Network properties of the
amygdala-hippocampus connectome
Supervisors: Natalie Cappaert, Niels van Strien
Nynke Boiten
BA-thesis Biomedical Sciences
Word count: 6079
Network properties of the amygdala-hippocampus connectome
1. Abstract
Connectomics is a subfield within neuroscience with the aim of understanding the “architecture of nervous system connectivity in all animals at all resolutions” (Saleeba et al., 2019). The field of connectomics has
profited from increasingly refined ways to visualize neuronal connections in the brain. One of these methods of visualization is tract tracing, in which a tracer substance is injected into a given brain region to demonstrate how that region is interconnected. Amygdalohippocampal interactions are known to influence both the encoding and consolidation of emotional memories (Phelps, 2004), but little is known on how these features are linked to the connectivity of the amygdalohippocampal network. In this study, tract tracing literature on amygdalohippocampal connections in the rat were collected and these connections were entered into an online database. Subsequently, these connections were extracted and a network was constructed using the brain connectivity toolbox developed by Rubinov and Sporns (2010). Multiple global and local network features were calculated and the role of different brain structures participating in the network was assessed. This study shows that the amygdalohippocampal network has a small world property. Furthermore, the lateral entorhinal area, subiculum, CA1 and amygdalohippocampal area were identified as putative hubs of the network. This supports the notion of the lateral entorhinal area being an important structure in amygdalohippocampal communication and its role as an intermediary in the enhancement of memory formation.
2. Introduction
Connectomics is a subfield within neuroscience with the aim of understanding the “architecture of nervous system connectivity in all animals at all resolutions” (Saleeba et al., 2019). Connectomics uses a graph
theoretical approach to construct a connectome, a representation of the connections between the brain structures under study. The level of analysis can range from the microscopic (single cells and synapses) to the
mesoscopic (groups of cells and circuits) or macroscopic (brain-wide) level (Saleeba et al., 2019; Schmitt et al., 2012a). The field of connectomics has profited from increasingly refined ways to visualize neuronal connections in the brain. One such method is tract tracing, in which a tracer substance is injected into a given brain region to visualize how that region is
interconnected. Since the origins of tract-tracing in the 50s and 60s progressively more selective tract tracing techniques and methods of tracer visualization have emerged (Lanciego and Wouterlood, 2011). This development has supported the construction of an increasingly clear picture of brain connectivity. Since tract-tracing data provides information at the scale of multiple cells or cell circuits, the resulting connectome gives a mesoscopic view of brain connectivity.
One brain network of great interest is the rat amygdalohippocampal network. This network includes the hippocampal formation (HF) and the bordering parahippocampal region (PHR) that together form the HF-PHR network (van Strien et al., 2009). The anatomical position of the HF and HF-PHR and the separate anatomical structures that form them are visualized in figure 1. Functions associated with the HF include spatial navigation
and memory formation (van Strien et al., 2009).
Figure 1. View of the HF and PHR as they are located in the brain (A), and visible in horizontal (Ba,Bb) and coronal sections (Bc,Bd). C shows an enlarged and more detailed view on section Bb. The rat hippocampus is a cortical structure stretching from the dorsal anteromedial part of the brain to the ventral posterolateral part in a C-shape. This long axis of HF is called the septotemporal axis (Cappaert et al., 2015). In two-dimensional coronal sections the hippocampus becomes visible as two wings below the corpus callosum that slowly grow into C-shaped structures when viewed in an anterior to posterior fashion. In horizontal sections the HF can be seen to spiral from proximal to distal, starting at the dentate gyrus (DG), followed by the hippocampal regions CA3 to CA1 before arriving at the subiculum (Sub). The PHR borders the subiculum and is generally divided into five subregions: presubiculum (PreS), parasubiculum (PaS), Entorhinal cortex (EC) consisting of the medial entorhinal area (MEA) and the lateral entorhinal area (LEA), the perirhinal cortex (PER) encompassing A35 and A36 (also known as BA35/BA36) and the postrhinal (POR) cortex (fig. 1). Van Strien et al. (2009)
The other structure involved in the network is the amygdala which is associated with memory, attention and the interpretation of the emotional significance of sensory stimuli (Pitkänen et al., 1997). The amygdala or
that are positioned in the anterior part of the medial temporal lobe. These nuclei and further amygdaloid regions are represented in figure 2. The amygdala is densely interconnected with the HF and PHR, with distinct amygdaloid nuclei projecting to distinct hippocampal subregions (McDonald and Mott, 2017). Understanding this complex network of amygdalohippocampal interconnections is essential for a detailed view of rodent brain connectivity, as well as uncovering the way in which information flows between these regions are
associated with the known functions of these regions.
Figure 2. The amygdaloid complex of the rat consisting of 13 divisions which can further be divided into subdivisions. In short, the deep nuclei include the lateral nucleus (L), basal nucleus (B) and accessory basal nucleus (AB). The superficial nuclei consist of the nucleus of the lateral olfactory tract (NLOT), the bed nucleus of the accessory olfactory tract (BAOT), anterior cortical nucleus (COa), medial nucleus (M), the periamygdaloid complex (PAC) and the posterior cortical nucleus. The remaining amygdaloid areas are the anterior amygdaloid area (AAA), central nucleus (CE), the amygdalohippocampal area (AHA) and the intercalated nuclei (I) (Pitkänen et al., 2006; Schmitt et al., 2012b). Some of these nuclei are frequently subdivided into separate divisions. Figure by Pitkänen et al. 1997
Amygdalohippocampal interactions are known to influence both the encoding and consolidation of emotional memories (Phelps, 2004). Previous studies demonstrate that the basolateral amygdala modulates memory consolidation of emotionally arousing events via projections to the hippocampus. Moreover, functional imaging studies in humans indicate that interactions between the basolateral amygdala and the entorhinal cortex, as well as anterior parts of the hippocampus, are involved in the formation of emotional memories (McDonald and Mott, 2017). One proposed mechanism is that the basolateral amygdala modulates memory consolidation through the enhancement of long-term potentiation (LTP) in the dorsal dentate gyrus via polysynaptic pathways, the most likely intermediary of this indirect modulatory pathway being the lateral entorhinal cortex (LEA) (McDonald and Mott, 2017). Although it is evident that amygdalohippocampal interconnections play a key role in the formation of emotional memories, it is still unclear what exact features the network possesses and how these are linked to different informational processes involved in network-associated functions, e.g. the consolidation of memories of emotional events.
