Anatomy and Function of the Primate Entorhinal Cortex

Abstract

The entorhinal cortex (EC) is a critical element of the hippocampal formation located within the medial temporal lobe (MTL) in primates. EC has received attention historically for being the primary mediator of cortical information going into and coming from the hippocampus proper. In turn, the hippocampus takes this information and manipulates it in the service of learning and memory. In this review, we highlight the significance of the EC as a major player in memory processing along with other associated structures in the primate MTL. The complex, convergent topographies of cortical and subcortical input to EC, combined with short-range intrinsic connectivity and the selective targeting of EC efferents to hippocampus, provide evidence for sub-regional specialization and integration of information beyond what would be expected if this structure was a simple conduit of information for the hippocampus. Lesion studies of this region provide evidence implicating this region as critical for the memory and flexible use of complex relational associations between experienced events. The physiology of this structure’s constituent principal cells mirrors the complexity of its anatomy. EC neurons respond preferentially to features of memory-dependent paradigms such as object, place, and time. EC neurons also show striking spatial representations of visual space, similar to those identified in rodents navigating physical space. In this review, we highlight the great strides that have made towards furthering our understanding of the primate EC, and we identify paths forward for future experiments to provide additional insight into the role of this structure in learning and memory.

1. INTRODUCTION

The entorhinal cortex (EC; Brodmann’s area 28) is a critical element of the hippocampal formation and receives a wide array of converging inputs from various poly-modal sensory processing streams, as well as subcortical arousal areas (1). In turn, the EC itself is the major source of cortical input into the hippocampus proper. The first reported investigations into the EC came from dissections in rodents performed by Santiago Ramon y Cajal in the early 20^th century. His findings highlighted a portion of rat ventral temporal cortex with substantial reciprocal connections with the hippocampus. Ramon y Cajal (2) posited that the rich inter-connectivity strongly implicated EC as necessary for proper hippocampal function in learning, memory, and navigation. This hypothesis inspired the current dogma of EC as a gatekeeper of cortical communication with the hippocampus and has shaped the ways in which this structure has been studied (3). As such, it will come as no surprise that the bulk of contemporary work on EC function has focused on elucidating how the physiology of this structure facilitates hippocampal processing. This field has been driven predominantly by research in rodents performing spatial navigation paradigms (4, 5). Through this work, medial (MEC) and lateral (LEC) subdivisions of the rodent EC have been identified that are distinguishable both by their anatomy and function (6). MEC projects to medial aspects of the dentate gyrus (7), and its cells, including grid cells, are largely driven by allocentric spatial reference frames (4, 8, 9). By contrast, LEC connectivity is biased towards more lateral aspects of DG (7) and its neurons have been reported to elicit temporally-modulated firing patterns (10), as well as responses to specific objects (11, 12).

While these findings in the rodent EC have had a significant impact on the trajectory of primate research, several incongruences arise when directly comparing this structure across species. Anatomically, EC takes a much more rostral position in the primate temporal lobe than in the rodent and sits largely anterior to the hippocampus. Additionally, whereas EC in rodents is more flat and ellipsoid, this structure adopts a C-like shape in primates (Figure 1A; 13). Primate EC contains more pervasive connectivity patterns with visual processing areas compared to those in the rodent (14, 15), a fact which is likely explained by the reliance of primates on vision as their primary exploratory modality, compared to olfaction in rodents (16). Given the cognitive deficits associated with Alzheimer’s-derived atrophy of EC — one of the initial neuropathologies in humans with this disease (17) — it is imperative that we continue to investigate the relationship between this structure’s physiology and higher-order cognitive processes, including episodic memory. In this review, we will synthesize the growing field of primate EC research aimed at doing just that. We will examine how recent research has both elaborated on findings from rodent studies while identifying key differences between species. First, we will review the anatomy of primate EC, highlighting the privileged position EC serves as a conduit to the hippocampus and as a remarkably efficient site of information convergence. We will discuss the diverse network of communications within the hippocampal formation, including interactions between the EC and hippocampus as well as the robust intrinsic connectivity within the cortical layers of the EC. Finally, we will discuss how these anatomical findings have motivated ideas regarding EC function, and we will review the lesion and neurophysiological studies that have examined various functional hypotheses. Clearly, the field has made significant headway in uncovering the function of the EC; however, the emergence of new tools for neurophysiological recordings and targeted manipulations have the potential to dramatically increase our understanding. We hope that this review will inspire and motivate further studies of this structure in the primate.

FIGURE 1.

A. Illustration showing the variation of anatomical placement of EC between rodents and primates (adapted from Strange et al. 2014). B. Photomicrograph of the ventral surface of the temporal lobe and flattened illustration of EC sub-fields as defined by Amaral et al. (1987). Note the change in Layer II and III cell density across the different sub-regions (adapted from Amaral et al. 1987). C. Schematic of Major EC connectivity (Adapted from Insausti et al. 1987a).

