Visual acuity and neural encoding of the fly Chrysomya megacephala

Author(s)

Nelson Mesquita Fernandes

• Document type: Doctoral Thesis
• Press: São Carlos.
• Institution: Universidade de São Paulo (USP). Instituto de Física de São Carlos (IFSC/BT)
• Defense date: 2010-03-12

Examining board members

Roland Koberle; Nestor Felipe Caticha Alfonso; Leonardo Paulo Maia; Antonio Carlos Roque da Silva Filho; Mauro Copelli Lopes da Silva

• Advisor: Roland Koberle

Abstract

We describe the practices of capturing, creation, and microsurgery of the flies Chrysomya megacephala. We present the procedures of stimulus generation and recording of the activity of the two H1 neurons in the lobula plate of its brain. One first result presented is related to its visual system acuity. We developed a method to compare its spontaneous firing rate with the H1s responses to excitatory and inhibitory stimuli. We show that the flys visual system is not only adapted to detect large optic flows but is also capable to detect small velocities about 1, 5o.s-1 with just 0, 25o of amplitude. These values show that the fly is capable to detect angular displacements much smallers than its ommatidial aperture, = 1 2o. Another relevant result is attained studying the processes of neural encode-decode. Some sensorial systems act as an analog-to-digital conversor, these systems encode the input stimulus S(t) in a sequence of action potential, spikes. The decoding process of the neural response consists of capturing this set of spikes and to generate an estimate Se(t) of the stimulus. This process requires the computation and subsequent inversion of high order correlations functions. The dimension of the matrixes that represent these functions can become prohibitively large. We present an efficient method to reduce these correlation functions. This approximation has low computational cost, avoids large matrixes inversion and give to us an excellent result to the stimulus reconstruction. We tested the reconstruction quality of rotational and translational stimuli. The contribution of second order stimulus reconstruction kernels is just 8% of first order kernels contribution. However, in specific times, the addition of these kernels may represent a 100% contribution. Finally, we investigate which stimulus features are codified by the H1 neurons. The stimulus space has a set of about 2 × 1096 elements. It is impossible to imagine that the system formed by the two H1 neurons could be able to encode efficiently this amount of elements. It is reasonable to consider that this system is at least able to encode an essencial characteristic of movement, its direction horizontal rotations to the right or to the left. Therefore, we presente two different stimuli to the fly, one which have velocities taken from a Gaussian distribution and another which contains just the direction of this movement. We obtain about 80 - 90% correlation between the estimates of both stimuli, estimates obtained through linear reconstruction methods. We obtain about 85% of efficiency in the prediction of stimulus direction. We find just a 10% difference between the information rate transmitted about the Gaussian stimulus and its reduced version using Information Theory. We conclude that the common attribute of these stimuli, the direction of movement, is the relevant attribute to be codified by the H1 neurons. (AU)