Biology subjects

Gjorgjieva, J.

Publications and source records attributed to Gjorgjieva, J..

At least 19 recordsLinked to original sources

bioRxiv2026

Heterogeneous plasticity rules generate diverse structures in recurrent networks

How do local changes at individual synapses give rise to structured recurrent connectivity? Experiments have revealed diverse forms of spike-timing-dependent plasticity (STDP) across the brain, making it difficult to predict which circuit structures different rules will generate. We developed an analytical framework mapping excitatory Hebbian, anti-Hebbian, and symmetric STDP rules to directed three-neuron motifs. Individual homogeneous rules generate restricted subsets of motifs, and no single rule produces all motifs. Combining different rules generates motifs inaccessible to individual rules and yields the full motif repertoire, and these predictions generalize to larger recurrent networks. The mapping is nevertheless highly non-unique, with distinct rules and parameters producing similar motif distributions, including distributions compatible with experimental connectivity. Thus, circuit structure constrains candidate plasticity rules but does not uniquely identify them. Finally, networks with similar motif distributions can nonetheless differ in their firing rates and pairwise correlations, suggesting that activity may provide an additional constraint on candidate plasticity rules.

neuroscience↗

bioRxiv2026

Functional temporal diversity in the fly reflects natural motion statistics

As animals move through the world, their visual input can change gradually or shift rapidly, spanning a broad range of temporal frequencies. Visual circuits therefore face the challenge of encoding inputs with an inherently multiscale temporal structure. From insects to vertebrates, peripheral visual neurons exhibit striking temporal diversity, yet the principles underlying this organization remain unclear. Here, we show that temporal diversity in the fly visual system supports efficient coding of natural motion by distributing encoding across complementary timescales. Using well-characterized fly motion-detection pathways as a model, we show that temporal filters within and across visual hierarchies occupy a low-dimensional feature space defined by filter polarity and characteristic timescale. Within an efficient-coding framework constrained by natural motion statistics, circuits that combine temporally distinct filters encode naturalistic inputs more efficiently than circuits composed of more similar filters, an advantage that largely disappears when changing the stimulus statistics. Measured filters of fly visual cell types closely match the optimal solutions predicted for natural motion, and these optima map onto both circuit architecture and function. Finally, extending beyond the motion pathway, we identify distinct correlation structures in circuits within and outside the motion pathway. Together, we identify temporal diversity in response filters as a circuit-level strategy for encoding natural motion, shaped by the multiscale statistics of the environment. Significance statementNeuronal diversity is ubiquitous across sensory systems, yet its functional role remains poorly understood. We investigate a prominent form of diversity: the wide temporal response profile of visual neurons, from transient detectors to slow integrators. Using the fly visual system, we uncover a coding function of temporal diversity. Neurons with complementary temporal properties span the multiscale structure of natural motion, achieving coding efficiency that no single neuron type can implement alone. Predictions from an efficient coding framework grounded in natural motion statistics closely match fly visual circuit organization, from cell-type response properties to anatomical wiring. Because similar temporal diversity exists in the vertebrate retina, we suggest this distributed temporal coding is a conserved design principle shaped by natural scene statistics.