2. WHAT IS THE ENTORHINAL CORTEX?

2.1. Etymology and Origin

In the primate, EC is situated in the anterior medial temporal lobe. It begins ventral to the amygdala and anterior to hippocampus. From there, it extends caudally underneath hippocampus for? approximately 5mm. The earliest usage of the term “entorhinal cortex” was given by Brodmann after his dissections of human tissue, during which he identified a cortical field extending medially from the rhinal sulcus (Brodmann 28) with distinct cytoarchitecture from the tissue surrounding the sulcus (perirhinal cortex, Brodmann 35) and that which extended laterally from it (ectorhinal/inferior temporal cortex) (1, 3)

2.2. Cytoarchitecture

Despite being recognized because of its unique cytoarchitecture, anatomists have been historically in disagreement over EC’s cytoarchitectonic composition. Ramon y Cajal (2) proposed a seven-layer architecture in his initial dissections, although follow-up investigations by his student, Lorente di No, reduced that number to six (3). After many subsequent years of disagreement, it is now broadly agreed that EC is in fact a six-layered structure in rodents (6) and primates (3, 18, 19). However, as we will discuss in sections three and four, the promiscuous connectivity of layers I-V has drawn the most functional intrigue. The proportionate layer sizes and constituent cell densities change across the various subdivisions of primate EC; however, the cell morphologies (or lack thereof) which constitute each layer remain consistent. Layer I is devoid of cell bodies and expectedly houses axially directed axons. Layer II is home to cells that are morphologically modified pyramidal cells typically referred to as stellate cells. Layer III is the canonical pyramidal cell layer often found in cortex, and, like layer II, it also changes in thickness and density across the various EC sub-regions. Layer IV is similar to layer I in that it is devoid of cell bodies, and accordingly, this area is typically labeled lamina dissecans. Like layer I, layer IV is a fiber-heavy zone. The next two layers perhaps explain the degree of disagreement with the overall layer organization of EC. Layer V, another thick pyramidal cell layer similar to layer III, is interesting in that its constituent cells undergo a change in density from the most superficial/ventral aspects to deep. Because of this cell density gradient, this layer is often sub-divided into sub-laminae Va, Vb, and Vc. Layer VI—perhaps the least studied portion of EC, houses variously sized neurons in bands of differing density. In contrast to the rest of EC, the sub-bands of this layer are interesting in that they take a much more curved shape compared to other layers (3).

2.3. Entorhinal Sub-Regions

The canonical sub-division of rodent EC into MEC and LEC has its basis in several facets of anatomy and physiology. The cytoarchitecture of LEC is consistent with the majority of cortex and has smooth, continuous cell layers, which is a stark contrast from the patchy, interrupted cell layers seen in MEC (6). Alongside this, differing patterns of connectivity (7) and unique neurophysiological responses to behavioral phenomena serve to distinguish these sub-regions (4, 8–10, 12, 20, 21). By contrast, our ability to sub-divide primate EC has also been steeped in a history of contradictions (3). In Brodmann’s initial dissections of the EC of various species, he opted not to subdivide the EC of primates despite delineating medial and lateral aspects in the brains of bats, hedgehogs, and other non-primate species (1). Subsequently, others have divided the primate EC into as many as 14 sub-areas in monkeys and 23 in humans based on cytoarchitectonics alone (3). Like the disagreements over cytoarchitecture, a subtle agreement has been reached within the field since the era of these initial studies. In rhesus monkeys, pivotal work by Van Hoesen and Pandya (18, 19) highlighted three predominant sub-fields—a caudal (28a), intermediate (28i), and rostral (28b) area. One of the main findings to support this separation was the changing cytoarchitecture between caudal EC and its more rostral elements. Specifically, in rostral area 28b, layer IV is virtually non-existent, and the stellate cells in layer II form small, discontiguous clusters of islands. In 28i, the layer II islets become more prominent and consistent moving caudally in conjunction with layer IV becoming increasingly visible. Finally, layer II is a continuous band of cells and the cell-less layer IV is significantly more prominent in caudal area 28a. These findings were largely corroborated by Amaral and colleagues (3) in cynomologus macaques; however, Amaral and colleagues also observed strong departures from the rhesus brain that led them to identify seven distinct EC sub-regions (Figure 1B). The most rostral element, olfactory EC (E_O), is the only sub-region delineated by its unique inputs. As its name suggests, this region receives strong, direct terminations from the olfactory bulb. Just behind E_O lies the region labeled rostral EC (E_R). E_R bares many similarities with 28b in the rhesus. It lacks the cell-less layer IV and contains patchy islets of cells in its superficial layers. Intermediate EC (E_I) sits directly behind E_R and serves as the transitionary bound between the interrupted cell islands in the more anterior portions of EC and the continuous layers in more posterior elements. Similar to rodents, a lateral portion of EC was identified as being cytoarchitectonically distinct, however, in primates, LEC is divided into two putative subdivisions, rostral LEC (E_LR) and caudal LEC (E_LC). These areas run along the medial edge of the rhinal sulcus, with E_LR sitting lateral to E_R and E_LC lateral to E_I. Layers II and III house thick islands of cells in both regions, with E_LC containing more dense clusters. The cell-less layer I increases in size in E_LR relative to E_LC, although there is no visible layer IV in either area. Interestingly, a rhesus homologue of these regions was identified by Van Hoesen and Pandya as independent prorhinal cortex (18, 19). Yet, Amaral opted to not label this an independent brain area because its projection patterns were similar to other subregions within the EC of cynomologous macaques. Caudal EC was divided into two fields, E_C directly behind E_I and the “caudal-limiting field” (E_CL) behind E_C. In both regions, the islands of layer II cells become thicker and more robust moving caudally, such that in the most posterior elements of E_C and all of E_CL, the layer looks like a uniform band of neurons (3).

