Functional differentiation in the transverse plane of the hippocampus: an update on activity segregation within the DG and CA3 subfields

Abstract

Decades of neuroscience research in rodents have established an essential role of the hippocampus in the processing of episodic memories. Based on accumulating evidence of functional segregation in the hippocampus along the longitudinal axis, this role has been primarily ascribed to the dorsal hippocampus. More recent findings, however, demonstrate that functional segregation also occurs along transverse axis of the hippocampus, within the hippocampal subfields CA1, CA2, CA3, and the dentate gyrus (DG). Because the functional heterogeneity within CA1 has been addressed in several recent articles, here we discuss behavioral findings and putative mechanisms supporting generation of asymmetrical activity patterns along the transverse axis of DG and CA3. While transverse subnetworks appear to discretely contribute to the processing of spatial, non-spatial, temporal, and social components of episodic memories, integration of these components also occurs, especially in the CA3 subfield and possibly downstream, in the cortical targets of the hippocampus.

Introduction

The hippocampus plays well-recognized roles in memory and affective behavior [1], which appear to be segregated along the longitudinal (also known as dorsoventral or septotemporal) axis (Fig. 1A) [1, 2]. The formation and systems consolidation of episodic memories critically depend on the dorsal hippocampus (DH), however, there is increasing evidence that the encoding of spatial, non-spatial, temporal, and social components of these memories is, at least in part, segregated. Such segregation occurs along the transverse (also known as proximodistal) axis of hippocampal subfields [3], which includes the supra- and infrapyramidal blades of the DG (also known as upper or interior and lower or exterior blades, respectively), and proximal and distal subregions of CA3 and CA1 (Fig. 1B). The differential processing of spatial and non-spatial information in proximal and distal subregions of CA1 has been supported by substantial evidence [4–11] and linked to the spatial organization of entorhinal cortical inputs to CA1 [3, 9, 12]. More recently, distinct patterns of transverse activity during processing of spatial, non-spatial, temporal, and social information have also been observed in the DG and CA3.

Figure 1. Cartesian axes of the hippocampus.

A. The diagram depicts the volume-filled hippocampal formation. Anatomic reference content was obtained from the Allen Brain Atlas [132]. The Cartesian axes of the hippocampus includes the anterior(A)-posterior(P) axis (also known as the longitudinal or septotemporal axis), the dorsal(D)-ventral(V) axis, and the medial(M)-lateral(L) axis (also known as and referred to herein as the radial axis). B. The schematic shows the transverse (also known as and referred to herein as the proximo-distal) axis of the hippocampus including the suprapyramidal and infrapyramidal blades of the dentate gyrus (DG) as well as the intermediary DG crest region, the proximal (p’ and p) and distal (d) aspects of CA3, and the proximal and distal aspects of CA1. Furthermore, the schematic shows the transverse intrahippocampal connectivity. The granule cells in the DG suprapyramidal blade project across the proximodistal extent of CA3, while the infrapyramidal blade and crest preferentially project to proximal CA3 (p’ and p)[107, 108]. In CA3, proximal CA3 projects back to the DG hilus [94, 96, 133, 134](not shown) and have fewer recurrent CA3 connections with minimal extent along the proximodisal axis [95]. In contrast, more distally located CA3 neurons make recurrent connections, which also extend further along the transverse axis [94–96]. From CA3 to CA1, proximal CA3 preferentially projects to distal CA1 (at levels dorsal to the origin), mid portions of CA3 project equally to levels dorsal and ventral to the origin, and distal CA3 preferentially projects to proximal CA1 (towards levels ventral to the level of origin) [94–97, 106, 108]. While the diversity of projections of CA2 along the transverse axis are not known, CA2 pyramidal cells provide stronger excitatory inputs onto deep (blue) compared to superficial (green) CA1 pyramidal cells [121], suggesting that they may form a functional circuit with the MEC, which also preferentially projects to deep, rather than superficial, CA1 pyramidal cells [122].

