2026 week 08

BCI Weekly Brief (week of 2026-02-16)

previous << weekly >> next

How this week was triaged

Frontiers in Human Neuroscience

Score: 0.82

Published: 2026-02-20T00:00:00+00:00

Tags: EEG, ERP, methods, electrophysiology

Uses 64-channel EEG at 1024 Hz and event-related potentials (ERPs) during matching-to-sample; direct electrophysiology and neural time-series methods.

Goal: First study to examine feedback-related ERPs during conditional discrimination learning in a matching-to-sample (MTS) paradigm using EEG
Participants: 11 young adults (6 male, 5 female; ages 20–29, mean 23.1 years) after excluding 5 of 16 recruits
Stimuli: 12 custom abstract visual stimuli organized into 4 equivalence classes (A–B–C linear series)
EEG setup: 64-channel wet-electrode system at 1,024 Hz sampling; 0.5–40 Hz bandpass filtering; 1,000-ms epochs with 200-ms pre-stimulus baseline
Phase I (Training): Established AB then BC relations (16 trials/block; 100% feedback; mastery = 16/16 correct)
Phase II (Maintenance): Mixed AB + BC trials (32/block; feedback faded from 100% → 50% → 0%; mastery = 30/32)
Phase III (Testing): Tested symmetry, transitivity, and equivalence relations with no feedback
Key finding 1: Incorrect feedback produced significantly larger ERP amplitudes than correct feedback — mean difference of 3.89 μV (p = 0.015, d = −0.88)
Key finding 2: Incorrect feedback produced significantly longer ERP latencies — mean difference of 109.2 ms (p = 0.0003, d = −1.65)
Key finding 3: Greater amplitude differences during training predicted lower Phase III equivalence test scores (p = 0.005, d = −0.95; ~2.3% score drop per μV increase)
Key finding 4: Latency differences were not significantly associated with test scores (p = 0.161)
ERP localization: Peaks consistently appeared at frontal electrode sites (Fp1, Fp2, FpZ, AF7), consistent with prefrontal involvement in error processing
Waveform interpretation: The observed component likely overlaps ERN, FRN, and P300 families; the authors interpret it conservatively as a feedback-related ERP
Limitation: Small convenience sample (n = 11), no reaction-time tracking, and possible motor-contamination of baseline window
Implication: Integrating EEG with behavior-analytic methods can illuminate covert neural processes during associative learning, with potential clinical relevance for neurocognitive disorders

Temporal Coding rather than Circuit Wiring allows Hippocampal CA3 Neurons to Dynamically Distinguish Different Cortical Inputs

bioRxiv Neuroscience

Score: 0.78

Published: 2026-02-20T00:00:00+00:00

Tags: electrophysiology, optogenetics, hippocampus, methods, computational

Dual-color optogenetic circuit mapping of MEC/LEC inputs to hippocampus and temporal coding; circuit-level electrophysiology and computational integration.

Title: “Temporal Coding rather than Circuit Wiring allows Hippocampal CA3 Neurons to Dynamically Distinguish Different Cortical Inputs”
Authors: Keelin O’Neil, Vincent Robert, Luke A. Arend, Buyong Kim, André A. Fenton, & Jayeeta Basu (NYU Langone Health / NYU)
Question: How do individual hippocampal CA3 neurons distinguish spatial (MEC) from contextual (LEC) entorhinal cortex inputs — via hard-wired anatomy or via temporal dynamics?
Approach: Dual-color optogenetic circuit mapping (ChrimsonR + Chronos in CaMKII-Cre mice) combined with ex vivo whole-cell patch-clamp recordings and in vivo large-scale extracellular electrophysiology
Convergence finding: Virtually all CA3 pyramidal neurons (20/21 in TTX+4-AP; 37/37 in ACSF) receive monosynaptic inputs from both MEC and LEC
Excitatory drive: MEC evoked ~2× larger monosynaptic EPSCs than LEC (~−40 pA vs. ~−20 pA, p = 0.0039), consistent with MEC axons terminating closer to the soma
E/I balance: Both pathways drove similar feedforward inhibition via PV+, CCK+, and VIP+ interneurons (but not SST+), yielding comparable E/I ratios < 1
Short-term plasticity divergence: At 10 Hz, both inputs facilitate; at 20–40 Hz, MEC continues facilitating while LEC depresses — the key biophysical distinction between the two pathways
Spike output ex vivo: Neither input alone drove CA3 spikes at low (monosynaptic) intensity; ~26% of neurons spiked only when polysynaptic/combined stimulation was used
Integration window: Optimal temporal summation of the two inputs occurs at ~0 ms ISI (coincident arrival), with a very narrow integration window
In vivo dentate spikes: MEC-originating dentate spikes (DS_MEC) drove significantly more CA3 spiking than LEC-originating ones (DS_LEC), regardless of relative timing or order between the two events
Theta-phase relationship: DS_MEC events preferentially occur at theta phases associated with higher CA3 excitability, amplifying their impact relative to DS_LEC
Core conclusion: CA3 distinguishes MEC from LEC not through distinct wiring or differential inhibitory microcircuits, but through frequency-dependent short-term plasticity and timing relative to network oscillations — i.e., temporal coding
Clinical/theoretical relevance: Findings challenge the canonical model of anatomical segregation of entorhinal inputs and suggest temporal dynamics are the primary mechanism for information multiplexing in memory circuits