neuroscience↗

bioRxiv2026

A right-hemispheric language network at single-neuron resolution

Human language depends on highly specialized left-hemispheric brain networks. Damage to these networks causes severe language impairments (aphasia), one of the most common, debilitating and costly consequences of left-hemispheric brain injury, especially stroke. The limited recovery of aphasia despite intensive rehabilitation efforts emphasizes the need to understand the basis of residual language abilities at the single-neuron level, which has remained unexplored so far. Here, we report large-scale microelectrode recordings with single-unit resolution over a period of ten months from the right-hemispheric prefrontal and parietal association cortex of an individual with stroke-induced chronic non-fluent aphasia. Single neurons exhibited regionally specific responses during comprehension, retrieval and articulation of words, the core operations of language. Distinct subpopulations encoded linguistic information in a task-specific manner, despite correlated firing patterns across tasks. Both single-neuron activity and temporally coordinated population dynamics were predicted by semantic and phonological embeddings derived from large language models (LLMs), revealing a regional dissociation in which semantic features preferentially accounted for prefrontal activity and phonological features for parietal activity. Our findings suggest that right-hemispheric circuits, homotopic to the left language network, can support language processing through structured, functionally organised activity at the level of single neurons. This study opens an avenue for developing mechanistically specific neurorehabilitation and neurorestorative strategies for aphasia, such as brain-computer interfaces (BCIs), that leverage right-hemispheric language resources.

neuroscience↗

bioRxiv2025

Synaptic density and relative connectivity conservation maintain circuit stability across development

As bodies grow during postembryonic and postnatal development, nervous systems must expand to preserve circuit integrity. To investigate how circuits retain stable wiring and function throughout development, we combined synaptic-level resolution electron microscopy (EM) with computational modeling in the Drosophila larval nociceptive system. Based on EM data, we generated the "contactome"--the set of synaptic membrane contacts--of this circuit across development to evaluate how different mechanisms contribute to wiring stability. Specifically, we investigated three mechanisms: correlation-based plasticity and synaptic scaling, which modify synaptic strength, and structural plasticity, which preserves synaptic density. We found that synaptic sizes remain largely stable across development, and synapses between the same pre- and postsynaptic neurons do not correlate in size, suggesting that synaptic scaling and correlation-based plasticity play a limited role in shaping connectivity. In contrast, dendritic synaptic density remains invariant despite a previously reported fivefold increase in neuron size and synapse number. This conservation requires increased axonal presynaptic density to compensate for unequal axonal and dendritic growth. As neurons grow, this adjustment is necessary to maintain the relative synaptic input associated with each presynaptic partner across development. Our EM analysis and modeling show that conserving relative connectivity and synaptic density is sufficient to maintain consistent postsynaptic responses across development, highlighting these conserved structural features as key contributors to circuit stability during growth.

neuroscience↗

bioRxiv2025

Assembly-based computations through contextual dendritic gating of plasticity

Neuronal assemblies - groups of strongly connected neurons - are considered the basic building blocks of perception and memory in the brain by encoding representations of specific concepts. Despite recent evidence for the biological basis behind the existence and formation of such assemblies, computational models often fall short of showing how assemblies can be flexibly learned and combined to perform real-world computations. A prominent problem is catastrophic forgetting, where learning a new assembly can disrupt existing connectivity structure and lead to forgetting previously learned assemblies. We propose a biologically plausible computational model, where dendritic compartments (instead of neurons) are the loci for learning and inhibition gates learning in a dendrite-specific manner, to flexibly learn new stimuli without forgetting of old ones. By learning stable projections from one brain region into another and associations between different brain regions, we demonstrate how the proposed assembly framework implements the basic building blocks for diverse computations. In a visual-auditory association task, we demonstrate how the context-specific assembly computations can be used to correctly separate ambiguous stimuli based on their dendritic representations. Our models provide unique insights and predictions for how hierarchically connected brain areas use their biological components to implement flexible yet robust learning.

neuroscience↗

bioRxiv2025

Cell-type-specific synaptic scaling mechanisms differentially contribute to associative learning