The aim of this review is to update the literature on the functionally dissociable DH transverse subnetworks, and discuss the mechanisms implemented by such subnetworks to generate discrete (asymmetrical) activity patterns in DH subfields. Because the transverse segregation of information coding across proximodistal CA1 has been thoroughly discussed elsewhere [12–17], here, we will update the literature to include novel insights on the transverse functional segregation in the DG and CA3, and suggest future work on CA2. Understanding these mechanisms will not only advance our fundamental understanding of memory organization, but also provide a substrate for memory dysfunction, especially in the face of traumatic stress, psychiatric, and neurodegenerative diseases.

Functional Segregation along the Transverse Axis of DG

Whereas segregation of hippocampal function along the longitudinal and dorsoventral axes could readily be observed with classical lesion approaches (e.g. [18]), evidence for functional heterogeneity within the transverse axis was predominantly generated by monitoring the expression of activity-dependent immediate early genes (IEG), in particular cFos and Arc. Many studies investigating IEG responses in the DG during various behavioral tasks provide evidence for asymmetrical activation of the DG suprapyramidal blade even though the individual blades were not directly compared [19–26]. For example, Chawla et al. reported that the proportion of IEG-positive neurons in the upper blade significantly increase following a spatial task, while cells in the lower blade exhibit virtually no behaviorally-induced changes [19]. In fact, in all of the aforementioned studies, the trend towards biased suprapyramidal blade activation is indirectly supported by the significant effect in the suprapyramidal blade but lack of effect in the infrapyramidal, though direct support of this conclusion would necessitate a statistical test on their difference [27].

A primary piece of evidence for the asymmetrical transverse patterning of the DG comes from a study which addressed the anatomical gradients of neurons activated during training in the Morris water maze, a well-established spatial task [28]. Notably, analyses of the IEG cFos revealed a main effect of blade, with spatial maze testing eliciting a greater density of cFos-positive neurons in the suprapyramidal blade, compared to the infrapyramidal [29]. These results demonstrate that spatial experience elicits an asymmetric response in the transverse DG. Recently, we found that social interactions are also accompanied by the asymmetric transverse patterning of activity in the DG and CA3, with predominant cFos responses in a suprapyramidal blade-distal CA3 transverse subnetwork (for intrahippocampal transverse connectivity, see Fig. 1B). This transverse asymmetry could be bi-directionally modulated with different manipulations: whereas intense stress disrupted social interactions and asymmetrical DG activity, both effects were restored via inhibition of oxytocin receptor (Oxtr)-positive hilar interneurons [30].Together, these two studies indicate that, across functional modalities (i.e. spatial, social), the suprapyramidal blade is the predominant transverse subnetwork which regulates behavior. Interestingly, the suprapyramidalblade activation preference seems to be restricted to the DH, while disappearing along the ventral/temporal axis [31].

Much of the hippocampal research over the last 50 years has converged upon a family of theories that collectively attribute the function of the DG to computations related to pattern separation [32–34]. Therefore, the asymmetric suprapyramidal blade activity needs to be addressed in the context of this body of work. The Barnes lab demonstrated, using in situ hybridization approaches, that distinct ensembles of cells in the suprapyramidal blade responded to familiar and novel environments [19]. This was confirmed by findings showing that distinct ensembles of cells in the suprapyramidal blade respond as a function of environmental dissimilarity, while the proportion of reactivated neurons remained unchanged in the infrapyramidal blade [35]. These studies support the view that the suprapyramidal blade is a more efficient pattern separator than the infrapyramidal blade, which may explain the preferential routing of information through the DG transverse subnetwork. It remains to be determined, however, whether and under which conditions there might be a further differentiation in the suprapyramidal blade with respect to different spatial and social information. So far, this issue has remained unresolved because some evidence suggests that distinct DG populations fire in different environments [19, 36], whereas others shows that the same sparse population is activated across multiple environments in DG while new populations are recruited in CA3 [37, 38].

