Connectivity of the Hippocampus

Abstract

The aim of this chapter is to extend the standard simplified diagram of the connectional organization of the hippocampus found in many current textbooks, by adding details on the connectivity of area CA2 and on entorhinal intrinsic wiring. In the chapter, some of the ‘traditional wisdoms’ on hippocampal connectivity are discussed, emphasizing the need for a more inclusive framework to model the hippocampus. The chapter focusses on intrinsic connections, and many of the well-known extrinsic connections of the hippocampus will not be covered in this chapter, for two reasons. First, the information is already available at a summarized (meta) level, and a new summary would not assist those who need anatomical details to contribute to the explanation of the functional outcome of a study. Second, this chapter is meant to provide a framework of knowledge to support computational modelling of the region, and therefore only the most relevant and quantitative data on the connectivity of the hippocampus are covered.

Overview

In the first edition of this book, I pointed to the increasingly complex intrinsic wiring diagram of the hippocampus and that new data are being added at an increasing speed. With the emergence of cell-specific viral tracing techniques, the potential for a data explosion has become eminent, going hand in hand with an increase of the potential for false-positive or incomplete data. The relevance of interneurons in the local network, as well as the fact that interneurons contribute to long-range projections, has been integrated into current conceptualizations of the ‘Connectivity of the Hippocampus’. Several comprehensive reviews have been published to which the reader is referred for many of the connections not covered in this chapter or for more details on the connections described here. An excellent, much more detailed resource can be found in a recent book chapter (Cappaert et al. 2015). Several online databases contribute to making this wealth of connectional data accessible as well (see further reading).

In contrast to this ever-expanding connectional knowledge base, many functional papers and reviews still use a simplified diagram of the connectional organization of the hippocampus as their reference, which we will here refer to as the standard view. The aim of this chapter is to extend this standard view, adding details that have been known for some time or have recently been provided, but apparently have not yet been incorporated in the commonly accepted connectional scheme for the region. For example, the increased insights on the connectivity of area CA2 are added in this second edition, as well as many new details on entorhinal intrinsic wiring.

I further aim to reinterpret some of the ‘traditional wisdoms’ on hippocampal connectivity, potentially pointing to the need of a changed functional framework for the hippocampus. The use will be made of a standardized scheme of connections which hopefully will facilitate easy dissemination of these adapted connectional concepts for the region. Many of the very well-known connections, such as all extrinsic connections of the HF and EC, will not be covered in this chapter, for two reasons. First, the information is already available at a summarized (meta) level, and a new summary would not assist those who need anatomical details to contribute to the explanation of the functional outcome of a study. Second, this chapter is meant to provide a framework of knowledge to support computational modelling of the region, and therefore I have selected what I consider the most relevant new data on the connectivity of the hippocampus, not of the brain.

Microscopical Anatomy and Nomenclature

Throughout the chapter, reference will be made to the hippocampal formation (HF) and the entorhinal cortex (EC) as the two main areas of interest. The HF in turn comprises three distinct subregions (Fig. 1): the dentate gyrus (DG), the hippocampus proper (consisting of CA3, CA2 and CA1) and the subiculum (Sub). The HF is a three-layered cortex that is easily differentiated from the EC, since the latter has more than three layers (see below). The deepest layer of the HF houses basal dendrites of principal cells and a mixture of afferent and efferent fibres and local circuitry – interneurons. Superficial to this polymorph layer is the cell layer, which is composed of principal cells and interneurons. On top, the most superficial layer or molecular layer contains the apical dendrites of the neurons and the large majority of axons that provide inputs. In the dentate gyrus, these layers are, respectively, referred to as the hilus, granular (cell) layer and molecular layer (stratum moleculare). In the CA-region, we find the deep polymorph layer (stratum oriens), followed by the pyramidal layer (stratum pyramidale), topped by the superficial or molecular layer. The latter is subdivided into a number of sub-layers. In CA3, three sub-layers are distinguished: stratum lucidum, representing the mossy fibre input from DG; stratum radiatum, i.e. the apical dendrites of the neurons in stratum pyramidale; and, most superficially, the stratum lacunosum-moleculare comprising the apical tufts of the apical dendrites. The lamination in CA2 and CA1 is similar to that in CA3, with the exception that the stratum lucidum is missing in CA1 and absent or much less evident in CA2. In Sub, the superficial layer is generally referred to as molecular layer, sometimes divided into an outer and inner portion, and the remaining two layers are referred to as the pyramidal (cell) layer (stratum pyramidale) and stratum oriens. The latter is very thin and quite often not specifically differentiated from the underlying white matter of the brain. The EC, commonly subdivided into a medial (MEC) and a lateral (LEC) part,¹ is generally described as having six layers, a molecular layer (layer I), the superficial cell layer (layer II), the superficial pyramidal cell layer (layer III), a cell-sparse lamina dissecans (layer IV), the deep pyramidal cell layer (layer V) and a polymorph cell layer (layer VI).²

