Critical Review Licensee Oa Publishing London 2013. Creative Commons Attribution License (cc-by)

For citation purposes: Hunsaker MR. Embracing complexity: using the attribute model to elucidate the role for distributed neural networks underlying spatial memory processes. Abstract Introduction Rodent models of spatial process­ ing have long been used as a model for human memory deficits, par­ ticularly as related to hippocampal function. What has been long absent in these analyses, however, is a thor­ ough description and rigorous study of the distributed neural networks asso ciated with spatial processing— both in the human and rodent. There is a trend emerging in research to expand beyond the hippocampus for evaluating spatial memory, but the thrust of the research still focuses on the role of the hippocampus as essential and other neural substra­ tes as performing subservient roles to the hippocampal processing. This review will describe spatial memory in terms of an attribute specificity model and demonstrate the nature of spatial processing in the rodent brain as well as describe a testable theoret­ ical model concerning the fundamen­ tal mechanisms whereby rodents perform spatial memory tasks. In particular, a pivotal role for the ret­ rosplenial cortex in spatial memory processing is outlined in the context of interacting memory systems. Conclusion Based on the proposed role for the retrosplenial cortex as an inter­ face between the event­based and


Introduction
Since Scoville and Milner 1 described a series of patients with amygdala/ hippocampal resections, the medial temporal lobe, and particularly the hippocampus, has been an area of intense interest for the study of memory in humans as well as animal models.In the late 1970s, Dave Olton and Robert Samuelson 2 developed a radial arm maze to study shortterm working memory and intermediate/ longterm reference memory in rats.In the early 1980s, Richard Morris provided a simple, easily replicable behavioural method to evaluate gen eral memory in rats using a spatial memory task [3][4][5] .The development of this task allowed researchers an opportunity to evaluate the role of the hippocampus for spatial learning, and thus directly test the hippocam pus as a cognitive map hypothesis proposed by O'Keefe and Nadel 6 con cerning the role of place specific fir ing observed in the hippocampus of freely moving rats 7 for associative memory function and spatial cogni tive mapping 8 .
Lesions to the rat hippocampus have been developed as an explicit model of the amnesia reported in patient H.M. The development of such models in rats was important as the development of similar models in nonhuman primates has proven difficult 9,10 .Since that time, there has been an explosion in the study of spatial memory in rat and mouse models-particularly as related to human memory dysfunction.To better design and interpret behav ioural paradigms evaluating memory processing, a number of theoreti cal models have been proposed that classify memory into different domains.This review will describe these models and focus on a highly parallel attribute specificity model as the most fle xible for the inter pretation of human and rodent research.This model will then be used to describe how the brain pro cesses spatial information and form spatial memory-the most common type of memory evaluated in rodents.

Discussion
The author has referenced some of his own studies in this review.These referenced studies have been conducted in accordance with the Declaration of Helsinki (1964) and the protocols of these studies have been approved by the relevant ethics committees related to the institution in which they were performed.All human subjects, in these referenced studies, gave informed consent to participate in these studies.Animal care was also in accordance with the institution guidelines.

Theoretical models of brain function
In order to understand the role of the hippocampus for these learning and memory processes, a number Embracing complexity: using the attribute model to elucidate the role for distributed neural networks underlying spatial memory processes MR Hunsaker* Behavioural memory that organise all the critical information.It is suggested that the specific, personal and temporal con text of a situation is coded in work ing memory 24 .This translates into memory for events that occur on a specific trial in a task, biasing mne monic coding toward the process ing of incoming data.In contrast, information concerning rules and procedures (general knowledge) of situations is coded in reference mem ory.This translates into memory for events that happen on all trials in a task, biasing mnemonic coding towards the processing of expec tancies based on the organisation of permanently stored memory repre sentations.The working memory model also states that hippocam pus mediates working, but not refe rence, memory.
The attribute memory model suggests that the hippocampus is directly involved in coding of all new spatialtemporal incoming informa tion that is likely to be relevant in trialunique situations but that could also be of importance for all trials within a learning task (eventbased memory).However, it is not involved in coding information based on expected nonvarying information in the form of maps, rules, strategies and procedures (knowledgebased memory; 23 ).
It is assumed by each of these models that in new learning situa tions the hippocampus is involved in consolidating new information, and in familiar situations, requiring attention or memory for trial unique stimuli, and is involved in temporarily maintaining information across time.It is likely that in both situations the hippocampus maintains this infor mation temporarily in an interme diate memory storage system.