In contrast to the behaviorally driven patterning of activity in the suprapryamidal blade, activation of the infrapyramidal blade appears to be more related to various pathologies and pharmacological treatments. For example, in a model of pilocarpine‐induced status epilepticus, after repeated spontaneous seizures it was observed that stimulus-evoked population responses were larger in the infrapyramidal blade, relative to the suprapyramidal blade [31]. In a kainic acid model of epilepsy, the infrapyramidal granule cells were no longer able to evoke canonical recurrent excitation of the suprapyramidal granule cells through their monosynaptic connections [39]. Additionally, infrapyramidal granule cells are more susceptible to hypoxia [40]. Furthermore, in aging and after cannabinoid treatment [41], or after 6-hydroxydopamine [42] or risperidone [43] treatment, there are greater changes in the infrapyramidal blade than the suprapyramidal blade.

Putative Mechanisms of Transverse Functional Segregation in the DG

Several observations have been proposed to account for the DG’s asymmetric activity [44]. Considering that predominant innervation of the DG comes from the entorhinal cortex and that the dual inputs from the lateral (LEC) and medial (MEC) subdivisions (Fig. 2A) are proposed to mediate transverse functional segregation in CA1, we will first explore whether these inputs could serve as a basis for transverse segregation in the DG. To date, the question of anatomically-biased afferents from the entorhinal cortex along the transverse DG axis has been a field of contention and debate [45]. While most tracing studies report no differences in blade input from the LEC and MEC [46–51], two groups found that the LEC preferentially projects to the suprapyramidal blade [52, 53]. Given that LEC and MEC inputs preferentially innervate the outer (OML) and middle molecular layers (MML) of the DG, respectively, others have put stronger emphasis on differences in dendritic arborization across DG blades and molecular layers. In rats, a selective increase of dendritic arborization in MML of the suprapyramidal blade indicated biased blade and layer innervation by the MEC, however mice had greater arborization in both suprapyramidal MML and OML, suggesting biased blade but not layer innervation by the LEC or MEC afferents [44, 54–57]. Furthermore, Witter [45] conducted a systematic analysis across 21 tracing experiments to address this issue and concluded that while individual injections of biotinylated tracers may display unequal density across blades, labeling tends to be present in both blades, and as a whole, the LEC/MEC projections homogenously innervate the DG. Thus, entorhinal innervation of DG alone cannot easily explain the transverse functional segregation of the DG. Alternatively, entorhinal cortical afferents activate the suprapyramidal blade more rapidly in response to a stimulus, which, acting along with recurrent inhibition, would allow this blade to out‐compete the infrapyramidal blade [58]. A role of neuroanatomical connectivity cannot be ruled out in general. For example, the supramammilary projections to DG, which were recently identified as generators of DG responses to novel stimuli [59], innervate twice as densely the suprapyramidal blade compared to the infrapyramidal blade [60]. Other sources of transverse activity patterning can be found in the organization of local inhibitory circuits (Fig. 2B). Using a transsynaptic anterograde tracer, we recently found that the majority of hilar interneurons, which co-express oxytocin receptors (Oxtr-HI)(66.7% of total population), significantly more densely innervates the suprapyramidal blade compared to the infrapyramidal blade [30](Fig. 2B, right). Additionally, it has been observed that greater numbers of interneurons are found in the suprapyramidal granule cell layer [61, 62](Fig. 2B, left), and Timm stains suggest that these suprapyramidal interneurons are more densely innervated by mossy fibers [63]. Both of these features could increase the strength of feedforward and recurrent inhibition in the suprapyramidal blade relative to the infrapyramidal blade. Further findings include the greater ratio of basket to granule cells [64] (Fig. 2A, left) and a larger axon plexus formed by calretinin(CR)‐immunoreactive neurons [65–67] in the suprapyramidal blade (Fig. 2B, right). Differences in inhibitory regulation are also indirectly supported by the finding that suprapyramidal neurons are more vulnerable to alcohol, which acts on GABA_A receptors [68]. Together, these studies support the thesis that the suprapyramidal blade receives more inhibition.

Figure 2. Segregated afferents of the transverse hippocampus.