Fig. 1

Schematic representation of the position of the HF and the EC and main topological axes. (a) Posterior view of the rat brain showing the position of the LEC (light green) and MEC (dark green; modified with permission from Fyhn et al., 2004, Science 305: 1258–1264. (b) Lateral view of a partially dissected brain showing the shape and position of the HF and the longitudinal or dorsoventral axis, as well as the position and extent of the pre- and parasubiculum (PrS and PaS, respectively) and entorhinal cortex (EC) (Modified with permission from Boccara et al. 2010 Nat Neurosci.13:987. (c) Schematic drawing of a horizontal section illustrating the main nomenclature. (d) Horizontal section stained for the neuronal marker NeuN, illustrating the main subdivisions of HF and the EC

In order to understand the anatomical organization, it is relevant to describe the coordinate systems that define position within the HF and PHR (Fig. 1). For the HF, there are three relevant axes: the long axis, the transverse or proximodistal axis, which runs in parallel to the cell layer, starting at the DG; and the radial or superficial-to-deep axis, which is defined perpendicular to the transverse axis. In the EC, a similar superficial-to-deep axis is used in addition to mediolateral (proximodistal) and anteroposterior (rostrocaudal) axes.

The Standard Connectional View

According to the standard view (Fig. 2), neocortical projections eventually reach the EC, which in turn provides the main source of input to DG of the hippocampal formation. All subregions of the hippocampal formation are sequentially connected by a serial chain of connections. In short, the dentate granule cells give rise to the mossy fibre pathway which targets the CA3. Axons from CA3 neurons form the so-called Schaffer collateral projection, targeting CA1 and lastly, CA1 projects to Sub. Output from the hippocampal formation arises in CA1 and the Sub and is directed to the parahippocampal region, mainly, but not exclusively to the deep layers of the EC. This series of unidirectional connections has also been referred to as the extended trisynaptic circuit. In a more complex version, EC mediates two parallel projection streams by way of LEC and MEC, respectively, that each reflect major input/output differences. The EC is the source of the perforant pathway, which projects to all subregions of the hippocampal formation. Entorhinal layer II projects to the dentate gyrus, CA3 and CA2, whereas layer III projects to CA1 and Sub. CA2 has been added to the circuitry. Whether or not CA2 receives mossy fibre input is still debated, but recent data indicate that species differences might exist. For now, we assume that the mossy fibre projection is a characterizing feature of CA3. In turn, CA2 has strong projections to both CA3 and CA1. The projections to CA1 and subiculum show a complex topographical organization (Fig. 3). In the following sections, each of the connections of the more extended scheme will be reviewed, detailed and when appropriate appended, starting with the entorhinal projections to the individual subdivisions of the HF.

Fig. 2

The standard view of the entorhinal-hippocampal network. Layer II of EC originates the perforant pathway to DG. DG in turn sends the mossy fibre projection to CA3, where neurons originate the autoassociative projection as well as Schaffer collaterals to CA1. CA1 projects to Sub and both of them send return projections back to layer V of EC

Fig. 3

The extended version of the standard view of the entorhinal-hippocampal network. CA2 was added to the network, as well as additional projections for Layer II of EC to CA3 and CA2 and the layer III projections to CA1 and Sub. The differential distribution of the projections from the LEC and the MEC along the transverse axis of the CA1 and the Sub has been included, as well as the organization of the projections from the CA1 to the Sub and from both back to LEC and MEC. The longitudinal topologies of neither connections are represented

Entorhinal-Hippocampal Projections

The elaborate Golgi studies of Ramon y Cajal and Lorente de Nó first demonstrated that EC is the origin of an immensely strong projection to HF. The latter became generally known as the perforant path(way). These observations were subsequently corroborated and extended in a seemingly continuous stream of tracing studies that drew attention to the many parallel entry routes for entorhinal inputs to HF, providing us with the contemporary image of EC projections to DG, the hippocampal fields CA1–CA3 and Sub. The total component of fibres was originally named the perforant (temporo-ammonic) pathway by Cajal,³ since the axons from EC perforated the pyramidal cell layer of the Sub. In the molecular layer of Sub, axons subsequently travel towards DG, crossing the hippocampal fissure, or course in stratum lacunosum-moleculare of CA1, CA2 and CA3, while making en passant synapses on the pyramidal neurons and interneurons in the CA fields, and continue into the tip of the molecular layer of the DG. There is an additional route for entorhinal fibres to reach targets in the hippocampus, referred to as the temporo-alvear tract. Axons in this pathway, which does not perforate the Sub, travel in the alveus and to some extent in stratum oriens below Sub and CA1–CA3 and will eventually traverse the pyramidal cells layer of the CA fields at specific points and continue to stratum lacunosum-moleculare where they terminate. Note that these axons target basal and apical dendrites of pyramidal cells as well as interneurons in strata oriens, pyramidale and radiatum.⁴

