Pattern separation in the hippocampus of similar object
stimuli. The potential influence of object-processing in
the perirhinal cortex.
Erik de Keijzer 10633294
Abstract
Pattern separation, the process of separating similar activity patterns to form non-overlapping output necessary to form distinct memories, is an important process that can occur in the hippocampus (HC) during the formation of episodic memory. Human imaging studies indicate that, in addition to activity in the HC, the lateral entorhinal cortex and perirhinal cortex (PRC) show increased activity during the separation of similar objects. The PRC is known to be important in object discrimination and object recognition memory. Therefore, the observed increase in activity might be related to computations contributing to pattern separation in the HC. How does the HC perform pattern separation on similar input activity? Do computations in the PRC influence pattern separation in the HC?
Based on literature, pattern separation in the HC involves a dispersion of signals on a five-fold higher number of neurons, changes in firing rates in the active subpopulation and the recruitment of newborn neurons resulting from adult neurogenesis. This is however predominantly shown for the separation of spatial information. The PRC is hypothesized to perform a similar computation during the separation of similar objects and images termed the ‘resolution of feature ambiguity’. Object representations formed and maintained in the PRC could be important for the separation of similar objects or influence pattern separation in the HC.
In conclusion, literature is at this point biased towards pattern separation of spatial information. The resolution of feature ambiguity in the PRC during object discrimination could inherently encompass the orthogonalization of similar input, but how the PRC interacts with the HC during object pattern separation remains to be elucidated.
Introduction
The simple process of parking your bike in the bicycle storage at the start of a working day, and successfully retrieving it after a busy working day, requires that each unique memory representation that is formed when parking your bike is separated from all the memories of every time you parked the same bike in the same storage before. The process that ensures that different occasions of similar events are stored separately, as distinct memories, is called pattern separation 1. It can be defined as
the computational process in which overlapping input to a brain region is separated in non-overlapping output 2. If the brain would be unable to perform pattern separation, and input is not
adequately disentangled, new memories could overwrite previously stored memories. Neural network models predict this would lead to abrupt forgetting upon learning of new information, so called catastrophic interference 3. Deficits in pattern separation have been observed in several
neurological diseases like Alzheimer’s disease, and have further been associated with cognitive impairments in schizophrenia. An inability to separate distinct occasions might also worsen the symptoms in post-traumatic stress disorder4. Deterioration in the ability to distinguish small changes
between environments or between similar stimuli is also thought to be the result of normal aging5
and has been observed in monkeys and rodents6,7, as well as in humans5.
Episodic memory –what happened when and where?—is usually classified as a form of declarative memory, i.e. the memory of facts and events 8. Spatial navigation has been implicated as a
behavioural read-out of declarative memory in animals 9. Two computations are predicted to be
important in the formation of episodic memory1. The hippocampal subregions are required to
orthogonalize similar events in order to store them separately, as well as retrieve previously stored events from partial cues. Pattern completion in the recurrent network of the CA3 subregion is thought to be important for the retrieval of a full episodic memory from degraded input 10. On the
other hand, the formation of a unique memory representation from similar input is believed to involve pattern separation occurring in the dentate gyrus and CA3 subregion of the HC 1.
Information to be stored in memory is conceived to reach the HC through the medial temporal lobe (MTL) in two information streams 11-14. Object-related information is thought to be processed
primarily by the PRC and lateral entorhinal cortex. Human imaging studies have indicated that the increase in brain activity during pattern separation is not confined to the HC. An increase of activity specifically in the lateral entorhinal cortex and PRC is observed when human subjects were requested to differentiate similar images while in an fMRI scanner14. This suggests the lateral entorhinal cortex
and PRC might also be engaged in the separation of object information. These afferent extra-hippocampal areas might thus already be involved in pattern separation-like computations on object information upstream of the HC.
The PRC is associated with the perception of objects and object recognition memory 15. It receives
highly processed information from all sensory areas, especially odour- and object related information
16. Loss of function of the PRC has been shown to result in deficits in the perception of images,
specifically when these are very similar 17. Together, the anatomical position of the PRC, indications
from human imaging studies and deficits in perception after ablation of the PRC imply it is possible that the PRC is involved in the orthogonalization of object information. So, do computations in the PRC influence pattern separation in the HC?
To answer this overarching research question, first the neural correlate of pattern separation as occurring in the HC will be investigated. Secondly, a potential role for the PRC in the perception of similar stimuli will be investigated. Subsequently, a number of scenarios of contributions of computations in the PRC to pattern separation in the HC will be discussed. First, orthogonalization of object information could take place in the PRC, whereas the HC could perform pattern separation independently. Lesions to one of these areas would then reveal stimulus domain-specific impairments. Secondly, computations in the PRC and pattern separation in the HC could reinforce each other. Either the PRC could reduce overlap in object-related input before non-overlapping representations are eventually formed in the HC 14, or the PRC could form and maintain
non-overlapping object representations 17,18, which could subsequently be important for pattern
The role of the hippocampus in the separate storage of
similar events
The hippocampus (HC) has long been implicated in spatial navigation and long-term memory 19.