The attribute model Memory systems for spatial memory
At the level of processing (Table 1), the eventbased memory system of competing theoretical models have been proposed.Around the same time the water maze was developed, Larry Squire described a dichotomous model for memory function 11,12 , declarative memory dependent upon the medial tem poral lobe and hippocampus, and nondeclarative memory associated with extramedial temporal lobe and extrahippocampal brain sub strates.Declarative memory, as the name implies, referred to memories whose source can be consciously recalled and declared, or reported to the experimenter.Nondeclarative memories refer to skills, habits, prim ing, simple classical conditioning and nonassociative learning.The binding feature of these types of memory is that there is not a source or moment in which the information was learned that can be consciously recalled.Similar models have been proposed that have similar dichotomies, albeit with different verbiage associated with the process of parsing function: hippocampus dependent explicit memory and nonhippocampus dependent implicit memory 13 ; or else a hippocampal dependent declara tive memory based on the represen tation of relationships among stimuli versus a nonhippocampal depen dent procedural memory based on the representation of a single stimu lus or nonrelational inflexible (con figural or conjunctive) configuration of stimuli, irrespective to conscious or subconscious processing 14,15 .
Olton 16,17 has suggested a differ ent dual memory system in which spatial memory can be divided into a hippocampal dependent short term working memory defined as mem ory for the specific personal spatial and temporal context of a situation and a nonhippocampal dependent intermediate or longterm reference memory, defined as memory for rules and procedures (general knowledge) of specific situations.Different terms have been used to reflect a similar distinction in humans; particularly episodic versus semantic memory 18 .Importantly, the working memory as defined by Olton is more akin to epi sodic memory or shortterm mem ory rather than the working memory as commonly studied in cognitive neuroscience dependent upon the frontal cortex.
However, memory is too complex and involves many neural systems in addition to the hippocampus to be simply parsed into a dichotomous system.To remedy this situation, Kesner [19][20][21] proposed a tripartite attribute based theoretical model of memory which is organised into eventbased (databased), know ledgebased (expectationbased) and rulebased memory systems.Each system is composed of the same set of attributes or domains of memory, characterised by a set of process ori ented operating characteristics and mapped onto interconnected neural circuits.
In the attribute memory model 22,23 , it is proposed that the hippocampus in animals codes only spatial and temporal attributes.In humans, the right hippocampus mediates spatial and temporal attribute representa tions, and the left hippocampus medi ates verbal and temporal attri bute representations.Other models (e.g., working, explicit and declarative memory models) propose that the hippocampus codes all infor mation, including spatial, temporal, non spatial (sensory cues) and egocentric spatial attributes [11][12][13][14][15][16][17] .Thus, models of hippocampal function vary greatly on the nature of information or attri butes that are processed, coded or represented in the hippocampus.
One view is that the hippocam pus is the storage site for spatial information 6 .A different approach was taken by the working, declara tive and attribute models of memory.The working/reference memory model 16,17  provides for temporary representa tions of incoming data concerning the present (i.e., online processing), with an emphasis upon data and events that are usually personal and that occur within specific external and internal contexts.The empha sis is upon the processing of new and current information.During initial learning, emphasis is placed on the eventbased memory sys tem, which will continue to be of importance even after initial learn ing in situations where unique or novel trial information needs to be remembered 26,27 .This system is akin to episodic memory 12,18,27 as for mulated to describe research using human subjects.The knowledgebased memory system provides for more permanent representations of previously stored information in longterm memory and can be thought of as one's gen eral knowledge of the world.The knowledgebased memory system would tend to be of greater impor tance after a task has been learned, or given that the situation has become invariant and/or familiar.This sys tem is akin to semantic memory 18 .
The rulebased memory system receives information from the event based and knowledgebased systems and integrates the information by applying rules and strategies for sub sequent action [28][29][30][31] .In most situations, however, one would expect a con tribution of all three systems with a varying proportion of involvement of one relative to the other depend ing primarily upon the demands of the task being performed.It is important to note that this is diffe rent than learning stimulusresponse or actionoutcome learning that underlies habit formation.

Specific attributes
The attribute model makes the basic assumption that the three memory systems are composed of the same forms, domains or attributes of memory.Even though there could be many attributes, the most impor tant attributes in the attribute model are space, time, response, sensory perception and reward value (affect).In humans, a language attribute is also added.
A spatial (space) attribute within this framework involves memory representations of places or rela tionships between places.It is exem plified by the ability to encode and remember spatial maps and to local ise stimuli in external space.Memory representations of the spatial attri bute can be further subdivided into specific spatial features including allocentric spatial distance, ego centric spatial distance, allocentric direction, egocentric direction and spatial location.
A temporal (time) attribute within this framework involves memory representations of the duration of a stimulus and the succession or tem poral order of temporally separated events or stimuli, and from a time perspective, the memory representa tion of the past.
A sensory/perceptual attribute within this framework involves memory representations of a set of sensory stimuli that are orga nised in the form of cues as part of a specific experience.Each sensory modality (olfaction, auditory, vision, somatosensory and taste) can be considered part of the sensory /perceptual attribute component of memory.
A response attribute within this framework involves memory repre sentations based on feedback from motor responses (often based on proprioceptive and vestibular cues) that occur in specific situations as well as memory representations of stimulus-response associations.
A reward value (affect) attrib ute within this framework involves memory representations of reward value, positive or negative emotional experiences, and the associations between stimuli and rewards.
Processes associated with each attribute Within each system attribute, infor mation is processed in different ways based on different operational characteristics.
For the eventbased memory system, specific processes involve (1) selective filtering or attenuation of interference of temporary memory representations of new information and is labelled pattern separation, (2) encoding of new information, (3) shortterm and intermediate term memory for new information, (4) the establishment of arbitrary associations, (5) consolidation or elaborative rehearsal of new infor mation and (6) retrieval of new  For the knowledgebased memory system, specific processes include (1) encoding of repeated informa tion, (2) selective attention and selective filtering associated with permanent memory representations of familiar information, (3) percep tual memory, (4) consolidation and longterm memory storage partly based on arbitrary and/or pattern associations and ( 5) retrieval of familiar information based on flexi bility and action.
For the rulebased memory system, it is assumed that infor mation is processed through the integration of information from the eventbased and knowledge based memory systems for the use of major processes that include (1) the selection of strategies and rules for maintaining or manipu lating information for subse quent decision making and action, (2) shortterm or working memory for new and familiar information, (3) development of goals, (4) pro spective coding, (5) affecting deci sion processes and (6) comparing actions with expected outcomes.