EC Projections to DG, CA3 and CA2

Cells in layer II of EC give rise to projections to DG, CA3 and CA2, and this observation has been made in most if not all species studied, including humans. It is likely that both the projections to DG and CA3 originate as collaterals from the same neuron and that the majority of neurons that project to DG and CA3 express marked levels of the protein Reelin, one of the two main cell markers for neurons in layer II. Details regarding the origin of CA2 projecting cells are unknown, but it is likely that these neurons also belong to the reelin-positive class of neurons. The other neuron class stains positive for the calcium-binding protein calbindin. These neurons give rise to widespread projections to the forebrain, but interestingly, about half of the population of EC layer II calbindin-positive neurons apparently issues local axon collaterals, contributing to an extensive, though yet now well-analysed intrinsic projection system. Only a small percentage of these neurons contribute to the projections to the hippocampus. Although the organization of the EC projection to DG has been described in much more detail than the EC to CA3 projection, the latter appears to follow organization principles like those that govern the projection from entorhinal layer II to DG. Generally, two components are differentiated which have their exclusive origin in LEC or MEC, respectively. Projections from LEC terminate in the outer half of the stratum moleculare of DG and the stratum lacunosum-moleculare of CA3 and CA2, whereas those from MEC terminate deep to the lateral fibres (Figs. 3 and 4). In DG, the entorhinal terminal zone occupies the outer two-thirds of the molecular layer, and in CA3/CA2, the entire radial dimension of stratum lacunosum-moleculare contains entorhinal fibres.⁵

Fig. 4

Wiring diagram, illustrating the organization of the projections from layers II, III and V of the MEC and the LEC to the various subdivisions of the HF. Note the laminar terminal distribution of the layer II component to the DG and the CA3 and the restricted transverse terminal distribution of the layer III projection to the CA1 and the Sub

There are conflicting papers on the transverse distribution of the layer II perforant path projection. Whereas in some studies no differences were reported, others reported that the lateral perforant pathway preferentially projects to the enclosed blade of the dentate gyrus and the medial component either does not show a preference or predominantly targets the exposed blade. In CA3 no indications have been found for a further transverse organization, although it should be mentioned that the distribution of apical dendrites makes it likely that neurons in the most proximal portion of CA3 are largely devoid of entorhinal input since their dendrites do not reach into the terminal zone in CA3. In the mouse and the monkey, no transverse organization has been described in either the DG or the CA3 projection.

EC Projections to CA1 and Sub

Layer III of the EC contributes a second component to the perforant path that selectively targets CA1 and the Sub (Fig. 3). Axons originating from the LEC and the MEC show strikingly different terminal patterns, but unlike the layer II projections, the difference is not along the radial axis but along the transverse axis. The projection that arises from the LEC selectively targets neurons in the distal part of CA1 (the part closest to the Sub) and in the adjacent proximal part of the Sub. In contrast, the projection from the MEC distributes selectively to the proximal CA1 and the distal Sub (Figs. 3, 4).⁶ In their respective target domain, entorhinal fibres completely cover the radial extent of stratum lacunosum-moleculare of CA1 and the other portion of the molecular layer of the Sub.

In addition to the main innervations arising from layers II and III in the EC, a projection originating from deep layers has been described as well. In the DG, this deep layer component preferentially distributes to the inner portion of the molecular layer, the granule cell layer as well as the subgranular, hilar zone, where it establishes asymmetrical synapses onto granule cell dendrites as well as on their somata and onto spine-free dendrites in the subgranular zone. The latter most likely represent dendrites of interneurons (Fig. 4). In the other divisions of the HF, details on the distribution of this deep pathway are lacking.

Also, weak inputs from the PrS and PaS reach all hippocampal subfields, where they terminate throughout stratum moleculare/lacunosum-moleculare, overlapping with the inputs from the EC. The CA1 and Sub receive additional inputs from the perirhinal (PER) and postrhinal cortices (POR). The inputs from the PER and POR show a topology along the transverse axis comparable to that seen in case of the projections from the LEC and MEC, respectively. However, both projections have a strong preference for the extremes, such that the PER project to the most distal part of CA1 and the most proximal part of the SUB and the projections from the POR favour the opposite extremes.

Synaptic Organization

In the rat, a majority of the terminals of the perforant path fibres (around 90%) form asymmetric synapses and thus likely are excitatory, and no major differences have been reported between the lateral and medial components of the pathway. Fibres contact most frequently dendritic spines of dentate granule cells or of pyramidal cells in the CA fields and the Sub. A small proportion of the presumed excitatory perforant path fibres terminate on non-spiny dendrites of presumed interneurons. In addition, a small proportion of the perforant path synapses is symmetrical, indicative of their inhibitory nature, and these likely target both interneurons and principal cells alike.