Spatial navigation has been used as a read-out of declarative memory (the memory of facts and events) in animals. During the formation of the memory of an event –what happened when and where?-- the HC is attributed an important role in separating similar occasions to form unique memory representations. How does the HC contribute to the separation of information that can be stored in long-term memory? From literature, there are three leading indications that the DG of the HC is especially suited to separate similar events.
The first indication that the DG might be important in the separation of information is its anatomical position. Input to the HC enters the DG from the entorhinal cortex (Figure 1) and is dispersed here on a five-fold higher number of dentate granule cells 20.
Slightly different patterns of activity in the entorhinal cortex likely engage different DG principal cells 21. These dentate granule cells show
sparse firing, but their mossy fibers terminate in powerful synapses on the CA3 pyramidal cells 21.
The sparse level of firing of dentate granule cells probably contributes to the separation of input in combination with sparse connections to the CA3 subregion to produces sparse, orthogonal output
20,22. The CA3 subregion of the HC has extensive
recurrent collaterals and it is thought the network in this subregion participates in the separation of memory information, but can also retrieve a full memory from partial cues 1.
Lesions of the HC result in severe deficits in spatial navigation, for instance in a radial arm maze 23. Most
lesion studies of the DG involve the dorsal DG, as this is the first part encountered when the DG is accessed along the dorsal-ventral axis. Lesions of the dorsal DG result in deficits in the discrimination between similar contexts, or between different object-place paired associations 24. It is thought the
dorsal DG is involved in the processing of object-context and object-place conjunctive information 25.
Rats with lesions to the hippocampus or DG are impaired in the acquisition and retention of object-in-place memories 24. However, ablation of the
dorsal DG does not result in deficits in the separation of similar objects, unless a spatial change
Figure 1: Schematical representation of areas converging on the hippocampus.
Inputs from multiple cortical regions and all sensory modalities converge on the entorhinal cortex, which serves as the principal input source of the
hippocampus. Forward projections are depicted by blue lines, green dashed lines depict backprojections from mainly the CA1 subregion. Adapted from (Rolls, 2016) MTL: medial temporal lobe.
is made to a complex object 25. This implies the dorsal DG is primarily involved in the processing of
spatial and contextual information.
The DG is one of three sites in the brain where adult neurogenesis takes place 26. Adult neurogenesis
means that, after the brain is fully developed, stem cells are developing into neurons that are subsequently integrated in the DG network 27. These newly developed neurons are labeled adult-born
or new-born neurons, whereas the population that is present from the developmental stage is called development-born. The new-born neurons comprise about 5% of the total population and distinguish from the development-born population (95%) in that they are more excitable 28. Mice in which
neurogenesis was disrupted, either by irradiation or blocking specific essential signaling pathways, showed impaired spatial discrimination compared to controls when the difference between two choice options was small but not when the difference was large 27. The limitation of an effect
specifically to situations with little spatial separation implies a detriment to the ability to separate similar events. Conversely, transgenic mice in which the new-born neuronal population was intact but the developmental population was functionally disrupted showed similar performance compared to controls 29. Furthermore, an improvement in object recognition has been observed when an increase
in adult neurogenesis was induced through exercise and running 30. Mice that were allowed to run in
their home cages showed an increase in new-born neurons in the DG as well as improved object recognition, specifically when objects were very similar 30. This study provides one of the few possible
links between neurogenesis and the separation of object information. Together these studies indicate that the separation of information in the HC depends on the new-born population of neurons in the DG.
On the neuronal level, the separation of different spatial contexts – the where? in an episodic memory-- was investigated by measuring changes in electrical activity in the HC. Place cells in the HC are known to form a cognitive representation of the environment 19. Hippocampal place cells fire in
specific locations in the environment called ‘place fields’. The subset of active neurons in a place field adapts its firing to changes in the environment. These changes elicit rate- and global remapping 20.
Rate remapping involves changes in the firing rate of an active subset of neurons. Global remapping on the other hand involves the recruitment of a largely different subset of active neurons, in addition to changes in firing rates 20. Remapping has been observed after changes to the shape of an enclosure 20,31 or the context in a contextual fear paradigm 21. Small changes to the shape of an enclosure causes
rate remapping in hippocampal place cells, whereas larger changes -like moving the enclosure to a different room- results in global remapping 20. Global remapping has also been observed in the DG 21.
A curved side-wall and different visual and olfactory cues differentiated a safe context from an unsafe context in a contextual fear paradigm. During the successful separation of the safe from the unsafe context, as indicated by lower levels of freezing behavior, a different subset of place cells was active in the DG.
Together, the anatomical position of the HC, the role of adult neurogenesis in the DG and the changes in electrical activity in the HC during the encoding of similar events imply that the HC, and more specifically the DG and CA3 subregions in the HC, play an important role in the separation of similar events.