A. The diagram illustrates color-coded afferent connections related to the parent brain regions with segregated projections into the transverse axis of the hippocampus. Not indicated are the afferent projections which non-discriminately innervate a given hippocampal subregion across the transverse axis. For example, not pictured are the lateral entorhinal cortex (LEC), medial entorhinal cortex (MEC), septal nucleus, locus coeruleus, and raphe nucleus projections which homogenously innervate the suprapyramidal and infrapyramidal blades of the DG [45, 135]. Of note, innervation may initially be asymmetric during development, illustrated by the biased projections from the LEC/MEC [136] and local DG Cajal–Retzius cells [137] to the suprapyramidal blade, but become symmetric over the course of development. First illustrated are the biased projections from the suprammaillary nucleus of the hypothalamus (SUM) to the inner molecular layer (IML) of the suprapyramidal blade of the DG [60]. Segregated inputs to transverse CA3 arrive from the entorhinal cortex (EC), with both the LEC and MEC preferentially innervating distal CA3. The CA3 neurons constrained by the blades of the DG (p’) have particularly weak perforant inputs from the EC [94, 95, 106]. The EC also provides segregated inputs to CA1, with distal CA1 afferents arriving from the LEC, while proximal CA1 receives preferential input from the MEC [16, 138]. Abbreviations: d, distal, infra., infrapyramidal blade, IML, inner molecular layer, MML, middle molecular layer, OML, outer molecular layer, p, proximal, p’, proximal, supra., suprapyramidal blade. B. From left to right, the DG suprapyramidal blade, as compared to the infrapyramidal, has more densely distributed gamma aminobutyric acid (GABA)- positive (yellow) and basket-type interneuron somata (green) in the granule cell layer (GCL), and receives more projections from oxytocin receptor-positive hilar neurons (Oxtr-HI; red) and has a greater distribution of axonal plexus formed by CR‐positive neurons (blue).

This, however, raises a different problem: if the infrapyramidal blade has less inhibition, one would expect stronger, rather than weaker or absent, activity under the various behavioral conditions discussed above. An easy answer to this question is not available at this time - on the contrary. Important controls for the experiments discussed above show that neurons across the transverse axis (suprapyramidal/infrapyramidal blades and proximal/distal CA3 and CA1) are equally capable of expressing IEGs [4, 8, 19]. Moreover, analysis of electrophysiological features between blades have not been observed in slices of adult mice [69, 70], whereas developmental sparsification of dentate granule cell responses to entorhinal cortical input has been primarily ascribed to the infrapyramidal blade [71]. Lastly, whereas the suprapyramidal blade appears to be more responsive under behaviorally relevant conditions, the situation is reversed in the presence of seizure-inducing agents. Scharfmann et al. (2002) demonstrated, by extracellular recordings from infrapyramidal sites in hippocampal slices of pilocarpine-treated rats, significantly larger population spikes and weaker paired-pulse inhibition in response to perforant path stimulation relative to suprapyramidal recordings. Moreover, stimulation of the infrapyramidal blade molecular layer evoked larger responses in area CA3 than suprapyramidal stimulation, suggesting that the infrapyramidal blade may play a greater role in activating the hippocampus. With this in mind, one would expect a tight inhibitory control of the infrapyramidal blade. It is not clear, however, whether such control is exerted by putatively unknown local inhibitory networks, long-range afferents, or both.

An additional explanation for the topographic differences of blade responsivity to stimuli, includes genetic diversity of discrete cellular populations between DG blades (Fig. 3)[72]. Whilst large scale sequencing efforts have led some groups to suggest that genetic diversity does not exist between the DG blades [73], small scale efforts have found differences. For example, retinoic acid, an important regulator of molecular patterning in development, is more strongly expressed in the infrapyramidal blade. In the suprapyramidal blade, trends of enhanced expression of dopamine type 1 [74] and beta-1 adrenergic receptors [75] have been observed. Furthermore, the suprapyramidal blade is more sensitive to fluctuating levels of the neuromodulators corticosteroid and mineralcortin, as demonstrated by greater cell death following adrenalectomy [76–79] and mineralcortin knockout [80], respectively. Interestingly, neuromodulatory inputs, themselves, appear to be organized without selectivity for either blade [81–85].