In the DG, entorhinal synapses make up at least 85% of the total synaptic population, and they target mainly apical dendrites of granule cells. Interneurons that are innervated are those positive for parvalbumin, as well as those positive for somatostatin and NPY. No details have been reported for the CA3, but on the basis of quantitative analyses on reconstructed single neurons (Matsuda et al., 2004), one may assume that a large majority of the excitatory entorhinal fibres terminate on spines, i.e. indicating synapses with pyramidal cells, and only a minor percentage terminate on shafts, taken to indicate presumed contacts with interneurons. Although in the stratum lacunosum-moleculare of the CA3 inhibitory terminals make up approximately 10% of the total population, it has not been established whether these all belong to local interneurons or whether part of them have an entorhinal origin. No studies to date have looked into possible interneuron targets for perforant path fibres in the CA3. In stratum lacunosum-moleculare of the CA1, about 15% of the total population of synapses is inhibitory, and the other 85% are excitatory. Unlike the situation in the DG and CA3 where most if not all of the synapses in stratum moleculare/lacunosum-moleculare are of entorhinal origin, in the CA1 the total population of excitatory terminals likely have three different origins, the EC, thalamic midline nuclei such as nucleus reuniens and the amygdala.⁷ Regarding entorhinal inputs, over 90% is asymmetrical, i.e. excitatory terminating on spines, and around 5% is excitatory terminating on shafts. Almost no symmetrical, i.e. inhibitory entorhinal fibres have been reported in CA1. The terminals on shafts likely indicate that interneurons are among the targets and recently interneurons that reside at the interface between strata lacunosum-moleculare and radiatum have been identified as recipients of entorhinal input.

In the Sub, the situation in the superficial half of the molecular layer is likely to be comparable to that in stratum lacunosum-moleculare of CA1 with the adding complexity of having even more inputs distributing here, including those from PrS and PER/POR. Of the entorhinal synapses, over 90% is excitatory and 80% terminates on spines and 10% on dendritic shafts, likely of interneurons, including those containing the calcium-binding protein parvalbumin, and the remaining are symmetrical terminals. The postsynaptic targets have not been identified anatomically, but electrophysiological data indicate that pyramidal cells that project back to EC are among the targets, an observation that has not been corroborated by anatomical findings (own unpublished data).

Projections from CA1 and Subiculum to Entorhinal Cortex

Transverse and Laminar Organization

The dentate gyrus and CA3 field of the hippocampus do not project back to EC. Thus, the recipients of the layer II projection do not have any direct influence over the activities of EC. It is only after the layer II and layer III projection systems are combined in CA1 and Sub that return projections to EC are generated. The return projections mainly terminate in the deep layers (V and VI) although a component ascends into the superficial layers. The main targets of these output projections are in layer V, where likely two or three different subgroups of principal neurons reside. The cells in the deeper part, referred to as layer Vb, stain positive for the transcription factor Ctip 2, whereas those in the superficial layer Va stain for Etv1. Projections from CA1 mainly target neurons in layer Vb, whereas subicular output seems to target both layers. Whereas neurons in Va originate the main extrinsic projection system of EC, those in Vb project preferentially intrinsically, targeting layers Va, III and II (Fig. 5). In case of the projections from the Sub, up to 93% of fibres form asymmetrical, i.e. excitatory synapses onto dendritic spines (68%) and onto shafts (23%). A small proportion (7%) forms symmetrical synapses, equally onto dendritic shafts and spines.

Fig. 5

Wiring diagram of some of the main intrinsic and extrinsic connections of EC. Reelin-positive layer II neurons project to DG, CA3 (and CA2), whereas neurons in layer III project to CA1 (and subiculum). Return projections from CA1 target mainly neurons in layer Vb, whereas those from the subiculum distribute to both Vb and Va. Neurons in layer Vb give rise to strong intrinsic connections to layers Va, III-I. Note that the intrinsic component originating from layer II calbindin-positive neurons is not indicated (Modified with permission from Witter et al. 2017a, Front Syst Neurosci. 11: 46)

In addition, electrophysiological evidence indicates that among the target cells are neurons in layer V that project to layers II and III of the EC (see section on “Entorhinal Associational and Commissural System”). It is relevant to point to the fact that the projections from CA1 and the Sub to the EC show a topology along the transverse or proximodistal axis. The projections from the proximal part of CA1 and the distal part of the Sub distribute exclusively to the MEC, whereas cells located in the distal part of CA1 and the proximal part of the Sub project mainly to the LEC. In this way the return projections thus maintain the topography displayed by the input projections from the MEC and the LEC (Figs. 3 and 4).

Longitudinal Organization

In addition to the radial and transverse organization of the layer II and layer III projections, respectively, as described above, all connections between the EC and the HF show a striking topology along the long axis of the HF. Both the projections from and to the EC follow the same principle in that lateral and posterior parts of the EC are connected to the dorsal portion of the HF, whereas increasingly more medial and anterior parts of the EC are connected to more ventral parts of the HF (Fig. 6). It is relevant to point out that this topographical organization is indeed a gradual one such that a small portion of the EC may distribute axons over up to 25–30% of the long axis of the HF and likewise a small part of CA1 and the Sub may distribute axons to a rather extensive area of the EC.