The computation that is thought to underlie the cognitive function of forming unique memory representations of very similar events is called pattern separation 1. The term ‘pattern separation’
originates from the field of computational neuroscience and describes the process of separating input (population activity) to a brain area to form non-overlapping output 2. On a behavioral level, pattern
separation is tested using similar stimuli and the discriminability between stimuli is parametrically changed 7,20,27,32. Such stimuli can for instance be squares on a touchscreen or baited arms in a radial
arm maze 27 in which the distance between the squares or arms is altered, or an environment in
which small changes have been made to the enclosure 20,21,31. To conclusively establish the occurrence
of pattern separation in a brain area, the input to this brain area should be measured as well as its output 32. If the output shows less overlap than the input, a pattern separation-like process has likely
occurred. Thus, the occurrence of pattern separation can only be established on a cellular/ molecular level by finding the neural correlate, a reduction in overlap in cell population activity compared to activity upstream of the investigated area. The above described studies did not measure both input and output to the investigated areas. Additionally, it is unknown whether rate- or global remapping preferentially occurs in the new-born cell population, as the identity of neurons in in vivo recording studies cannot be easily determined 33, although it is postulated the more excitable newborn cells are
likely engaged 29.
In the investigation of pattern separation, an important methodological issue is the presence of discriminating cues, as these can provide prominent extra discriminative stimuli. It is important to parametrically change the ‘discriminability’ between two contexts, objects, or other stimuli. Only then does an experiment presumably tap into the putative underlying computation of pattern separation. For instance, the presence of a grid floor in a contextual fear paradigm has been reported to provide a prominent environmental cue 34. The presence of this cue could undermine the necessity
for pattern separation in the DG. In the above described contextual fear study, the grid floor was only uncovered in the session in which the mice received a shock 21. However, as the mice displayed
sustained freezing behavior in the unsafe context with floor cover, but not in the safe context, the addition of a plastic floor cover was seemingly a small enough change to not elicit a aberrant behavioral response whereas the changes to the context were large enough to differentiate it from the unsafe context. The occurrence of global remapping during pattern separation can however not be established from this single study, and as the similarity between the contexts was not an explicit parameter, we can only assume the changes were small enough to necessitate pattern separation. In conclusion, experimental evidence indicates that the HC contributes to the independent storage of overlapping stimuli in memory through pattern separation in the DG. The DG is thought to be an important site for pattern separation because input from the entorhinal cortex is dispersed here on a five-fold larger number of sparse-firing dentate granule cells. This alone is however likely not sufficient to explain the occurrence of pattern separation. Additionally, easily excitable newborn granule cells and remapping through changes in firing rates of the active subset of neurons are likely to play a role. It is probable that pattern separation also involves the activation of different subset of neurons in the DG (global remapping in the DG) as highly similar inputs engage different dentate granule cells. Research regarding pattern separation in animals is mostly concentrated on the separation of similar spatial input. Pattern separation of objects with overlapping features receives less attention.
The role of the perirhinal cortex in the separation of similar
images or objects
Further investigation of the potential contributing role of the PRC to hippocampal pattern separation requires elaboration on the role of the PRC. The PRC is considered to be part of the ventral visual stream, which is involved in the identification and recognition of visual information. The medial temporal pathway of the ventral visual stream, towards the PRC, is specifically implicated in the perception of objects 13. The PRC is also considered important for recognition memory 35. Recognition
memory is often divided into two components: recollection and familiarity 15. Familiarity could be
described as ‘the feeling that something has been encountered before’, whereas recollection involves the memory of a specific event –what happened when and where?-- in which an object or person has been previously encountered, including contextual details 15,36. Object recognition has been used as
another important test for declarative memory in animals, like spatial navigation and its neural correlate in the HC. Before it reaches the HC, object information is considered to be processed in the PRC 24,37-39. How does the PRC differentiate during the perception of similar images or objects?
An important model 17,18 for a function of the PRC in separating object information regards ‘feature
ambiguity’ which arises when a feature of a complex object or an image is rewarded in one case, but not in the other. Lesions of the PRC result in deficits in object discrimination. These deficits in discriminating objects or images that are paired with a reward only become apparent when objects are very similar 17. Object discrimination is impaired when monkeys have to discriminate images
consistent of two pictures or images comprised of two combined ‘morphed’ images (Figure 2). Only one stimulus is rewarded, but the rewarded image shares features (or an entire picture) with the unrewarded stimulus. Thus, selection of the rewarded image becomes more difficult with an increase in overlap with other, unrewarded images. This overlap has been termed ‘feature ambiguity’ and discrimination of images is thought to become more difficult with an increase in feature ambiguity
17,18. A pattern separation-like process might be taking place in the PRC to decrease overlap between
the images, which has been coined the ‘resolution of feature ambiguity’ 18,40.
Figure 2: Images with different levels of feature ambiguity. (a) Image pairs share one picture
with the unrewarded stimulus in none (minimum), two (intermediate) or all (maximum) of the four possible combinations of stimulus pairs. FA, feature ambiguity. (b) A rewarded (+) and unrewarded (-) stimulus morphed to different degrees. Numbers depict the position of a picture in a range from 1 to 40 of overlap between the two starting images. Pictures 20 and 21 are closest together and share the highest degree of overlap. (Bussey & Saksida, 2005)
B A
The above described findings after ablation of the PRC have led to the representational-hierarchical model of object discrimination. The representational-hierarchical model proposes that discrete content is formed in response to ambiguous but different stimuli 17,40. This model places the PRC atop
a functional hierarchy in the ventral visual stream where increasingly complex representations are formed. This model argues the ‘resolution of feature ambiguity’ takes place in the PRC and proposes that distinct multimodal representations might be formed in the PRC when a subject is confronted with stimuli with a high level of feature ambiguity. The representational-hierarchical model of increasingly complex representations throughout the ventral visual stream is in accordance with electrophysiological measurement in neighbouring visual cortical area TE in monkeys 41. Area TE feeds
information to the PRC and the PRC is, in this account, responsible for the composition of distinct representations comprising multiple modalities 40. However, the existence of conjunctive multimodal
object representations in the PRC, as predicted by the representational hierarchical model, has not been conclusively established.