One of the network properties that has gained popularity within the neuroscience community is the small-network property. Networks with this property, also called small-world networks, are associated with high speed of transmission and optimal communication efficiency (Papo et al., 2016). Since the discovery that the complete connectome of C. elegans is a small-world network, small-world properties have been implicated for cat and macaque connectomes as well as for structural and functional connectomes of the human brain (Bassett and Bullmore, 2017; Schmitt et al., 2012a). Small-worldness appears to be a nearly universal property
of nervous systems that informs us about e.g. the integration of information or the management of energy costs (Bassett and Bullmore, 2017).
To study amygdalohippocampal connectivity in the rat, graph analysis was performed on connectivity data extracted from tract-tracing data from the existing literature. Data collection encompassed creating an inventory of the relevant literature, extracting the anatomical connections from the literature and entering these into an online database. Subsequently, network analysis was performed on the collected connectivity data. This paper address whether the amygdalohippocampal network is a small-world network and what that means in the light of small-worldness. Additionally, other network features, such as the presence and location of highly connected nodes or ‘hubs’, will be assessed to study the role of separate brain structures involved in the network.
3. Materials and Methods
A brief overview of the steps taken for the network analysis is presented in Figure 3.
Figure 3. Network analysis workflow. Literature search was centred around tract-tracing experiments performed on rat species and featuring injections or observed labelling in the HF, PHR and/or amygdaloid complex. The database was constructed according to the criteria described on the website: https://dev.temporal-lobe.com/.
3.1. Literature search
The literature from which the connections were extracted was largely provided by the supervisor and selected on the basis of relevance for the connectome. A paper was said to be relevant if:
a. its experiment involved non-transsynaptic neuroanatomical tract tracing b. a rat species was used as an animal model
c. results of healthy (control) cases were mentioned in the report
d. some, but not necessarily all of the connections fell within the scope of the amygdalohippocampal network. This included the aforementioned areas of the HF, PHR and amygdaloid complex
literature search
database
construction
extraction of
connections
(15915 entries from 132 publications)pre-analysis
(10757 entries from 132 publications)graph analysis
using the network
No limits were set on the date of publication.
3.2. Data extraction and database construction
For a connection to the included in the database the following criteria had to be met: a. the connection was reported in a healthy animal or healthy control case
b. both the region of injection and region of the observed termination had to lie within the previously defined scope
c. the existence or non-existence of a connection had to be explicitly mentioned or could be derived from the text (noted as an implicit connection)
To give an idea of how the information was entered into the database and structured, some exemplary entries are highlighted in figure 4. All connections from the database recorded in entries like these (figure 4) were downloaded into an excel file. The resulting file contained 15915 extracted connections, based on 132 publications.
Figure 4. screenshot showing some exemplary database entries. Every connection entry includes a publication ID, experiment ID, the brain structure names of origin and termination, terms of anatomical position within that structure, the given layer, an
injection/projection rating giving the quality of an injection/projection and illustration indicating where evidence for the connection is given within the publication; for every connection the category ‘explicit’ notes whether a connection was explicitly mentioned by the author (0 = implicit, 1 = explicit) and ‘exists’ indicates whether a connection exists ( 0 = no , 1 = yes). The screenshot was taken on https://dev.temporal-lobe.com/wiredbrain/data-entry
3.3. Pre-analysis
The extracted connections were filtered in Excel. Connections meeting the following properties were retained: a. ipsilaterality (ipsi, left-left and right-right)
b. data was obtained from rat species
c. the tracer direction was unidirectional (retrograde or anterograde) d. the connection was explicitly mentioned in the paper
e. injection ratings were higher than 2 and projection ratings were higher than 3
The decision to only include connections that were explicitly mentioned by the author and connections with higher quality ratings was made to ensure quality control. Applying these exclusion criteria yielded 11297 records of amygdalohippocampal connections.
Another pre-processing step ensured that all connections were entered with the site of origin first and site of termination second. Anterograde tracing involves an injection at the origin of a projection (at the cell bodies) after which the labelling substance can be observed at sites where the labelled projection terminates. In contrast, retrograde tracing involves an injection at the site of termination enabling tracing the back the projection to its site of origin. Because both connections recorded with anterograde and retrograde tracing methods were entered site of injection first and site of projection second, the connections recorded with retrograde tracing had to be ‘flipped’. This way, both the anterograde and retrograde connections were oriented from their structure of origin to their structure of termination.
Since the remaining connections were described in different nomenclatures, the brain structure names were standardized according to the nomenclature used by van Strien et al. (2009) for the HF and PHR, and Pitkänen et al. (1997) for the amygdaloid complex. Border structures, such as CA1/CA2 were split into the two encompassing structures resulting in a connection with CA1 and one with CA2. In the case that only labelling at the border was observed, one structure was mentioned with ‘near ….’ Or ‘bordering ….’ as a specification of the position. During standardization, connections mentioning structures that were termed too global, such as ‘hippocampus’ or ‘amygdaloid complex’ without further specification, wereremoved. Similarly, structures that could not be strictly translated to the standard nomenclature were removed, for instance because the original structure name encompassed two distinct structures within this standard nomenclature. The terms denoting the anatomical position within a brain structure were also standardized so that only terms specifying the placement within the rostro-caudal, dorso-ventral and latero-medial axes were used.