Fig. 6

Longitudinal organization of entorhinal-hippocampal connectivity. A dorsolateral–to–ventromedial gradient in the EC (left-hand side; magenta to blue) corresponds to a dorsal–to–ventral gradient in the HF (right-hand side). Note that the topology in the EC cuts across the MEC-LEC border indicated with the yellow line (left-hand side, modified with permission from Fyhn et al. 2004, Science 305: 1258–1264; right-hand side, modified with permission from Amaral and Witter 1989, Neuroscience 31: 571–591)

When taking the transverse and longitudinal organization into account, the important point emerges that these return projections from CA1 and the Sub are exactly in register, i.e. they are point-to-point reciprocal, with the entorhinal inputs to these areas. This remarkable topography confirms the critical role of the EC with respect to the input to, and output from, the HF.

Entorhinal Associational and Commissural System

The EC harbours an extensive, well-developed, yet largely underestimated network of intrinsic connections. There are three prevailing organizational principles that govern the overall organization. First, columnar-like projections emanate from layer Vb pyramidal cells distributing to the superficial layers Va, III-I (Fig. 5). This projection consists mainly of asymmetrical synapses (95%) which target presumed principal neurons and interneurons in almost equal proportions. Second, longitudinal connectivity prevails over transverse connectivity. The longitudinal projections that originate from a particular layer will preferentially innervate more superficial layers and they tend to be stronger from posterior (i.e. MEC) to anterior (i.e. LEC) than those that travel into the reverse direction. These longitudinal connections seem to originate preferentially from layers Vb and the calbindin-positive neurons in layer II. The transverse connections are much more restricted, and mostly confined to the layer of origin. Fairly strong homotopic commissural projections exist that terminate predominantly in layers I and II.

Connections of the Dentate Gyrus

Mossy Fibre Projections to Hilus and CA3

Dentate granule cells issue a massive projection of so-called mossy fibres to the entire transverse or proximodistal extent of CA3. Mossy fibres provide en passant presynaptic terminals that are unique with respect to size, anatomical complexity and the fact that they are correlated with likely complex postsynaptic specializations called thorny excrescences. On their way to field CA3, these fibres contact a fairly large cell type in the hilus called mossy cells. They also give rise to many small collaterals that target a wide variety of presumed interneurons in the hilus (Fig. 7).

Fig. 7

Wiring diagram, extended version of Fig. 4, illustrating the organization of the intrinsic connections of the HF. Indicated are the mossy fibre projections to the hilus and to the CA3, the return projection from proximal CA3 to the DG and the diminished contribution of proximal CA3 cells to the associate projection (stippled line). Also indicated is the proximodistal organization of the CA1-to-Sub projections as well as the calbindin-positive associative connection in the CA1. Finally, the intrinsic entorhinal connections from layer V to more superficial layers are indicated. (Note that additional associative networks in the CA1 and the Sub have not been indicated; see text for further details)

The projections from a single neuron or from a small group of neighbouring neurons distribute axons within a fairly limited longitudinal extent that hardly ever covers more than 400–500 μm and coincides with their level of origin. There is however a noticeable exception, in that mossy fibres abruptly change their course from an overall transverse orientation to a longitudinal one, once they reach the distal end of CA3. The extent of the longitudinal component depends on the dorsoventral level of origin in that granule cells at dorsal levels distribute mossy fibres ventrally for about 2 mm. The more ventral the origin, the less the longitudinal projection is developed such that granule cells at the ventral DG have little or no longitudinal component. The longitudinal component of the mossy fibre projection appears synaptically indifferent from the transverse component.

The anatomical organization of the mossy fibre projections along the transverse axis indicates that the influence exerted by granule cells on CA3 pyramidals depends on the position along the transverse axis of DG, since the proximal portion of CA3 is innervated preferentially by neurons in the exposed (infrapyramidal) blade, the crest and the adjacent portion of the enclosed (suprapyramidal) blade of the DG. The distal portion of CA3 receives mossy fibre input preferentially from granule cells in the enclosed blade of DG.

Current conceptions of CA3 as having a homogeneously wired architecture are incorrect or at least incomplete. Cells at different transverse positions receive inputs from cells in the DG that in turn are either different in their connectivity or functionally. In addition, at the most distal end of the dorsal part of CA3, a population of CA3 pyramidal cells most likely integrate inputs from the entire dorsal tip of the DG, a feature which is absent at proximal and mid-transverse levels as well as at ventral CA3 levels.

The DG Associational and Commissural System

The mossy cells in turn give rise to axons that bilaterally innervate the inner molecular layer of the DG, thus providing a powerful excitatory input to the proximal dendrites of the dentate granule cells. Interesting feature of this associational/commissural connection is that it may innervate as much of 65% of the long axis of the DG, but the innervation is weak at the level of origin and increases in density with increasing distance from the origin. Local hilar interneurons provide an inhibitory projection to the outer portions of the molecular layer, and this innervation is largely restricted to the level of origin, thus complimenting the excitatory associational system (see also section on “Neurons, Numbers and Connections”).