Discovering macroscale functional organization on the structure of brain-like recurrent neural networks

bioRxiv Neuroscience

Score: 0.75

Published: 2026-02-20T00:00:00+00:00

Tags: computational, methods, RNN, structure-function

Links structure to function in brain-like RNNs with parcellations and modules; computational neuroscience and interpretable brain models.

Title: “Discovering macroscale functional organization on the structure of brain-like recurrent neural networks”
Authors: Peiyu Chen, Zaixu Cui, & Christos Constantinidis (Vanderbilt University; Chinese Institute for Brain Research, Beijing)
Question: Can macroscale functional organization (modules, gradients, hierarchies) observed in the human cortex emerge in artificial neural networks when brain-like structural constraints are imposed?
Model — BrainRNN: A recurrent neural network with 128 units embedded on a hemispheric manifold via Fibonacci lattice, with designated visual input units (posterior, 32 units), motor output units (anterior, 16 units), and association units (80 units)
Training: Networks trained on 22 cognitive tasks spanning visuomotor, working memory, match–nonmatch, and decision-making families; training stopped at >95% accuracy across all tasks
Wiring-cost constraint: L1 regularization penalizing connection strength × inter-unit distance (parameter λ), mimicking the brain’s economical wiring principle
Finding 1 — Connectivity: Increasing λ produced a stronger negative correlation between inter-unit distance and connection strength, replicating the distance-dependent decay of anatomical connectivity seen in the brain
Finding 2 — Activation patterns: Higher wiring-cost constraints selectively reduced activated association units while visual/motor unit activation remained stable, paralleling the brain’s reliance on association cortex for higher-order cognition
Finding 3 — Cognitive hierarchy: Simple visuomotor tasks remained at ceiling under strong constraints; working memory, match–nonmatch, and decision-making tasks degraded, linking association unit recruitment to cognitive capacity
Finding 4 — Structure–function coupling: BrainRNNs showed a significant negative correlation between spatial distance and functional connectivity, and a positive correlation between structural and functional connectivity — closely mirroring human MRI data from HCP
Finding 5 — Functional modules: Functional modules in BrainRNNs were neither purely spatially localized nor randomly distributed, matching the complementary localized-and-distributed character of canonical resting-state networks (e.g., Yeo parcellations) in the human cortex
Finding 6 — Module alignment: Functional modules significantly overlapped with structural network modules (AMI = 0.24, p < 0.0001) and predefined visual–motor–association segregation (AMI = 0.31, p < 0.0001), paralleling human brain organization
Finding 7 — Functional gradients: Macroscale functional gradients emerged along both topographic (posterior–anterior) and topological axes, resembling the principal sensorimotor-to-association gradient reported in human cortex
Control comparisons: Conventional RNNs (CRNNs) and spatially-embedded RNNs without areal interfaces (SpRNNs) showed weaker structure–function coupling, confirming the importance of brain-like input/output interfaces
Core conclusion: Brain-like structural constraints — spatial embedding, areal interfaces, and wiring-cost economy — are sufficient to give rise to macroscale functional organization in artificial networks, supporting the principle that function can be discovered from structure

From Mary Shelley to Netflix: a Pan-European perspective on public communication of neuroscience and neurotechnology

Frontiers in Neuroscience

Score: 0.72

Published: 2026-02-20T00:00:00+00:00

Tags: neurotechnology, communication, policy

Explicitly addresses public communication of neuroscience and neurotechnology; relevant for neurotechnology analyst context.