Consequently, every description of anatomical position using the septo-temporal axis or proximo-distal axis was translated to the three aforementioned axes using atlas figures of the given brain areas, e.g. fig. 1. In total, these steps yielded a total of 10757 entries with six columns of information on the structure of origin
(structure, division, three axes indicating the anatomical position and layer) and structure of termination. Because information on the layers involved in the connection and the precise anatomical position were often missing, only the information on the anatomical structure and division of that structure were included in the network analysis.
3.4. Network analysis
Network analysis was performed with MATLAB 2020b using the networks analysis toolkit developed by Rubinov and Sporns (Rubinov and Sporns, 2010). In network analysis a brain network is represented by a graph in which the nodes stand for brain structures and the edges represent connections between them. A script was written to create a MATLAB structure with a list of unique brain structure names and a connectivity matrix. The connectivity matrix was constructed by writing a function called Conmat (names, connections), which takes a list of unique brain structure names and unique connections to create a binary directed connectivity matrix. This means that the recorded connections are unweighted and that their direction is retained. The resulting MATLAB structure was used as input for the subsequent network analysis. The following network measures were calculated from the amygdalohippocampal network: node degrees, global and local clustering, network
density, assortativity, distances and average path length, global and local efficiency, as well as edge and node betweenness. In the next section these measures will be described in more detail.
The most fundamental network measure is the node degree of a given node, which indicates the number of edges it participates in (Bullmore and Sporns, 2009). All node degrees of nodes within the network form a degree distribution which can be visualized as a histogram. All node degrees in a random network are equally probable resulting in a Gaussian distribution of node degrees, whereas non-random networks are usually non-Gaussian and have a long tail towards higher degrees (Bullmore and Sporns, 2009). The assortativity measure is a correlation coefficient between the degrees of any two nodes sharing an edge. Therefore, a positive assortativity measure shows that nodes tend to be connected to nodes with a same or similar degree (Rubinov and Sporns, 2010). Since the network in this study is a directed network, in- and outdegrees were calculated indicating the number of edges going to and away from a structure, respectively. In and out degrees can be seen to represent the number of input and output connections of a given brain structure. A joint degree matrix was calculated to visualize the number of nodes with specific indegree and outdegree measures.
Network density is the ratio of the actual number of edges versus the number of all possible edges for the network. It can be seen as the simplest estimator for the physical cost of a network (Bullmore and Sporns, 2009). The global clustering coefficient measures the degree of clustering in the network and is calculated as the average of all local clustering coefficients. Local clustering coefficients measure the probability of the two neighbours j and k of a given node i being directly connected to each other (Bassett and Bullmore, 2017). A shortest path or shortest distance is a purely topological measure demonstrating the minimal number of nodes separating any two nodes (Bullmore and Sporns, 2009). Topological here means that the shortest path measure does not represent actual anatomical distances between regions but only distances in terms of the
arrangement of nodes within the network. The average path length gives the average of all shortest paths between the network. As an alternative to the average path length measure, the global efficiency measure was calculated as the average of all local efficiencies, which is the inverse of a shortest path.
To estimate the importance of specific nodes and edges for the network, node betweenness and edge betweenness centrality measures were calculated. Node betweenness quantifies the number of shortest paths running through a given node, while edge betweenness quantifies the number of shortest paths traversing a given edge. Nodes with high node betweenness centrality are important for efficient communication within the network (Bullmore & Sporns, 2009). Such nodes with high centrality or high node degrees are called hubs of the network. Consequently, node betweenness and total node degree measures were used to identify putative hubs of the network.
Small-worldness was assessed by comparing averaged network measures of 1000 random networks to the network measures of the amygdalohippocampal network. This was achieved by randomizing the directed network 1000 times while preserving the node degree distribution and maintaining connectedness of the network. Small-world networks tend to have high global clustering and a short average path length (Bassett and Bullmore, 2017; Papo et al., 2016). These aspects are generally associated with a trade-off between local connectivity and segregation, the formation of separate highly interconnected clusters, and global network
integration, a capacity for the transmission of information within the whole network (Vecchio et al., 2017). Subsequently, global clustering measures and average path lengths were calculated across all random networks, averaged and compared to the same measures for the amygdalohippocampal network. The following formula was employed to calculate small-worldness measure:
S=(C /Crand )/(PD /PDrand)
where S is the small worldness measure, C and PD and Crand and PDrand are the global clustering coefficients and average path lengths of the network and the random networks respectively (Bassett and Bullmore, 2017).
4. Results
4.1. Global parameters
The resulting network consists of 59 nodes and 793 edges. The connections are summarized in the adjacency matrix (Fig. 3). The density amounts to 0.2278 for the present network. The assortativity is -0.2439 meaning that nodes tend to form edges with nodes of a dissimilar degree. This hints at the existence of more central, heavily connected nodes and more peripheral, sparsely connected nodes. The global efficiency is 0.4861. On average each node can be reached via 1.9068 edges (characteristic path length). The global clustering measure is 0.7583 and represents the mean of the local clustering coefficients (table 1). Both the average path length and clustering coefficient were found to be higher than the mean of 1000 connected random networks with the same number of networks and nodes. These values result in a small-worldness parameter of 1.4015 which shows that the amygdalohippocampal network is a small-world network.
4.2. Network
The adjacency matrix shows dense clusters of interconnections between the HF/PHR and the accessory basal, basal and lateral nuclei, as well as all subdivisions of the PAC (figure 5). Looking at connections within the amygdala, most clusters seem to involve connections having the central nucleus and the PAC as structures of origin, whereas the HF and PHR seem to be more densely interconnected without separate structures emerging as the origin for wider spread connections to almost all other structures within the HF/PHR. Connections between the amygdala and the hippocampus will be discussed in more detail in the following section.