Connections of CA3

The CA3 to Dentate Projections

In contrast to the well-accepted view that projections within the hippocampal formation are largely if not exclusively unidirectional, implying that CA3 does not project to the DG, there is now substantial evidence to support such projections. These connections have not been described in the initial Golgi and subsequent tracing studies. However, intracellular filling consistently showed that pyramidal cells in the most proximal portion of CA3 embedded within the blades of the dentate granule cell layer issue collaterals that reach the hilar region (Fig. 7). Described as sparse, true at more dorsal levels of the hippocampal formation, at more ventral levels, CA3 neurons actually densely innervate the DG; not only the hilus but numerous CA3 axon collaterals also terminate in the most inner portions of the dentate molecular layer. The increase in density of projections to the DG at ventral levels goes hand in hand with a decreased contribution to the more traditionally known projections to CA1 (see below). Note that also GABAergic projections from CA3 to the DG have been reported.

The CA3 Associational/Commissural System

Local axon collaterals of CA3 axons make preferentially asymmetrical, thus most likely representing excitatory synapses, contacting dendrites of interneurons and importantly also spines of pyramidal cells, thus forming the strong autoassociative network considered to be the characteristic feature of the CA3 network (Fig. 7). The organization of the associational projections from CA3 to CA3 follows a few systematic principles that have been described essentially in two detailed tracing studies using either larger injections of anterogradely transported tracers or intracellular filling of individual CA3 pyramidal cells. Density and extent of local connectivity in CA3 is inversely related to the origin along the proximodistal axis. Irrespective of their dorsoventral position, CA3 pyramidals embedded within the extent of the DG that, as described contribute to the projection to the DG, do not seem to contribute much to the intrinsic associative system. The associative fibres that do emerge from these CA3 cells are restricted both along the proximodistal axis and along the longitudinal (dorsoventral) axis to the level of the parent cell(s). Cells with an increasingly more distal position in CA3 tend to exhibit increased associational axonal collaterals, extending several hundred microns anterior and posterior to the cell body but restricted along the transverse axis to the region of the parent cell body. The proximodistal origin also apparently relates to the radial distribution of the axons, such that proximal neurons preferentially project to stratum radiatum, whereas axons from increasingly more distal cells distribute more to stratum oriens. To further complicate the connectional matrix, the transverse-radial relation varies along the longitudinal axis.

Single pyramidal cells in CA3 not only distribute axonal branches ipsilaterally but also contralaterally. The detailed topography of the commissural connections has not been as thoroughly investigated as the ipsilateral connections, but it appears an image of the ipsilateral organization for both the projections to CA3 as to CA1 (see below). Also, the synaptic organization of both ipsilateral and commissural projections is quite similar. Note that species differences are present with respect to whether or not the commissural connections are present and if present, how they are organized with respect to their longitudinal and radial distribution.

The CA3 to CA1 System: Schaffer Collaterals

Comparable to the situation in CA3, the postsynaptic targets in CA1 for CA3 fibres comprise both interneurons and pyramidal cells. CA3 projections distribute in stratum radiatum and stratum oriens of CA1, whereas almost no fibres are present in the pyramidal cell layer (Fig. 7). Almost without exception, the longitudinal extent of the projections to CA1 is larger than that of the corresponding associative CA3 projections. Irrespective of the level of origin, projections do extent to levels both dorsal and ventral to the level of origin; however there is a preferential direction of the projections that relates to the transverse level of origin. Neurons with a proximal location, close to or inside the hilus, preferentially project to more dorsal levels, whereas more distal origins result in a shift to more ventral levels. Irrespective of the location of the neuron of origin though, the projections exhibit differences in radial distribution along the long axis of CA1. At more dorsal levels, collaterals tend to be located deeper in stratum radiatum and in stratum oriens, whereas at progressively more ventral levels, the fibres shift towards a more superficial position in stratum radiatum and less dense innervation in stratum oriens. This pattern is thus similar to that described above for the associative CA3-CA3 projections. The transverse position of the parent CA3 neuron does relate, at and around the level of origin to two other features. First, proximal projections tend to distribute somewhat more distally in CA1, and more distal CA3 cells project with some preference to more proximal portions of CA1. Furthermore, proximally originating projections terminate more superficially in stratum radiatum than distal projections, which distribute deeper in strata radiatum and oriens.

Connections of CA1

The CA1 to CA3 Projection

No excitatory projections from CA1 have been described that systematically target neurons in CA3. All the projections that run counter to the traditional unidirectional view apparently arise from a specific group of long-range GABAergic neurons that are prominently present in CA1. These neurons also provide projections of the DG, the EC and the lateral septum.

The CA1 Associational and Commissural System

Although much weaker than in CA3, there are recurrent connections in CA1. Anterograde tracing and intracellular filling date all consistently show that pyramidal cells in CA1 issue collaterals that distribute throughout strata oriens, pyramidale and radiatum of CA1. Of similar interest are reports on a narrow calbindin-positive bundle of fibres located at the exact border between lacunosum-moleculare and radiatum most likely emerging from calbindin-positive pyramidal cells in the distal part of CA (Fig. 7). The physiological nature and terminal distribution of any of these associational connections need further study before any functional inferences are intelligible.