Attributes map onto neural substrates
On a neurobiological level (Table 2; Figure 1) each attribute maps onto a set of neural regions and their interconnected neural circuits.For example, within the event based memory system, it has been demonstrated that in animals and humans the hippocampus supports memory for spatial, temporal and language attribute information; the entire basal ganglia, particularly the dorsolateral striatum (in rodents) mediates memory for response attribute information; the amygdala and nucleus accumbens subserves memory for reward value (affect) attribute information; and the peri rhinal, extrastriate visual cortex, and all other sensory cortices support memory for visual object sensory/ perceptual attribute.
Within the knowledgebased memory system, it has been demon strated that in animals and humans the posterior parietal cortex sup ports memory for spatial attributes; the dorsal and dorsolateral prefron tal cortex and/or anterior cingulate support memory for temporal attri butes; the premotor, supplementary motor, and cerebellum in monkeys and humans and precentral cor tex and cerebellum in rats support memory for response attributes; the orbital prefrontal cortex and amyg dala support memory for reward value (affect) attributes; the infero temporal cortex in monkeys and humans and TE2 cortex in rats sub serve memory for sensory/percep tual attributes (e.g., visual objects); and the parietal cortex, Broca and Wernicke's areas subserve memory for the language attribute.
Within the rulebased memory system it can be shown that diffe rent subdivisions of the prefrontal cor tex (and rodent homologues; [32][33][34][35] ) support different attributes.For example, the dorsolateral and vent rolateral prefrontal cortex in humans support spatial, object, and language attributes and the infralimbic and prelimbic cortex in rats supports spatial and visual object attributes; the premotor and supplementary motor cortex in monkeys and humans and precentral cortex in rats sup port response attributes; the dorsal, dorsolateral, and mid dorsolateral prefrontal cortex in monkeys and humans and anterior cingulate in rats mediate primarily temporal attributes; and the orbital prefrontal cortex in monkeys and humans and agranular insular cortex in rats sup port affect attributes.

Interactions among attributes
Despite the relative independence and parallel processing of the diffe rent attributes, it bears to mention that the neural systems that under lie memory of all forms interact and contain similar nodes (i.e., hippocam pus processes space and time and in special cases can process sensory/ perceptual information and affectbut all these attributes are processed in larger networks made up of dis parate elements).To provide a con crete example, the interaction among temporal, spatial, and sensory/ perceptual attributes will be dis cussed based on different interac tions based on task demands.
The nature of the interactions between memory systems can be evaluated to dissect out the processes involved in both episodic and non episodic behavioural experiments.For illustration, two hippocampus dependent tasks involving specific and easily identifiable sensory/ perceptual stimuli (what), spa tial information (where-computed from a combination of sensory/ perceptual and temporal attrib utes), and temporal relationships between the stimuli (when) will be compared and contrasted.One task will require eventbased memory processes and the other task can be solved via knowledge based memory processes.The nature of the interac tions between these three attributes corresponding to what, when, and where will be analysed to differen tiate between the two tasks.
The knowledgebased memory task requires that a pair of asso ciations be acquired over multiple training trials.It is an object-traceplace pairedassociate task involving sensory/perceptual stimuli (what), a temporal stimulus in the form of a temporal discontinuity (trace inter val; when) and spatial information (where).This task is designed as Figure 1: Neural networks underlying spatial attribute processing organized by memory system.In this model, the event based memory system (green diamonds) process incoming sensory/perceptual information to calculate the spatial attrib ute in an online manner.This can be simplified as an encoding process involving spatial pattern separation and rapid trial unique arbitrary associations.The knowledge based memory system (blue rounded rectangles) uses already processed spatial information from the eventbased memory system to further process space.This can be simplified to a retrieval process involving spatial pattern completion, consolidation and multiple trial learned paired patterned associations, as well as retrospective coding.The rulebased memory system (red rectangles) provides the event and knowledgebased memory systems with goals and reward contingencies to guide online spatial processing and retrieval processes.This can be simplified as a prospective coding process.The retrosplenial cortex (highlighted as a pink rectangle), appears based on anatomical connectivity to provide an interface for all three neural systems [112][113][114] .