Title: “From Mary Shelley to Netflix: a Pan-European perspective on public communication of neuroscience and neurotechnology”
Authors: Ángeles Consuelo Gallar Martínez & Alicia De Lara González (Miguel Hernández University, Elche, Spain)
Type: Original research article in Frontiers in Neuroscience (Vol. 18, 2024); part of the NeurotechEU project
Aim: Explores 10 distinct European cultural and media formats that communicate neuroscience and neurotechnology to the public, analyzing how they shape public understanding
Scope: Examples drawn from 10 NeurotechEU member countries (France, Germany, Hungary, Iceland, Netherlands, Romania, Spain, Sweden, Türkiye, UK)
Format 1 — Fiction novel: Mary Shelley’s Frankenstein (1818, UK) as the first mass-distributed cultural product reflecting public fear of science and early bioethical concerns
Format 2 — Autobiography: Santiago Ramón y Cajal’s Recollections of My Life (1901, Spain) as a tool for constructing the public image of the scientist-hero and promoting “scientific patriotism”
Format 3 — Documentaries: Hungarian films Alzheimer (2020) and Hüség (2022) illustrate how long-form documentary can accurately convey the lived experience of neurological disease while building empathy and support for research
Format 4 — Art & brain imaging: Refik Anadol’s Connectome Architecture (Türkiye) demonstrates how neuroscience data visualization transcends traditional art to give tangible shape to the intangible concept of the mind
Format 5 — TED Talks: An Icelandic TEDx talk by a spinal-cord-injury patient shows how patients can become science communicators, though the format tends to emphasize personal narrative over scientific content
Format 6 — Children’s literature: Bräkiga bokstäver (Noisy Letters, 2018, Sweden) communicates dyslexia to young audiences, empowering neurodiverse children and building peer empathy without technical jargon
Format 7 — Museums/exhibits: The Humania exhibit at NEMO Science Museum (Netherlands) uses interactive installations to engage visitors in understanding the human body and brain
Format 8 — Science fiction TV: Netflix’s Osmosis (2019, France) sparks public debate on ethical “red flag” areas in neurotechnology — privacy of brain data, mind-reading, and consciousness manipulation
Format 9 — Feature film: Fritz Lang’s Metropolis (1927, Germany) introduced the first cinematic robot and remains a seminal work raising questions about AI, class struggle, and the creation of artificial beings
Format 10 — Short documentary/robotics: Erica: Man Made (2017, Romania/Japan) explores whether a human-like robot constitutes a “new category of being,” blending neuroscience with philosophical questions about consciousness and self-awareness
Key theme — Fear of science: From Shelley’s creature to modern AI anxiety, public communication consistently reflects societal fears about scientific overreach
Key theme — Neuromyths & accuracy: Neuroscience communication is prone to misinformation, oversimplification, and publication bias favoring positive results; media outlets frequently amplify inaccuracies
Key theme — Democratization: Social media, TED Talks, and children’s literature are expanding who communicates neuroscience (patients, artists, educators), though institutional accounts still struggle to reach beyond the scientific community
Conclusion: Cultural products — literature, film, art, interactive exhibits — serve as powerful vehicles for neuroscience communication, sparking ethical debate and shaping public understanding even when they sacrifice technical accuracy

Conditioned Graph Reconstruction of Brain Functional Network Connectivity Reveals Interpretable Latent Axes of Sex and Fluid Intelligence

bioRxiv Neuroscience

Score: 0.68

Published: 2026-02-20T00:00:00+00:00

Tags: methods, functional connectivity, computational

Generative framework for brain functional connectivity and network metrics with demographic/cognitive variables; methods for connectivity and interpretable encoding.