The CA1 to Subiculum Projection

Principal cells in CA1 give rise to a strong projection to the Sub, terminating on proximal distal and apical dendrites of subicular pyramidal cells, not innervating the outer half of the molecular layer. Both intracellular fills and tracing studies have convincingly shown that this projection shows a marked topology along the transverse axis such that a cell or group of cells in the proximal one-third of CA1 project to the distal one-third of the Sub. Vice versa, cells in the distal CA1 will exclusively target cells in the proximal portion of the Sub, and cells in the centre of CA1 will reach cells in the centre of the Sub. Note that although a single cell will provide a set of axonal collaterals spanning about one-third of the transverse extent of the Sub, a cell with a slightly shifted position will also slightly shift its axonal pattern slightly in the opposite direction (Fig. 7).

The CA1 to Entorhinal Projection

The projection from all parts of the CA1 to the EC and the complex transverse and longitudinal topology have been dealt with already (see section “Projection from CA1 and Subiculum to Entorhinal Cortex”). Also the striking similarities with respect to these topologies with the reciprocal EC-to-CA1 projection have been mentioned (see also below in the next section on “Connections of the Subiculum”).

Connections of Subiculum

The Subiculum to CA1 Projection

According to at least two studies, neurons in the pyramidal cell layer of the Sub send axon collaterals into all layers of CA1. The origin of this projection includes superficial pyramidal cells and is likely to form both excitatory and inhibitory terminals on spines and dendritic shafts, respectively. Although no detailed information is available on spread along the transverse or longitudinal axes, the data indicate no marked transverse topography and a restricted longitudinal spread, comparable to the CA1 to CA3 projection.

The Subiculum Associational System

There are at least two types of pyramidal cell types in the Sub that both belong to the group of projection neurons. Both types, the so-called bursters and regular-spiking neurons, contribute to an extensive intrinsic innervation in the Sub. Intracellular filling of electrophysiologically identified bursting cells reveals an axonal distribution that remains within the region circumscribed by their apical dendrites. In contrast, the regular-spiking cells give rise to an axon that shows more widespread distribution along the transverse axis. Since these data have been generated in in vitro slices, it is not known whether similar differences exist with respect to a possible longitudinal spread. The longitudinal spread of the average population of neurons covers approximately 0.5–0.7 mm which is about 7% of the long axis.

The Subiculum to Entorhinal Projection

The projections from the Sub to the EC and the complex transverse and longitudinal topology have been described already (see section “Projection from CA1 and Subiculum to Entorhinal Cortex”). Also the striking similarities with the topological organization of the reciprocal EC-to-Sub projection have been referenced.

Neurons, Numbers and Connections

A number of estimates are available on how many neurons there are in the different areas of the HF and the EC as well as on total dendritic length and number of synapses leading to a number of published attempts addressing questions like how many cells converge on to a single cell and what is the level of divergence for a single cell axons. Although far from complete, in the following section, an attempt is made to summarize those data in rats (Table 1). Note that possible age and strain differences as well as methodological differences are not taken into account.

Table 1 Quantitative data on principal neurons in the HF and the EC

Numerical estimates have indicated that the population of granule cells may carry a total number of 4.6 × 10⁹ spines of which 77%, i.e. 3.542 × 10⁹, would belong to entorhinal synapses. Taken the total number of entorhinal layer II cells, each of them could potentially contact 32,200 spines. If we would know how many entorhinal inputs target a single granule cell, we would be able to estimate how many granule cells would be innervated by a single layer II cell in the EC, i.e. we should have a numerical estimate of the divergence of this connection. By using published estimations of the number of spines on granular cell dendrites (4600), we could estimate that each granule cell can receive input from maximally 0.77 × 4600 is 3542 cells in EC or 3542/110.000 is 3.2% of the total layer II population (based on Amaral et al. 1990). Using comparable lines of reasoning, it has been inferred that a single mossy fibre can make as many as 37 synaptic contacts with dendrites of a single CA3 pyramidal cell, a single granule cell may innervate 15 CA3 pyramidal cells, and a single CA3 pyramidal cell may receive convergent input from 72 granule cells. A single CA3 neuron might be innervated by 6000 other CA3 neurons, and a single CA1 cell receives input from 5500 CA3 cells. Details on the Sub and the EC are currently lacking. A final word of caution would be in place since all these numerical estimates assume homogeneity of the network, which most likely will turn out to be a false assumption. For example, it is known that the absolute numbers as well as the percentages of the total population of principal cells and interneurons vary along the long axis of the hippocampus. Also differences in numbers of neurons are obvious for the LEC versus the MEC (Table 1).

