Visual information processing and the natural input

We are interested how individual and small groups of neurons arranged into microcircuits break sensory patterns into parallel, highly specific representations of the outside world. We use a combination of 2-photon imaging of genetically encoded biosensors, electrophysiology, patterned light stimulation and computational modelling to probe the visual processing of individual and networks of neurons in the retina and brain of zebrafish and mice. We also study the distribution of visual features in the natural visual world to better understand how different visual systems have evolved to optimally process their own natural input.

Visual feature extraction and population coding

Population imaging of retinal ganglion cells

Somata of the ganglion cells layer labelled with synthetic calcium dye OGB-1 (green) with tdtomato expressed in specific populations (red).

The retina breaks light patterns into increasingly specific internal representations of the visual world1–3. At the level of spike trains being sent through the optic nerve, only a very small fraction of the visual information sampled by the photoreceptor is retained. Depending on the species, visual information picked up by only a handful of photoreceptor types is mapped via bipolar cells onto several dozen types of retinal ganglion cells (RGCs). Here, each type of RGC carries its own, unique and highly specific aspect of the original input to higher visual centres. Within the retinal network, interactions with interneurons fundamentally shape the response properties of individual pathway elements. By monitoring the visual responses to a defined and ecologically relevant set of visual stimuli at each processing level we study the gradual evolution of feature extraction as visual information trickles through the retinal network4–7.

Vision in its natural context

The mouse's natural visual world

The horizon divides the world into two very different visual environments.

Sensory systems have evolved to specifically sample features in the outside world critical to an animal’s survival8,9. Depending on the lifestyle of an animal, the importance of individual features differs widely. For example, in many terrestrial species, one such evolutionary adaptation is particularly prominent: Rather than uniformly sampling visual information over the entire visual field, the visual system features a pronounced separation of “green” and “blue” light sensitive regions of the visual field, presumably matched to differentially sample visual information above and below the horizon4. Such sampling and processing adaptations to the natural visual world are a fundamental feature of vision in all species, and we are interested in understanding which computational knobs and gears are tuned as an animal grows or evolves to explore new visual niches. Here, zebrafish are an excellent model: they undergo two distinct life-stages – larva and adult – with distinct lifestyles in different visual environments and hence different feature-detection requirements. Comparison of processing strategies employed by different life-stage visual systems of zebrafish with that of other species will lead to an increasingly generalised understanding of biological vision.

Synaptic computation

Bipolar cells of the mouse retina

Optical imaging allows non-invasive access to inner retinal neurons at single synapse resolution.

The computational power of retinal networks is deeply rooted in their synaptic architecture10–16. Within the retina’s two synaptic layers, highly specialised ribbon-type synapses of photoreceptors and bipolar cells respectively do not merely “forward” the activity of their parent neuron to their postsynaptic partners. Instead, each synapse receives a myriad of feedback and feed-forward signals through the network. Here, retinal bipolar cells (BCs) are of particular importance: They are only neurons that communicate with every other class of retinal neuron through their dendrites and axonal terminals. In the inner retina, they form complex synapses with inhibitory amacrine cells (AC) to ultimately drive retinal ganglion cells (RGCs) and thus retinal output. This computationally powerful synapse is probably the most important yet yeast understood computational locus of the circuit. Using 2-photon imaging of voltage, calcium and synaptic release signals in vivo we aim to probe the visual response properties of BCs towards a better understanding of visual signal transformations occurring locally within individual BC synapses.

Why retina?

The retinal circuitory

The mammalian retina. By T. Euler

How are networks consisting of anatomically and functionally distinct neuronal elements adapted to sample, filter and forward information contained in a sensory stimulus or in central neuronal ensemble activity towards their specific computational goals?  Faced with the staggering complexity of even the simplest neuronal circuits, addressing such questions continues to present a daunting task. One very promising approach will lie in promoting our understanding of computations performed by specific, very well described neuronal ensembles with a clear function, input-output organisation and experimental accessibility, such as the vertebrate retina.

