Rhythm hoppers: How the auditory neural network in crickets has adapted and evolved to recognize the diversity of male calling song features across multiple timescales

Crickets sing in the chorus alongside almost every other species in the animal kingdom, and their songs communicate important information, most critically pertaining to mating. A central goal of this thesis is understanding how constituent computations limit and shape neural network responses into the diversity of behavioral data observed across cricket species at multiple time scales. As this thesis will demonstrate, the song recognition network of crickets is responsible for recognizing cues at both the pulse and chirp timescale. Preference functions define the specificity of the recognition of these cues, and are a critical insight into neural mechanisms that can be measured at the level of behavior. On the pulse scale, preference functions are typically unimodal–meaning that there is a single contiguous peak, and closed–meaning that all edges of the function have a well-defined boundary. In contrast, on the chirp scale, preference functions exist on a continuum from open to closed depending on chirp duty cycle. In order to understand how a common network is able to shape preference functions on two time scales, wherein the two scales differ in the characteristic type of tuning that defines individual functions within each scale; it is necessary to examine the characteristics of the computations that make up the song recognition circuit. The first part of this thesis examines the interplay between various sources of suppression that shape the delay and coincidence detection mechanism at the heart of the song recognition circuit. A cricket species with an unusual resonant recognition phenotype, in contrast to the single peaked or unimodal preference functions normally observed on the pulse scale, provides a valuable source of in- sight to how multiple sources of suppression function within the song recognition network to eliminate off-target responses and ensure simple response functions with highly specific tuning. The specific pattern of period and duty cycle tuning characterizing each of the peaks of this resonant response provides significant insight into the mechanics of how losses of suppression within song recognition networks appear at the level of behavior; and furthermore how downstream adaptation might suppress resonances by selecting between resonant peaks to regain the more desirable unimodal type. The question of why and how resonances may be typically suppressed in most observed phenotypes is discussed from an evolutionary and mechanistic perspective, including reasoning for why the remaining rare resonance types may encourage speciation. These resonances, now consequential at the level of behavior, may be useful for allowing a nonlinear, saltatory evolutionary paradigm in which a single instance of adaptation selects from among the multiple peaks, quickly transitioning female preference to a new calling song type. The second part of this thesis examines how a specific kind of adaptation in the network may help determine the characteristics of chirp tuning, revealing the network to be dual purpose–shaping pulse and chirp preferences simultaneously. Relationships between tuning profiles on the pulse and chirp scales provide evidence for the notion of a single network that serves the role of shaping the preference functions of both timescales simultaneously. A five neuron network model for song recognition described for Gryllus bimaculatus is used to demonstrate how adaptation in the ascending neuron AN1 and the coincidence detection neuron LN3 are the critical computations required for chirp tuning. The mechanistic in- sights gained from the study of this network lead to a model of how chirp preferences for 16 other cricket species arise, ultimately demonstrating that chirp preferences exist on a continuum from bandpass-DC to highpass-DC, depending on the strength of adaptation in the song recognition circuit. This analysis of the neural mechanisms governing each timescale leads to the conclusion that more tightly constrained preference functions on the pulse scale are correlated with bandpass- type preference functions on the chirp scale, indicating that adaptation is a critical link between timescales across species. Overall, this thesis reveals how careful examination of the unique characteristics of behavioral preference functions on both timescales can provide insight into the mechanisms within the song recognition network that are necessary and sufficient to explain the diversity of tuning for song traits observed among many female cricket species.