3. NON-HIPPOCAMPAL ENTORHINAL CONNECTIVITY

3.1. Sub-Cortical Connectivity

The EC receives substantial input from sub-cortical areas, including the amygdala, claustrum, striatum, basal forebrain, and others (22). Interestingly enough, with rare exception, there is no strong evidence of topographic bias of subcortical projections onto EC. Additionally, subcortical sites appear to project only ipsilaterally to EC as opposed to bilaterally (22). However, the diversity of these diffuse sub-cortical inputs combined with the immense topography of cortical termination patterns onto EC (see Section 3.2) amplifies the high degree of specialized information this structure can convey to downstream hippocampus. The number of sub-cortical sites which project to EC is tremendous, however here we will highlight a few key areas that specifically communicate with the primate EC in the service of learning and memory. In rodent, the medial septal nucleus (MS) and diagonal band of Broca (DBB), components of the basal forebrain, are considered among the most critical cholinergic and GABA-ergic inputs into both the EC and hippocampus because of their strong influence on the rhythmic mesoscopic properties of neuronal activity in these regions (23–25). In monkey, there exists strong projections between the MS/DBB and EC. However, in contrast to findings in rodents suggesting a strong topography of connectivity (26), the primate MS/DBB efferents to EC appear to be considerably more diffuse across the structure (22). In addition, strong cholinergic projections (27) from the nucleus basalis of Meynert terminate onto the primate EC (22, 28, 29). Projections from this region are virtually non-existent in the rodent (26), suggesting an additional excitatory drive into EC that is primate-specific. This largely unexplored interaction between nucleus basalis and EC in primates may explain some of the observed differences in the neurophysiology of this region across species (see section 5.3). Surprisingly, the largest sub-cortical projections to EC come not from the basal forebrain nuclei, but rather by way of the amygdala, which has been implicated in numerous behavioral phenomena including fear conditioning, reward processing, and memory-associated processes (30–32). Most of the projections from amygdala to EC originate in the lateral nucleus of the amygdala, with sparser projections arising from the accessory basal nucleus and periamygdaloid complex (22, 33). In contrast to other subcortical termination patterns, these projections are biased towards rostral elements of EC (22), with the heaviest sub-regions innervated by amygdala being E_o, E_r, and E_LR. In these regions, retrograde tracing experiments highlighted robust terminations specifically in layers I, III, and V, whereas amygdalar input is restricted to layer I in more intermediate and caudal EC regions (22).

With respect to inputs from diencephalic structures, the primate EC receives considerable input from both the thalamus and hypothalamus (22, 28). Thalamic input to EC is dominated mostly by the central lateral nucleus, which has been implicated in rodent studies to be responsive to noxious, visceral stimulation (34). Retrograde tracing also suggests noticeable EC terminations by neurons originating from the paraventricular, parataenial, and pulvinar nuclei (22, 28). Although efferents from the nucleus reuniens are also observed consistently in labeled EC neurons, the density of this projection appears sparser to that in rodents (22, 35). In hypothalamus, substantial connectivity exists between the supramammillary nucleus and EC (22, 28). The supramammilary area is another region that contributes to rhythmic synchrony in the neurophysiology of rodent EC, both through its direct influences on the structure as well as through interactions with the MS/DBB (36). Its promiscuous connectivity in the primate EC suggests an analogous synchronizing function with the hippocampal formation of this species. Several neuromodulatory systems in the brainstem also innervate the EC. Considerable projections from the dorsal raphe nucleus and ventral tegmental area have been observed through retrograde tracing (22). In the most anterior EC sub-region, E_O, additional retrograde labeling can be seen in the substantia nigra and parabrachial nucleus, among others, suggesting a unique interaction between reward-related and visceral brainstem areas with the most rostral element of EC.

3.2. Sensory Afferents

EC serves as the main sensory relay of cortical input into the hippocampus, and the strongest of these projections is sent directly from the olfactory bulb to the most rostral element of EC, E_O (14, 37). As previously mentioned, this is the only entorhinal sub-area defined on the basis of its cortical inputs, and no other unimodal sensory area has been shown to project directly to EC in primates. Retrograde tracing studies have definitively shown that neither auditory cortices nor high-level visual cortices such as area TE in inferior temporal cortex project to any entorhinal subfield (14).

As may be expected then, the majority of sensory input into the primate entorhinal cortex comes from polymodal association cortices. Two of the heaviest projections into EC come from perirhinal cortex (area 35/36) and parahippocampal cortex (TF/TH) (14, 19). While these areas have been implicated in the processing of somatosensory and auditory stimuli, they receive predominantly visual input (38–40). Activity in the perirhinal cortex has been associated with the processing of complex visual objects (41–44), and, in parahippocampal cortex, this culmination of sensory processing is geared towards spatial information processing (45). In rodents, homologous regions project to distinct parts of EC and form discrete functional domains by way of parahippocampal-MEC and perirhinal-LEC interactions (46). In primates, the topography is not as distinct, although still existent. Retrograde tracing experiments showed that both perirhinal and parahippocampal cortices project to all sub-regions of the EC apart from the most rostral E_O and lateral elements E_LR and E_LC (14). While this suggests a more diffuse projection pattern of these regions onto primate EC, follow-up experiments have shown that perirhinal cortex projects more strongly to rostro-lateral elements of EC, whereas parahippocampal cortices are biased towards more caudo-medial portions (Figure 1C; 47, 48). Thus, rodent and primate EC appear similar in that distinct portions of EC are innervated differentially by the perirhinal and parahippocampal cortices, although in primates there appears to be more intermixing of information.

