematurely attempt reward collection, suggesting impaired time/distance estimation. Conversely, the second CA1 subpopulation shows opposite dynamics - an initial rapid suppression followed by a gradual ramp-up. Inactivating parvalbumin-positive (PV) interneurons diminished this initial suppression, resulting in transient attempts to collect reward near integration onset. These findings reveal parallel hippocampal circuits that initiate and maintain time or distance encoding, controlled by PV and SST interneurons, respectively, and provide insights into the neural computations supporting goal-directed navigation.

Foraging animals often sample options that yield rewards with different probabilities. In such scenarios, many animals exhibit “matching”, whereby they allocate their choices such that the fraction of rewarded samples is equal across options. While matching can be optimal in environments with diminishing returns, this condition alone is not sufficient to determine optimality. Moreover, diminishing returns arise when resources deplete and replenish over time, but their form depends on the temporal structure and statistics of replenishment. Here, we investigate how these environmental properties influence whether matching is optimal. We consider an agent that samples options at fixed rates and derive the resulting reward probabilities across different types of environments. This allows us to analytically determine conditions under which the optimal policy exhibits matching. When all options share the same replenishment dynamics, matching emerges as optimal across a wide range of environments. However, when dynamics differ across options, optimal policies can deviate from matching. In such cases, the rank-ordering of observed reward probabilities depends only on the qualitative nature of the replenishment process, and not on the specific replenishment rates. As a result, the optimal policy can exhibit under- or over-matching depending on which options are more rewarding. We use this result to identify environments where performance differs substantially between matching and optimality. Finally, we show that fluctuations in replenishment rates—representing environmental stochasticity or internal uncertainty—can amplify deviations from matching. These findings deepen our understanding of the relationship between environmental variability and behavioral optimality, and provide testable predictions across diverse settings.

Nervous systems can process information in serial or in parallel, trading off efficiency for flexibility and speed. How these network architectures are implemented across sensorimotor pathways to control behavior is unclear. We investigate this tradeoff directly in Drosophila by comparing neuronal circuits underlying landing and takeoff, behaviors transforming similar visual cues to whole-body motor output. Using a whole-CNS connectome, electrophysiology, and behavioral analysis, we reconstruct the complete feedforward pathway for landing, including visual feature detectors, a dedicated ensemble of descending neurons (DNs), and a core premotor circuit in the nerve cord. Comparison to the takeoff pathway reveals that, despite encoding the same sensory feature and engaging similar muscle groups, neuronal circuits controlling the two behaviors are separated at every sensorimotor level. Extending this analysis to the complete DN population reveals a blueprint for descending motor control: DNs across the behavioral space utilized by the fly are organized as a set of parallel, loosely-overlapping ensembles that form a continuum from command-like control, with individual DNs determining behavioral output, to population coding, with multiple DNs controlling behavior synergistically. Distinct combinations of sensory feature detectors differentially recruit DN ensembles to enable flexible, context-dependent behavioral control.

Many plasticity rules rely on adjusting the strength of synapses between pairs of cells based on their coincident activity. We uncovered a new mechanism for coincidence detection in the Drosophila head direction network. To maintain an accurate sense of direction, head direction neurons that signal orientation during navigation must learn to anchor to relevant external sensory cues in novel environments. Yet the synaptic mechanism for this form of unsupervised learning is unknown in any organism. In Drosophila, GABAergic visual inputs converge onto head direction neurons, and these inhibitory synapses change strength with experience to learn the relationship between visual landmarks and head direction. However, how coincident pre- and postsynaptic activity is detected across this inhibitory synapse is not understood. We discovered that neurons which release the monoamine octopamine close a feedback loop that conveys postsynaptic head direction activity onto presynaptic terminals of visual inputs. This octopamine pathway is required for anchoring the head direction network to visual cues. Furthermore, pairing structured activation of octopamine neurons with a visual cue is sufficient to drive rapid plasticity, even without postsynaptic head direction cell activity. Previous work has extensively characterized coincidence detection mechanisms at excitatory synapses; our work defines a novel mechanism for coincidence detection at an inhibitory synapse, in which postsynaptic activity is relayed via a neuromodulatory neuron onto presynaptic terminals