Ambiguity arises for instance when a feature of a complex object or an images is rewarded in one case, but not in the other. Like in the above described task in which a picture could appear in the rewarded and unrewarded stimulus-pair. The process of resolving feature ambiguity differs from pattern separation in that it does not consider a process of separating overlapping inputs into non-overlapping outputs inasmuch as it considers the formation of non-non-overlapping content. The above described model proposes the formation of increasingly complex representations of in particular visual information across brain regions 17,40. This idea could be conceived as an inherent process that
takes place during the formation of distinct representations in a high dimensional space formed by the population code 41. Different ‘ambiguous stimuli’ would still occupy a different subspace in this
high dimensional representational space, as other neurons fire for different image features. The conjunctive, multimodal representation in the PRC would then encode positive valence for a rewarded object in one case, but not for a highly similar object represented in another representational subspace. Thus, the representational-hierarchical model proposes the PRC is involved in the process of resolving feature ambiguity through the composition of separate patterns of activity for objects and images upstream of the HC 17,18,40.
The resolution of feature ambiguity was tested on a behavioral level in the above described experiment, using tests with and without a memory component 18. The discrimination between
similar images was impaired when lesioned animals had to acquire stimulus-reward associations of images with a high degree of overlap (i.e. feature ambiguity was high). However, object discrimination was also impaired when the animals learned to discern stimuli with low feature ambiguity during sample sessions, but subsequently had to discriminate versions of those images with high feature ambiguity in the test phase 18. This points to an effect of lesions of the PRC on
object perception and not object recognition memory, as the discrimination of images with a high degree of overlap is readily impaired when they are viewed for the first time.
As stated above, the presence of conjunctive multimodal representations of objects in the PRC has not yet been established. Electrophysiology studies did find that single units in the PRC can show ‘object fields’ 39, analogous to the place fields of place cells in the HC. Single neurons in the PRC
increase their firing in the immediate vicinity of an object and are silent in other locations of the environment. When objects are absent, the proportion of non-selectively firing PRC neurons is increased 39. This points to a stable subset of PRC principal cells that is dedicated to encoding
object-related spatial information. Even brief object exploration activates a stable amount of PRC neurons and once established, spatial firing patterns in the PRC can be maintained for at least two hours 42.
single units showed specificity in firing for a particular object 39. Meanwhile most of the PRC principal
cells involved retained their firing fields regardless of object identity. This could still mean these are important in encoding spatial information with regard to objects, but might make it less likely they are involved in the discrimination between individual objects.
Tests with similar images in humans that are designed to tap into the process of pattern separation do implicate the PRC is involved in the discrimination of these images. The PRC shows increased activity with the identification of lures 14. These lures are images that resemble images encountered during a sample phase, but are presented in a different location on the screen (spatial lures) or are similar but slightly different pictures (object lures). The PRC and LEC are, together with the DG/CA3 region, more active during the rejection of object lures. Together, lesion studies in animals and human fMRI studies provide indications the PRC is involved in orthogonalizing similar images. The same task using object and spatial lures has been used to identify subclinical object-specific memory deficits in aged individuals 5. The PRC is suggested to be especially susceptible to the deteriorating
effects of aging 5,7. Aged individuals show a significant lower performance in the rejection of object
lures compared to spatial lures. This is hypothesized to reflect a general age-associated decline in pattern separation abilities in older individuals. This age-associated deterioration in pattern separation abilities is in turn also observed in aged animals 7 and can be linked to a deterioration of
object-related spatial activity in the PRC 42. In older animals a lower proportion of PRC neurons is
active and neuronal ensembles show reduced firing patterns and lower information content 42, as well
as less ensemble overlap between separate trial phases with the same objects 43. A deterioration of
(spatial) object-related signaling in the PRC could be of importance for the computation of separate object representations 7,39. A lower proportion of active PRC neurons can reduce the discriminative
power of the population which might impair object recognition of highly similar stimuli.
However, other studies on lesions of the PRC pose findings that are in contradiction with the representational-hierarchical model discussed. In contrast to the above described experiments, lesions of the PRC can also result in severe impairments in object recognition memory. Recognition memory for objects is often tested in a spontaneous object recognition (SOR) task. Animals receive two objects to explore during a sample phase. After a retention delay, they re-encounter a copy of the object encountered during the sample phase and one novel object. During this test phase, animals display a natural tendency to explore the novel object more 15. Loss of function of the PRC
only results in deficits in the SOR-task if the retention delay exceeds approximately 10 minutes 15,44.