Figure 3. Genetic diversity of the transverse hippocampus.

(Top) In the DG, retinoic-acid (RA)-regulated transcription is stronger in the infrapyramidal blade, as detected using a LacZ reporter mouse [72]. Additionally, there is an apparent trend of enhanced dopamine type 1 (D1D3) and beta-1 adrenergic receptors (β1-adrenoceptor) in the suprapyramidal blade, as compared to the infrapyramidal [74, 75]. In CA3, proximal aspects express more Titin (ttn-1) [111], while distal aspects express more oxytocin receptor-positive neurons [89]. In CA2, distal aspects are characterized by elevated transcription factor Sox5 [118]. For additional genetic diversity across the transverse axis of CA2, see Fig. 2b. In CA1, proximal aspects are characterized by greater TENM3, which is also expressed in multiple topographically interconnected areas of the hippocampal region, including proximal CA1, distal subiculum, and the MEC [139]. Proximal CA1 can also be characterized by Calbindin 1 (CALB1), Cartilage acidic protein 1 (CRTAC1), while distal CA1 can be characterized by N-Deacetylase And N-Sulfotransferase 4 (Ndst4) and SIM BHLH Transcription Factor 1 [140]. Abbreviations: d, distal, p, proximal, p’, proximal (top). Genetic diversity can also be used to characterize the proximal and distal subregions of CA2. Across thres studies, it has been shown that gradients of PCP4-, α-Actinin2-, Wfs1-, CR-, and CgA-positive neurons can be used to characterize subregions across the transverse axis of CA2 (bottom) [118–120]. Abbreviations: D, dorsal, L, lateral, M, medial, V, ventral.

In summary, biased local and long-range connectivity, including greater innervation by DG interneurons and the supramammilary nucleus, as well as enhanced responsiveness to entorhinal cortical afferents and the main stress mediators, and distinct interblade genomic expression patterns, are some of the potential mechanisms underlying the transverse segregation of behaviorally-relevant information towards the suprapyramidal blade. Discrete mechanisms are likely to contribute to further activity patterning of DG, however the extent of segregation of different types of information (e.g. spatial, social, etc.), remains to be determined.

Functional Integration and Segregation Along the Transverse Axis of CA3

Before discussing the transverse activity of CA3, it is important to note that there are some inconsistencies regarding the classification of CA3 subnetworks. Lorente de No originally suggested that CA3 can be parsed into distinct regions: distal (CA3a, nearest to the CA2), intermediate, and proximal (CA3c, nearest to the DG) [86], though he also included a vaguely-defined term “CA4”, which appears to have been intended for the part of CA3 that inserts within the DG and is enclosed by the blades (Fig. 2A, see CA3 p’). However, even Lorente de Nó appears to have sometimes used the term CA4 while referring to the polymorphic layer of the DG, leading many groups to define proximal CA3 by the edge of the DG blades [8, 87–89](Fig. 2A, see CA3 p). These inconsistencies are taken into consideration as we compare data across studies.

The majority of reports indicate that task-related information processing, including spatial-, novelty-, and temporal-related aspects of episodic memory, is uniformly distributed across the proximal (p) and distal subregions of CA3. For example, contextual fear conditioning and retrieval, temporal-related object recognition, random reward, and exposure to similar and dissimilar contexts all elicit homogenous responsivity across proximal, intermediate, and distal CA3 subregions [8, 11, 90, 91]. Together, these reports would suggest that, in contrast to the DG, functional information is not segregated along the transverse axis in CA3, providing a basis for the view that CA3 is a homogenous circuit [92, 93]. The observed functional homogeneity across transverse CA3 might seem surprising in light of the known differences that exist along the CA2 transverse axis, including segregated input from the DG [94], differences in cell type, size, and dendritic length [94–96], differences in recurrent collaterals [95–97]; and differences of electrophysiological properties [91, 98, 99]. However, the dense recurrent collaterals of CA3 neurons appear to trigger generalized responses to different inputs from DG, resulting in more uniformed activity and unique role of CA3 in pattern completion.