The dorsal polymodal areas of the superior temporal gyrus also send projections to EC, and tracing studies implicate a bias towards more caudal aspects (14, 48). Similar to the collective functions of perirhinal and parahippocampal cortices, these inputs from the superior temporal sulcus provide an additional avenue by which the caudal EC receives highly processed visual information for object recognition and biological motion processing (49, 50). Rostral EC receives unique polysensory input from the insula, and anterior portions of this region (the agranular insula) innervate the anterior EC sub-regions E_R, E_LR, and anterior E_I (14, 48). These inputs have been reported to carry audio-visual information, although the insula has been implicated in widespread sensory and cognitive functioning (51). For all sensory areas, the strongest sites of laminar termination were located in layers I-III (Figure 1C; 48). Thus, with the exception of olfactory area E_o, the data suggest that all areas of EC receive largely visually-modulated information about putative objects and relevant motion information from neocortical sites.

3.3. Frontal Afferents

One of the largest inputs from the frontal cortices in primates to EC originates in the orbitofrontal cortex (OFC, Area 13/13a) (14, 19, 48, 52, 53). These inputs are biased towards the more rostral elements of EC, such as E_O, E_R, and E_I, and these projections have been shown to most strongly emanate from caudal elements of Areas 13/13a in retrograde tracing experiments. Sparser projections have also been observed in more rostral orbitofrontal areas 11 and 12, ventromedial area 14, and dorsolateral frontal cortices (14, 48). Cingulate cortices such as infralimbic area 25 and retrosplenial cortex provide similarly robust input to the more caudal elements of EC such as posterior E_I, E_C, and E_CL (14, 48). These frontal projections are similar to inputs from sensory areas (section 3.2) in that they terminate preferentially onto layers I-III, although in the case of retrosplenial cortex and some of the caudo-medial portions of orbitofrontal cortex, the laminar distribution of projections included deep layers V and VI (48, 52). Orbitofrontal areas have long been associated with behaviors such as value-based decision making and reward-prediction errors (54), whereas cingulate cortices have been more associated with emotion-based and contextual learning (55). The parcellation of these inputs onto distinct rostro-caudal sites suggests unique functional specialization of frontal cortex decision-making information along the longitudinal axis of EC.

4. ENTORHINAL CONNECTIVITY WITHIN THE HIPPOCAMPAL FORMATION

As mentioned above, EC is the main source of cortical input to the hippocampus. In all species, the perforant pathway provides the strongest connection between these two regions (53). In macaque brains, layers II and III of EC comprise the majority of perforant path projections, with deeper layers V and VI providing only sparse efferents (18, 53, 56). Layers II and VI communicate specifically to DG and CA3, whereas Layers III and V projections are more biased towards CA1 and the subiculum (56). In turn, feedback from CA1 and the subiculum in the hippocampus terminates predominantly onto the deep layers V and VI, with very sparse projections onto layer III (Figure 1C; 56).

The various connections between hippocampus and EC are not homogenously distributed and actually span three distinct topographies. First, cells located laterally in layers II and III of EC project to more rostral aspects of hippocampus, while more medially situated cells terminate caudally (56, 57). In addition, two distinct topographies have been identified along the longitudinal, rostral-caudal axis of EC. Efferent projections to DG and CA3, which arise largely from layer II, are oriented such that rostral connections from EC synapse onto more superficial aspects of the dendrites of the principal cells within each area. Specifically, rostral projections terminate onto the dendrites of granule cells in superficial aspects of the molecular layer of the DG compared to more caudal projections, and these projections also terminate onto pyramidal cell dendrites more distally in the strata lacunosum moleculare within CA3 (18, 56). Projections to CA1 from layer III of EC also have a rostrocaudal termination pattern such that rostral efferents target distal CA1 near its border with CA3 and proximal subiculum, with increasingly caudal projections synapsing onto more proximal CA1 and distal subiculum (47). In contrast, the recurrent projections from hippocampus to the deep layers of EC appear more homogenous in their distribution (56).

Alongside its myriad projections to the hippocampus, the diverse intrinsic connectivity of EC adds an additional level of complexity when interpreting the role of this structure within the hippocampal formation. As in rodents, the EC of primates can be distinguished as having at least three discrete bands of interconnectivity (58). Rostro-medial layer V and VI neurons synapse onto caudo-medial layer II and III cells (and vice versa for caudal-medial layer V and VI cells), and similar recurrent projection patterns are observed for intermediate and lateral aspects of EC. However, unlike rodents, the bands do not span the entire length of the primate EC. Instead, they are limited to between one-third to half the length of EC (~3–5mm), with rostral bands being shorter and less dense than more caudal elements (58). Interestingly, the deep-to-superficial orientation of intrinsic EC connectivity has been identified to be largely excitatory and asymmetric, with 56% of the projections from layer V and VI being glutamatergic and synapsing onto stellate and pyramidal cells, respectively (58). The remaining 44% synapse onto the dendritic shafts of putative interneurons, however, suggesting parallel mechanisms of gain modulation between deep and superficial layers of EC. Only 5% of total identified projections in superficial layers were identified as recurrent in monkey EC. Additionally, layer II cells of lateral EC (E_LC and E_LR) send projections to layer II of all other sub-regions of the structure (58).