Title: “Conditioned Graph Reconstruction of Brain Functional Network Connectivity Reveals Interpretable Latent Axes of Sex and Fluid Intelligence”
Authors: Ishaan Batta, Meenu Ajith, & Vince D. Calhoun (TReNDS Center: Georgia State, Georgia Tech, Emory)
Journal target: Network Neuroscience (preprint posted Feb 20, 2026)
Goal: Develop a generative framework that models brain functional connectivity as graphs while encoding and reconstructing condition-specific differences (biological sex, fluid intelligence)
Method — C-GVAE: A conditional graph variational autoencoder that encodes static functional network connectivity (sFNC) into a latent space, conditioned on demographic/cognitive variables, and reconstructs sFNC matrices
Encoder: GATv2 (Graph Attention Network v2) selected as the best-performing encoder over GCN, GAT, and GIN, due to its dynamic attention mechanism capturing complex node-pair interactions
Data: >20,000 subjects from UK Biobank resting-state fMRI, split into two independent sub-cohorts (UKB1 and UKB2); validated on HCP dataset
sFNC construction: Pearson correlations among 53 Neuromark brain components across 7 functional subdomains; thresholded at τ = 0.1 for sparse, signed, weighted adjacency matrices
Reconstruction fidelity: Sex-conditioned model achieved correlation ~0.887, MSE ~0.027, Frobenius norm ~7.9, consistently outperforming standard CVAE baselines
Graph property preservation: Across 15 network statistics (density, clustering, k-core, community count, etc.), NRMSE ranged 0.148–0.156, demonstrating faithful topological reconstruction
Sex differences preserved: Reconstructed data retained significant male vs. female sFNC differences (FDR-corrected), with a false positive rate of only ~1.5%
Fluid intelligence associations preserved: Correlations between sFNC features and fluid intelligence scores in reconstructed data closely matched those in real data after FDR correction
Predictive utility: SVM classifiers trained on reconstructed data and tested on real data (and vice versa) achieved sex classification accuracy comparable to real-on-real training, suggesting enhanced separability in the generative latent space
Latent space — sex: Significant group differences between males and females were found across a substantial proportion of latent dimensions; one-hot probing revealed sex-related connectivity patterns in subcortical regions (caudate–subthalamus, caudate–hippocampus)
Latent space — fluid intelligence: Discriminative latent dimensions decoded to positive correlations between visual cortex/precuneus (DMN) and interactions between frontal regions and cerebellum, consistent with literature on higher-order cognitive performance
Key sex finding: Females showed greater within-DMN connectivity; males showed stronger task-positive/frontoparietal network connectivity — aligning with established literature
Generalizability: Unconditional model tested on independent HCP dataset achieved correlation 0.86 and Frobenius norm 7.70, confirming cross-dataset robustness
Conclusion: C-GVAE provides a scalable, network-informed framework for studying individual differences in brain functional connectivity, with potential applications in characterizing functional signatures for mental health conditions

The parafascicular nucleus of the thalamus orchestrates coordinated skeletomotor, autonomic, and aversive state transitions

bioRxiv Neuroscience

Score: 0.65

Published: 2026-02-20T00:00:00+00:00

Tags: optogenetics, circuits, methods

Optogenetics in mice with high-resolution continuous measures of skeletomotor and autonomic outputs; circuit-level neural manipulation.

Title: “The parafascicular nucleus of the thalamus orchestrates coordinated skeletomotor, autonomic, and aversive state transitions”
Authors: Marina Roshchina & Henry H. Yin (Duke University, Departments of Psychology/Neuroscience and Neurobiology)
Preprint: bioRxiv, posted Feb 20, 2026; funded by NIH R01NS121253
Question: What are the full behavioral outputs controlled by the parafascicular nucleus (Pf) of the thalamus beyond its known role in turning?
Approach: Optogenetic stimulation of Vglut2+ (glutamatergic) Pf projection neurons in male and female mice using Cre-dependent CoChR, combined with DeepLabCut tracking, pupillometry, and ECG recording
Animals: 21 mice (15 male, 6 female); Vglut2-ires-Cre and C57BL/6J lines; 12 opsin, 9 control
Finding 1 — Ipsiversive turning: Unilateral Pf stimulation reliably produced rapid ipsilateral head turns (latency ~98 ms), driven primarily by head reorientation independent of body locomotion, with power-dependent angular velocity (RM ANOVA, p < 0.0001)
Finding 2 — Facial/whisker movements: In head-fixed mice, Pf stimulation evoked rapid nose and whisker movements time-locked to individual laser pulses (up to 10 Hz), with whisker protraction increasing with laser power
Finding 3 — Pupil constriction: Pf stimulation produced robust pupil constriction in both eyes; even a single 10-ms pulse triggered constriction (onset ~224 ms, peak ~679 ms, ~19.6% from baseline); response scaled with pulse number (RM ANOVA, p < 0.0001)
Finding 4 — Heart rate reduction (bradycardia): ECG in freely moving mice showed significant heart rate decrease ~1.5–2 s after stimulation onset at 4–6 mW/20 Hz, with a ceiling effect at 15–20 pulses (RM ANOVA, p < 0.0001)
Finding 5 — Strong aversion: In a real-time place avoidance task, CoChR mice rapidly avoided the stimulation-paired compartment (average visit ~1.6 s vs. 33.2 s in controls); avoidance persisted after laser offset (two-way ANOVA interaction, p < 0.001)
Finding 6 — Distinct temporal profiles: Whisker movements were fast and pulse-locked (skeletomotor); pupil and heart rate responses were slow, smooth, and parasympathetic — indicating concurrent but separable motor and autonomic output pathways
Finding 7 — Topographic analysis: Posterior Pf stimulation sites produced larger turning amplitudes (p = 0.036), but pupil constriction and heart rate effects showed no dependence on fiber position along the mediolateral or anteroposterior axis
Finding 8 — No covariation: Motor output magnitude (angular velocity) did not correlate with autonomic output magnitude (pupil constriction, heart rate reduction) across stimulation sites, suggesting partially independent downstream pathways
Conclusion: The Pf serves as a major integrative hub coordinating skeletomotor outputs (turning, orofacial movements), parasympathetic responses (pupil constriction, bradycardia), and aversive motivational states — broadening its known role beyond attention and action selection