4.3. Amygdalohippocampal interconnectivity
Out of all HF areas, only the subiculum, CA1, and CA2 target the amygdaloid complex, with the subiculum being the most interconnected, followed by the CA1 (figure 5, table1). The subiculum projects extensively to the amygdala, including the accessory basal nucleus, the cortical nuclei, the basal nucleus, the central nucleus, the intercalated nuclei, the medial lateral nucleus, the medial nucleus, and the nucleus of the lateral olfactory tract (figure 5). The CA1 does not target the cortical nuclei, intra-amygdalar division of the bed nucleus of the stria terminalis or anterior amygdaloid area. Looking at amygdala-to-HF projections, six amygdaloid areas were found to project to the whole extent of the HF: 1) the accessory basal nucleus, 2) the amygdalohippocampal
area, 3) the basal nucleus, 4) the lateral nucleus, 5) the periamygdaloid cortex, and 6) the posterior cortical nucleus. Out of these, all amygdaloid areas except for the amygdalohippocampal area project to CA2.
The PHR is also heavily interconnected with the amygdaloid complex (figure 5). The areas connected to multiple amygdalar regions include the lateral and medial entorhinal areas, as well as the perirhinal cortices 35 and 36. The LEA is the most interconnected with a node degree of 68, followed by the medial entorhinal area (table1). Both the entorhinal cortices project to the accessory basal nucleus, the basal nucleus, the central nucleus, the lateral nucleus, the amygdalohippocampal area, olfactory amygdala, and the periamygdaloid cortex. While the LEA and MEA have similar indegrees, 36 and 37 respectively, the LEA has a much higher outdegree (OD = 36) than the MEA (OD = 16) mostly because the LEA projects to many more amygdaloid areas than the MEA (figure 5, table 1). Almost all amygdaloid areas project back to the PHR, most notably the MEA, LEA and the perirhinal cortex.
Figure 5. Binary directed connectivity matrix of 59 HF, PHR and amygdaloid areas as well as their subdivisions. The edges represented by the black squares run from the structures listed along the y-axis to the structures listed on the x-axis on top. Colours are used to represent the more global brain regions these structures belong to: hippocampal formation (green), parahippocampal region (red) and
amygdaloid complex (purple).
4.4. Local parameters
All local network parameters, except for edge betweenness and path distances, are listed in table 1. The edge betweenness and path distance measures are visualized using colour-coded matrices (Fig.11 and 10). The LEA is connected with the highest number of other nodes (n = 68), followed by the amygdalohippocampal area (n=67) and the subiculum (n=63). The amygdalohippocampal area targets a high number of other areas (OD=43) compared to the amount of afferent connections it receives (ID = 24) (figure 5, table 1). Other structures demonstrating higher OD values than ID values are the parvicellular division of the accessory basal nucleus, the anterior cortical nucleus, all the subdivisions of the PAC, and the cortical amygdaloid nucleus (figure 5). In contrast, most notably the DG, CA1 and CA2, the PER, POR, the anterior amygdaloid area, the intercalated nuclei, and all subdivisions of the medial nucleus have a higher number of input than output connections (table 1 and fig 7). The node degree distribution shows a wide spread of total node frequencies indicating that there are few nodes with very little and few nodes with many connections, which hints at the existence of highly interconnected nodes or ‘hubs’.
Figure 7. Bar chart highlighting the difference between in- (ID) and outdegrees (OD) per structure (min = -25 (PACm)), max = 21 (MEA))
Figure 8. Node degree distribution of total node degrees of 59 nodes and 793 edges
Short path distances (1-2) are observed between all nodes of the HF-PHR region with the exception of the perirhinal cortex. Moreover, the accessory basal, basal, central and lateral nuclei, and all PAC subfields demonstrate short distances to most of the HF-PHR regions. In the other direction, most HF-PHR regions have short path distances to the
amygdalohippocampal,
amygdalostriatal and amygdalopiriform areas, the anterior cortical nucleus, all divisions of the basal nucleus, the capsular
Figure 9. Joint degree matrix giving the number of edges between nodes of indegree n and outdegree m for every pair (n, m)
and medial divisions of the central nucleus, all divisions of the medial nucleus, as well as the olfactory amygdala.
The structure with the highest number of shortest paths running through it is the
amygdalohippocampal area (BC = 521.071300), followed by the LEA (BC= 370.682118), the subiculum (BC = 293.748897), and CA1 (BC = 202.276861) (tab. 1, fig. 5). These are also the four structures with the most edges (figure 6). 16 structures have a node betweenness centrality of 0 and have therefore no shortest paths passing through them. Among these, the medial amygdalohippocampal area, amygdalostriatal zone, all divisions of the bed nucleus of the stria terminalis, piriform amygdaloid area, and postpiriform transition area have a node degree of 2 or lower (tab. 1), meaning they are sparsely interconnected and are peripheral structures within the network.
Figure 10. Distance matrix indicating the shortest path between two areas. The y-axis shows structures of origin, whereas the x-axis shows structures of termination. Blue stands for shorter distances between two areas while yellow stands for relatively longer
distances.
Figure 12. Edge betweenness of the edges between 59 areas. The y-axis shows structures of origin, whereas the x-axis shows structures of termination. Yellow hues indicate a higher edge betweenness measure, thus a high number of shortest paths running through the given edge. In contrast, blue indicates a low edge betweenness measure.