5. FUNCTION OF ENTORHINAL CORTEX

5.1. Functional Implications from Anatomy

Considering the high degree of anatomical selectivity of cortical input to EC, EC projections to hippocampus, and within the intrinsic connectivity of EC, it is reasonable to assume that there are distinct processing domains within EC that enable it to send unique information to downstream hippocampus. For example, given the topography of projections from the perirhinal and parahippocampal cortices to EC, we might expect that the rostral and caudal portions of this structure process information about objects and space, respectively (14, 48). Because rostral projections from EC terminate on proximal CA1, the object-associated information carried through EC from perirhinal cortex exclusively influences CA1 near its border with CA3 (56). Similarly, spatial information carried by parahippocampal cortex is sent to distal CA1 closer to the subiculum (56). Thus, it would seem that there are two parallel information streams processed by primate EC similar to what has been observed in rodents—one centered around object processing and another with more spatial representations. However, in contrast to the rodent EC, the intrinsic circuitry across different rostro-caudal elements of monkey EC is discontinuous throughout the length of the structure (58). This suggests more distinct functional specialization within the primate EC by virtue of the fact that there is virtually no feedback from portions of EC that receive input from one of these cortices onto those that communicate with the other.

Similar hypotheses can be posited regarding the top-down influences of the frontal cortices. Both the OFC and cingulate cortex are associated with decision-making. However, whereas OFC has been more strongly implicated in value-based decision-making (54), the cingulate cortex is more associated with contextual and emotional learning (55). With respect to EC connectivity, as described in section 3.3, OFC synapses primarily onto rostral elements of EC, and cingulate cortex terminates onto caudal sites. Again, because of limited EC feedback between different sections along its longitudinal extent (58), this anatomical evidence suggests functionally distinct domains by virtue of their different frontal influences. Because of the difficulty in accurately targeting putative EC sub-areas, it has been difficult to test these hypothesized functional distinctions. As electrophysiological recordings and elegant manipulation paradigms continue to evolve, it will be important to assess how the diverse connectivity into and within EC facilitates the specificity of the information it sends to hippocampus in the service of learning and memory.

5.2. Lesion Results

Over the past two decades, lesion studies have directly examined the role of EC in various visually-guided memory-dependent behaviors. The most ubiquitous of these paradigms are the delayed match-to-sample task (DMS) and its non-match variant (DNMS), in which subjects are rewarded for choosing a stimulus that either matches or is distinct from an initially viewed sample after a delay period. Monkeys are remarkably skilled at performing these behaviors and achieve scores significantly above chance levels, even when the delays between the sample and choice phase are extended to beyond ten minutes (59–62). Interestingly, unlike recognition memory deficits observed in animals with lesions of the hippocampus alone (62–64) or lesions that include perirhinal and parahippocampal cortex (41, 42, 44, 65) animals with EC lesions show only transient impairment, with normal performance on tests of recognition memory observed after ~1 year post-lesion. (59). Leonard and colleagues (59) showed through histological analyses that perirhinal cortex projections to CA1 become significantly more robust in animals with EC lesions; providing a potential mechanism by which plasticity can induce the observed behavioral recovery. Perirhinal involvement in object recognition observed through permanent lesions (41, 42, 44, 65) has been corroborated by recent work using optogenetic techniques in which neurons in this region were manipulated to bias recognition of a series of objects (66). These transient and highly local manipulations were effective in biasing animals’ perception of novel objects as familiar and familiar objects as novel when neurons were either activated or inhibited, respectively. Whether acute manipulations of EC produce a deficit in object recognition remains to be tested.

