Abstract:
More than 30 functional types of retinal ganglion cells (RGCs) compute in parallel distinct features of the visual world and send this information to the brain. Little is known, however, about which RGC types project to the dorsolateral geniculate nucleus (dLGN) of the thalamus, and how the different RGC channels recombine there. Interest in these questions has been fuelled by recent estimates of retinogeniculate convergence obtained by anatomical work, which far exceeded those obtained in electrophysiological recordings.
To get insights into the nature of retinal input to the dLGN, we conditionally expressed the genetically encoded Ca2+ indicator GCaMP6f in dLGN-projecting (dLGN-p) RGCs, followed by in vitro retinal two-photon Ca2+ imaging of light-evoked responses. Visual stimuli matched those in a previously published survey of mouse functional RGC types (Baden et al., 2016). We then assigned each dLGN-p RGC to the best-matching RGC type with the best-matching response properties. We found that most functional RGC types seem to innervate dLGN, with certain types, such as ON- and OFF alpha cells or OFF supressed cells, showing clear overrepresentations.
In a separate set of experiments, we characterized the responses of dLGN neurons to the same visual stimuli using in-vivo extracellular multi-electrode recordings in the dLGN of awake, head-fixed mice. We quantitatively assessed the degree of diversity in the dLGN responses by using sparse non-negative matrix factorization (NNMF), which decomposed the dLGN population response into a rich and highly diverse set of components.
Finally, we linked the functionally characterized population of dLGN-projecting RGCs and geniculate neurons, via computational modelling to provide a quantitative account of the transformations in visual representation between RGCs and dLGN neurons. We found that responses of dLGN neurons could be best predicted as a sparse linear combination of responses from 3-7 different RGC types.
In conclusion, this study provides fundamental insights into how the representation of visual information changes along the first stages of the retino-geniculo-cortical pathway, suggesting that the precortical basis of vision displays an unexpectedly rich functional diversity of retino- geniculate projections and thalamic features that can be modelled by a sparse feed-forward model.