However, several exceptions to this mode of activity have been noted. The most proximal aspect of CA3 that is enclosed by the blades (p’), was found to perform pattern separation of spatial-related information [90, 99, 100], functioning as a coordinated unit with the DG [100, 101].

It has also been observed that the processing of nonspatial information in an olfactory association task is segregated across transverse CA3, preferentially activating a proximal (p) CA3-distal CA1 subnetwork [8]. These findings support and extend previous findings indicating that the proximal and distal parts of CA1 exhibit differential coherence with the entorhinal cortex during this task [9], and that the learning of this task requires the LEC-CA1 circuit, which preferentially projects to distal CA1, but not the MEC-CA1 circuit, which preferentially projects to proximal CA1 [10]. Furthermore, the contribution of this putative LEC-distal CA1 circuit mechanism, falls in line with anatomical evidence regarding preferential projections from olfactory regions to the LEC, but not MEC [102–105]. These findings provide further support for the role of entorhinal cortical projections in the transverse segregation of nonspatial information [10, 12, 13]. However, while biased temporammonic projections from the entorhinal cortex can readily explain nonspatial segregation in CA1, they cannot account for the observations with CA3. Although the proximal CA3 receives more entorhinal cortex inputs than distal CA3 [94, 95, 106], differences in proximal afferents from LEC vs. MEC have not been observed. This raises the question of whether biased LEC-proximal CA3 innervation exists. Alternatively, based on anatomical connectivity [107, 108], it can be inferred that the upstream DG infrapyramidal blade preferentially processes non-spatial information, which could give way to the subsequent asymmetric pattern in proximal CA3, however this remains to be experimentally documented.

We recently demonstrated that social information is segregated across transverse CA3. Specifically, we observed that social interactions are accompanied by biased activation of a suprapyramidal blade-distal CA3 subnetwork [30]. If transverse segregation was retained as information is propagated through the trisyanptic circuit, the biased activation of distal CA3 would presuppose propagation to anatomically connected ventral proximal CA1 and caudal lateral septum [95, 109–111] through the transverse subnetwork DG suprapyramidal blade-distal CA3- proximal ventral CA1-caudal lateral septum. This is consistent with observations that ventral, but not dorsal, CA1 makes a substantial contributions to social behavior [112–114], as does the caudal lateral septum [30, 115–117].

Together, this evidence suggests that unlike the homogenous response of transverse CA3 to novelty-, reward-, temporal-, and possibly spatial-related information, non-spatial and social information are asymmetrically processed throughout proximal and distal CA3 and likely retain their segregation as they are propagated downstream. This transverse heterogeneity in CA3 could support and extend the previous thesis that the CA1 network operates between alternating states. Specifically, it was suggested that CA1 operations alternate between a state where the network is primarily responsive to functionally segregated direct inputs from entorhinal cortex and a state where cells are predominately controlled by the integrated inputs from CA3 [15]. Based on the findings discussed above, a state of CA1 where the network is primarily responsive to functionally segregated direct inputs from CA3 needs to be accounted for as well.

Functional Segregation of CA2 in the Transverse Plane: An Avenue for New Research

