Electron microscopy remains critical technology in neuroscience because the potential diversity in network topologies is so high that only anatomical ground truth can produce a connectome (Sporns et al., 2005). Though the mammalian retina is far simpler than those of most other vertebrates (Marc, 2008), it still expresses no fewer than 70 unique cell classes (Masland, 2001; Wässle, 2004). The flow of signals from cone photoreceptors to ganglion cells involves a stereotyped set of networks that seem simple: bipolar cells synaptically drive ganglion and amacrine cells, while amacrine cells provide feedback and feedforward network control (Marc, 2004; Masland, 2001; Wässle, 2004). However, a vast number of synaptic motifs can be produced from even a limited neuron set. A small network of two different bipolar cells driving two separate ganglion cell channels, interconnected by one amacrine cell class can be connected in 90 distinct motifs assuming lumped-parameter circuitry. Assuming distributed parameter circuitry (Weiss, 1996) explodes this to over 2000 potential motifs.

This calculation does not consider differential synaptic weights, molecular diversity, coupling by gap junctions or even the most common connection in the retina: two different amacrine cells forming serial synaptic chains. In fact another way to express the problem is to allow all cells to be connected in a 012345x pattern and simply alter the weights of the connections. In that mode, the network diversity is infinite. In stark contrast, we know that the outflow of signals from the mammalian retina is represented by only 15-20 ganglion cell classes (Marc and Jones, 2002; Rockhill et al., 2002), each representing a filter channel. But which network topology does each channel use?

Even rich models (Hennig et al., 2002) that mimic physiologic data acquired over limited spatiotemporal domains predict little about network topology or emergent features. Despite a broad view of the bounds of biophysical performance provided by physiology, models derived from physiology are essentially degenerate: not unique to any one network topology. In addition, remodeling and reprogramming of neural networks in retinal disease strongly argues that network scrambling is a key pathology (Marc et al., 2007). Network motif diversity is analogous to genetic diversity: many connective motifs (gene sequences) are possible, but only a subset form good filters (proteins), and mutating motifs generates neural malfunction (genetic disease).

Mammalian noctural vision is a prime example of the success of electron microscopy in informing and guiding physiological investigations. In mammalian retina, the main scotopic signal flows is:

Rod photoreceptor ·> rod bipolar cell > rod amacrine cell ... which then bifurcates into two arms:

(i) rod amacrine cell :: ON cone bipolar cell > ON ganglion cell;

(ii) rod amacrine cell ·> OFF cone bipolar cell > OFF ganglion cell;

> sign-conserving synapses, ·> sign-inverting synapses, :: gap junctions.

This unusual motif was by discovered by E.V. Famiglietti Jr. and Helga Kolb (Famiglietti and Kolb, 1976) using serial section electron microscopy. No other physiological study directly predicts this network, apart from pharmacologic analyses guided by the anatomic data (Muller et al., 1988; Protti et al., 2005). Even genetic interruption of ON arm of the network in the Cx36-/- knockout mouse (Deans et al., 2002) was informed by the motif and would not have, in itself, yielded the correct topology. Despite five decades of physiologic study of rod pathways, the discovery of a secondary direct rod > OFF bipolar cell synapse was based on anatomy, and its display is highly variable (Protti et al., 2005) . Finally, the view that rods couple to cones, though influenced by physiologic work, is largely dependent on anatomic evidence (Raviola and Gilula, 1973). In contrast, the absence of ground truth for the networks serving directionally selective ganglion cells has been a major hurdle in vision research. A range of physiologic and pharmacologic features have long been known for these cells, but none bound a unique synaptic model (Poznanski, 2005).

Deans MR, Volgyi B, Goodenough DA, Bloomfield SA, Paul DL. 2002. Connexin36 Is Essential for Transmission of Rod-Mediated Visual Signals in the Mammalian Retina. Neuron 36(4):703-712.

Famiglietti EV, Jr., Kolb H. 1976. A Bistratified amacrine cell and synaptic circuitry in the inner plexiform layer of the retina. Brain Research 84(2):293-300.

Hennig MH, Funke K, Worgotter F. 2002. The Influence of Different Retinal Subcircuits on the Nonlinearity of Ganglion Cell Behavior. J Neurosci 22(19):8726-8738.

Marc RE. 2004. Retinal Neurotransmitters. In: Chalupa LM, Werner J, editors. The Visual Neurosciences. Cambridge, MA: MIT Press. p 315-330

Marc RE. 2008. Functional Neuroanatomy of the Retina. In: Albert D, Miller J, Azar D, Blodi B, editors. Albert and Jakobiec's Principles and Practice of Ophthalmology. 3d ed: Elsevier. p in press.

Marc RE, Jones BW. 2002. Molecular phenotyping of retinal ganglion cells. Journal of Neuroscience 22(2):in press.

Marc RE, Jones BW, Anderson JR, Kinard K, Marshak DW, Wilson JH, Wensel T, Lucas RJ. 2007. Neural reprogramming in retinal degeneration. Investigative Ophthalmology & Visual Science 48(7):3364-3371.

Masland RH. 2001. Neuronal diversity in the retina. Current Opinion in Neurobiology 11(4):431-436.