Licensee OA Publishing London 2013. Creative Commons Attribution License (CC-BY)
Competing interests: none declared.Conflict of interests: none declared.
All authors contributed to the conception, design, and preparation of the manuscript, as well as read and approved the final manuscript.
All authors abide by the Association for Medical Ethics (AME) ethical rules of disclosure.For citation purposes: Hunsaker MR.Embracing complexity: using the attribute model to elucidate the role for distributed neural networks underlying spatial memory processes.OA Neurosciences 2013 Sep 01;1(1):2.
follows: when a particular spatial location (a) and a Garfield toy (1) are paired across a 10 s trace interval, the animal is rewarded (a_1+).Also, if a different spatial location (b) and a truck toy are matched (2), the animal is rewarded (b_2+).If spatial location (a) and the toy truck are paired, there is no reward (a_2-), similarly for spa tial location (b) and (a) Garfield toy (b_1-).The trace interval separates the sensory/perceptual stimulus and the presentation of the spatial location 36 .If the association pre sented during a trial were rewarded, the rat would receive a reward upon displacing a block in the correct spatial location, which is then rep resented by the affect attribute, sig nalling a correct choice.This should bind the sensory/perceptual stimu lus and spatial location association across the temporal discontinu ity (an association involving what, when, and where).Then, the animal is presented with a new sensory/ perceptual stimulus and spatial loca tion association.If rewarded, then the process continues as before; if not rewarded, the animal does not receive any reward, and the affect attribute signals an error.Learning this task within only the eventbased memory system would be difficult because the eventbased memory system is susceptible to trialbytrial interference.Both temporally adja cent (e.g., subsequent) and spatially adjacent (e.g., occurring in the same or very similar spatial locations irre spective of temporal contiguity) epi sodes would interfere and degrade each other during acquisition.
Learning these associations involves comparing accumulated behavioural episodes or events within the knowledgebased and rulebased memory systems to develop appropriate rules, goals and schemas to perform the task effi ciently.In other words, repeated tri als must be presented if the rule and knowledgebased memory systems are to be engaged in learning and performing the task.Also, these two latter systems generate and apply abstract rules and generalise tempo ral, spatial, and internal contexts.In other words, the knowledgebased and rulebased systems read the accumulated behavioural episodes, clarify the relevant contextual infor mation, and apply this information to guide future actions.Once the knowledgebased and rulebased memory systems have processed the data and generated the sche mas necessary to perform the task, the eventbased memory system does not significantly contribute to performance of this task since the four discriminations or associations (a_1+, b_2+, a_2-, b_1-) have been efficiently encoded and only need to be discriminated from each other 37,38 .
In contrast to the above bicondi tional discrimination, a task devel oped by Morris 39 and modified by Kesner and colleagues 40 allows rats to perform a very similar sensory/ perceptual stimulus and spatial loca tion association in an eventbased manner.The critical aspect of this task that requires the eventbased memory system is the fact that each trial is unique, and as such there are no repeated trials across which the knowledge and rulebased memory systems can be recruited.During the study phase, the animal receives two rewarded object-place pairings (i.e., single sensory/perceptual stimu lus in a spatial location defined by the sum total of sensory/perceptual stimuli in the environment) sepa rated by a short temporal interval.Since there are two distinct behav ioural episodes in close temporal proximity to each other, information pertaining to temporal relationships between stimuli (e.g., temporal conti guity) discriminates the two episodes and facilitates retrieval 27 .During the test phase, the animal is provided with a retrieval cue.The animal has to learn that the sensory/perceptual stimulus provided as a retrieval cue is a signal to displace a neutral block in the corresponding spatial loca tion previously paired with the cue (or to the sensory/perceptual stimu lus cued by a spatial location).Since none of the 50 sensory/ perceptual stimuli and 48 spatial locations are frequently paired (there are nearly 2,500 possible combinations), each pairing is trial (or behavioural epi sode) unique.Since the animal receives two distinct behavioural episodes followed by a retrieval cue to signal which of the two episodes needs to be recalled, the animal not only has to remember the relevant episode to receive reward, but also to discriminate between the relevant episode and the episodes presented either immediately before or after the relevant episode, as well as all previous episodes that occurred in the same or a similar spatial context.
The critical difference between the two tasks is not the cuedrecall nature of the latter task per se, but that the associations to be remembered are trial unique.This allows each beha vioural episode to be coded as unique, but increases potential inter ference from previous or subsequent behavioural episodes.To overcome this interference and to guide effi cient recall of the correct beha vioural episode, this task is performed with the contribution of the knowledge based and rulebased systems such as traditional biconditional discrimi nation tasks, but the trial specific epi sodes make it necessary to depend on the eventbased memory system to compare the retrieval cue to the stored episodes to efficiently recall the correct, and only the correct, behavioural episode to guide behav ioural decisions and actions.

Definitions critical to understanding spatial memory
Spatial memory is a general term that is made up of a number of com ponent processes that need to be Licensee OA Publishing London 2013.Creative Commons Attribution License (CC-BY) Competing interests: none declared.Conflict of interests: none declared.
All authors contributed to the conception, design, and preparation of the manuscript, as well as read and approved the final manuscript.
All authors abide by the Association for Medical Ethics (AME) ethical rules of disclosure.For citation purposes: Hunsaker MR.Embracing complexity: using the attribute model to elucidate the role for distributed neural networks underlying spatial memory processes.OA Neurosciences 2013 Sep 01;1(1):2.defined prior to proceeding into how the attribute model can be used to describe these processes.
The first of these spatial features is the distinction between egocen tric (or idiothetic or individual cen tred) and allocentric (or allothetic or world centred) reference frame.The allocentric reference frame typically refers to processing of space with reference to a constel lation of landmarks or stimuli or else with respect to a standard refe rence angle/ direction (for full mathematical description 41 .A fully learned allocentric representation of an environment is what is com monly referred to as a cognitive map as defined by O'Keefe and Nadel 6 .An egocentric frame of reference can mean a number of things, but most parsimoniously it is a reference frame in which space is computed to coor dinates that centre on the observer or the focus of the observer's atten tion.This can mean a representation linked to a single attended to beacon or landmark in space, or to observer centred coordinates proper.A cor relate to processing space using an egocentric reference frame is that the spatial representation must obligato rily be updated constantly to account for and change in the position of the observer, as once the observer has moved any remembered egocentric representation is now incorrect and no longer behaviourally relevant.
In performing calculations of space, there are four general classes of cues that can be used to generate mental representations of space that can be acted upon.The first two are the distinction between proximal and distal cues.Proximal cues refer to cues available for the subject to inter act with and that can be explored from all sides.Distal cues, however, are commonly referred to as envi ronmental cues or spatial context, and refer to cues 'out there' in space that form a background or general scene within which the other types of cues are located.Importantly, it has been shown that the hippocampus and entorhinal cortex preferentially respond to changes to distal, but not proximal, cues and the parietal cor tex shows the opposing pattern 42,43 .
The third type of cue is actually a mental construct generated from the relationship among proximal cues.If there is a general shape made up by the proximal cues, then the parie tal lobe will compute the topological relationships among the cues and the resulting general map of spatial relationships will be used to guide behaviour.In other words, if the objects are in a general kite shape, then rodents will preferentially use the general shape the cues gene rate rather than any features of the indi vidual cues to guide behaviour 44 .This process is disrupted by pari etal, but not hippocampus lesions 45 .Additionally, neuronal firing in the parietal cortex has been shown to fire specifically to topological information 46,47 .The fourth type of cue is the use of selfmotion cues to generate idiothetic cues.
Also important for the understand ing of spatial memory are the subtle distinction between spatial location memory, spatial distance memory, and memory for spatial context.Spatial location (or locale) memory refers to the ability to process a given specific portion of space as defined by the constellation of proximal and dis tal cues present in the environment.This has been shown to be performed in parallel by the hippocampus, and parietal cortex, as well as the ros tral cortices (infralimbic/prelimbic among others; [28][29][30]48 ). Memry for spatial context refers to the ability to discriminate the constellation of proximal and distal cues present in the environment from similar cues and cue configurations in different environments.The latter is a much more computationally intensive pro cess that has been strongly linked to hippocampus function [49][50][51] .
Memory for spatial distance actu ally refers to a computation per formed by the hippocampus at the level of the dentate gyrus-called metric processing [43][44][45] .This metric information is initially encoded by the dentate gyrus as a collection of egocentric distances and angles from the individual's eyecentred field of view.This collection of egocentric distances is then combined into a spatial representation that becomes allocentric in nature.In other words, with time and hippocampal pro cessing, what was once a collection of angles and distances between a rat and objects in space becomes independent to the rat's location and the precise angles and dis tances among the items in proximal space and distal space are directly represented 43 .Similarly, grid cells in the medial entorhinal cortex have been implicated in providing the information necessary for the calcu lation of spatial distance.
One of the burning questions in the early days of spatial memory research was just how does an indi vidual return to a starting point in a direct way after taking a non direct route in space.This process is called path integration.It is the ability of an animal to integrate the distances travelled, as well as the directions taken and use that information to return to a starting point (or nest/den in the case of for aging species; [52][53][54].At present, it is presumed that path integration in mammals makes use of the vestibu lar organs and other idiothetic inputs to detect and compute the animal's movements in the three dimensions.This information is then put together with motor effe rent signals, where the motor system tells the rest of the brain which movements were commanded, and optic flow, where the visual sys tem signals how fast the visual world moves past the eyes.Information from other senses such as echoloca tion and magnetoception may also be Licensee OA Publishing London 2013.Creative Commons Attribution License (CC-BY) Competing interests: none declared.Conflict of interests: none declared.
All authors contributed to the conception, design, and preparation of the manuscript, as well as read and approved the final manuscript.
All authors abide by the Association for Medical Ethics (AME) ethical rules of disclosure.For citation purposes: Hunsaker MR.Embracing complexity: using the attribute model to elucidate the role for distributed neural networks underlying spatial memory processes.OA Neurosciences 2013 Sep 01;1(1):2.integrated in animals that have access to those information.The hippocam pus is the part of the brain that inte grates linear and angular motion to encode a mammal's relative position in space and the medial entorhinal cortex appears to contribute to this process as well due to a robust rep resentation of idiothetic data in the medial entorhinal neurons 55,56 .

