Project Details
Projekt
The impact of network architecture on evolvability and robustness: Comparing the song recognition networks of grasshoppers and crickets
Applicants
Professor Dr. Jan Benda; Professor Dr. Jan Clemens
Subject Area
Cognitive, Systems and Behavioural Neurobiology
Term
since 2020
Project identifier
Deutsche Forschungsgemeinschaft (DFG) - Project number 430158535
Animal behavior is highly diverse, even among closely related species. For instance, acoustic signals used during mating are species-specific and diverge during speciation to maintain species separation. This implies that the neural networks that recognize these songs are highly evolvable. At the same time, song recognition needs to be robust to environmental noise and to genetic mutations that perturb network parameters. The observed evolvability and robustness (E&R) of neural networks seem to be in conflict – evolution is a plastic response to perturbations, while robustness is a compensatory response to perturbations. However, studies on gene regulatory networks have shown that E&R can coexist in high-dimensional and nonlinear parameter spaces. We therefore propose to study E&R of song recognition not at the level of human-defined song properties, but at the level of the feature and parameter space of the recognition networks themselves. We will compare the evolvability and robustness of song recognition in two insect groups– grasshoppers and crickets. Both groups produce pulsed songs but recognize these songs using different network architectures. Grasshopper have a parallel network architecture: A species-shared peripheral network produces a high-dimensional representation of song features in 20+ neurons. The species-specific songs are then recognized based on a linear combination of features in a species-specific manner in the brain. E&R is likely achieved through redundancy in the high-dimensional feature space. By contrast, crickets recognize songs using a serial network architecture: five serially connected neurons in the brain recognize a song through a series of nonlinear temporal computations. The network’s small size prevents redundancy or modularity at the cellular or feature level, but may support E&R through a nonlinear and high-dimensional parameter space. Based on existing and physiologically plausible models of the song recognition networks in crickets and grasshoppers, we will test whether and through which properties the networks are robust and evolvable. We will quantify E&R in both networks using sensitivity analyses to determine to what extent both properties co-exist in these networks. We will then assess how the network structure constrains and supports phenotypic transitions and diversity within each species group. Lastly, we will explore the limits of phenotypic flexibility, by testing to what extent each network can discriminate the songs of the other species group. This will reveal to what extent differences in the songs – crickets sing simple, unmodulated pulses while grasshoppers produce songs with complex, modulated pulses – arise from limitations of the recognition mechanism in the two species groups.
DFG Programme
Priority Programmes