Intriguingly, EC lesions produce deficits on tasks that require the flexible manipulation of learned associations (61), suggesting a role for this structure in supporting the relational organization of memory. Buckmaster and colleagues (61) examined both monkeys with EC lesions and intact monkeys in a series of memory-related paradigms that probed the animals’ ability to form associational relationships between objects and flexibly access those relationships. The first of these was the paired-associates (PA) task, in which subjects are rewarded for choosing the correct object from two choices, based on a centrally-located visual cue (Figure 2). The correct choice item can then serve as a sample for a proceeding pair of objects, one of which is again rewarded by association with the cue. Monkeys with EC lesions were just as capable of learning these two premise pairs as unoperated controls, although they did take slightly longer to reach the established acquisition criterion. When presented with another pair of problems, the initial training in controls facilitated much more rapid acquisition. In stark contrast, monkeys with EC lesions were just as slow at learning the associations in these new problems as during initial training. Conducting the PA task with overlapping stimuli allowed for a critical inference test in which the sample from the very first trial was presented with the choice stimuli from the second trial. Because these objects have never been shown together, the monkey must draw upon the overlapping relationships between the experienced objects to solve this inherently novel problem. During these probe trials, monkeys with EC lesions were incapable of linking the sample to the indirectly associated choice object. By contrast, control monkeys were able to flexibly use their previously acquired associations to correctly solve the probe trials. Similar results were observed in the transitive inference (TI) paradigm. In TI, overlapping pairs of objects form a transitively associated hierarchy (i.e. A>B>C>D>E). If EC is critical for either creating the overall reward hierarchy or flexibly manipulating the relationships that constitute it, then monkeys with EC lesions should fail at an inference test where discontinuous objects in the hierarchy are presented together (i.e. B vs. D). This is exactly what was observed in the study by Buckmaster et al. (61). Although monkeys with EC lesions were not impaired at learning the initially presented pairs, they were incapable of successfully employing the learned associations between items to discriminate between non-adjacent items. By contrast, when presented with the first and last objects of the hierarchy (the only items consistently rewarded and not rewarded, respectively), monkeys with EC lesions performed normally, suggesting that they were able to leverage the consistent reward history of the objects to solve the problem. Similar findings have been observed in monkeys with combined lesions of the hippocampus and fornix, although the performance impairment observed in these animals might be explained by the unintended damage of subcortical-EC projections running through the fornix (67). Given this apparent role of EC in flexibly linking experiences with objects, one might wonder if this structure plays a role in associating objects with other physical features such as spatial location. Buckmaster and colleagues tested this potential function using a serial-delayed recognition span (SDRS) task, which rewarded monkeys for choosing one item from an identical set on a multi-site board when that item was in a novel spatial location. Results showed that monkeys with EC lesions were incapable of holding even one spatial location in memory. Taken together, lesion studies suggest that although EC is not critical for remembering putative objects or even simple associations between experienced objects, it is necessary for employing existing relational frameworks between objects to navigate novel experiences.

FIGURE 2.

A. Paired Associates (PA) task schematic. Monkeys are trained to select one stimulus in the presence of a centrally-located cue. In the following trial, the previously chosen stimulus becomes the sample cue for a novel pair of choice stimuli. The critical inference test utilizes the initially cued stimulus with the latter pair of choice stimuli, and animals are rewarded for using the indirect relationship between objects to infer the rewarded associations. B. Performance scores for the PA task show that monkeys with EC lesions (E; black bars) perform just as well as unoperated controls (C; white bars) at the initial premise pair training. However, monkeys with EC lesions are significantly impaired at discerning the correct paired stimulus during inference probes (adapted from Buckmaster et al. 2004).

5.3. Physiological Findings

One of the neurophysiological hallmarks of the rodent hippocampal formation is high amplitude, rhythmic, neuronal activity that occurs in its constituent structures as animals actively explore an environment. This activity is typically present as a consistent, theta-band (6–12Hz) oscillation in recorded local field potentials (LFP) as animals translocate (25, 68), and it is strongly correlated with olfactory exploratory behaviors such as sniffing and whisking (69). By contrast, rhythmic activity in the theta band observed in the monkey hippocampal formation, including the EC, is not observed as a sustained oscillation but instead occurs in interrupted bouts (70–76). These discontinuous bouts of theta are also observed in the hippocampus proper of humans (77–79) and in both the hippocampus and EC of bats (80–83). Additionally, whereas non-REM sleep in rodents is characterized by a lack of sustained theta activity in MTL structures (68, 84), evidence suggests that structures within the hippocampal formation in primates, including EC, paradoxically elicit stronger and longer bouts of theta-band activity during transitions from awake behavior into sleep (75). Future work will be critical to advance our understanding of the synchronized network states that organize neuronal activity in the primate hippocampus and EC.

In addition to research focused on the mesoscopic physiology of this region, analysis of the responses of EC neurons during memory-dependent paradigms has begun to elucidate how the distinct neurophysiology of EC facilitates its function as proposed from anatomical (sections 3, 4, 5.1) and lesion investigations (section 5.2). EC receives input from neurons in the most anterior portions of the ventral visual stream, which do not themselves respond to simple stimulus features (85). It is therefore not surprising that EC neurons do not show selective responses to particular features of object stimuli, such as color or contour (86). Instead, single-unit recordings in the EC have identified memory-related responses to visual stimuli. For example, EC neurons show stimulus-specific increases in firing rate in response to the presentation of the sample stimulus in a DMS task, (86, 87). If EC were simply relaying object identity to downstream hippocampus, then one would expect the firing rates of sample-selective neurons to be similar between repeated presentations of the same stimulus. However, during match trials, EC neurons demonstrated modulations in firing rates, showing both increases and decreases (Figure 3A-B; 86–88), and these match enhancement and suppression effects on firing rate are observed before the animal makes its response during the test phase. Similar patterns of activity have been observed in the hippocampus (76), perirhinal cortex (89), and prefrontal cortex (90) and suggest that EC is potentially tuning its responses to facilitate memory-based decision making. Alternatively, given the dopaminergic inputs from the VTA mentioned in section 3.1, stimulus novelty may drive the responses in this structure. In fact, when unique non-matching stimuli are presented between the sample and matching test item, EC neurons elicit elevated firing (86). Interestingly, during inter-stimulus delays, EC neurons show sustained firing across numerous delay lengths, and in the majority of cases these responses are sample-selective, suggesting a possible working memory influence on these neurons (Figure 3C; 86). These delay-sustained responses stand in contrast to responses of upstream perirhinal cortex in which any delay activity was eliminated upon presentation of intervening stimuli (85). Although lesion studies suggest only a transient role of EC in simple object recognition (see section 5.2), these memory-related neural responses may underlie the role of EC in the flexible manipulation of learned associations (61).