Application of the attribute model to spatial processing
The attribute model can be used to tease apart the roles for different neural networks underlying spatial processing.The role for each mem ory system as well as their interac tions will be discussed and a model for performance on spatial memory tasks will be posited.

Event-based spatial memory processes
For processes associated with event based memory, the focus will be on the role of circuitry associated with the hippocampus for spatial memory.With respect to specific spatial fea tures, such as allocentric spatial dis tance, egocentric spatial distance and spatial location, it has been shown in both rats and humans with bilateral hippocampal damage that there are severe deficits in eventbased mem ory for these spatial features [57][58][59][60][61][62][63] .
The spatial direction feature of spatial memory has also been inves tigated.Based on a delayed match ingto sample task for assessing memory for direction in rats, it was shown that hippocampal lesions dis rupt memory for direction.It should be noted that medial caudate nucleus lesions also produced an impair ment in memory for direction.It has been suggested in the literature suggesting that hippocampal lesions disrupt memory for the 'direction' to the goal because of impaired tri angluation with the help of exter nal landmarks, while caudate lesion disrupt memory for the direction to a beacon 64 .Additionally, it has been demonstrated that through out the neural networks subserving eventbased spatial memory (e.g., pre and post subiculum and medial entorhinal cortex among other regions) there are cell populations that show direction specific firing, called head direction cells that can be used by the rodent as an orienting mechanism [65][66][67] .
The determination of a spa tial pattern separation process has been developed extensively by computational models of the sub regions of the hippocampus with a special emphasis on the dentate gyrus [68][69][70] .Based on the empirical findings that all sensory inputs are processed by the dentate gyrus sub region of the hippocampus, it has been suggested that a possible role for the hippocampus might be to provide for sensory markers to demarcate spatial locations, such that the hippocampus can efficiently process spatial information 51,70 .It is thus possible that one of the main process functions of the hippocam pus is to encode and separate spa tial events from each other.This would ensure that new highly processed sensory information is organised within the hippocam pus and enhances the possibility of remembering and temporarily storing one place as separate from another place.It is assumed that this is accomplished via pattern separa tion of event information, so that spatial events can be separated from each other and spatial interference reduced.This process is akin to the idea that the hippocampus is invol ved in orthogonalisation of sensory input information 70,71 , in representa tional differentiation 72 , and indirectly in the utilisation of relationships 14 .
Within the eventbased memory system, the hippocampus com putes overall spatial relationships within the allocentric frame by com puting and combining egocentric calculations into a coherent spatial representation (perhaps via CA3 recurrent circuitry binding the ego centric and idiothetic inputs into a view invariant representation 51,70 ).This is particularly important for the generations of overall space anchored to distal, rather than proxi mal cues.
The hippocampusbased repre sentations involved mathematically rich information such as precise metric relationships among sti muli (Figure 2).The mathematical blueprint is generated using infor mation from the medial entorhi nal cortex and parahippocampal gyrus/postrhinal cortex.As such, this blueprint designated nodes occupied by elements, but does not specify the precise identity of the objects other than rough sketch information.
Next, this mathematical or archi tectural blueprint generated by the hippocampus is then populated with individual objects through interac tions with the rhinal cortices, par ticularly the perirhinal cortex and lateral entorhinal cortex.This can be viewed as a conjunctive process whereby the object and contextual information become bound into a single representation in the dentate gyrus, and as a relational process in CA3 wherein the context and object information remain separately rep resented to maximise flexibility in memory recall 14,40,51,73 .