Excitatory synaptic scaling regulates network dynamics by proportionally adjusting excitatory synaptic strengths after sensory perturbations. During associative learning, blocking excitatory scaling in conditioned taste aversion paradigms prolongs generalized aversive responses and delays memory specificity. Recent evidence also implicates inhibitory synaptic scaling in the regulation of network dynamics. Specifically, parvalbumin (PV)-expressing inhibitory neurons, targeting perisomatic regions of excitatory (E) pyramidal neurons, and somatostatin (SST)-expressing neurons, targeting distal dendrites, exhibit distinct scaling responses. This leaves open the question of how complex plasticity mechanisms regulate recurrent excitatory-inhibitory circuit dynamics in associative learning. Using computational approaches, we demonstrate that Hebbian plasticity drives memory generalization to novel stimuli not presented during conditioning. Following conditioning, diverse synaptic scaling mechanisms progressively induce memory specificity, which can be regulated by top-down inputs. Our results reveal that, in the absence of excitatory scaling, PV-to-E scaling can effectively compensate and rescue memory specificity, highlighting the presence of degenerate mechanisms in the brain. Notably, in the process of establishing memory specificity, excitatory scaling and PV-to-E scaling function synergistically, while concurrently opposing SST-to-E scaling. The synergistic and antagonistic plasticity mechanisms are orchestrated to shape the temporal evolution of memory representations, from generalized to precise. Significance statementAssociative learning is a fundamental brain function that allows us to link experiences, adapt behavior, and form lasting memories. During this process, memory representations are shaped by synaptic scaling, a homeostatic plasticity mechanism that provides slow, negative feedback to regulate synaptic strengths and adjust network excitability. Operating at the synapses of diverse excitatory and inhibitory cell types, multiple forms of homeostatic plasticity influence the dynamics of associative learning. Here, we demonstrate that synergistic and antagonistic cell-type-specific synaptic scaling mechanisms operate at different types of inhibitory synapses to jointly govern the temporal evolution of memory representations. Through their interaction, they guide the transition from generalized to precise memories.

neuroscience↗

bioRxiv2024

Structured stabilization in recurrent neural circuits through inhibitory synaptic plasticity

In cortical circuits, inhibitory interneurons play a dual role: they regulate overall activity levels to prevent runaway excitation, and contribute to diverse computations. While unstructured inhibitory synaptic connections achieve the first role by homeostatically regulating firing rates, computational tasks often require structured excitatory-inhibitory (E/I) connectivity. Here, we consider a broad class of pairwise inhibitory spike-timing dependent plasticity (iSTDP) rules, demonstrating how inhibitory connections can self-organize to both stabilize excitation and generate functionally relevant connectivity motifs--a process we call "structured stabilization". We show that in both small E/I circuits and large spiking recurrent neural networks the choice of iSTDP rule can lead to either mutually connected E/I pairs, or to lateral inhibition, where an inhibitory neuron connects to an excitatory neuron that does not directly connect back to it. In a one-dimensional ring network with two inhibitory subpopulations following these distinct iSTDP rules, the effective connectivity within the excitatory units self-organizes into a Mexican-hat-like profile, with excitatory influence in the center and inhibitory influence away from the center. This leads to emergent network responses such as contextual modulation effects as in the visual cortex and spatially modular activity characteristic of developmental spontaneous activity. Our theoretical work introduces a family of rules that retains the broad applicability and simplicity of spike-timing-based plasticity, while stabilizing activity and promoting specific connectivity motifs which support emergent network computations. Significance StatementNeural circuits are faced with the dual challenge of keeping activity stable while modifying synaptic strengths during development and learning. A prominent mechanism to keep excitation in check is inhibitory synaptic plasticity. In recurrent circuit models, we show that plasticity rules based on spike timing at synapses from inhibitory to excitatory neurons can stabilize activity while reinforcing specific connectivity motifs. The shape of these rules dictates the resulting motifs, and determines whether inhibitory neurons form strong reciprocal connections with their excitatory partners or lateral connections onto excitatory neurons that do not connect back. In recurrent networks with local excitatory connectivity, these motifs generate surround suppression as in the visual cortex, and long-range spatial correlations characteristic of developmental spontaneous activity.

neuroscience↗

bioRxiv2024

Spontaneous emergence and drifting of sequential neural activity in recurrent networks