The retina is a beautiful neural circuit, and presents a powerful model system to study many general aspects of neuronal network computation, including sampling optimization, filtering as well as compression of the incoming stream of visual information into highly selective parallel streams towards transmission to the brain. First, the retina operates almost exclusively as a feed-forward network, receiving little or no fast feedback from the brain. Accordingly, great control can be maintained over network input (the visual stimulus) while retinal output is readily recorded in single unit or population measurements of retinal ganglion cells (RGCs), the retina’s output neurons. Second, the anatomical basis of the retina of different species ranks amongst the most well-described ‘complete’ networks, with ready availability of specific transgenic lines and molecular markers. Third, individual retinal microcircuits tile the retinal surface, such that the same computations are effectively performed at every retinal position – this specific organizational motif greatly reduces the parameter space that needs to be explored when aiming to complement our understanding of retinal networks using computational models. Indeed, it is often possible to derive comparatively simple models of computations performed by individual microcircuits that capture the vast majority of the variance observed even when presenting complex visual stimuli. Taken together, the vertebrate retina may be one of the very few model systems in neuroscience today that promises to hold a complete functional understanding within arm’s reach. Detailed comparison between processing strategies employed between the retinas of different species, in the context of each system’s natural input, then may yield more general insights into how neuronal networks can evolve to best serve an animal’s needs.


  1. Masland, R. H. The fundamental plan of the retina. Nat. Neurosci. 4, 877–86 (2001).
  1. Wässle, H. Parallel processing in the mammalian retina. Nat. Rev. Neurosci. 5, 747–57 (2004).
  1. Euler, T., Haverkamp, S., Schubert, T. & Baden, T. Retinal Bipolar Cells: Elementary Building Blocks of Vision. Nat. Rev. Neurosci. 15, 507–519 (2014).
  1. Baden, T. et al. A tale of two retinal domains: near-optimal sampling of achromatic contrasts in natural scenes through asymmetric photoreceptor distribution. Neuron 80, 1206–17 (2013).
  1. Baden, T., Berens, P., Bethge, M. & Euler, T. Spikes in Mammalian Bipolar Cells Support Temporal Layering of the Inner Retina. Curr. Biol. 23, 48–52 (2012).
  1. Franke, K*., Berens, P*., Schubert, T., Bethge, M., Euler, T., Baden, T. (in preparation). A glutamate map of the inner retina.
  1. Baden, T.* Berens, P*., Franke, K*., Roman Roson, M., Bethge, M., Euler, T. (in press). What the mouse eye tells the mouse brain: Recording the entire RGC output at a single retinal location. Nature.
  1. Land, M. & Nilson, D.-E. Animal Eyes. (Oxford University Press, 2012).
  1. Cronin, T. W., Johnsen, S., Marshall, N. J. & Warrant, E. J. Visual Ecology. (Princeton University Press, 2014).
  1. Abbott, L. F. & Regehr, W. G. Synaptic computation. Nature 431, 796–803 (2004).
  1. Masland, R. H. The neuronal organization of the retina. Neuron 76, 266–80 (2012).
  1. Baden, T., Euler, T., Weckström, M. & Lagnado, L. Spikes and ribbon synapses in early vision. Trends Neurosci. 36, 480–8 (2013).
  1. Baden, T. & Euler, T. Early vision: where (some of) the magic happens. Curr. Biol. 23, R1096–8 (2013).
  1. Diamond, J. S. & Copenhagen, D. R. The contribution of NMDA and Non-NMDA receptors to the light- evoked input-output characteristics of retinal ganglion cells. Neuron 11, 725–738 (1993).
  1. Dunn, F. a, Lankheet, M. J. & Rieke, F. Light adaptation in cone vision involves switching between receptor and post-receptor sites. Nature 449, 603–6 (2007).
  1. DeVries, S. H., Li, W. & Saszik, S. Parallel processing in two transmitter microenvironments at the cone photoreceptor synapse. Neuron 50, 735–48 (2006).