A quantitative census of millions of postsynaptic structures in a large electron microscopy volume of mouse visual cortex

bioRxiv Neuroscience

Score: 0.78

Published: 2026-02-20T00:00:00+00:00

Tags: methods, connectomics, EM, circuits

Dense EM connectomics and subcellular synaptic targeting methods; direct relevance to neural circuit and methods for large-scale neuroanatomy.

Title: “A quantitative census of millions of postsynaptic structures in a large electron microscopy volume of mouse visual cortex”
Authors: Benjamin D. Pedigo, Bethanny P. Danskin, Rachael Swanstrom, Erika Neace, Sven Dorkenwald, Nuno Maçarico da Costa, Casey M. Schneider-Mizell, & Forrest Collman (Allen Institute; MIT)
Status: Draft preprint on bioRxiv, posted Feb 20, 2026
Goal: Develop a scalable, cost-efficient pipeline for classifying postsynaptic targets (soma, dendritic shaft, or spine) across an entire large-scale EM connectome
Dataset: MICrONS mouse visual cortex — 75,241 neurons with 208.6 million input synapses total
Core method — Heat Kernel Signatures (HKS): Features derived from heat diffusion on the intrinsic geometry of neuronal mesh surfaces; heat dissipates slowly from spines (isolated) vs. quickly from somas, creating naturally separable feature vectors
Classifier: Random forest trained on HKS features using 110,946 manually labeled synapses; achieved weighted F1 score of 0.961 (96.8% accuracy on held-out neurons)
Scalability optimizations: Mesh simplification, overlapping mesh chunking, robust Laplacian, band-by-band eigendecomposition, and mesh agglomeration for lossy compression — yielding 42× speedup and 27× storage reduction with no accuracy loss
Cost: Entire MICrONS dataset processed in the commercial cloud for less than $500
Validation: Macro-average F1 = 0.977 on excitatory neurons and 0.925 on inhibitory neurons in independent validation across cell types
Finding 1 — Spine input rates: Excitatory neurons received 70.3% of inputs onto spines vs. 19.8% for inhibitory neurons; Layer 2/3 and Layer 4 IT cells had the highest spine proportions (~75–76%)
Finding 2 — Cell-type exceptions: Layer 5 near-projecting (NP, 42.1%) and Layer 6 corticothalamic (CT, 62.5%) excitatory cells had notably lower spine input proportions; among inhibitory cells, Martinotti cells had the highest (28.8%)
Finding 3 — Output targeting: Excitatory neurons targeted spines of other excitatory neurons 83.3% of the time; L5 NP and L6 CT cells were exceptions, often targeting excitatory shafts instead
Finding 4 — Within-type variability: Substantial cell-to-cell variation in spine targeting existed even within a single cell type (e.g., L5 ET ranged from 32.6% to 73.3% spine input)
Finding 5 — Multi-input spines: Frequency of spines receiving multiple synaptic inputs varied widely across cells (well-described by a lognormal distribution), even within a cell type, and was not strongly correlated with spine size, distance from soma, or presynaptic cell type
Finding 6 — Excitatory + inhibitory co-innervation: Most multiply-innervated spines received one excitatory and one inhibitory synapse; excitatory cleft size correlated strongly with spine volume (r = 0.82), but inhibitory cleft size did not (r = 0.10)
Public resource: All postsynaptic target predictions are publicly available via CAVE; pipeline code is open-source on GitHub (meshmash, cloud-mesh)
Broader significance: Demonstrates that mesh-derived geometric features (HKS) are a powerful, scalable primitive for characterizing neural morphology without processing raw imaging data