Repeating sequences of neural activity exist across brain regions of different animals are thought to underlie diverse computations. However, their emergence and evolution during ongoing synaptic plasticity remain unclear. To mechanistically understand this process, we investigated the interaction of biologically-inspired activity-dependent synaptic plasticity rules in models of recurrent circuits to produce connectivity structures that support neuronal sequences. Under unstructured inputs, our recurrent networks developed strong unidirectional connections, resulting in spontaneous repeating sequences of spikes. During ongoing plasticity these sequences repeated despite turnover of individual synaptic connections, a process reminiscent of synaptic drift. The turnover process occurred over different timescales, with sequence-promoting connectivity types and motif structures being strengthened while others weakened, leading to sequences with different degrees of volatility. Structured inputs could reinforce or retrain the resulting connectivity structures underlying sequences, enabling stable yet flexible encoding of inputs. Our model unveils the interplay between synaptic plasticity and sequential activity in recurrent networks, providing insights into how the brain might implement reliable and flexible computations.

neuroscience↗

bioRxiv2024

Brain-wide microstrokes affect the stability of memory circuits in the hippocampus

Cognitive deficits affect over 70% of stroke survivors, yet the mechanisms by which multiple small ischemic events contribute to cognitive decline remain poorly understood. In this study, we employed chronic two-photon calcium imaging to longitudinally track the fate of individual neurons in the hippocampus of mice navigating a virtual reality environment, both before and after inducing brain-wide microstrokes. Our findings reveal that, under normal conditions, hippocampal neurons exhibit varying degrees of stability in their spatial memory coding. However, microstrokes disrupted this functional network architecture, leading to cognitive impairments. Notably, the preservation of stable coding place cells, along with the stability, precision, and persistence of the hippocampal network, was strongly predictive of cognitive outcomes. Mice with more synchronously active place cells near important locations demonstrated recovery from cognitive impairment. This study uncovers critical cellular responses and network alterations following brain injury, providing a foundation for novel therapeutic strategies preventing cognitive decline.

neuroscience↗

bioRxiv2024

Correlated spontaneous activity sets up multi-sensory integration in the developing higher-order cortex

To perceive and navigate complex sensory environments, animals combine sensory information from multiple modalities in specialized brain circuits. Known as multisensory integration, this process typically depends on the existence of co-aligned topographic connections from several sensory areas to downstream circuits exhibiting multimodal representations. How such topographically co-aligned connectivity necessary for multisensory integration gets set up in early stages of development is still unknown. Inspired by the role of spontaneous activity in refining topographic connectivity between early sensory circuits, here we investigated the potential of such spontaneous activity to also guide the co-alignment of multiple sensory modalities in RL, a higher-order associative cortical area rostro-lateral to V1. Analyzing spontaneous activity simultaneously recorded in primary visual and somatosensory cortex and area RL at different developmental ages before sensory experience, we identify candidate features of this activity to guide the emergence of co-aligned topographic multisensory projections with somatosensory leading the visual projection. We confirm this hypothesis using a computational model of activity-dependent circuit refinement, and show that the correlation of spontaneous activity between the visual and somatosensory primary cortex can establish an optimal fraction of multisensory neurons in RL for stimulus decoding. Our model provides an exciting new computational perspective of the role of spontaneous activity in the emergence of topographically co-aligned multimodal sensory representations in downstream circuits, specialized for the processing of rich sensory environments.

neuroscience↗

bioRxiv2024

Pre-training artificial neural networks with spontaneous retinal activity improves motion prediction in natural scenes