Looking at edges between amygdalar areas and structures of the HF-PHR network the following edges projecting to the amygdaloid areas show high edge betweenness centrality: CA2-to-medial central amygdaloid nucleus, CA1-to-amygdalohippocampal area, CA1-to-postpiriform transition area,
subiculum-to-amygdalohippocampal area (AHA), subiculum-to-capsular central nucleus, subiculum-to-piriform amygdaloid area, medial entorhinal area (MEA)-to-medial amygdalohippocampal area, MEA-to-accessory basal nucleus, lateral entorhinal area (LEA)-to-medial central nucleus, LEA-to-posterior cortical nucleus, and perirhinal cortex 36-to-accessory basal nucleus (figure 12). In the direction amygdaloid area to HF-PHR less edges with
noticeably high edge betweenness centrality are observed: AHA-to-presubiculum, amygdalopirform area-to-LEA, central nucleus-to-area-to-LEA, and rostral medial nucleus -to-MEA. Thus, CA1, subiculum, area-to-LEA, and AHA do not only have the highest node centralities (figure 11), but also participate in the edges with the highest edge centralities, demonstrating their importance for the network.
5. Discussion
5.1. Amygdalohippocampal interconnectivity
HF projections to the amygdaloid complex involve the subiculum, CA1, and in most cases CA2. This supports previous findings in which the CA1 and subiculum were shown to provide the most substantial input to the amygdala (Pitkänen et al., 2006). Contrary to previous findings from McDonald and Mott (2017), no connections to the divisions of the medial nucleus were present (McDonald and Mott, 2017). However, it is important to note here that the lack of an observed connection between the CA1 and the medial nucleus in the data set does not necessarily mean that no connection exists, since the connections extracted from the database often did not mention the subdivisions of the amygdalar areas. This is also apparent in the low total degrees of the amygdalar nuclei subdivisions, especially in the case of the medial nucleus.
In contrast to the subiculum and CA1, the CA2 is often not studied as a separate structure in the tract tracing literature resulting in little to no mention of CA2 connections and neglect of the structure’s function, possibly because of its small size. This might account for the fact that often no CA2-to-amygdala connections were reported, leading to the conclusion that the CA2 only receives input from the amygdala but does not project to the amygdala (McDonald and Mott, 2017). However, the connectivity data in this study shows that the CA2 targets multiple amygdaloid areas, namely the basal amygdaloid nucleus, the medial division of the central amygdaloid nucleus, as well as the periamygdaloid cortex. Shedding light on CA2 connectivity and function might support the relevance of it being studied. The CA2 has recently been associated with social memory by Hitti and Siegelbaum (2014), who showed that targeted inactivation of CA2 pyramidal neurons in transgenic mice led to loss of social memory with no loss of hippocampus-related memory functions, such as spatial or contextual memory. It would be interesting to see whether and how social memory associated with the CA2, e.g. memory of a previous social encounter, is mediated by connections within the
amygdalohippocampal network.
The amygdaloid innervation of HF subfields might be less regionally segregated than previously thought. While Pitkänen et al. (2006) report single connections from the basal nucleus of the amygdala to the CA2 and CA3 regions, this network shows the existence of connections to CA2 and CA3 originating in multiple
amygdaloid regions, including the accessory basal nucleus, basal nucleus, lateral nucleus, and periamygdaloid cortex. This finding supports the view that the CA2 and CA3 integrate information from multiple amygdaloid sources, which in turn receive information from other areas, such as the HF, frontal cortex, and sensory areas (Pitkänen et al., 1997). Moreover, the basal-nucleus-to-CA2 connection appears to be reciprocal, which means that input from the HF to the basal nucleus might be regulated via a connection back to the CA2.
5.2. Identification of putative hubs and edges crucial for network functionality
Within the amygdalohippocampal network the LEA has the highest number of connections and the highest node betweenness, indicating that the it might be a connector hub of the network. This view is supported by
the LEA being the HF-PHR structure that is the most interconnected with the amygdaloid area, harbouring both efferent projections to and afferent projections from the amygdala (McDonald and Mott, 2017). Furthermore, the LEA is vastly interconnected with other structures of the HF-PHR network, having reciprocal connections to all HF areas, as well as the MEA and both perirhinal cortices. This has multiple consequences. First, this reciprocal connection with the MEA supports a notion of the EC not as a mere input-output structure, but as a structure involved in complex associations that already take place at the EC level, as put forward by van Strien et al. (2009). Second, the LEA also participates in some of the edges with the highest edge betweenness, including connections to the medial central nucleus, posterior cortical nucleus and connections from the amygdalopirform area and the central nucleus. Given that the posterior cortical nucleus projects to most amygdaloid areas, this route might be an important structure for hippocampal input to the amygdaloid complex. Moreover, not only does the central nucleus receive substantial projections from the hippocampal and parahippocampal areas, such as the LEA, but also projects back to the LEA. Given that the LEA is one of the hippocampal structures with most connections to the amygdala, and has extensive intra-hippocampal
connections, the central nucleus might regulate HF input to itself and other connected amygdaloid areas via its connection to the LEA.
The centrality of both the subiculum and the CA1 in the network is supported by their high node betweenness and node degree measures reinforcing that the subiculum and CA1 are hubs of the network. The central role of the subiculum and CA1 is also corroborated by these structures being the ‘endpoint’ of the polysynaptic pathway originating in the DG (van Strien et al., 2009). That the subiculum and CA1 are network hubs is unsurprising, since they are known to provide the main part of the HF output to the PHR (van Strien et al., 2009), as well as being the main HF structures directly projecting to the amygdala (McDonald and Mott, 2017). Thus, they provide information already processed by the polysynaptic pathway.
One of the amygdaloid targets of the CA1 and subiculum is the amygdalohippocampal area. The high outdegree of the amygdalohippocampal area suggests that it plays a great role in information distribution, receiving additional input from the LEA, accessory basal nucleus, central nucleus, anterior cortical nucleus and the PAC and projecting further to most structures of the HF, PHR, and all amygdaloid nuclei. In conclusion, the node betweenness centrality and node degree measures for structures of the amygdalohippocampal network suggests that the subiculum, LEA, amygdalohippocampal area, and CA1 are hubs for the integration of information.
