P096 Coding and information processing with firing threshold adaptation near criticality in E-I networks
Mauricio Girardi-Schappo*1, Leonard Maler2, André Longtin3
1Departamento de Física, Universidade Federal de Santa Catarina, Florianopolis SC 88040-900, Brazil
2Department of Cellular and Molecular Medicine, University of Ottawa, Ottawa ON K1H 8M5, Canada
3Department of Physics, University of Ottawa, Ottawa,ON K1N 6N5, Canada
*Email: girardi.s@gmail.com
Introduction The brain can encode information in output firing rates of neuronal populations or spike patterns. Weak inputs have limited impact on output rates, which challenges rate coding as a sole explanatory mechanism for sensory processing. Spike patterns contribute to perception and memory via sparse, combinatorial codes, enhancing memory capacity and information transmission [1, 2]. Here, we compare these two forms of coding in a neural network with and without threshold adaptation of excitatory neurons, including or excluding inhibitory neurons. This extends our previous study and assess the impact of inhibition on coding properties of adaptive networks. Methods We model a recurrent excitatory network incorporating an inhibitory population of neurons, which, in line with biological evidence, acts as a stochastic process independent of immediate excitatory spikes [3-5]. Networks with and without threshold adaptation are compared using measures of pattern coding, rate coding, and mutual information [6]. We examine whether threshold adaptation maintains its functional advantages when weakly coupled inhibitory inputs are introduced. The results are analyzed in the light of self-organized (quasi-)criticality [7], and a new theory for near-critical information processing is proposed. Results In the limit of weak inhibition, threshold adaptation maintains its ability to enhance coding of weak inputs via firing rate variance. Adaptive networks facilitate a smooth transition from pattern to rate coding, optimizing both coding strategies. This dynamic is lost in non-adaptive networks, which require stronger inputs for pattern coding. Constant-threshold networks rely on supercritical states for pattern coding, whereas adaptation allows robust coding through a near-critical dynamics. The threshold recovery timescale of 100ms to 1000ms is found to favor the pattern coding of weak inputs, matching experimental observation in dentate gyrus neurons [5]. However, the dynamic range of adaptive networks matches the subcritical regime of constant-threshold networks, contrary to what would be expected by the theory of self-organized criticality alone. Discussion
Threshold adaptation is a biologically relevant mechanism that enhances weak stimulus processing by pattern coding, while keeping the capacity to perform rate coding of strong inputs. The optimal recovery timescale aligns with observations in the hippocampus and other brain regions. Adaptation improves information transmission, feature selectivity, and neural synchrony [8], supporting its role in sensory discrimination and memory tasks. Our findings reinforce the idea that weakly coupled inhibition does not disrupt threshold adaptation’s advantages, suggesting it is a robust coding mechanism across diverse neural circuits.
Acknowledgements The authors thank financial support through NSERC grants BCPIR/493076-2017 and RGPIN/06204-2014 and the University of Ottawa’s Research Chair in Neurophysics under Grant No. 123917. M.G.-S. thanks financial support from Fundacao de Amparo a Pesquisa e Inovacao do Estado de Santa Catarina (FAPESC), Edital 21/2024 (grant n. 2024TR002507). References 1.https://doi.org/10.1016/j.conb.2004.07.007 2.https://doi.org/10.1523/JNEUROSCI.3773-10.2011 3.https://doi.org/10.1038/s41583-019-0260-z 4.https://doi.org/10.1152/jn.00811.2015 5.https://doi.org/10.1101/2022.03.07.483263 6.https://doi.org/10.1007/s10827-007-0033-y 7.https://doi.org/10.1016/j.chaos.2022.111877 8.https://doi.org/10.1523/JNEUROSCI.4906-04.2005