Cholinergic—dopaminergic interplay underlies prediction error broadcasting

bioRxiv Neuroscience

Score: 0.72

Published: 2026-02-20T00:00:00+00:00

Tags: computational, neuromodulation, reinforcement-learning

Neuromodulatory systems and coordinated encoding of prediction errors; core computational neuroscience and reinforcement learning.

Title: “Cholinergic–dopaminergic interplay underlies prediction error broadcasting”
Authors: Bálint Király, Vivien Pillár, Írisz Szabó, Dániel Schlingloff, Panna Hegedüs, Krisztián Szigeti, Yulong Li, & Balázs Hangya (HUN-REN Institute of Experimental Medicine, Budapest; Medical University of Vienna; University of Copenhagen; Peking University)
Preprint: bioRxiv, posted Feb 20, 2026
Question: Do basal forebrain cholinergic (ACh) and midbrain dopaminergic (DA) systems cooperate or compete to jointly control prediction error signaling during reinforcement learning?
Task: Head-fixed mice performed a psychometric auditory operant conditioning task with fixed and novel tone–outcome associations of varying difficulty, generating learning psychometric curves
Methods: Dual cell-type optogenetic tagging (ChAT-Cre × DAT-Cre mice), simultaneous single-neuron recordings in HDB (basal forebrain) and VTA (midbrain), GRAB-ACh3.0/GRAB-DA2m fiber photometry in BLA and ventral striatum, channelrhodopsin-assisted circuit mapping, and chemogenetic (DREADD) suppression
Finding 1 — Synergistic reward coding: Both ACh (in BLA) and DA (in ventral striatum) were released phasically to reward-predicting cues and rewards, scaling with prediction error — showing synergistic co-activation for positive outcomes
Finding 2 — Antagonistic punishment coding: Punishments increased cholinergic neuron firing but suppressed ~half of dopaminergic neurons (T1-DANs), revealing opposite valence responses; trial-by-trial activity showed negative noise correlations between BFCNs and T1-DANs
Finding 3 — Two DAN subtypes: T1-DANs showed anti-correlated activity with BFCNs (suppressed by punishment, negative noise correlation); T2-DANs showed positive correlations with BFCNs, consistent with synergistic coding
Finding 4 — Disynaptic inhibitory circuit: No direct BFCN→VTA projection exists; instead, BFCNs excite GABAergic HDB neurons that project to VTA and inhibit T1-DANs (HDB_ACh → HDB_GABA → VTA_DA pathway), confirmed by slice electrophysiology showing iPSCs in ~half of TH+ VTA neurons
Finding 5 — Cholinergic suppression impairs learning: Chemogenetic silencing of BFCNs (hM4Di + C21) virtually eliminated learning of first novel associations and slowed subsequent learning, with intermediate-difficulty associations most affected
Finding 6 — Cholinergic suppression disrupts DA prediction errors: BFCN silencing reduced go-cue-evoked DA release (disrupting reward anticipation signal) while simultaneously increasing DA levels after punishments (relieving punishment-induced suppression) — confirming dual synergistic/antagonistic roles
Finding 7 — Cholinergic activity anticipates behavior: BFCN firing preceded and predicted trial-by-trial behavioral performance and T1-DAN responses during learning, suggesting a causal upstream role
Core conclusion: Prediction error signaling is not the province of a single neuromodulator; instead, it is jointly implemented through coordinated synergistic (reward) and antagonistic (punishment) interactions between cholinergic and dopaminergic systems — challenging the textbook view of their functional independence
Clinical implication: Cross-system neuromodulatory coordination may be relevant to neuropsychiatric diseases involving disrupted reinforcement learning (e.g., addiction, depression, Parkinson’s disease)

bioRxiv Neuroscience

Score: 0.65

Published: 2026-02-21T00:00:00+00:00

Tags: neural circuits, Drosophila, computational, behavior

Identifies neural circuit for decision to flee during fighting in Drosophila; circuit-level mechanisms and evolutionary game theory align with systems/computational neuroscience.