Likewise, the performance of non-human primates in a delayed match-to-sample task is impaired in a delay-dependent fashion after ablation of the PRC 36. As there are no deficits with a short retention
delay, perception seems to be intact. This likely indicates that loss of function of the PRC results in a deficit in the recognition memory for the objects. Furthermore, the representational-hierarchical model is based mainly on observations from tests with images, as is the potential participation of the PRC in pattern separation in humans. It is however likely that objects and images are not processed similarly in the brain 16. For instance, the well-known repetition suppression effect, in which the
repeated presentation of an image results in reduced activity in the MTL compared to the first viewing of that image, does not occur for 3D objects 16. A simple 2D image and a complex 3D object in
its spatial location and context might engage different brain areas or similar brain areas differently.
In short, the PRC is important in the recognition, perception and in maintaining representations of objects. However, the exact role of the PRC in the resolution of feature ambiguity for individual
images and objects remains to be unraveled. To date, little is known about the formation of distinct object representations in the PRC and their possible role in the discrimination of objects that share many features.
Interactions between the PRC and the hippocampus in the
processing of object information
The hippocampus is considered important for the separation of mnemonic information through pattern separation in the DG (Chapter 1). The PRC could play a role in the perceptual separation of similar images and objects through the formation of conjunctive representations (Chapter 2). Apart from these roles in the separation of mnemonic and perceptual information, the hippocampus and perirhinal cortex are both attributed important roles in declarative memory: the memory of facts and events 8. Declarative memory is often tested in animals using spatial navigation tasks for tests of
hippocampal functioning 2,9 and object recognition tasks for tests of perirhinal cortex functioning 15,36.
The formation of an episodic memory, which is classified as a form of declarative memory, is thought to involve the separation of similar events through pattern separation in the HC (see chapter 1). Because lesion studies of the PRC reveal deficits in the resolution of feature ambiguity and human imaging studies show increased activity in the PRC during the rejection of object lures, computations in the PRC have been proposed to be important for separation of object stimuli. The PRC and HC are interconnected via the intermediate entorhinal cortex as well as via direct connections 28. The
entorhinal cortex is the principal input source to the HC and divided in a medial and a lateral part that feed information to the DG 11,12. The medial- and lateral entorhinal cortex are thought to differentially
process visual and spatial information; the lateral entorhinal cortex (LEC) preferentially processes object-related information along with the PRC whereas the medial entorhinal cortex (MEC) preferentially processes spatial information along with the postrhinal cortex (or parahippocampal area in humans) 11,12,14. As part of the MTL, the PRC is more strongly connected with the LEC, whereas
the MEC shares more connections with the postrhinal cortex, such that the division between MEC and LEC is prolonged throughout the MTL 12,13. Although a functional and anatomical dichotomy
between the two processing streams is fairly well-established, multiple theories have been proposed regarding the contributions of MTL circuitry to computations in the HC. If the PRC would aid in the separation of object-related information, how could the PRC and HC interact to contribute to the separation of object-related information?
In the first scenario, the HC and PRC could operate entirely independently 36,45. The PRC and HC could
subserve different kinds of memory, wherein the PRC is important for object recognition and the association between objects, whereas the postrhinal cortex and HC encode the spatial and situational context 45. In this framework, the PRC and HC are part of two dissociable networks, termed the
anterior temporal system and the posterior medial network respectively. The PRC in the anterior temporal system is hypothesized to encode items in a multidimensional space, such that similar objects that differ in a single dimension are represented differently. If object and spatial information are processed by independent loops involving the PRC and HC respectively, lesions of the PRC would leave pattern separation in the HC intact and in turn, lesions of the HC would not affect object recognition memory. A double dissociation between the function of the HC and PRC has been established for spatial and object recognition memory 23. Bilateral excitotoxic lesions of the HC lead to
selective impairments of spatial memory while leaving object recognition memory intact and vice versa for lesions of the PRC plus postrhinal cortex. Object information from the PRC and context information from the postrhinal cortex is thought to converge in the HC 45,46, where these can be
bound into an episodic memory. However, recent lesioning experiments reveal the entorhinal cortex already processes object- as well as contextual information 46. Although selective lesions of the
medial- and lateral entorhinal cortex lead to selective impairments in object recognition when the identity of the object was changed (LEC) or the context in which it was presented (MEC), no deficits were observed when both the object and the context changed simultaneously, which indicates both
the LEC and MEC are capable of processing object and context related information. What is more, a spatial signal has been established in the PRC 16,39(see chapter 2) and the PRC has also been reported
to represent contextual information 38 which seems to contradict a strict division in mnemonic
function.