FIGURE 3.

A. Average firing-rate histogram (left) and time-course (right) of stimulus-selective cells in the monkey EC that show match suppression during DMS with various intervening stimuli. B. Average firing rate histogram (left) and time-course (right) of cells that exhibit match enhancement during DMS under the same conditions. C. Firing-rate histogram of EC neuron with significant delay period activity. As memory load increases with intervening irrelevant stimuli, the firing rate of this neuron increases (adapted from Suzuki et al. 1997).

The most prominent neural response that has been identified in the rodent hippocampal formation relates to allocentric spatial representations (4, 8, 20, 91), and one of the major hypotheses of the function of the hippocampal formation is that of a cognitive map, which represents not just physical space but also relational object space (92–95). During a spatial task in which monkeys are rewarded for responding to a stimulus located in the same position as it was during a sample phase (delayed match to place), a proportion of EC neurons elicited biased responses for particular locations (86). Recent studies have also explored the extent to which EC neurons show place-specific firing relative to visual exploration. In rodents, exploration in physical space elicits a range of neuronal responses in EC including firing in symmetric, triangular patterns (grid cells) (4, 8, 20), firing in relation to experienced physical boundaries (border cells) (9), and firing relative to an animal’s orientation within an environment (head direction cells) (8). Work from our own lab in head-fixed animals viewing images on a computer screen shows that some cells in primate EC elicit spatial responses based on locations of gaze including grid-like firing fields and responses that are selective to fixations made at the borders of visual stimuli (Figure 4A-C; 72). In addition, and potentially analogous to rodent head-direction cells, a population of primate EC neurons were identified that were tuned to fire when the animal made saccades in a particular direction (Figure 4D; 96). Grid cells in the monkey EC display a gradient of grid field size and spacing such that both parameters increase in more medial recording sites. These findings are consistent with the gradient of changing grid field structure along the longitudinal extent of EC that has been observed in both rodents (4) and bats (80). In more recent work, a large proportion of non-grid but spatially-responsive EC neurons have been identified during visual exploration (97). The firing fields of these neurons are spatially reliable but more amorphous compared to grid cells. These findings are consistent with a recently described population of rodent EC cells (98). Recordings in the monkey EC identified multiple frames of reference among the population of non-grid spatial cells (97). When monkeys viewed an image that shifted to multiple positions on a screen, the firing patterns of one population of EC neurons remained the same relative to the image regardless of its position within the screen. By contrast, a second population of EC neurons elicited egocentric firing preferences. As images shifted on the screen, these neurons stably represented where on the screen the animal was looking as opposed to boundaries established by the visual image. Finally, spatial responses in the monkey EC were also identified relative to the location of attention, even in the absence of any movement (99). In this study, monkeys were trained to maintain fixation on a small cross presented at the center of the screen and covertly attend to a small dot moving in the periphery. Monkeys were rewarded for releasing a bar in response to a subtle luminance change in the dot. The luminance change was titrated to ensure that selective attention to the cue was required for successful performance. The responses of EC neurons were examined relative to the location of the monkey’s attention on successful trials, and a significant proportion of EC neurons demonstrated grid-like responses. These findings, taken together with recent similar findings in humans (100, 101), suggest that primate EC neurons may have the ability for regular tiling of experience, independent of specific sensory or environmental dimensions.

FIGURE 4.

A. Example image presented on a computer screen during head-fixed visual exploration with an example visual scan path overlaid (yellow trace). B. Combined eye trace trajectories (gray traces) with spikes from a putative EC grid cell overlaid (red dots). Note the regular interval of spike clusters that is also apparent in the adjacent firing rate map of this cell and exemplified best by the spatial autocorrelation map. C. Firing rate map (left) and autocorrelation map of a border cell isolated during the same visual exploration paradigm (adapted from Killian et al. 2012). D. Example polar histograms of EC neurons that elicited strong firing rate modulations for saccades in a particular direction made immediately before (green trace) or after (blue trace) spikes were collected (adapted from Killian et al. 2015).

In addition to responding to particular places and things during an experienced event, one might expect that a structure implicated in episodic memory would also process information about the temporal order of those experiences, and several forms of timing-related responses have been observed in the primate EC. When monkeys are rewarded for remembering the order of a list of individually presented objects, EC neurons demonstrate selective responses for objects in a particular position within the presented list (Figure 5A-B; 102, 103). In contrast to neurons recorded from ventral prefrontal cortex (103), EC neurons also elicit conjunctive object-list position specificity. Recent work has demonstrated timing responses in the monkey EC as monkeys perform a free-viewing task (104). Many neurons in the EC responded to image onset, showing large changes relative to baseline shortly after image onset. Interestingly, these neurons showed a wide variety of rates with which they relaxed back to baseline, extending from hundreds of milliseconds to over 5 seconds (Figure 5C-D). Results from a linear discriminant analysis suggested that elapsed time could be decoded from the population, along with the identity of the presented image. Taken together, these neurophysiological investigations provide evidence that information about what, when, and where is represented within the monkey EC.