Knowledge-based spatial memory processes
The posterior parietal cortex com putes spatial relationships in two reference frames: a topological refe rence frame with reference to the local cue configurations irrespective to the distal cue configurations as well as in the egocentric frame of ref erence, primarily in reference to the sagittal midline of the individual, but also to single landmarks or beacons in the environment [74][75][76] (Figure 2).

Licensee OA Publishing London 2013. Creative Commons Attribution License (CC-BY)
Competing interests: none declared.Conflict of interests: none declared.
All authors contributed to the conception, design, and preparation of the manuscript, as well as read and approved the final manuscript.
All authors abide by the Association for Medical Ethics (AME) ethical rules of disclosure.For citation purposes: Hunsaker MR.Embracing complexity: using the attribute model to elucidate the role for distributed neural networks underlying spatial memory processes.OA Neurosciences 2013 Sep 01;1(1):2.).(B) Depicts the neural computations carried out by the eventbased memory system as defined in Figure 1 with particular focus on the contributions from the hip pocampus.Spatial relationships among stimuli are defined mathematically in terms of raw angles and distanced from the rat.(C) Depicts the neural computations undertaken by the knowledgebased memory system, particular the posterior parietal cortex.Spatial relationships are defined by only the most general geometric relationships (i.e., geometric shape defined by the elements-connectedness and neighbourhood), but without regard to the vantage of the rat for this processing or specific regard to which object is located at which node. (D) depicts the processing by the rulebased memory system, particularly the infralimbic/prelimbic cortices.The rulebased memory system used affective information to guide exploratory behav iour and decisions undertaken by the rat.Importantly, the rulebased memory system can switch between the event and knowledgebased memory systems as needed to guide behavioural performance.(E) depicts an active interaction among the three memory systems, focusing on the proposed role for the retrosplenial cortex [112][113][114] .The retrosplenial cortex, having access to raw data from the thalamus, as well as the three memory systems, can perform computations in concert with the rule based memory system, as well as independently, and then signal that update to the rulebased memory system.Impor tantly, the retrosplenial cortex simplifies the spatial inputs to those relevant to guide task performance; this process provides a mechanism whereby an animal may compute egocentric position within an allocentric frame as proposed by other authors.

Licensee OA Publishing London 2013. Creative Commons Attribution License (CC-BY)
Competing interests: none declared.Conflict of interests: none declared.
All authors contributed to the conception, design, and preparation of the manuscript, as well as read and approved the final manuscript.
All authors abide by the Association for Medical Ethics (AME) ethical rules of disclosure.For citation purposes: Hunsaker MR.Embracing complexity: using the attribute model to elucidate the role for distributed neural networks underlying spatial memory processes.OA Neurosciences 2013 Sep 01;1(1):2.
Counterintuitively, however, the pos terior parietal cortex is also involved in spatial retrie val for allocentric information.
Rats with posterior parietal cor tex lesions display deficits in both the acquisition and retention of spatial navigation tasks that are presumed to measure the opera tion of a spatial cognitive map within a complex environment [77][78][79] .However, strong these lesion data may appear, it is worth noting that to date that there are not neuro physiological evidence for cognitive mapping in the rodent parietal lobe.They also display deficits in the acquisition and retention of spatial recognition memory for a list of five spatial locations.In a complex dis crimination task in which a rat has to detect the change in location of an object in a scene, rats with pos terior parietal cortex lesions are profoundly impaired 80 , yet on less complex tasks involving the dis crimination or shortterm memory for single spatial features includ ing spatial location, allocentric and egocentric spatial distance 81 there are no impairments.When the task is more complex, involving the asso ciation of objects and places (com ponents of a spatial cognitive map), then posterior parietal cortex plays an important role 81,82 .
Comparable deficits are found within an egocentricallocentric distance paired associate task 81 , but there is no deficit for an object object or objectodour paired asso ciate task, suggesting that spatial features are essential in activating and involving the posterior parietal cortex.Finally, it should be noted that in rats, neurons have been found within the posterior parietal cortex that encode spatial location and head direction information and that many of these cells are sensitive to multiple cues including visual, propriocep tive, sensorimotor and vestibular cue information 83,84 .