The central position of the LEA within the amygdalohippocampal network, both extensively connected to HF, as well as amygdaloid regions also puts it in an ideal place to mediate memory enhancement in between the basolateral amygdala and the DG. Functional imaging studies in humans have shown that formation of emotional memories involve interactions between the basal nucleus, EC, and the anterior hippocampus (McDonald and Mott, 2017). Another finding supporting this mechanism of memory enhancement is the position of the basal nucleus within the amygdalohippocampal network. On the one hand, the basal nucleus receives input almost exclusively from the central nucleus and periamygdaloid nucleus, with the central nucleus possibly representing a later stage in information processing in which information from different
other hand, the basal nucleus projects almost exclusively to the HF-PHR network suggesting a tight-knit interaction with this network. As an alternative to the indirect basal-nucleus-to-LEA-to-DG connection, a direct connection between the basal nucleus and the DG seems to be present, which is not mentioned by McDonald and Mott (2017). However, their finding that no direct projection to the dorsal part of DG exists can still be true, since these results do not specify to which part of the DG the basal nucleus projects. If it turns out that the basal amygdala does project to the dorsal dentate gyrus, that could open up the possibility of the basal amygdala directly influencing LTP in the dentate gyrus. Then the question arises whether LTP enhancement requires the indirect route via the LEA and if so, what role the LEA plays in the basal nucleus-to-dentate gyrus pathway.
5.3. Methodological considerations
Although a large number of tract-tracing papers have already been included in the temporal lobe database, it is important to note that the database does not yet fully cover all of the available tract-tracing literature of (para-)hippocampal and amygdaloid connections. Especially connections including solely amygdaloid regions were less than half in number compared to connections involving the HF and PHR regions even though the number of amygdalar structures was higher (out of 10757 only 1836 were intra-amygdaloid connections vs 3865 were intra-hippocampal connections). This possibly accounts for the fact that fewer regions of high connectivity are observed for connections between amygdaloid regions than connections involving the HF or PHR (fig. 5). Consequently, this network gives a more accurate view on connections within and between the HF/PHR, as well as between the amygdala and HF/PHR than on connections between distinct amygdaloid regions.
Furthermore, the database also includes connections categorized as ‘non-existent’ meaning that the absence of a connection was reported. These absent connections were not included in the analysis to avoid conflict within the data. However, this also means that priority is given to existent connections over non-existent connections, seen as it is enough for a connection to be recorded once to be included in the network regardless of the potential number of entries supporting the non-existence of that connection. This is the case because a connection noted as ‘non-existent’ just means a connection was not observed, which does not prove the non-existence of that connection. It is far more probable that a connection was not observed and was therefore recorded as absent, than that a connection was observed and recorded as existent but falsely so. Furthermore, the exclusion of entries with low injection and projection ratings ensure the quality of the results ensuring the accuracy of the recorded connection.
Although all nodes have at least one connection, some areas had a total degree of 2 or lower, indicating that too little data might have been present on the connectivity of that area (table 1). Some of these areas are amygdalar regions that are peripheral to the amygdaloid network, e.g. bed nuclei of the stria terminalis, and are therefore sometimes even not classified as part of the amygdala resulting in little entries including these areas in the database. Other regions were subdivisions of amygdalar regions, such as the medial division of the amygdalohippocampal area, which was rarely part of connections because little intra-amygdalar connections
are present in the database and HF/PHR-to-amygdala connections often do not describe amygdalar regions in high detail. Thus, regions often not classified as being part of the amygdala and some amygdalar subdivisions featured less heavily in the database, which probably influenced the role of these structures in the
amygdalohippocampal connectome.
5.4. A critical note on small-world networks
Small-worldness is a property that is commonly found in nervous systems (Bassett and Bullmore, 2017) and has been used to characterize various real-world networks in multiple fields, such as biology, physics, and medicine (Bialonski et al., 2010). Thus, it is unsurprising that the amygdalohippocampal network shares this common property of many nervous systems. The near-universal appearance of small-world networks in complex natural systems and it being a purely topological quantity has led to the question what valuable inferences can be drawn from the small-worldness of a network. How does network topology relate to underlying neurological processes and how might biological features be related to the small-worldness property?
One possible starting point to understand the importance of the small-world property is to view the brain as economically embedded in a three-dimensional anatomical space, where economical means that nodes are located as to minimize the wiring cost. Since brains are physically expensive to build and maintain, some properties of brain anatomy seem to be organized to minimize these wiring costs (Bassett and Bullmore, 2017). The trade-off represented by small-worldness is then between the wiring cost and the global efficiency of the network, which represents the efficiency of information transfer between nodes that are far away in anatomical space (Bassett and Bullmore, 2017). Second, small-worldness is known to support computations in neural circuits, for example by lowering the time a brain area needs to reach synchronization or by supporting the trade-off between the networks ability to learn new information and to retain memories (Bassett and Bullmore, 2017). Thus, the small-worldness property tells us something about general principles of brain network organization, which can then be studied, e.g. from an economical point of view (Bullmore and Sporns, 2012).
However, while small-worldness can shed light on computational properties of brain networks, it is much more difficult to couple these computational properties to the underlying biological aspects of a network, such as inhibitory and excitatory connections, the use of different transmitters or changes in synaptic strength. The question then is less whether small-worldness and related network properties are valuable per se but how these properties emerge from the evolution of the brain, as well as the underlying dynamic molecular processes of brain structure and activity. The construction of brain networks and network analysis is only the first step in creating a dynamic functional model of the brain, which starts at uncovering the nearly-universal, structural properties of neural networks.