The ability to process visual stimuli rich with motion represents an essential skill for animal survival and is largely already present at the onset of vision. Although the exact mechanisms underlying its maturation remain elusive, spontaneous activity patterns in the retina, known as retinal waves, have been shown to contribute to this developmental process. Retinal waves exhibit complex spatio-temporal statistics and contribute to the establishment of circuit connectivity and function in the visual system, including the formation of retinotopic maps and the refinement of receptive fields in downstream areas such as the thalamus and visual cortex. Recent work in mice has shown that retinal waves have statistical features matching those of natural visual stimuli, such as optic flow, suggesting that they could prime the visual system for motion processing upon vision onset. Motivated by these findings, we examined whether artificial neural network (ANN) models trained on natural movies show improved performance if pre-trained with retinal waves. We employed the spatio-temporally complex task of next-frame prediction, in which the ANN was trained to predict the next frame based on preceding input frames of a movie. We found that pre-training ANNs with retinal waves enhances the processing of real-world visual stimuli and accelerates learning. Strikingly, when we merely replaced the initial training epochs on naturalistic stimuli with retinal waves, keeping the total training time the same, we still found that an ANN trained on retinal waves temporarily outperforms one trained solely on natural movies. Similar to observations made in biological systems, we also found that pre-training with spontaneous activity refines the receptive field of ANN neurons. Overall, our work sheds light on the functional role of spatio-temporally patterned spontaneous activity in the processing of motion in natural scenes, suggesting it acts as a training signal to prepare the developing visual system for adult visual processing. Author summaryBefore the onset of vision, the retina generates its own spontaneous activity, referred to as retinal waves. This activity is crucial for establishing neural connections and, hence, ensuring the proper functionality of the visual system. Recent research has shown that retinal waves exhibit statistical properties similar to those of natural visual stimuli, such as the optic flow of objects in the environment during forward motion. We investigate whether retinal waves can prepare the visual system for motion processing by pre-training artificial neural network (ANN) models with retinal waves. We tested the ANNs on next-frame prediction tasks, where the model predicts the next frame of a video based on previous frames. Our results showed that ANNs pre-trained with retinal waves exhibit faster learning on movies featuring naturalistic stimuli. Additionally, pre-training with retinal waves refined the receptive fields of ANN neurons, similar to processes seen in biological systems. Our work highlights the importance of spatio-temporally patterned spontaneous activity in preparing the visual system for motion processing in natural scenes.

neuroscience↗

bioRxiv2023

Sensory experience steers representational drift in mouse visual cortex

Representational drift - the gradual continuous change of neuronal representations - has been observed across many brain areas. It is unclear whether drift is caused by synaptic plasticity elicited by sensory experience, or by the intrinsic volatility of synapses. Here, using chronic two-photon calcium imaging in mouse primary visual cortex, we find that the preferred stimulus orientation of individual neurons slowly drifts over the course of weeks. By using cylinder lens goggles to limit visual experience to a narrow range of orientations, we show that the direction of drift, but not its magnitude, is biased by the statistics of visual input. A network model suggests that drift of preferred orientation largely results from synaptic volatility, which under normal visual conditions is counteracted by experience-driven Hebbian mechanisms, stabilizing preferred orientation. Under deprivation conditions these Hebbian mechanisms enable adaptation. Thus, Hebbian synaptic plasticity steers drift to match the statistics of the environment.

neuroscience↗

bioRxiv2023

Top-down modulation in canonical cortical circuits with inhibitory short-term plasticity

Cortical dynamics and computations are strongly influenced by diverse GABAergic interneurons, including those expressing parvalbumin (PV), somatostatin (SST), and vasoactive intestinal peptide (VIP). Together with excitatory (E) neurons, they form a canonical microcircuit and exhibit counterintuitive nonlinear phenomena. One instance of such phenomena is response reversal, whereby SST neurons show opposite responses to top-down modulation via VIP depending on the presence of bottom-up sensory input, indicating that the network may function in different regimes under different stimulation conditions. Combining analytical and computational approaches, we demonstrate that model networks with multiple interneuron subtypes and experimentally identified short-term plasticity mechanisms can implement response reversal. Surprisingly, despite not directly affecting SST and VIP activity, PV-to-E short-term depression has a decisive impact on SST response reversal. We show how response reversal relates to inhibition stabilization and the paradoxical effect in the presence of several short-term plasticity mechanisms demonstrating that response reversal coincides with a change in the indispensability of SST for network stabilization. In summary, our work suggests a role of short-term plasticity mechanisms in generating nonlinear phenomena in networks with multiple interneuron subtypes and makes several experimentally testable predictions.