In the second scenario, the PRC and HC could interact to successfully separate object-related information. The modular approach in which different brain regions are viewed as performing distinct mnemonic functions is opposed by the representational-hierarchical model in which the formation of increasingly complex multimodal representations is proposed 17,18,40. If the PRC holds conjunctive
multimodal object representations, these could be important input for the separation of similar objects in the HC. The formation of increasingly complex content in this model is hypothesized to extent up to the HC 40. Unfortunately, as this view is based mainly on lesion studies of the PRC, how
these representations contribute to pattern separation in the DG is unspecified. Third, information could be
pre-separated in the PRC, before distinct representations are formed in the HC. Findings in human images studies regarding the separation of similar images have led to the proposition of a hierarchical model of incremental reduction of overlap between representations of objects throughout the MTL 14. In this
framework, overlap is reduced along the input pathways to the HC, before the HC forms distinct representations through pattern separation in the DG. The observed differences between MEC/ parahippocampal and PRC/ LEC pathways is explained as a division in processing across domains. The activation of the MEC and parahippocampal cortex during the rejection of spatial lures and of the PRC and LEC during rejection of object lures led to the hypothesis that information in the spatial and contextual domain is separated in the dorsal visual stream and information in the images and
objects domain in the ventral visual stream, whereafter the DG acts as domain-agnostic general pattern separator 14. Domain-selective reduction of overlap could aid in a stronger cross-domain
reduction of overlap. As these ideas are based mainly on changes in the activation of brain areas, how the domain-selective reduction in overlap is achieved is not specified further. However, findings after lesions of the dorsal DG seem at odds with this idea. If the DG would act as a domain-agnostic pattern separator, damage to the DG is expected to impair the separation of object and spatial information equally. After ablation of the dorsal DG in rats, these animals by contrast show no deficits in the detection of a novel component in a combined object made up of two parts, which is thought to
Picture 3: A model of incremental reduction of interference for spatial and object information.
Overlapping spatial- and object-related information is pre-processed separately throughout the dorsal and ventral visual streams and in the medial temporal lobe. The dentate gyrus subsequently acts as a domain-agnostic general pattern separator. (Reagh & Yassa, 2014)
require pattern separation 25. Pattern separation was only impaired when the novel part was placed
2cm away from the other part of the combined object or a 2cm gap was introduced between two parts of a combined object. This could indicate pattern separation in the DG only separates object information when there is a spatial component involved 25, which would be in opposition to the
notion of a domain-agnostic general pattern separator.
In conclusion, the PRC and HC could operate independently, with separate processing of objects in the PRC and contextual information and object-context associations in the HC. Alternatively, increasingly complex content could be formed throughout the MTL up to the HC. How these representations contribute to pattern separation in the HC is unknown. Overlap in representations could be reduced along the input pathways to the HC before the formation of distinct representations through pattern separation in the HC. How overlap is reduced along the input pathways is not further specified. Most models consider the predominant processing of spatial information in the postrhinal cortex in conjunction with the MEC and of object-related information in the PRC and LEC. Therewithal, object representations in the PRC are generally contributed an important role in object memory.
Discussion
The hippocampus is thought to separate overlapping input (activity patterns in the principal input source, the entorhinal cortex) through pattern separation. The process of pattern separation likely involves the dispersion of the incoming signal on a five-fold higher number of dentate granule cells 20,
it can be detected as changes in firing rates of the active population (rate remapping) 20,31 and the
newborn cell population in the DG seems critically involved 27,29. The PRC has been hypothesized to
perform a similar computation during the perception of similar objects 17,18,40. In humans and animals
it has been observed the PRC is activated 14 and required 18 to separate similar images. The
computation performed by the PRC is however not thought to be pattern separation, but the formation of conjunctive, multimodal representations of objects 18,40. Little is known about the
contributions of pattern separation in the hippocampus or the resolution of feature ambiguity in the PRC to the behavioural result of successfully selecting the rewarded option out of very similar objects or images. Do computations in the PRC influence pattern separation in the hippocampus? Three options were discussed: the PRC and HC could operate independently 36,45 or they could interact, with
distinct multimodal conjunctive object representations 17,18,40 providing important input for PS in the
HC or a gradual reduction of overlap between object representations along the input pathway to the HC 14.
The research regarding pattern separation is mostly concentrated on separation of spatial inputs in the dorsal DG. The most common example of pattern separation: “Where did you park your bike today and where did you park that same bike yesterday?” is a reflection of this emphasis. This could be due to the accessibility of the dorsal DG and the discernibility of hippocampal place cells. Pattern separation of objects with overlapping features receives less attention and controversy remains whether pattern separation of objects in the DG requires a spatial component, as animals without dorsal DG are unimpaired in normal object recognition or detection of change in one half of a combined object 10. Furthermore, few studies explicitly manipulate the discriminability between
choice options. In some occasions it is thus difficult to gauge the necessity for an underlying process of pattern separation. It is for instance controversial whether global remapping in the DG already occurs with small changes in input 21 as a role for the dispersion of an incoming signal on a high
number of neurons during pattern separation and thus the recruitment of a largely different cell population would imply. Even with the use of comparable behavioural tasks that use similar stimuli and manipulate the similarity between stimuli (‘pattern separation tasks’) to test the function of the PRC 7,18 or HC 20,27 there is a discrepancy between the fundamentally different underlying
computations that are thought to generate the behavioural results (pattern separation vs. the resolution of feature ambiguity). Lastly, it is unknown whether the changes in firing rates observed in the DG occur in the newborn population. It has been hypothesized this easily excitable population is likely engaged 29, but identification of the active population during measurement of activity is difficult 33.