Rule-based spatial memory processes
Medial prefrontal cortex ( infralimbic/ prelimbic cortex in rodents) com putes the rulebased memory sys tem data to determine the rules of the task and provide that infor mation to the parietal and tempo ral lobe spatial processing systems (Figure 2).These processes are associated with shortterm work ing memory processes, cross modal switching, goaloriented control and prospective coding [85][86][87] .
The anterior cingulate cortex and agranular insula compare the behavioural outputs with expected rewards, affective contingencies, etc. and signal any mismatch to the infralimbic/prelimbic cortex for the rule to be modified 88,89 .
Rats with lesions of the infralimbic/ prelimbic cortex, but not of the ante rior cingulate and precentral cor tex, fail to acquire an objectplace association 88 .In a subsequent study Lee and Solivan 90 showed that tempo rary inactivation of the infralimbic/ prelimbic cortex led to profound impairments in an objectplace paired association task.Furthermore, impairment in a novelty detection paradigm using an objectinplace learning task has been observed in rats with infralimbic/prelimbic cor tex lesions 91 .In the context of other types of paired associate learning, Petrides 92 has shown that humans with prefrontal cortex lesions have difficulty in learning a paired asso ciate task and Pigott and Milner 93 reported that frontal lobe damaged patients are impaired for objects and places in a complex visual scene task.
The infralimbic/prelimbic cor tex appears to play an important role in working memory for visual object and spatial location infor mation.Supporting evidence is based on the findings that lesions of the infralimbic/prelimbic cor tex produce deficits in working memory for spatial information 88,89 , and working memory for visual object information 89,94,95 .However, infralimbic/ prelimbic cortex lesions do not produce a deficit in working memory for a food reward 96,97 .Further support of this conclusion was reported by Chang and colleagues 97 , who found sustained neural firing in the infralimbic/prelimbic cortex during the delay within a delayed matchingtoposition task, and Baeg and colleagues 98 who recorded from the infralibmic/prelimbic cor tex in a spatial delayed alternation task reported an increase in neu ral firing during the delay period.Importantly, it is becoming clear that the infralimbic/prelimbic cortices influence cellular firing in the hip pocampus by biasing the firing pat terns to goal locations or other task relevant factors other than place per se 100 .These findings are interpreted as the infralimbic/prelimbic cortices being critical for the computation of planned trajectories and modifi cation of behavioural output along those preplanned goals [101][102][103][104][105] .

Independence of memory systems
Rats with posterior parietal cortex lesions are impaired in an implicit spatial repetition priming experi ment but perform without diffi culty in processing positive priming for features of visual objects and a shortterm or working memory for a spatial location experiment 106 , sug gesting that the posterior parietal cortex plays a role in spatial percep tual memory within the knowledge based memory system, but does not play a role in spatial memory within the eventbased memory system.
Additionally, GoodrichHunsaker and colleagues 45 demonstrated that the hippocampus, but not the poste rior parietal cortex processed met ric relationships among proximal stimuli.However, the posterior parie tal cortex, but not the hippocampus, processed topological relationships among proximal stimuli.These find ings functionally doubly disso ciated Licensee OA Publishing London 2013.Creative Commons Attribution License (CC-BY) the event and knowledgebased memory system spatial processes.GoodrichHunsaker and colleagues 106 also demonstrated that the parie tal lobe mediates the processing of topological relationships among ele ments that make up stimuli using a task modified from one used to study topological processing deficits in Balint's syndrome 108 .
In humans there is a general loss of topographic sense, which may involve loss of longterm geo graphical knowledge as well as an inability to form cognitive maps of new environments.Using positron emission tomography scanning and functional magnetic resonance imaging data, it can be shown that complex spatial information results in activation of the parietal cortex 109 .Furthermore, in patients with parietal lesions and spatial neglect, there is a deficit in spatial repetition priming without a loss in short term or working memory for spatial information 110 .Additionally, Keane and colleagues 110 reported that a patient with occipitallobe damage (extending into parietal cortex) showed a deficit in percep tual priming but had no effect on rec ognition memory, whereas a patient with bilateral medial temporal lobe damage (including hippocampus) had a loss of recognition memory, but no loss of perceptual memory.

Integration among memory systems
The rulebased memory system sends projections into the elements processing the event and knowledge basedmemory systems (i.e., poste rior parietal cortex and parahippo campal gyrus/hippocampus).
Importantly, the retrosplenial cor tex is in a unique location to integrate head direction information from the thalamus and subicular complex (pre/parasubiculum), metric/ distal allocentric space from the hippocam pus and eventbased memory system and topologic/proximal space from the parietal cortex and knowledge based memory system as well as reciprocal connectivity with the ante rior cingulate, infralimbic/ prelimbic cortices, agranular insula and rodent homologue to the dorsolateral pre frontal cortex (Figure 1).This allows the retrosplenial cortex to not only act upon incoming sensory informa tion, but also to send the necessary signals to the rostral cortices to modulate/influence top down sig nals that guide behaviour This puts the retrosplenial cortex in a unique location to not only inte grate the event and knowledge based memory systems or to compute ego centric location in allocentric space as has been proposed 112,113 ; but also to actively switch among pure ego centric, pure allocentric, and integra tions involving combinations of the twocatered to the demands of each particular behavioural context 114 (Figure 2).
The obvious benefit of having a structure in this location is that it allows full integration of multiple spa tial reference frames in a manner that is capable of changing among many different states or forms depending upon the behavioural context 115,116 .Additionally, it is likely that the ret rosplenial cortex, in performing this integration removes redundancy from the spatial representation, which results in a parsimonious map sufficient to drive behavioural out put, but not containing the resolution of the spatial map computed within the eventbased memory system or elegant geometric representation of the knowledgebased memory system.Tasks requiring higher reso lution metric information or demand ing topological representations will result in the rulebased memory sys tem biasing the integration to favour those modalities.This is not a trivial point because the dentate gyrus in the hippocam pus will always compute an orthogo nal representation of space based on a metric representations of distal cues and mathematical relationships among the local cues and the distal cues in relation to the location of the rodent, whether the behavioural situation demands such a map or not.In parallel, the parietal lobe will always compute a topological, or ego centric space with a particular focus on proximal cue configurations, even when doing so may be disruptive to performance of the behavioural or spatial task at hand.As such, neither the hippocampus nor parietal read out is sufficient to guide performance on behavioural tasks, even with the input from the rostral cortices with rulebased information, as neither the posterior parietal cortex nor the hippocampus have the others' spatial information.
Additionally, the retrosplenial cortex receives information per taining to the sensory/perceptual and temporal attributes, as well as affective information via the infralimbic/ prelimbic cortices.These information from nonspatial attributes facilitate the identifica tion of beha vioural context, time of day or affective motivation; all infor mation useful for the optimisation of task performance 115 .
The retrosplenial cortex, however, having a robust connectivity with both hippocampus as well as pari etal cortex, has relatively complete access to both types of information, as well as independently derived information pertaining to head direction from the thalamus and pre/ parasubiculum (leading to retrosple nial neurons showing directionbased firing themselves 115 ).As such, with inputs from the anterior cingulate and infralimbic/prelimbic, the ret rosplenial cortex can bias the inte grated map toward topologic over metric information or ego over allocentric information as needed, or vice versa as situations demand (for evidence for infralimbic/ hippocampus cellular activity 101,105 ).Additionally, the retrosplenial cortex can transmit information pertaining to the map being computed to the rostral cortices (esp.anterior cingu late, infralimbic/prelimbic, anterior insula and rodent homologue to the dorsolateral prefrontal cortex) to inform the rule based memory sys tem pertaining to the maps being used to guide behaviour 113 .

