A1 Towards a complete representation of visual information at a single retinal location
Many theoretical studies have focused on neural population codes for which individual neurons exhibit very similar tuning properties. Different retinal ganglion cells (RGCs), however, exhibit very different nonlinearities and spatio-temporal tuning properties, rendering the analysis of population codes more difficult. We plan to develop and analyse computational models for optimal stimulus reconstruction in order to determine which visual information is preserved and which is already discarded at the retina level. Within such a stimulus reconstruction framework, we can describe systematically how the different RGC types contribute to the representation of spatio-temporal information in a local image patch (“information fingerprint”).
Data for the modelling is being acquired by optically recording light-evoked calcium activity at single-cell resolution from mouse RGCs using two-photon microscopy. We have developed a battery of visual stimuli towards a physiological characterization of all cells in the ganglion cell layer (GCL) in an image patch. Currently, our database contains >10,000 GCL cells from >30 retinas. Out of those, the best responding cells were used as the input to our Mixture of Gaussians Clustering model. Based on their responses to our light stimulus set, we can distinguish between 15 and 20 retinal ganglion cell clusters, several of which are readily matched to known morphological/genetic types. We also find a number of amacrine cell clusters that roughly matches the number of types predicted by anatomy.
We currently refine our clustering algorithm and are in the process of adding data from transgenic animals as well as further immuno data to verify the clustering and to link clusters to known morphologically and genetically identified cell types. Moreover, we continue to collect single unit extracellular recordings followed by intracellular filling of individual RGCs to generate the link between functional clusters, spike output and cell morphology.
From this project we expect a more complete view of what visual information is preserved in the retina and how it is decomposed into the different channels. Such a local retinal “information fingerprint” should be very interesting, not only for our understanding of neuronal computations in the healthy retina, but also as a research tool for evaluating specific functional deficiencies in degenerating retina.
Project leaders: Thomas Euler, Matthias Bethge
Coworkers: Tom Baden, Philipp Berens*, Katrin Franke, Miroslav Rezac
* funded by BMBF
- Baden T, Behrens P, Bethge M, Euler T. (2013) Spikes in Mammalian Bipolar Cells Support Temporal Layering of the Inner Retina. Curr Biol. 23(1):48-52.
- Auferkorte ON, Baden T, Kaushalya SK, Zabouri N, Rudolph U, Haverkamp S, Euler T. (2012) GABA(A) receptors containing the α2 subunit are critical for direction-selective inhibition in the retina. PLoS One 7(4):e35109.
- Macke JH, Zeck G, Bethge M (2007), Receptive fields without spike-triggering, Advances in Neural Information Processing Systems 20: Proceedings of the 2007 Conference, 969-976. (Eds.) Platt JC, Koller D, Singer Y, Roweis S, MIT Press, Cambridge, MA, USA