In conclusion, the amygdalohippocampal network is a small-world network. Within the network, the LEA, subiculum, amygdalohippocampal area, and CA1 have the highest number of efferent and afferent connections and were identified as putative hubs.
Since this study did not perform quantitative analyses on in- and outdegree measures, a further step could be to analyse whether some structures have significantly higher in-or outdegrees compared to
simulations of random networks and can therefore be regarded as primarily and input or output structures. Furthermore, great differences in connectivity have previously been observed between layers of certain regions, as well as between different subregions which were not mentioned in this study (van Strien et al., 2009). With more specific information on the layers, as well as the anatomical positions of structures of origin and termination, a more detailed connectome could be constructed to observe patterns of connectivity for the distinct sublayers and subregions of all amygdalohippocampal structures. This could shed more light on where information flows within the amygdalohippocampal region are spatially segregated and where they converge. Furthermore, this connectome could be supplemented with information on the cell types involved in a connection. Knowing whether a connection is largely inhibitory or excitatory could provide valuable insights into the functional interactions of amygdalohippocampal regions. Regarding the role of amygdalohippocampal interactions in memory formation and consolidation, the different possible pathways through which the basal nucleus might modulate LTP in the DG could be studied. Selective inhibition of structures in these pathways, e.g. the LEA, during fear conditioning could provide information on whether these structures are functionally involved. Additionally, structures involved in possible pathways could be selectively inhibited and the basal nucleus orthodromically stimulated to observe whether a certain structure is specifically involved in DG activation.
7. Bibliography
Bassett, D.S., Bullmore, E.T., 2017. Small-World Brain Networks Revisited. The Neuroscientist 23, 499–516. https://doi.org/10.1177/1073858416667720
Bialonski, S., Horstmann, M.-T., Lehnertz, K., 2010. From brain to earth and climate systems: Small-world interaction networks or not? Chaos Interdiscip. J. Nonlinear Sci. 20, 013134.
https://doi.org/10.1063/1.3360561
Cappaert, N.L.M., Van Strien, N.M., Witter, M.P., 2015. Hippocampal Formation, in: The Rat Nervous System. Elsevier, pp. 511–573. https://doi.org/10.1016/B978-0-12-374245-2.00020-6
Hitti, F.L., Siegelbaum, S.A., 2014. The hippocampal CA2 region is essential for social memory. Nature 508, 88– 92. https://doi.org/10.1038/nature13028
Lanciego, J.L., Wouterlood, F.G., 2011. A half century of experimental neuroanatomical tracing. J. Chem. Neuroanat. 42, 157–183. https://doi.org/10.1016/j.jchemneu.2011.07.001
McDonald, A.J., Mott, D.D., 2017. Functional neuroanatomy of amygdalohippocampal interconnections and their role in learning and memory: Amygdalohippocampal Interconnections. J. Neurosci. Res. 95, 797– 820. https://doi.org/10.1002/jnr.23709
Papo, D., Zanin, M., Martínez, J.H., Buldú, J.M., 2016. Beware of the Small-World Neuroscientist! Front. Hum. Neurosci. 10. https://doi.org/10.3389/fnhum.2016.00096
Phelps, E.A., 2004. Human emotion and memory: interactions of the amygdala and hippocampal complex. Curr. Opin. Neurobiol. 14, 198–202. https://doi.org/10.1016/j.conb.2004.03.015
Pitkänen, A., Pikkarainen, M., Nurminen, N., Ylinen, A., 2006. Reciprocal Connections between the Amygdala and the Hippocampal Formation, Perirhinal Cortex, and Postrhinal Cortex in Rat: A Review. Ann. N. Y. Acad. Sci. 911, 369–391. https://doi.org/10.1111/j.1749-6632.2000.tb06738.x
Pitkänen, A., Savander, V., LeDoux, J.E., 1997. Organization of intra-amygdaloid circuitries in the rat: an emerging framework for understanding functions of the amygdala. Trends Neurosci. 20, 517–523. https://doi.org/10.1016/S0166-2236(97)01125-9
Rubinov, M., Sporns, O., 2010. Complex network measures of brain connectivity: Uses and interpretations. NeuroImage, Computational Models of the Brain 52, 1059–1069.
https://doi.org/10.1016/j.neuroimage.2009.10.003
Saleeba, C., Dempsey, B., Le, S., Goodchild, A., McMullan, S., 2019. A Student’s Guide to Neural Circuit Tracing. Front. Neurosci. 13. https://doi.org/10.3389/fnins.2019.00897
Schmitt, O., Eipert, P., Philipp, K., Kettlitz, R., Fuellen, G., Wree, A., 2012a. The Intrinsic Connectome of the Rat Amygdala. Front. Neural Circuits 6. https://doi.org/10.3389/fncir.2012.00081
Schmitt, O., Eipert, P., Philipp, K., Kettlitz, R., Fuellen, G., Wree, A., 2012b. The Intrinsic Connectome of the Rat Amygdala. Front. Neural Circuits 6. https://doi.org/10.3389/fncir.2012.00081
van Strien, N.M., Cappaert, N.L.M., Witter, M.P., 2009. The anatomy of memory: an interactive overview of the parahippocampal–hippocampal network. Nat. Rev. Neurosci. 10, 272–282.
https://doi.org/10.1038/nrn2614
Vecchio, F., Miraglia, F., Maria Rossini, P., 2017. Connectome: Graph theory application in functional brain network architecture. Clin. Neurophysiol. Pract. 2, 206–213.