neuroscience↗

bioRxiv2023

Multifaceted luminance gain control beyond photoreceptors in Drosophila

Animals navigating in natural environments must handle vast changes in their sensory input. Visual systems, for example, handle changes in luminance at many timescales, from slow changes across the day to rapid changes during active behavior. To maintain luminance-invariant perception, visual systems must adapt their sensitivity to changing luminance at different timescales. We demonstrate that luminance gain control in photoreceptors alone is insufficient to explain luminance invariance at both fast and slow timescales and reveal the algorithms that adjust gain past photoreceptors in the fly eye. We combined imaging and behavioral experiments with computational modeling to show that, downstream of photoreceptors, circuitry taking input from the single luminance-sensitive neuron type L3 implements gain control at fast and slow timescales. This computation is bidirectional in that it prevents underestimation of contrasts in low luminance and overestimation in high luminance. An algorithmic model disentangles these multifaceted contributions and shows that the bidirectional gain control occurs at both timescales. The model implements a nonlinear interaction of luminance and contrast to achieve gain correction at fast timescales and a dark-sensitive channel to improve the detection of dim stimuli at slow timescales. Together, our work demonstrates how a single neuronal channel performs diverse computations to implement gain control at multiple timescales that are together important for navigation in natural environments. Significance StatementLuminance varies slowly throughout the day, or rapidly during movements. Yet, animals show luminance-invariant behaviors to visual cues across these scenarios. Combining in vivo imaging, behavioral analyses and computational modeling, we here show that diverse luminance gain control operations required for robust visual processing exist in the fly visual circuitry beyond photoreceptors. Despite relying on a single luminance-sensitive neuron type, this luminance gain control is multifaceted, in that it achieves both gain increments and decrements at two different timescales. Overall, the study sheds light on a circuit gain control strategy operating at multiple timescales, which together challenge luminance-invariant visual behaviors in natural environments.

neuroscience↗

bioRxiv2022

Inhibition stabilization and paradoxical effects in recurrent neural networks with short-term plasticity

Inhibition stabilization is considered a ubiquitous property of cortical networks, whereby inhibition controls network activity in the presence of strong recurrent excitation. In networks with fixed connectivity, an identifying characteristic of inhibition stabilization is that increasing (decreasing) excitatory input to the inhibitory population leads to a decrease (increase) in inhibitory firing, known as the paradoxical effect. However, population responses to stimulation are highly nonlinear, and drastic changes in synaptic strengths induced by short-term plasticity (STP) can occur on the timescale of perception. How neuronal nonlinearities and STP affect inhibition stabilization and the paradoxical effect is unclear. Using analytical calculations, we demonstrate that in networks with STP the paradoxical effect implies inhibition stabilization, but inhibition stabilization does not imply the paradoxical effect. Interestingly, networks with neuronal nonlinearities and STP can transition non-monotonically between inhibition-stabilization and non-inhibition-stabilization, and between paradoxically- and non-paradoxically-responding regimes with increasing excitatory activity. Furthermore, we generalize our results to more complex scenarios including networks with multiple interneuron subtypes and any monotonically increasing neuronal nonlinearities. In summary, our work reveals the relationship between inhibition stabilization and the paradoxical effect in the presence of neuronal nonlinearity and STP, yielding several testable predictions.

neuroscience↗

bioRxiv2022

A local activity-dependent learning rule shapes dendritic morphologies and synaptic organization