FIGURE 5.

A. Schematic of the temporal order memory task from Naya et al. (2011, 2017). Monkeys were shown two cues separated by brief delays (Encoding Phase). During the response phase, monkeys were rewarded for selecting the first, then second observed object from a cluster of simultaneously presented items. B. Percentage of order- and item-specific cells isolated from various regions in the medial temporal lobe. Note that in contrast to the strong temporal selectivity in hippocampal neurons and item selectivity in perirhinal cortex and inferior temporal lobe neurons, cells in EC are heavily mixed in their temporal order- and stimulus- selectivities (Adapted from Naya & Suzuki 2011). C. Raster plots and firing-rate histograms of putative temporal context cells acquired during a visual exploration paradigm (see Fig. 4A) that differentially modulate their firing rate relative to a pre-stimulus baseline with differing rates of relaxation. D. Normalized firing rate map of 128 EC neurons sorted by their time of return to baseline firing rate. All recorded cells elicited little variance in their response latency to the presented image. However, their variance in relaxation time over the course of presentation suggest the ability of EC neurons to track the passage of time at different timescales (Adapted from Bright et al. 2019).

6. PARALLELS WITH HUMAN ENTORHINAL CORTEX

There is a growing body of evidence suggesting that EC in humans closely mirrors what has been observed in non-human primates in both its anatomy and function. Although tracing experiments are not possible in humans, homologous sub-regions of human EC were identified in functional connectivity experiments using fMRI (105, 106). Specifically, strong blood oxygen level-dependent (BOLD) signal was observed in the rostro-lateral EC and caudo-medial EC that was coherent with the BOLD signal measured in the perirhinal and parahippocampal cortices, respectively. As described in section 3.2, tracing studies in non-human primates also strongly imply topographic biases of projections from perirhinal cortex to rostro-lateral EC and parahippocampal cortex to caudo-medial EC (47, 48). Unique functional connectivity between proximal and distal portions of the human subiculum with the rostro-lateral and caudo-medial EC was also observed, which is consistent with the patterns of projections observed through tracing experiments in non-human primates (47, 105, 106).

The rostro-lateral and caudo-medial aspects of the human EC also exhibit behavioral selectivity in their respective BOLD activity preferences, with the rostro-lateral elements being more object-responsive and the caudo-medial portions eliciting more spatially correlated BOLD responses (106). While additional research is needed to confirm homologous functional specialization of EC sub-regions in non-human primates, these observed functional specializations in humans parallel the respective object- and spatial- specificity of LEC and MEC in rodents (4, 8–10, 12, 20, 21). Deep brain stimulation of human EC during virtual spatial navigation has shown a critical, albeit still not fully understood, influence of this structure in remembering previously visited locations (107, 108). As mentioned in section 5.2, EC lesions in non-human primates produce profound deficits in spatial memory (61). Neurophysiological recordings and imaging studies have also identified neural firing motifs within the human EC homologous to those observed in other species (see Section 5.3). Specifically, fMRI reveals macroscopic activity patterns with hexadirectionally selective, grid-like regularity in healthy subjects navigating a virtual-reality open field maze (109). Like rodent grid cells, these grid-like BOLD signals are anchored to allocentric cues, and certain signals were conjunctively modulated by the direction of virtual movement. These hexadirectionally modulated patterns are also present when subjects are tasked with imagining a route to a target stimulus (110), during visual scan behaviors (111), and when subjects are required to conceptually organize information along two orthogonal discriminative axes (112), suggesting a generalized coding scheme within the human EC for flexibly encoding and retrieving continuous information across behaviors. These imaging findings have been corroborated by single-unit recordings taken from human epileptic patients, in which EC neurons with grid-like firing fields (113) and those with direction-selective firing patterns (114) were identified during virtual reality navigation paradigms. Taken together, the growing body of converging evidence suggests that both the anatomical and functional features of EC are largely conserved across species.

7. SUMMARY

While the primate EC provides the majority of the cortical input to the hippocampus, data from lesion and neurophysiological studies suggest that EC functions as much more than a simple information relay. The impairments observed following lesions restricted to the primate EC are distinct from those following hippocampal or perirhinal lesions, and the diverse neural activity observed among EC neurons is similarly distinct from its neighboring regions. In particular, EC is only transiently necessary for simple object recognition but becomes critically important when the relationships between indirectly related objects must be inferred and when previously experienced locations must be remembered. Along with object-related delay activity and timing responses that have been observed in EC, these neurons also display robust and reliable spatial responses, similar to the exquisite spatial representations identified in the rodent EC. Anatomically, EC is a site of convergence of multiple cortical information streams, many with their own topographical projection patterns. Combined with its own limited intrinsic connectivity and topographically-biased projections to different sub-regions within the hippocampus, EC is a computationally diverse area with elegant targeting of downstream structures. However, to date, the functions of these distinct processing streams have not been fully characterized. Novel technologies, including large-scale chronic recordings to densely sample neurons across EC sub-regions simultaneously with other brain regions, >1000-site laminar probes to more thoroughly investigate network oscillations across laminae, and targeted modulation through cell-type specific optogenetic manipulations are becoming accessible in primate research. Such tools will be critical to advancing our understanding of the computations EC performs intrinsically and in concert with other memory-associated machinery within the hippocampal formation.