Talk:Models of hippocampus
DEAR REVIEWERS - SEE END OF REVIEW SECTION FOR RESPONSES. Dear Alessandro and Anonymous reviewer, I made extensive revisions in response to your comments in the text of the article. I also described my response to your comments in a section that appears at the bottom of this review section. Thanks for your excellent and useful comments. Mike Hasselmo
This is a thoughtful and well balanced review in Mike Hasselmo's usual clear style. I prefer therefore to refrain from tampering with his text, and I only comment on a few issues that Mike may want to include in a revision.
1) Although most "models of hippocampus" indeed simulate hippocampal function and activity, researchers attempt to use models also with a somewhat different goal: to understand relationships that cannot yet be probed with experimental techniques. Examples are the adapted extensions of the Hopfield model used to assess storage capacity, mentioned in point 2) below, but also my attempt to contrast actual and potential hippocampal circuitry (Treves, 2004) to understand the significance of the CA3-CA1 differentiation in mammals. Therefore I would say at the start that some models of the hippocampus also attempt to infer quantitative relations between structural and biophysical traits and parameters on the one hand and functional measures on the other, in domains not yet accessible to direct experimentation.
2) The notion that the hippocampus includes an associative memory network of limited storage capacity plays a central role in attempts to understand the complementarity among memory systems relying on different substrates to retain often the very same information. There are theories that deem neocortex unable to "do memory" all on its own because of insufficient connectivity and theories like McClelland's, that see the main neocortical limitation in the requirement for slow back-propagation-style learning; but both regard storage capacity as the crucial hippocampal constraint. Because of its limited storage capacity, the hippocampus is viewed as an intermediate term or "buffer" store. Therefore assessments of this storage capacity are an essential contribution by hippocampal models. For discrete memory patterns, capacity estimates have been based on analytical techniques borrowed from spin-glass physics, and first applied to the Hopfield model by Amit, Gutfreund and Sompolinsky (1987). These techniques were later extended by Francesco Battaglia to estimate hippocampal capacity for the storage of multiple continuous attractors, as representations of multiple 1D and 2D environments (Battaglia and Treves, 1998). These results obtained by using mathematical models are still several years ahead of potential experimental falsification.
3) An important concept in discussing hippocampal function is self-organization, in that hippocampal representations are considerably less constrained than those in cortical regions more interfaced with the input and output periphery. We and others see the dentate gyrus, of course, as the ultimate random organizer, determining the representations set up in CA3; but most of us regard also entorhinal grid cells as emerging out of a self-organizing process, whether due purely to single-unit adaptation (Kropff and Treves, 2008), to oscillatory mechanisms or to recurrent connectivity.
4) The distinction between a storage and a retrieval mode in hippocampal memory function gives a completely different flavour to pattern separation and pattern completion, with respect to connectionist models that do not consider such a distinction. The distinction opens the door for applying to the hippocampus Mike Hasselmo's very own ideas about the role of acetylcholine, and for analyses of dentate function that go beyond the crude pattern separation notion (Lassalle et al, 2000; Lee and Kesner, 2004; and see Treves et al, 2008).
Generally, I agree with R1 that this is well-thought out and clear review. I have added some minor notes in the text (as "<review></review>").
In general, I think this review could be improved by making the history of the models more clear. I think that's particularly useful for an encyclopedic reference like scholarpedia. For example, the effect of damage by ischemia on hippocampus long precedes Eichenbaum's 2003 paper. (I usually cite the early Squire papers.) In fact, most of the models of such effects are from the 1990s and 2000s, such as the indexing theory (Teylor, DiScenna 1995/1996; Alvarez, Squire 1994).
I also think the review could spend some more time comparing models to data. Are there any of the models that have been accepted/rejected? How have the models changed our interpretation of the data? Hippocampus is (I think) one of the few places where there has been a close interaction between theory, model, and data. A nice example is the author's classic work on acetylcholine, hippocampus, and associative memory.
I do think this review is missing a section on the models that predicted the multiple map explanation for non-spatial responses (Redish 1999; charts, Samsonovich, McNaughton 1997; reference frames, Touretzky, Redish 1996; Redish, Touretzky 1997). Also, there are a number of models that tried to identify the function of place cells by asking how hippocampal dynamics could be used for navigation. (Such as the goal-orientation models of Foster, Morris, Dayan 2000, or Burgess, Recce, O'Keefe 1998, or the self-localization models of Redish and Touretzky.) Also, there were the replay models of replay functionality (Alvarez and Squire 1994; Johnson, Redish 2005; Redish, Touretzky 1998, Foster, Wilson 2006).
Historically, there were a lot of place field models that attempted to explain phenomena of place fields (like the non-spatial responses mentioned above) and directional place fields (e.g. Sharp 1991). For example, at one time, these models were very much at the center of the debate on the role of path integration (dead reckoning) in place field formation (and what the anatomical substrate of that path integration system was). Although I think that it is true that much of the debate that centered on these models has been resolved, I think some more time could have been spent on these models, particularly their role in helping resolve that debate.
Author : RESPONSE to reviewer comments
REVIEWER #1 - Alessandro Treves Comment #1 – I have addressed the alternate use of modeling using some of the wording provided by Alessandro in the section on Types of models. I also added discussion of the Treves, 2004 paper in the section on region CA1.