While the behavior-related functional diversity of CA2 along the transverse axis has remained unexplored, histological and synaptic studies suggest that CA2 can also be subdivided into proximal and distal aspects. In one study, it was demonstrated that gradients of Purkinje Cell Protein 4 (PCP4), α-Actinin2, and Wolframin ER Transmembrane Glycoprotein 1 (Wfs1) could be used to classify transverse CA2. In proximal CA2, most glutamatergic neurons were PCP4/α-Actinin2-positive (65%), while a minority were Wfs1/α-Actinin2-positive (<5%). This is in contrast to distal CA2 where 50% of glutamatergic neurons were Wfs1/α-Actinin2-positive (Fig. 3, bottom left)[118]. In another study, it was shown that the percentage of PCP4-positive cells overlapping with CR-positive mossy fibers differs between the proximal and most distal ends (Fig. 3, bottom middle)[119]. Furthermore, even in the human hippocampus a case can be made that CA2 can be subdivided along the transverse plane. This is based upon findings that proximal CA2 can be identified as having a thick layer of chromogranin A (CgA) positive neurons, with the perikarya of the deep layer more intensely stained than the superficial. Meanwhile, intermediate CA2 can be characterized by CgA-positive perikarya accounting for most of the neurons and covering the entire extent of stratum pyramidale, whereas distal CA2 can be characterized by the dwindling of CgA-positive neurons which intermingle with CA1 pyramids (Fig. 3, bottom right)[120]. In addition to this genetic support of transverse trends in CA2, through the use of intracellular and extracellular recordings, Fernandez-Lamo [118] found marked differences in synaptic activity, subthreshold membrane potentials, and phase-locked firing coupled to theta and gamma oscillations. Although the functional significance underlying the topography of gene expression and synaptic features in the transverse CA2 remains to be determined, these studies begin to define the proximal from the distal subdomains and call to question whether this heterogeneity could allow for different functional capabilities. Lastly, while the diversity of projections of CA2 along the transverse axis are not known, CA2 pyramidal cells provide stronger excitatory inputs onto deep (blue) compared to superficial (green) CA1 pyramidal cells [121], suggesting that they may form a functional circuit with the MEC, which also preferentially projects to deep, rather than superficial, CA1 pyramidal cells [122]. The functional and anatomical segregation across proximodistal CA2 is an avenue of research ripe for future exploration.

Implications for Human Research

How the three-dimensional anatomical organization governs segregation and integration of information processing is less well understood in humans. This is about to change given the intensified research in this are using various neuroimaging techniques. Most human studies have confirmed the functional segregation along the longitudinal axis (anteroposterior corresponding to ventrodorsal in rodents), confirming the predominant involvement of the posterior hippocampus in episodic memory [123–125]. However, little differentiation along both the longitudinal and transverse axes has been found with respect to specific memory content [123]. Moreover, the differentiation of pattern completion and separation functions in the human hippocampus has been proposed to occur along the longitudinal rather than transverse axis [126].

Analysis of segregation and integration of memory content is not only relevant for advancing our fundamental understanding of memory but also has a strong impact on numerous psychiatric conditions. Well-established episodic memory deficits were recently confirmed in sufferers from schizophrenia [127], major depression [128], and post-traumatic stress disorder [129]. Moreover, recollection deficits are sometimes content-dependent as shown recently for social episodic memory deficits in first-episode and chronic schizophrenia patients [130]. In light of this findings, the disrupting effects of stress on the asymmetrical activation in DG and CA3 related to social behavior may be particularly relevant, as is the potential to restore normal asymmetric patterns by manipulations of DG hilar interneurons [30]. Although human imaging approaches lack the spatial resolution of single cell analysis perform with rodent approaches, rapid technological advances are likely to overcome some of these limitations and open this area to scientific examination.

Conclusion

The presented evidence shows that the hippocampus displays functional segregation along the transverse plane, with some functional modalities retaining transverse segregation as information is propagated through the transverse axis of the trisynaptic circuit (i.e. non-spatial information via proximal CA3-dorsal distal CA1 and social information via DG suprapyramidal blade-distal CA3-ventral proximal CA1 transverse subnetworks), whilst others are integrated across proximodistal CA3. Furthermore, it is important to appreciate that different mechanisms might mediate segregation of each task-type modality in each transverse subregion. As functional segregation of proximodistal CA1 seems to arise as a consequence of differential connectivity with the MEC and LEC (Fig. 4, spatial and non-spatial), respectively, segregation across DG blades is likely due to biased innervation by suprammilary afferents and local interneurons. Further functional segregation could be performed by distinct modular ensembles embedded within blades, although mosaic patterns are also conceivable. The fact that distinct neuronal ensembles within a memory engram mediate different functions [131] is consistent with the former proposal, while the fact that individual cells can contribute to representations in multiple and putative unrelated environments is consistent with the later.

