Mechanisms of robust feature extraction in early visual processing
Mechanisms of robust feature extraction in early visual processing
by Sebastian Mauricio Molina Obando
Date of Examination:2022-02-09
Date of issue:2022-02-24
Advisor:Prof. Dr. Marion Silies
Referee:Prof. Dr. Marion Silies
Referee:Prof. Dr. André Fiala
Persistent Address:
http://dx.doi.org/10.53846/goediss-461
Files in this item
Name:PhD_Thesis_Sebastian_Molina_print.pdf
Size:14.8Mb
Format:PDF
Description:Ph.D. Thesis
The following license files are associated with this item:
Abstract
English
The detection of changes in signal intensity, namely contrast, happens across senses. Visual systems extract contrast information at very early stages. The first cells to “sense” light, the photoreceptors, carry contrast information which is processed by their downstream neuronal circuitry. Visual systems have evolved to ensure robust ON and OFF detection in distinct ON- and OFF-selective or pathways. Downstream of a common photoreceptor input, the molecular mechanism to implement the split in ON and OFF pathways has been described in the vertebrate retina and involves a signal inversion in the ON pathway through inhibitory glutamatergic synapses at individual synaptic connection (Masu et al., 1995). In the first study of this thesis, using cell-type specific manipulations and pharmacogenetics, I demonstrate that the extraction of ON selectivity in Drosophila is rather a distributed and multisynaptic computation involving glutamatergic and GABAergic inhibitions. This raises the possibility that using more specific manipulation in other systems, including the vertebrate retina, might reveal a similar distributed coding strategy. Contrast extraction by early visual processing serves as basis for later computations and eventually to guide motor behavior. In a dynamically changing world, rapid changes in illumination challenges proper contrast extraction in early stages when photoreceptors cannot adapt fast enough. Corrections downstream of photoreceptors are necessary to detect the same contrast equally well across distinct illuminations, namely in a luminance-invariant manner. In the second study, this thesis explores the correction mechanism behind ON and OFF luminance-invariant behaviors in Drosophila. In both ON and OFF pathways, luminance-sensitive signals scale contrast signals to achieve luminance-invariant behaviors. For that, distinct lamina first-order interneurons diversify and asymmetrically distribute luminance and contrast signal across pathways. This second study changes the current understanding of the role of these three first-order interneurons, previously thought to be specific inputs for either the ON or the OFF pathway. In the first study and second study we show that contrast extraction is supported by parallel inputs, involving different inhibitory connections and distinct lamina input neurons, respectively. In the third study of this thesis, I show that a parallel input architecture leads to robust contrast extraction. Individual input pathways can be sufficient but not necessarily required to implement ON selectivity. These findings align with the idea of neuronal circuits being degenerate, in which structurally different elements can implement the same function, leading to robustness. Since comparative connectomics reveals parallel connectivity as a common trait across animals and sensory systems (Barsotti et al., 2021), a distributed feature extraction might be a general strategy present in other systems too to achieve functional robustness.
Keywords: Neuroscience; Vision; Drosophila; Contrast extraction; Feature detection; Robustness