Conclusion
Spatial processing is a complex pro cess mediated by a large network of structures located throughout the brain.Using the attribute model one can parse the networks into smaller subnetworks that can be studied in turn.The advantage of using a theo retical model such as the attribute model is that the resulting descrip tions of information processing lend themselves to empirical study since the model assumes mas sively par allel processing across memory systems.
Using the attribute model, a role for the retrosplenial cortex for spa tial processing can be proposed and tests of the theory may be designed.Specifically, following proposals of Aggleton and colleagues as well as Petrides and colleagues, the retro splenial cortex is assumed to pro vide a twoway interface among the eventbased, knowledgebased and rulebased memory system in such a manner as to optimise behav ioural performance.Importantly, the proposed role for the retrosplenial cortex for spatial processing can be easily tested using behavioural methods.Based on the proposed role for the retrosplenial cortex as an interface between the eventbased and knowledgebased memory sys tems, it seems logical that evaluat ing spatial processing using tests of eventbased, knowledgebased, and tests of the eventbased and knowledgebased memory systems' interactions would serve to tease apart the role for the retrosplenial cortex more precisely than previ ously possible.
information based on flexibility, action and pattern completion.

Figure 2 :
Figure2: Model of spatial processing by interacting memory systems.(A) Depicts an environment in which a rat is placed to explore in a spatial memory task(GoodrichHunsaker and colleagues, 2005, 2008).(B) Depicts the neural computations carried out by the eventbased memory system as defined in Figure1with particular focus on the contributions from the hip pocampus.Spatial relationships among stimuli are defined mathematically in terms of raw angles and distanced from the rat.(C) Depicts the neural computations undertaken by the knowledgebased memory system, particular the posterior parietal cortex.Spatial relationships are defined by only the most general geometric relationships (i.e., geometric shape defined by the elements-connectedness and neighbourhood), but without regard to the vantage of the rat for this processing or specific regard to which object is located at which node. (D) depicts the processing by the rulebased memory system, particularly the infralimbic/prelimbic cortices.The rulebased memory system used affective information to guide exploratory behav iour and decisions undertaken by the rat.Importantly, the rulebased memory system can switch between the event and knowledgebased memory systems as needed to guide behavioural performance.(E) depicts an active interaction among the three memory systems, focusing on the proposed role for the retrosplenial cortex[112][113][114] .The retrosplenial cortex, having access to raw data from the thalamus, as well as the three memory systems, can perform computations in concert with the rule based memory system, as well as independently, and then signal that update to the rulebased memory system.Impor tantly, the retrosplenial cortex simplifies the spatial inputs to those relevant to guide task performance; this process provides a mechanism whereby an animal may compute egocentric position within an allocentric frame as proposed by other authors.
prelimbic cortices influencing Licensee OA Publishing London 2013.Creative Commons Attribution License (CC-BY) Competing interests: none declared.Conflict of interests: none declared.All authors contributed to the conception, design, and preparation of the manuscript, as well as read and approved the final manuscript.All authors abide by the Association for Medical Ethics (AME) ethical rules of disclosure.For citation purposes: Hunsaker MR.Embracing complexity: using the attribute model to elucidate the role for distributed neural networks underlying spatial memory processes.OA Neurosciences 2013 Sep 01;1(1):2.

Table 1 Description of the processes performed by different memory systems used in the attribute theory as applicable to research using rodents Event-based Knowledge-based Rule-based
Hunsaker MR.Embracing complexity: using the attribute model to elucidate the role for distributed neural networks underlying spatial memory processes.OA Neurosciences 2013 Sep 01;1(1):2.

Table 2 Primary neuroanatomical correlates underlying each attribute in rodents Attribute Event-based Knowledge-based Rule-based
Murine homologs of † inferior temporal cortex, *medial prefrontal cortex, # orbitofrontal cortexLicensee OA Publishing London 2013.Creative Commons Attribution License (CC-BY)Competing interests: none declared.Conflict of interests: none declared.All authors contributed to the conception, design, and preparation of the manuscript, as well as read and approved the final manuscript.All authors abide by the Association for Medical Ethics (AME) ethical rules of disclosure.For citation purposes: Hunsaker MR.Embracing complexity: using the attribute model to elucidate the role for distributed neural networks underlying spatial memory processes.OA Neurosciences 2013 Sep 01;1(1):2.
Competing interests: none declared.Conflict of interests: none declared.All authors contributed to the conception, design, and preparation of the manuscript, as well as read and approved the final manuscript.
All authors abide by the Association for Medical Ethics (AME) ethical rules of disclosure.For citation purposes: Hunsaker MR.Embracing complexity: using the attribute model to elucidate the role for distributed neural networks underlying spatial memory processes.OA Neurosciences 2013 Sep 01;1(1):2.