Dendritic branching and synaptic organization shape single-neuron and network computations. How they emerge simultaneously during brain development as neurons become integrated into functional networks is still not mechanistically understood. Here, we propose a mechanistic model in which dendrite growth and the organization of synapses arise from the interaction of activity-independent cues from potential synaptic partners and local activity-dependent synaptic plasticity. Consistent with experiments, three phases of dendritic growth - overshoot, pruning, and stabilization - emerge naturally in the model. The model generates stellate-like dendritic morphologies that capture several morphological features of biological neurons under normal and perturbed learning rules, reflecting biological variability. Model-generated dendrites have approximately optimal wiring length consistent with experimental measurements. In addition to establishing dendritic morphologies, activity-dependent plasticity rules organize synapses into spatial clusters according to the correlated activity they experience. We demonstrate that a trade-off between activity-dependent and -independent factors influences dendritic growth and synaptic location throughout development, suggesting that early developmental variability can affect mature morphology and synaptic function. Therefore, a single mechanistic model can capture dendritic growth and account for the synaptic organization of correlated inputs during development. Our work suggests concrete mechanistic components underlying the emergence of dendritic morphologies and synaptic formation and removal in function and dysfunction, and provides experimentally testable predictions for the role of individual components.

neuroscience↗

bioRxiv2022

Stability and learning in excitatory synapses by nonlinear inhibitory plasticity

Synaptic changes underlie learning and memory formation in the brain. But synaptic plasticity of excitatory synapses on its own is unstable, leading to unlimited growth of synaptic strengths without additional homeostatic mechanisms. To control excitatory synaptic strengths we propose a novel form of synaptic plasticity at inhibitory synapses. We identify two key features of inhibitory plasticity, dominance of inhibition over excitation and a nonlinear dependence on the firing rate of postsynaptic excitatory neurons whereby inhibitory synaptic strengths change in the same direction as excitatory synaptic strengths. We demonstrate that the stable synaptic strengths realized by this novel inhibitory plasticity achieve a fixed excitatory/inhibitory set-point in agreement with experimental results. Applying a disinhibitory signal can gate plasticity and lead to the generation of receptive fields and strong bidirectional connectivity in a recurrent network. Hence, a novel form of nonlinear inhibitory plasticity can simultaneously stabilize excitatory synaptic strengths and enable learning upon disinhibition.

neuroscience↗

bioRxiv2022

Synapse-type-specific competitive Hebbian learning forms functional recurrent networks

Cortical networks exhibit complex stimulus-response patterns that are based on specific recurrent interactions between neurons. For example, the balance between excitatory and inhibitory currents has been identified as a central component of cortical computations. However, it remains unclear how the required synaptic connectivity can emerge in developing circuits where synapses between excitatory and inhibitory neurons are simultaneously plastic. Using theory and modeling, we propose that a wide range of cortical response properties can arise from a single plasticity paradigm that acts simultaneously at all excitatory and inhibitory connections - Hebbian learning that is stabilized by the synapse-type-specific competition for a limited supply of synaptic resources. In plastic recurrent circuits, this competition enables the formation and decorrelation of inhibition-balanced receptive fields. Networks develop an assembly structure with stronger synaptic connections between similarly tuned excitatory and inhibitory neurons and exhibit response normalization and orientation-specific center-surround suppression, reflecting the stimulus statistics during training. These results demonstrate how neurons can self-organize into functional networks and suggest an essential role for synapse-type-specific competitive learning in the development of cortical circuits. Significance StatementCortical circuits perform diverse computations, primarily determined by highly structured synaptic connectivity patterns that develop during early sensory experience via synaptic plasticity. To understand how these structured connectivity patterns emerge, we introduce a general learning framework for networks of recurrently connected neurons. The framework is rooted in the biologically plausible assumption that synapses compete for limited synaptic resources, which stabilizes synaptic growth. Motivated by the unique protein composition of different synapse types, we assume that different synapse types compete for separate resource pools. Using theory and simulation, we show how this synapse-type-specific competition allows the stable development of structured synaptic connectivity patterns, as well as diverse computations like response normalization and surround suppression.

neuroscience↗