Any model that would attempt to explain the role of the PRC in the separation of object information should take into account the functional and anatomical division between the two information pathways in the MTL 11-14. Stronger object-related information representation has been observed in
the PRC and LEC, whereas spatial information is stronger in the MEC and postrhinal cortex (parahippocampal cortex in humans) 47. This is however presumably not a strict division, as the PRC
also contains a spatial signal 16,39 and the medial and lateral entorhinal cortex are interconnected 11. A
possible role of the PRC would most likely involve the formation or maintenance of object representations 15,39. How these representations are formed, which are hypothesized to be
recognition tasks and electrophysiological measurements of the spatial component of an object representation in a small subset of PRC principal cells are not sufficient to elucidate the formation of these complex conjunctive multimodal representations. Representational similarity analysis in the PRC could help to clarify this matter. Analysis of differences in representations in the MEC/parahippocampal area and LEC/PRC have revealed object representations in the PRC for instance depict different objects in the same location, whereas the MEC does not seem to retain separate representations of different objects in the same location 38. Representational similarity
analysis could likewise reveal differences in representations of overlapping object input (similar 3D objects) in the PRC.
Conclusion
The investigation of the putative underlying process of pattern separation in the hippocampus has thus far mainly involved behavioural tasks that require the separation of spatial information. Less is known about the separation of object information which is thought to be processed differently throughout the MTL. The PRC could be involved in the separation of object information, although the proposed process of the resolution of feature ambiguity is fundamentally different from pattern separation and could better be described as ‘pattern composition’. An interaction of this process with pattern separation or a role for object representations in the PRC in pattern separation in the hippocampus remains to be established.
References
1. Yassa MA, Stark CE. Pattern separation in the hippocampus. Trends Neurosci. 2011;34(10):515-525. 2. O'reilly RC, McClelland JL. Hippocampal conjunctive encoding, storage, and recall: Avoiding a trade ‐ off. Hippocampus. 1994;4(6):661-682.
3. McClelland JL, McNaughton BL, O'Reilly RC. Why there are complementary learning systems in the hippocampus and neocortex: Insights from the successes and failures of connectionist models of learning and memory. Psychol Rev. 1995;102(3):419.
4. Van Hagen B, van Goethem N, Lagatta D, Prickaerts J. The object pattern separation (OPS) task: A behavioral paradigm derived from the object recognition task. Behav Brain Res. 2015;285:44-52. 5. Reagh ZM, Ho HD, Leal SL, et al. Greater loss of object than spatial mnemonic discrimination in aged adults. Hippocampus. 2016.
6. Burke SN, Wallace JL, Nematollahi S, Uprety AR, Barnes CA. Pattern separation deficits may contribute to age-associated recognition impairments. Behav Neurosci. 2010;124(5):559.
7. Burke SN, Wallace JL, Hartzell AL, Nematollahi S, Plange K, Barnes CA. Age-associated deficits in pattern separation functions of the perirhinal cortex: A cross-species consensus. Behav Neurosci. 2011;125(6):836.
8. Tulving E. Elements of episodic memory. . 1985.
9. Tulving E, Markowitsch HJ. Episodic and declarative memory: Role of the hippocampus.
Hippocampus. 1998;8(3):198-204.
10. Neunuebel JP, Knierim JJ. CA3 retrieves coherent representations from degraded input: Direct evidence for CA3 pattern completion and dentate gyrus pattern separation. Neuron. 2014;81(2):416-427.
11. Knierim JJ, Neunuebel JP, Deshmukh SS. Functional correlates of the lateral and medial entorhinal cortex: Objects, path integration and local-global reference frames. Philos Trans R Soc Lond B Biol Sci. 2013;369(1635):20130369.
12. Deshmukh SS, Knierim JJ. Representation of non-spatial and spatial information in the lateral entorhinal cortex. Frontiers in behavioral neuroscience. 2011;5:69.
13. Deshmukh SS, Johnson JL, Knierim JJ. Perirhinal cortex represents nonspatial, but not spatial, information in rats foraging in the presence of objects: Comparison with lateral entorhinal cortex.
Hippocampus. 2012;22(10):2045-2058.
14. Reagh ZM, Yassa MA. Object and spatial mnemonic interference differentially engage lateral and medial entorhinal cortex in humans. Proc Natl Acad Sci U S A. 2014;111(40):E4264-73.
15. Winters BD, Saksida LM, Bussey TJ. Object recognition memory: Neurobiological mechanisms of encoding, consolidation and retrieval. Neuroscience & Biobehavioral Reviews. 2008;32(5):1055-1070.
16. Burke SN, Barnes CA. The neural representation of 3-dimensional objects in rodent memory circuits. Behav Brain Res. 2015;285:60-66.
17. Bussey TJ, Saksida LM. Object memory and perception in the medial temporal lobe: An alternative approach. Curr Opin Neurobiol. 2005;15(6):730-737.
18. Bussey TJ, Saksida LM, Murray EA. The perceptual-mnemonic/feature conjunction model of perirhinal cortex function. The Quarterly Journal of Experimental Psychology Section B. 2005;58(3-4):269-282.
19. O'keefe J, Nadel L. The hippocampus as a cognitive map. Oxford University Press, USA; 1978. 20. Leutgeb JK, Leutgeb S, Moser MB, Moser EI. Pattern separation in the dentate gyrus and CA3 of the hippocampus. Science. 2007;315(5814):961-966.
21. Deng W, Mayford M, Gage FH. Selection of distinct populations of dentate granule cells in response to inputs as a mechanism for pattern separation in mice. Elife. 2013;2:e00312.
