Why a Sparse Synapse-Resolution Brain Connectivity (SSRBC) Atlas is Needed

'One thing must be stressed quite firmly: henceforth functional localisation of the cerebral cortex without the lead of anatomy is utterly impossible in man as in animals. In all domains, physiology has its firmest foundations in anatomy. Anyone wishing to undertake physiological localisational studies will thus have to base his research on the results of histological localisation. And today with greater reason than ever, one must recall the words of the past master of brain research, Bernhard Gudden, spoken three decades ago in the face of a dangerous tendency to specialise in extirpation experiments: "Faced with an anatomical fact proven beyond doubt, any physiological result that stands in contradiction to it loses all its meaning ... So, first anatomy and then physiology; but if first physiology, then not without anatomy".'

-Korbinian Brodmann in Localisation in the Cerebral Cortex (1909) Translated by Laurence Garey

This quote from Brodmann (written almost 100 years ago) nicely sums up the overarching reason a Synapse-Resolution Brain Connectivity Atlas is needed today. Neuroanatomy provides the structural framework within which physiological experiments and cognitive theories are grounded. Our understanding of neuroanatomy (at the regional, microcircuit, and neuronal ultrastructural levels) has of course improved immeasurably in the interim; however, the "physiological" disciplines (for our purposes; cell recording, functional imaging, systems modeling, and neuronal modeling) have continued to outstrip the anatomical information upon which these disciplines must rely[1].

Brain experimentalists and theorists today are not content to ask questions like “In what regions of the brain is visual information processed?” They have moved to deeper, computationally oriented questions like, “How do the ventral visual stream, dorsal visual stream, and prefrontal cortex interact dynamically to mediate view-invariant object recognition in cluttered scenes?” Experimentalists and theorists attempting to answer this type of question must come to grips with the inherently multi-scale nature of the brain. For example, a particular neuron in V4 is part of a neuronal circuit spanning vast regions of cortex (as well as some subcortical structures) including connections with V2, V4, IT, PPC, and many other regions.

The computation this V4 neuron is performing is intimately tied to the details of the synaptic connections it has relative to projections from these other regions. The synapses themselves are many orders of magnitude smaller than the neuronal circuit they help define. Today’s neuroanatomical techniques can provide glimpses of this complex circuit at the scale of the whole brain (e.g. axonal tracer experiments showing which region connects to which), at the regional level (e.g. Golgi stained brain slices), at the neuronal level (e.g. confocal microscopic 3D reconstruction of a dye injected neuron), and even at the synapse level (e.g. 3D serial TEM reconstruction). However, because these separate experiments are performed on different animals in different, non-compatible imaging protocols, these scales have proven to be incredibly difficult to stitch together into a coherent picture of the complete circuit.

Even such a basic question as “What is the retina->LGN->V1 anatomical circuit underlying a V1 simple cell’s computation?” today cannot be answered at the neuroanatomical level. A standard circuit model for this has existed for several decades (the linear connection of several on-center LGN cells projecting onto a common V1 simple cell) but this is neuroanatomical speculation grounded in physiological results, exactly what is warned of in the quote from Brodmann above. More importantly, as we move beyond simple primary sensory cortical responses and ask how the more complex responses of IT cells are computed, our intuition fails. We need a detailed, multi-scale neuroanatomical map of the underlying circuitry to guide our experiments and theorizing, not the other way around.

A Synapse-Resolution Brain Connectivity Atlas would provide exactly this type of integrated, multi-scale neuroanatomical data by providing a 3D voxel representation of cell membrane structure (via Osmium staining) of an entire brain at a resolution sufficient to visualize individual synapses (perhaps 10nm). A neuroanatomist specializing in the visual system’s circuitry could use such an atlas to trace back all classes of projections onto a particular V4 neuron thus providing a complete picture of the information flows with which this type of neuron interacts. Other specialists could follow up directly on this mapping work, tracing projections onto the very same neurons that were found to synapse on the original V4 neuron. In this way, complete neuronal circuits can be traced both within and between regions, and follow-up tracing experiments can be performed on the very same brain and the very same circuits, thus obviating the traditional problems of integrating work between experiments and laboratories[2].

To fully appreciate the improved pathway mapping abilities that a Synapse-Resolution Brain Connectivity Atlas could allow, imagine starting a pathway mapping experiment by following an afferent fiber from a retinal ganglion cell, then following this axon until it synapses on a relay neuron of the LGN, then following the axon of this LGN relay neuron to V1, and further following the axon of this V1 neuron where it leads, etc. In this way, the flow of visual information in the brain could be traced from region to region, eventually completing its journey in primary motor cortex neurons projecting to spinal motor units. Applying a combination of depth-first-search and breath-first-search at branch points in this pathway tracing experiment would trace out all the parallel information flows and information loops of the brain.

[1] Today's neuroanatomical mapping and tracing protocols are quite advanced. For an in-depth look at how "traditional" neuroanatomical mapping is applied today using multiple tracers, immunological stains, and advanced data analysis software see the following two papers:

Lewis JW, Van Essen DC. 2000. Architectonic parcellation of parieto-occipital cortex and interconnected cortical regions in the macaque monkey. J Comp Neurol 428:79–111.

Lewis JW, Van Essen DC. 2000. Architectonic parcellation of parieto-occipital cortex and interconnected cortical regions in the macaque monkey. J Comp Neurol 428:112–137.

These papers review an experiment designed to determine the architectonically defined regions of the primate intraparietal sulcus (IPS). The experiment was additionally designed to map brain connectivity by determining all cortical regions projecting axonal connections to the IPS. These papers provide a good look at the state-of-the-art in neuroanatomical mapping; providing insights into its great power as well as its inherent limitations.

[2] Of course, this structural information alone cannot answer the entire question of what computation is being performing in a particular neuron or in the overall system. To answer this larger question, pharmacological data, physiological recording data, and a whole host of other sources of information must be brought to bear.

Why a Sparse Synapse-Resolution Brain Connectivity (SSRBC) Atlas is Needed?