Title: “Neural circuits regulating social dominance implement a strategy predicted by evolutionary game theory”
Authors: Donovan Ventimiglia, Elizabeth Chamiec-Case, Claire S. Lee, Bruce Ruff, & Kenta Asahina (Salk Institute; UC San Diego; Johns Hopkins)
Preprint: bioRxiv, posted Feb 21, 2026; funded by NIH R35GM119844
Question: Are the competitive strategies predicted by evolutionary game theory (War of Attrition model) embedded in the neural circuits that regulate aggression and defeat in animals?
Model organism: Male Drosophila melanogaster, which engage in weaponless, wrestling-like persistence-based contests
Key prediction tested: The War of Attrition (WOA) model predicts defeat onset timing should follow an exponential probability distribution (p(x) = (1/v) \exp(-x/v)), where (v) = payoff
Finding 1 — Exponential defeat distribution: Both spontaneous and optogenetically induced fights produced defeat onset times that followed an exponential distribution, exactly matching the WOA prediction
Finding 2 — Tk-GAL4FruM neuron suppression: Defeat occurs through inhibition of Tk-GAL4FruM neurons, which promote male aggressive arousal; losers showed ~50% reduction in calcium responses in these neurons compared to winners and naïve flies
Finding 3 — Increasing Tk-GAL4FruM drive prolongs fighting: Higher optogenetic stimulation power on Tk-GAL4FruM neurons extended defeat onset time in losers, consistent with the model
Finding 4 — PPL1 dopaminergic neurons required: Silencing aversive PPL1 DANs (projecting to mushroom body vertical lobes) prevented defeat formation in 55% of contests; both flies kept fighting
Finding 5 — V2 MBONs required: Silencing V2 mushroom body output neurons (MBON-18/19 specifically) blocked defeat in 51% of contests; these neurons are classically associated with aversive learning
Finding 6 — V2 MBON activation induces rapid defeat: Co-activating V2 MBONs with Tk-GAL4FruM neurons caused flies to lose 94% of contests (n=69), with a left-shifted (faster) exponential defeat distribution — consistent with increased perceived cost in the WOA model
Finding 7 — PAM reward DANs promote winning: Activating reward-encoding PAM dopaminergic neurons biased flies toward winning, revealing a dual role for dopamine — aversive (PPL1) drives defeat, reward (PAM) promotes persistence
Finding 8 — Olfaction not required: Antenna removal and Orco neuron silencing had no effect on defeat onset, showing the PPL1–V2 MBON circuit operates independently of olfactory conditioning
Finding 9 — Internal state modulates the distribution: Hunger (starvation) and resource availability shifted the exponential defeat distribution, consistent with the WOA prediction that payoff ((v)) modulates fighting persistence
Finding 10 — Prolonged suppression after defeat: Losers’ Tk-GAL4FruM neurons entered a sustained suppressed state lasting hours, during which optogenetic activation no longer triggered aggression
Core conclusion: Natural selection has embedded game-theoretic competitive strategies directly into neural circuit architecture — the mushroom body integrates internal states and generates probabilistic outputs governing the exponential defeat distribution predicted by the War of Attrition model

Neural Noise

Browse

BCI Weekly Brief (week of 2026-02-16)

Temporal Coding rather than Circuit Wiring allows Hippocampal CA3 Neurons to Dynamically Distinguish Different Cortical Inputs

Discovering macroscale functional organization on the structure of brain-like recurrent neural networks

From Mary Shelley to Netflix: a Pan-European perspective on public communication of neuroscience and neurotechnology

Conditioned Graph Reconstruction of Brain Functional Network Connectivity Reveals Interpretable Latent Axes of Sex and Fluid Intelligence

The parafascicular nucleus of the thalamus orchestrates coordinated skeletomotor, autonomic, and aversive state transitions

A quantitative census of millions of postsynaptic structures in a large electron microscopy volume of mouse visual cortex

Cholinergic—dopaminergic interplay underlies prediction error broadcasting

Graph View

Table of Contents

Backlinks