Progress in Neurobiology 79 (2006) 1–48 www.elsevier.com/locate/pneurobio A computational theory of hippocampal function, and empirical tests of the theory Edmund T. Rolls a,*, Raymond P. Kesner b a University of Oxford, Department of Experimental Psychology, South Parks Road, Oxford OX1 3UD, United Kingdom b University of Utah, Department of Psychology, 380S 1530E Room 502, Salt Lake City, UT 84112, USA Received 23 September 2005; received in revised form 23 March 2006; accepted 28 April 2006 Abstract The main aim of the paper is to present an up-to-date computational theory of hippocampal function and the predictions it makes about the different subregions (dentate gyrus, CA3 and CA1), and to examine behavioral and electrophysiological data that address the functions of the hippocampus and particularly its subregions. Based on the computational proposal that the dentate gyrus produces sparse representations by competitive learning and via the mossy fiber pathway forces new representations on the CA3 during learning (encoding), it has been shown behaviorally that the dentate gyrus supports spatial pattern separation during learning. Based on the computational proposal that CA3–CA3 autoassociative networks are important for episodic memory, it has been shown behaviorally that the CA3 supports spatial rapid one-trial learning, learning of arbitrary associations where space is a component, pattern completion, spatial short-term memory, and sequence learning by associations formed between successive items. The concept that the CA1 recodes information from CA3 and sets up associatively learned backprojections to neocortex to allow subsequent retrieval of information to neocortex, is consistent with findings on consolidation. Behaviorally, the CA1 is implicated in processing temporal information as shown by investigations requiring temporal order pattern separation and associations across time; computationally this could involve temporal decay memory, and temporal sequence memory which might also require CA3. The perforant path input to DG is implicated in learning, to CA3 in retrieval from CA3, and to CA1 in retrieval after longer time intervals (‘‘intermediate-term memory’’). # 2006 Elsevier Ltd. All rights reserved. Keywords: Autoassociation; Recall; Episodic memory; Cortical backprojections; CA3; CA1; Dentate granule cells Contents 1. Introduction . .................................................................................. 2 2. The theory . .................................................................................. 3 2.1. Systems-level functions of the hippocampus . ..................................................... 3 2.1.1. Evidence from the effects of damage to the hippocampus . ........................................ 3 2.1.2. The necessity to recall information from the hippocampus ........................................ 4 2.1.3. Systems-level neurophysiology of the primate hippocampus . .................................... 5 2.1.4. Systems-level anatomy ................................................................. 7 2.1.5. Perirhinal cortex, recognition memory, and long-term familiarity memory. ............................ 7 2.2. The operation of hippocampal circuitry as a memory system. ............................................ 8 2.2.1. Hippocampal circuitry. ................................................................. 8 2.2.2. Dentate granule cells . ................................................................. 9 2.2.3. CA3 as an autoassociation memory . .................................................... 11 2.2.4. CA1 cells ......................................................................... 22 2.2.5. Backprojections to the neocortex—a hypothesis ............................................... 24 * Corresponding author. Tel.: +44 1865 271348; fax: +44 1865 310447. E-mail address: [email protected] (E.T. Rolls). URL: http://www.cns.ox.ac.uk 0301-0082/$ – see front matter # 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.pneurobio.2006.04.005 2 E.T. Rolls, R.P. Kesner / Progress in Neurobiology 79 (2006) 1–48 2.2.6. Backprojections to the neocortex—quantitative aspects . ......................................... 26 2.2.7. Simulations of hippocampal operation...................................................... 27 3. Tests of the theory . ........................................................................... 27 3.1. Dentate gyrus (DG) subregion of the hippocampus. .................................................. 27 3.1.1. Spatial pattern separation .............................................................. 28 3.1.2. Encoding versus retrieval .............................................................. 28 3.2. CA3 subregion of the hippocampus . .......................................................... 29 3.2.1. Rapid encoding of new information . ...................................................... 29 3.2.2. The types of information associated in CA3 ................................................. 30 3.2.3. Acquisition/encoding associated with multiple trials . ......................................... 30 3.2.4. Pattern completion . .................................................................. 31 3.2.5. Recall . ........................................................................... 31 3.2.6. Mossy fiber versus direct perforant path input into CA3 ......................................... 32 3.2.7. Pattern separation . .................................................................. 32 3.2.8. Short-term memory, and path integration. .................................................. 33 3.3. CA1 subregion of the hippocampus . .......................................................... 33 3.3.1. Sequence memory . .................................................................. 34 3.3.2. Associations across time . .............................................................. 34 3.3.3. Order memory ...................................................................... 35 3.3.4. Intermediate memory . .............................................................. 36 3.4. CA1 and CA3: interactions and dissociations. ...................................................... 37 4. Comparison with other theories of hippocampal function . .................................................. 39 Acknowledgement . ........................................................................... 41 References ................................................................................... 41 1. Introduction Willshaw and Buckingham (1990). Early work of Gardner- Medwin (1976) showed how progressive recall could operate in In this paper a computational theory of hippocampal a network of binary neurons with binary synapses. Rolls (1987) function developed by Rolls (1987, 1989a, 1989b, 1989c, produced a theory of the hippocampus in which the CA3 1996b), Treves and Rolls (1992, 1994) and with other neurons operated as an autoassociation memory to store colleagues (Rolls and Stringer, 2005; Rolls et al., 2002)is episodic memories including object and place memories, and refined and further developed, and tests of the theory based the dentate granule cells operated as a preprocessing stage for especially on subregion analysis of the hippocampal system are this by performing pattern separation so that the mossy fibres described, with further refinement of the theory then added. The could act to set up different representations for each memory to relation of this theory to other computational approaches to be stored in the CA3 cells.2 He suggested that the CA1 cells hippocampal function is described. operate as a recoder for the information recalled from the CA3 The theory was originally developed as described next, and cells to a partial memory cue, so that the recalled information was preceded by work of Marr (1971) who developed a would be represented more efficiently to enable recall, via the mathematical model, which although not applied to particular backprojection synapses, of activity in the neocortical areas networks within the hippocampus and dealing with binary similar to that which had been present during the original neurons and binary synapses which utilised heavily the episode. This theory was developed further (Rolls, 1989a, properties of the binomial distribution, was important in 1989b, 1989c, 1989d, 1990a, 1990b), including further details utilizing computational concepts.1 The model was assessed by about how the backprojections could operate (Rolls, 1989b, 1989d), and how the dentate granule cells could operate as a competitive network (Rolls, 1989a). Quantitative aspects of the 1 Marr (1971) showed how a network with recurrent collaterals could theory were then developed with A. Treves, who brought the complete a memory using a partial retrieval cue, and how sparse representations expertise of theoretical physics applied previously mainly to could increase the number of memories stored (see also Willshaw and Buck- understand the properties of fully connected attractor networks ingham, 1990). The analysis of these autoassociation or attractor networks was developed by Kohonen (1977) and Hopfield (1982), and the value of sparse representations was quantified by Treves and Rolls (1991). Marr (1971) did not specify the functions of the dentate granule cells versus the CA3 cells versus the 2 McNaughton and Morris (1987) at about the same time suggested that the CA1 cells (which were addressed in the Rolls (1989a, 1989b, 1989c, 1989d) CA3 network might be