Comment #2 – I’ve added the following statement to the section on region CA3. The modeling of attractor dynamics builds on extensive earlier analysis of the memory capacity of recurrent networks (Amit et al., 1987) that was applied to hippocampal function (Treves and Rolls, 1992; 1994; Battaglia and Treves, 1998). The capacity limits of hippocampal circuits could constrain its role to an intermediate term episodic memory that must eventually update neocortical representations.
Comment #3 – In response to the comment about self-organization, I have added a sentence about self-organization for pattern separation to the section on dentate gyrus and discussion of models of self-organization for formation of place cell responses in region CA1 and of grid cell responses in entorhinal cortex.
Comment #4 – I have shortened the titles of the subregion functions, so that the titles just list the names of the subregions. This avoids overemphasizing the very simple notions of separation and completion. I’ve also provided more discussion of the encoding and retrieval model in a new section on medial septum and have provided citations on the role of neurogenesis in the dentate gyrus.
REVIEWER #2 – I have gone through and directly implemented all of the comments of this reviewer that were inserted into the text. These responses are marked with *** in the COMMENT and RESPONSE section below.
Comment #1 – Historical perspective. For the lesion effects, I have cited the earlier historical papers such as Scoville and Milner, 1957 and Zola-Morgan et al., 1986. I also inserted a citation to Teyler and DiScenna, 1986 and Teyler and Rudy, 2007 in the comment about linking together different sensory inputs.
Comment #2 – Relationship of models to data. I have added discussion of how models showed the importance of the Hebb rule before hebbian synaptic modification was demonstrated in the hippocampus. I’ve also added a section on the role of the medial septum and the influence of cholinergic modulation on setting dynamics of encoding and retrieval. In addition, as noted in the text, I’ve addressed the paper by Harvey et al. as a test of different models of theta phase precession. I also added a section on entorhinal cortex and modeling of grid cells and testing of predictions from these models, and about the prediction of boundary vector cells that was recently supported by data.
Comment #3 – I have added some discussion about models using path integration to generate place cell responses and accounting for remapping using multiple maps, adding citations to the work by Touretzky and Redish, 1996, 1998, 1999 and Samsonovich and McNaughton, 1997. I have expanded the section about the use of place cell representations for goal-directed behavior in Burgess et al., 1997; Touretzky and Redish, 1996; Redish and Touretzky, 1998 and Foster et al., 2000. I have added a sentence about the use of replay to enhance representations of the environment for goal-directed behavior, citing Johnson and Redish, 2005 and Foster and Wilson, 2006.
Comment #4 – I have expanded the discussion of different models of the formation of place cell responses, including the early models, the models of place cells based on path integration, the modeling of place cell remapping with multiple maps, and the contribution of these early models to later models of grid cell responses.
COMMENT: Also input from subcortical structures, particularly GABAergic from septum and neural modulators from several structures. I think it important to mention these as they have played a role in some of the hippocampal models. RESPONSE: ***A reference to the septal input and other subcortical modulatory input has been added.***
COMMENT: Also, it might be worth mentioning models of hippocampal function, which assume place fields and try to predict behavioral results from that. (For example, the suggestion that place fields can drive planning and goal cells and the like. There were a bunch of these types of models in the 1980s and 1990s, e.g. Touretzky, Redish 1996; Burgess, Recce, O'Keefe 1994; Johnson, Redish 2005; Foster, Morris, Dayan, 2000, and lots of others). RESPONSE: ***A section further down addresses these types of models. Additional citations have been inserted for Touretzky and Redish, 1996; Johnson and Redish, 2005 and Foster et al., 2000. I could not find a Burgess et al., 1994 reference but feel the 1997 citation is sufficient for that***
COMMENT: but Recent Some typos?***I fixed this typo***
COMMENT: A lot of the early models tried to suggest that hippocampus tied together diverse sensory information. RESPONSE: ***I have added a mention of this with citation to Teyler***
COMMENT: Shouldn't this be Samsonovich and McNaughton 1997? RESPONSE: ***Thanks. I have corrected the citation and added citations to Redish 1999 and to Doboli, Minai and Best in the section on dentate gyrus as well as mentioning the relationship to the attractro models of grid cells.*** COMMENT: Also, I think one should include the Redish and Touretzky models in attractor dynamics to model place cell activity, e.g. Redish 1999. As well as the more recent Fuhs and Touretzky (200?) and Doboli and Minai (2001) models.) RESPONSE: ***These citations have been included.
COMMENT: Shouldn't this include O'Reilly and McClelland (1994)? RESPONSE: ***Though O’Reilly and McClelland talk about interaction of EC and DG input to CA3, I could not find a mention of comparison function in region CA1 in their paper.***
COMMENT: I think the citations here could be historically improved. RSEPONSE: ***I added citations to Green and Arduini, 1954; O’Keefe and Nadel, 1978 and Buzsaki et al., 1983***
COMMENT: Much of this has been resolved with the recent intracellular recordings from awake animals (Harvey, Collman, Dombek, Tank (2009). RESPONSE: ***I’ve added description of the intracellular data in relation to these models.***