APPENDIX A
Table 1. Table showing local parameters for every structure: in degree (IN), outdegree (OD), total node degree (ID+OD), local clustering (LC), local efficiency (LE) and node betweenness (BC)
Structure name Indegree
(ID) Outdegree (OD) Total node degree (ID+OD) local clustering (LC) local efficiency (LE) Node Betweenness (BC) Hippocampal formation (HF) dentate gyrus (DG) 26 8 34 0.652803 0.875000 1.499117 CA3 23 13 36 0.635266 0.807692 5.231869 CA2 21 12 33 0.580769 0.796717 17.732633 CA1 28 26 54 0.478092 0.607179 202.276861 subiculum (Sub) 30 33 63 0.450828 0.581439 293.748897 Parahippocampal region (PHR) presubiculum (PrS) 26 12 38 0.536075 0.717172 115.294704 parasubiculum (PaS) 28 10 38 0.597262 0.894444 56.357646
medial entorhinal area (MEA) 37 16 53 0.461115 0.708333 169.919645
lateral entorhinal area (LEA) 36 32 68 0.418331 0.595598 370.682118
perirhinal cortex (PER) 15 0 15 0.723810 0.000000 0.000000
perirhinal cortex area 35 (PER35)
28 22 50 0.514463 0.693362 80.221978
perirhinal cortex area 36 (PER36)
29 22 51 0.484127 0.693362 132.537377
postrhinal cortex (POR) 29 15 44 0.504283 0.728571 31.837796
retrosplenial cortex A29ab 3 4 7 1.138.889 0.791667 0.000000
retrosplenial cortex A29ac 3 4 7 0.789474 0.291667 2.954167
retrosplenial cortex A30 3 2 5 0.833333 0.000000 1.742059
Amygdaloid complex
accessory basal nucleus (AB) 16 33 49 0.540052 0.639205 127.025264
accessory basal nucleus (AB) parvicellular division (ABpc)
7 29 36 0.483280 0.610530 3.497650
accessory basal nucleus (AB) magnocellular division (ABmc) 7 12 19 0.728070 0.878788 0.997973 amygdalohippocampal area (AHA) 24 43 67 0.382930 0.617617 521.071300 amygdalohippocampal area medial division (AHAm)
1 0 1 Inf 0.00000 0.000000 amygdalopiriform transition area (APir) 5 4 9 1.014286 0.875000 0.076923 amygdalostriatal transition zone (AStr) 2 0 2 2.000000 0.00000 0.000000
anterior amygdaloid area (AAA)
14 1 15 0.875000 0.000000 0.000000
anterior cortical amygdaloid nucleus (Coa)
15 31 46 0.551910 0.628047 43.334006
basal amygdaloid nucleus (BA)
14 9 23 0.662602 0.763889 8.782737
basal amygdaloid nucleus (BA) parvicellular division
16 15 31 0.624729 0.819048 10.410620
basal amygdaloid nucleus (BA) intermediate division
11 12 23 0.636364 0.878788 7.815748
(BA) magnocellular division bed nucleus of the accessory olfactory tract (BAOT)
12 3 15 0.865385 0.666667 0.981923
bed nucleus of the stria terminalis (BST) posterior division
1 1 2 Inf 0.00000 0.00000
bed nucleus of the stria terminalis (BST) dorsomedial nucleus
0 1 1 Inf 0.00000 0.00000
bed nucleus of the stria terminalis (BST) transverse nucleus
0 1 1 Inf 0.00000 0.00000
bed nucleus of the stria terminalis (BST)
2 1 3 2.25000 0.00000 0.00000
central amygdaloid nucleus (CE)
19 3 22 0.609649 0.416667 1.743722
central amygdaloid nucleus (CE) capsular division (CEc)
16 23 39 0.574760 0.525033 48.287235
central amygdaloid nucleus (CE) lateral division (CEl)
11 28 39 0.521233 0.483245 10.964348
central amygdaloid nucleus (CE) intermediate division (CEi)
7 25 32 0.533742 0.467500 2.304376
central amygdaloid nucleus (CE) medial division (CEm)
15 26 41 0.521605 0.446923 76.853239
intercalated nuclei of the amygdala (I)
16 0 16 0.766667 0.000000 0.000000
intra-amygdaloid division of the bed nucleus of the stria terminalis (BSTIA)
2 0 2 2.000000 0.00000 0.000000
intra-amygdaloid division of the bed nucleus of the stria terminalis (BSTIA) posterior division
0 2 2 1.500000 0.750000 0.000000
lateral amygdaloid nucleus (L)
18 16 34 0.656987 0.744444 14.483592
lateral amygdaloid nucleus (L) dorsolateral division (Ldl)
6 12 18 0.614379 0.878788 6.808572
lateral amygdaloid nucleus (L) ventrolateral division (Lvl)
6 12 18 0.614379 0.878788 6.808572
lateral amygdaloid nucleus (L) medial division (Lm)
8 12 20 0.646277 0.878788 7.067314
medial amygdaloid nucleus (M)
16 19 35 0.618803 0.670565 48.364614
medial amygdaloid nucleus (M) rostral division
12 1 13 0.807692 0.000000 0.368572
medial amygdaloid nucleus (M) central division
14 0 14 0.785714 0.000000 0.000000
medial amygdaloid nucleus (M) caudal division
16 3 19 0.698225 0.666667 3.359035
nucleus of the lateral olfactory tract (NLOT)
16 3 19 0.800000 0.000000 0.000000
olfactory amygdala (OA) 15 0 15 1.000000 0.750000 0.621523
periamygdaloid cortex (PAC) 13 24 37 0.568598 0.630435 40.865464
periamygdaloid cortex (PAC) PAC subfield
11 35 46 0.518049 0.638445 13.308549
periamygdaloid cortex (PAC) medial subfield (PACm)
11 36 47 0.515873 0.640939 13.811879
periamygdaloid cortex (PAC) sulcal subfield (PACs)
11 34 45 0.512232 0.632724 13.217640
posterior cortical nucleus (Cop)
15 38 53 0.492313 0.631046 57.603619
