Updating hippocampal representations: CA2 joins the circuit

The hippocampus integrates the encoding, storage and recall of memories, binding the spatio-temporal and sensory information that constitutes experience and keeping episodes in their correct context. The rapid and accurate processing of such daunting volumes of continuously changing data relies on dynamically assigning different aspects of mnemonic processing to specialized, interconnected networks corresponding to the anatomical subfields of dentate gyrus (DG), CA3 and CA1. However, differentially processed information ultimately has to be reintegrated into conjunctive representations, and this is unlikely to be achieved by unidirectional, sequential steps through a DG-CA3–CA1 loop. In this Review, we highlight recently discovered anatomical and physiological features that are likely to necessitate updates to the hippocampal circuit diagram, particularly by incorporating the oft-neglected CA2 region.

Introduction

Adaptations of the hippocampus that are likely to reflect the demands of memory processing are immediately apparent in its gross histology: the dense hippocampal cell layers are precisely arranged in a circuit of subfields encompassing the arrowhead of dentate gyrus (DG) and the curve of CA1–3. No single, homogeneous neural network can process all aspects of episodic memory simultaneously and, indeed, anatomical, neurophysiological and behavioural studies over the past two centuries or more have informed influential models of these subfields as specialized processing modules, each contributing to different facets of hippocampal function.

In piecing together this jigsaw of hippocampal subfields and connections, the collective tendency has been to start with the DG and build around a trisynaptic circuit to CA3 and then CA1 (Figure 1a,b). Most models emphasize sequential steps of information processing in this circuit: layer II principal cells of the entorhinal cortex (EC) project to the granule cells of the DG through the perforant path (PP), the granule cells project to CA3 pyramidal cells through mossy fibers (MF), CA3 pyramidal cells synapse onto CA1 pyramidal cells via the Schaffer collaterals (SC), then CA1 outputs to subiculum, deep-layer EC pyramidal cells and related parahippocampal and frontal neocortical regions. Prominent examples of differential information processing include pattern separation in DG (granule cells are abundant and sparse firing, hence different patterns of EC inputs are highly unlikely to activate identical subsets of granule cells and may be ‘orthoganolized’ at this stage) followed by pattern completion in CA3 (where dense, recurrent, excitatory projections within its own pyramidal cell population endow ‘auto-associative’ properties) (1, 2, 3, 4, 5; see also [6] in this Issue). The neat hippocampal loop has therefore been presumed to allow integration and processing of information provided via association cortex, then subsequent feedback to the cortex via CA1.

However, as the resolution of anatomical knowledge reaches the subcellular level and the nature of hippocampal network activity during a diverse behavioral repertoire of encoding, processing, storage and recall is increasingly well documented, simplifying models inevitably become more complex (Box 1). Here, we review recent discoveries that are likely to necessitate updates to the prevailing hypotheses, with particular emphasis on the potentially unique contributions made by the oft-neglected subfield, CA2.