Figure 4. Schematic of functional differentiation in the transverse plane of the hippocampus.

Schematic of a proposed theoretical model regarding hippocampal activity across the transverse axis in the regulation of modality-specific behaviors, including spatial (black), non- spatial (red), temporal (yellow), and social (green)-related tasks. An emerging view is that task-specific information types are patterned within functionally-discrete transverse subdivision in individual hippocampal subfields. Furthermore, asymmetrically transverse patterned activity within a transverse subfield may retain its transverse segregation as information is propagated downstream within the trisynaptic circuit, forming a functionally-dissociable transverse subnetwork. Experimentally-derived data regarding segregation mechanism (solid lines) are denoted from hypothetical retention of segregation (dashed lines) based upon known anatomical connectivity. The segregation of spatial/non-spatial information in proximal and distal CA1 are ascribed to the parallel inputs from preferential medial entorhinal cortex (MEC) and lateral entorhinal cortex (LEC) inputs, respectively. Direct support of these upstream segregating mechanisms includes anatomical connectivity [16] and a study demonstrating that inactivation of the LEC, but not MEC, disruptions non-spatial learning [10]. In contrast, when spatial information enters the hippocampus via the perforant path, it is segregated to the suprapyramidal blade of the DG, though appears to become distributed throughout CA3. Non-spatial information, on the other hand, is segregated to the proximal CA3. In addition to spatial/non-spatial segregation in transverse CA3 and CA1, it has also been shown that discrimination of temporal-related task information is correlated with IEG levels in distal, but not proximal, CA1 [11] and that temporal sequence coding is significantly higher in intermediate CA1 [7]. Recently, it has also been demonstrated that the processing of social information also patterns hippocampal activity in the transverse pane of the hippocampus. Specifically, it was demonstrated that oxytocin-receptor positive hilar interneurons (Oxtr-HI) regulate the assymetrical transverse segregation of social information towards the suprapyramidal blade of the DG and distal CA3. Distal CA3 and ventral, but not dorsal, CA1, have been shown to mediate social discrimination. Together, this studies suggest that social information may retain its transverse segregation as it is propagated from CA3 towards CA2 and ventral proximal CA1 networks, whose biased efferent to the caudal LS have also been shown to regulate social behavior. Abbreviations: d, distal, p, proximal, p’, proximal.

In CA3, transverse functional segregation may reflect the asymmetrical transverse intrahippocampal circuity, including the innervation of distal CA3 by the DG suprapyramidal blade, but not infrapyramidal. It remains to be determined whether the asymmetrically patterned processing of non-spatial information in proximal CA3, arises as a consequence of predominant upstream processing in DG infrapyramidal blade, or reflects a biased innervation of the LEC to proximal CA3, or something entirely different. In contrast to the functional segregation of some information-type modalities across proximodistal CA3, the integration of others (i.e. temporal, spatial) may arise from the transverse-spanning recurrent collaterals.

In summary, newfound evidence supports that the hippocampus contains more than two transverse functional subnetworks. These modality-specific transverse subnetworks arise from distinct segregation mechanism, allowing the hippocampus to process different contents of episodic memories.

Highlights.

• The aim of this review is to update the literature on the functionally dissociable dorsal hippocampus transverse subnetworks, and discuss the mechanisms implemented by such subnetworks to generate discrete (asymmetrical) activity patterns in subfields.

• Across functional modalities (i.e. spatial, social), the suprapyramidal blade is the predominant transverse subnetwork of the dentate gyrus which regulates behavior.

• In contrast, task-related information processing, including spatial-, novelty-, and temporal-related aspects of episodic memory, is uniformly distributed across the proximal and distal subregions of CA3.

• Lastly, we discuss that the hippocampus displays functional segregation along the transverse plane, with some functional modalities retaining transverse segregation as information is propagated through the transverse axis of the trisynaptic circuit, whilst others are integrated across proximodistal CA3.