22. Kesner RP. Role of the hippocampus in mediating interference as measured by pattern separation processes. Behav Processes. 2013;93:148-154.
23. Winters BD, Forwood SE, Cowell RA, Saksida LM, Bussey TJ. Double dissociation between the effects of peri-postrhinal cortex and hippocampal lesions on tests of object recognition and spatial memory: Heterogeneity of function within the temporal lobe. J Neurosci. 2004;24(26):5901-5908. 24. Byun J, Lee I. Disambiguation of similar object-place paired associations and the roles of the brain structures in the medial temporal lobe. Experimental neurobiology. 2010;19(1):15-22.
25. Kesner RP, Taylor JO, Hoge J, Andy F. Role of the dentate gyrus in mediating object-spatial configuration recognition. Neurobiol Learn Mem. 2015;118:42-48.
26. Voss MW, Vivar C, Kramer AF, van Praag H. Bridging animal and human models of exercise-induced brain plasticity. Trends Cogn Sci (Regul Ed ). 2013;17(10):525-544.
27. Clelland CD, Choi M, Romberg C, et al. A functional role for adult hippocampal neurogenesis in spatial pattern separation. Science. 2009;325(5937):210-213.
28. Vivar C, Potter MC, Choi J, et al. Monosynaptic inputs to new neurons in the dentate gyrus.
Nature communications. 2012;3:1107.
29. Nakashiba T, Cushman JD, Pelkey KA, et al. Young dentate granule cells mediate pattern separation, whereas old granule cells facilitate pattern completion. Cell. 2012;149(1):188-201.
30. Bolz L, Heigele S, Bischofberger J. Running improves pattern separation during novel object recognition. Brain Plasticity. 2015;1(1):129-141.
31. Leutgeb JK, Leutgeb S, Treves A, et al. Progressive transformation of hippocampal neuronal representations in “morphed” environments. Neuron. 2005;48(2):345-358.
32. Knierim JJ, Neunuebel JP. Tracking the flow of hippocampal computation: Pattern separation, pattern completion, and attractor dynamics. Neurobiol Learn Mem. 2016;129:38-49.
33. Neunuebel JP, Knierim JJ. Spatial firing correlates of physiologically distinct cell types of the rat dentate gyrus. J Neurosci. 2012;32(11):3848-3858.
34. McHugh TJ, Jones MW, Quinn JJ, et al. Dentate gyrus NMDA receptors mediate rapid pattern separation in the hippocampal network. Science. 2007;317(5834):94-99.
35. Ho JW, Poeta DL, Jacobson TK, et al. Bidirectional modulation of recognition memory. J Neurosci. 2015;35(39):13323-13335.
36. Brown MW, Aggleton JP. Recognition memory: What are the roles of the perirhinal cortex and hippocampus? Nature Reviews Neuroscience. 2001;2(1):51-61.
37. Murray EA, Richmond BJ. Role of perirhinal cortex in object perception, memory, and associations. Curr Opin Neurobiol. 2001;11(2):188-193.
38. Keene CS, Bladon J, McKenzie S, Liu CD, O'Keefe J, Eichenbaum H. Complementary functional organization of neuronal activity patterns in the perirhinal, lateral entorhinal, and medial entorhinal cortices. J Neurosci. 2016;36(13):3660-3675.
39. Burke SN, Maurer AP, Hartzell AL, et al. Representation of three-dimensional objects by the rat perirhinal cortex. Hippocampus. 2012;22(10):2032-2044.
40. Kent B, Hvoslef-Eide M, Saksida L, Bussey T. The representational–hierarchical view of pattern separation: Not just hippocampus, not just space, not just memory? Neurobiol Learn Mem. 2016;129:99-106.
41. Lehky SR, Tanaka K. Neural representation for object recognition in inferotemporal cortex. Curr
Opin Neurobiol. 2016;37:23-35.
42. Burke SN, Maurer AP, Nematollahi S, Uprety A, Wallace JL, Barnes CA. Advanced age dissociates dual functions of the perirhinal cortex. J Neurosci. 2014;34(2):467-480.
43. Burke SN, Hartzell AL, Lister JP, Hoang LT, Barnes CA. Layer V perirhinal cortical ensemble activity during object exploration: A comparison between young and aged rats. Hippocampus. 2012;22(10):2080-2093.
44. Ennaceur A, Neave N, Aggleton JP. Spontaneous object recognition and object location memory in rats: The effects of lesions in the cingulate cortices, the medial prefrontal cortex, the cingulum bundle and the fornix. Experimental Brain Research. 1997;113(3):509-519.
45. Ranganath C, Ritchey M. Two cortical systems for memory-guided behaviour. Nature Reviews
Neuroscience. 2012;13(10):713-726.
46. Hunsaker MR, Chen V, Tran GT, Kesner RP. The medial and lateral entorhinal cortex both contribute to contextual and item recognition memory: A test of the binding ofitems and context model. Hippocampus. 2013;23(5):380-391.
47. McKenzie S, Keene CS, Farovik A, et al. Representation of memories in the cortical–hippocampal system: Results from the application of population similarity analyses. Neurobiol Learn Mem.