A complementary approach would be to look at the overall distribution of the individual connections that make up the region of interest. A single entorhinal neuron may distribute its axon along approximately 25% of the long axis of the HF. It has been estimated that in adult animals, this axis extends for up to 10 mm, so a single axon targets 2.5 mm of the length of the HF. Axons from granular cells are fairly limited in their longitudinal distribution, extending for about 400 μm in CA3c-b but up to 1.5 mm in CA3a. The associational projection from the hilus back into the inner molecular layer extends over 6.5 mm, exhibiting a dramatic drop in density around the level of its origin. Note that some other hilar projections, such as those originating from somatostatin-positive interneurons to the outer portions of the molecular layer, fill that gap. The subsequent projection from CA3 to CA1 shows a longitudinal extent similar to that of the DG association system, whereas the autoassociative connections are slightly more restricted. The projections from a single cell in CA1 to the Sub extend up to 2 mm along the long axis forming a slab-like innervated strip. The associative connections within CA1 apparently are rather restricted along the long axis, whereas currently no data are available for the associational connections in the Sub, be it that they at least extend for 400 μm. Finally the projections from CA1 and the Sub to the EC cover a narrow strip in EC that extends for at least along 60% of the longitudinal extent of either the LEC or the MEC, depending on whether the injection is in proximal or distal part of the Sub, respectively.

Experimental Techniques

Most of what is known today about the pathways that connect neurons in different brain regions has been discovered by using neuroanatomical tract-tracing techniques. Tracers are molecules that are either applied extracellularly or intracellularly. In case of extracellular application, the tracer is taken up by neurons at the injection site and transported or diffused within cells. A tracer substance can be transported anterogradely (e.g. Phaseolus vulgaris leucoagglutinin), from the soma towards the axon terminals, retrogradely (e.g. Fast Blue) and from the axon terminals towards the soma, or it can be transported in both directions (e.g. horseradish peroxidase). In case of intracellular application, both autofluorescent dyes (e.g. Lucifer yellow, Alexa dyes) and biotin-conjugated dyes are most often used, since they can be easily visualized for fluorescent or transmitted light microscopy (LM). All these methods can be analysed using a variety of microscopical techniques, including to some extent electron microscopy (EM). In the latter case, one quite often combines them with lesions. Small lesions (mechanical, toxins, electrolytic) will result in local degeneration of axon terminals that show up as electron dense material in the EM.

Standard light and confocal techniques, when applied to extracellular tracer deposits, allow the visualization of distribution patterns, including laminar distribution, topologies as well as the identification of likely synaptic relationships. They are poor with respect to quantitative resolution since it is very difficult to control or estimate the number of neurons that take up and transport the tracer. A much more reliable but very time-consuming method is the intracellular filling of single neurons in vivo and the subsequent complete reconstruction of its dendritic and axonal arborizations. This technique can be combined with anterograde or retrograde tracing to identify projection targets and synaptic inputs. A recently added tool is to make use of retrograde labelling with genetically modified viruses that carry the genes for certain fluorescent proteins such that infected cells express the protein throughout their dendritic and sometimes even axonal arborizations. The viral toolkit has expanded enormously and now comprises a number of ways to selectively visualize monosynaptic inputs to identified neurons both at the population and single cell level.

Another powerful approach is to use in vitro slice electrophysiology combined with viral expression of light-sensitive channels. This facilitates the analysis of postsynaptic targets of distant inputs, i.e. those that are difficult, if not impossible to maintain in a slice preparation.

Electron microscopy can be used to visualize whether a presynaptic axon contacts a postsynaptic identified neuron. This is a very accurate but time-consuming method because only small pieces of tissue can be examined at one time. A promising development may be the use of automated systems to do serial reconstructions, at the EM level but also at the LM level, but in all instances, our limited understanding of the mechanisms underlying labelling and transport of tracers seriously hampers our aims to generate quantitative data. The only exception is the high standard of unbiased methods to count number of cells, synapses and actually any identifiable element in the nervous system, using stereological approaches. But even when applying such sophisticated methods, one has to be aware of differences between strains, effects of age, environment and gender on quantitative estimates. A recent and promising addition is the use of serial EM techniques which have now provided the first very large quantitative datasets on connectivity of small volumes of brain, for example, a recent reconstruction of part of the entorhinal cortex (Schmidt et al. 2017, Nature 549:469).

The Future: Open Questions and How to Address Them

Many conceptual or theoretical accounts and modelling attempts use a rather simple and generalized representation of what we actually know as their starting point. This may eventually lead to disuse of available data, such that these data eventually will be forgotten. Our attempts to understand structure needs to take into account all the subtle differences in topology and densities of projections, the many parallel pathways that are so characteristic of the brain and the many different levels of integration that may occur within the different networks that constitute the HF. It will become relevant to integrate our current insights in the connectional organization of the hippocampus into a new functional framework. To give one example, the current traditional view of the hippocampal network as a unidirectional, sequential series of connections does not credit the wealth of data on parallel EC inputs to all components of HF and the well-established backprojections in the system. The addition of CA2 to the network provides a further complicating factor.

During the last decade, we have learned a lot about EC, the types of principal neurons and their specific connectivity. In addition to data on layer II, more recently data on layer V have become available. Data on layer III are still very sparse as are data on intralaminar interactions. By adding the complexity of the very many specific types of interneurons, our task to describe, to model and to understand EC is still a major challenge. The combination of anatomical and electrophysiological studies, with the use of promising new genetic tools and computational modelling, will provide the foundation for further detailed functional studies in freely behaving animals, which in turn form the ground work to understand the human hippocampus, both when it is healthy and when it starts to break down, as seen in ageing and in several neurodegenerative diseases.