Abstract
The parahippocampal region is thought to be critical for memory and spatial navigation. Within this region lies the parasubiculum, a small structure that exhibits strong theta modulation, contains functionally specialized cells, and projects to layer II of the medial entorhinal cortex (MEC). Thus, it is uniquely positioned to influence firing of spatially modulated cells in the MEC and play a key role in the internal representation of the external environment. However, the basic neuronal composition of the parasubiculum remains largely unknown, and its border with the MEC is often ambiguous. We combine electrophysiology and immunohistochemistry in adult mice (both sexes) to define first, the boundaries of the parasubiculum, and second, the major cell types found in this region. We find distinct differences in the colabeling of molecular markers between the parasubiculum and the MEC, allowing us to clearly separate the two structures. Moreover, we find distinct distribution patterns of different molecular markers within the parasubiculum, across both superficial-deep and DV axes. Using unsupervised cluster analysis, we find that neurons in the parasubiculum can be broadly separated into three clusters based on their electrophysiological properties, and that each cluster corresponds to a different molecular marker. We demonstrate that, while the parasubiculum aligns structurally to some to general cortical principals, it also shows divergent features in particular in contrast to the MEC. This work will form an important basis for future studies working to disentangle the circuitry underlying memory and spatial navigation functions of the parasubiculum.
SIGNIFICANCE STATEMENT We identify the major neuron types in the parasubiculum using immunohistochemistry and electrophysiology, and determine their distribution throughout the parasubiculum. We find that the neuronal composition of the parasubiculum differs considerably compared with the neighboring medial entorhinal cortex. Both regions are involved in spatial navigation. Thus, our findings are of importance for unraveling the underlying circuitry of this process and for determining the role of the parasubiculum within this network.
Introduction
The parasubiculum (PaS) is a small, elongated structure within the parahippocampal region, lying between the presubiculum (PrS) and the medial entorhinal cortex (MEC). The parahippocampal region is thought to play an integral role in spatial navigation and has attracted attention since the discovery of grid cells (Hafting et al., 2005). While the majority of work studying spatial navigation has focused on the hippocampus and MEC, the discovery of functional cell types in both the PrS and PaS has led to an increasing interest in these structures too. The PaS contains grid cells, head-direction cells, and border cells (Taube, 1995; Boccara et al., 2010), but it is the position of the PaS in the circuitry that perhaps makes it most interesting for the study of spatial navigation. The PaS receives inputs from the anterior thalamus (van Groen and Wyss, 1990; Ding, 2013), where an abundance of head-direction cells is found (Taube, 2007; Winter et al., 2015). Additionally, PaS receives input from the medial septum (Tang et al., 2016), a major generator of theta oscillations, as well as the subiculum, and CA1 (van Groen and Wyss, 1990; Ding, 2013; O'Reilly et al., 2013). The PaS sends output projections to layer 2 (L2) of the MEC (Köhler, 1985; van Groen and Wyss, 1990; Caballero-Bleda and Witter, 1993), where the greatest abundance of grid cells are found (Sargolini et al., 2006), as well as to the lateral entorhinal cortex, ipsilateral and contralateral PrS, and the contralateral PaS (van Groen and Wyss, 1990; Agster and Burwell, 2013). Current research is addressing how grid cells develop their distinctive firing patterns and how this circuit combines information from functionally different cell classes to encode spatial information. Thus, given the upstream and downstream targets of the PaS, it is uniquely positioned to influence both the development and continuing modulation of grid cells in the MEC. Moreover, lesion studies have shown that damage to the PrS and PaS impairs spatial location memory (Kesner and Giles, 1998; Liu et al., 2004), highlighting the importance of the PaS in the navigation circuit.
To address exactly how the PaS is involved in navigation and memory processes, it is important that we first understand what the basic components of this brain region are. Despite our increasing knowledge of the functional cell types within the PaS, our understanding of the basic, cellular composition of this region remains sparse. In this study, we use a combination of immunohistochemistry and electrophysiology to identify the major cell types within the mouse PaS. The PaS has previously been reported to express high levels of the transmembrane protein Wolframin, encoded by the gene WFS1 (Luuk et al., 2008; Ramsden et al., 2015; Ray et al., 2017). Here, we confirm these findings and show that WFS1-expressing cells comprise the principal, putative excitatory population of neurons and make up the majority of neurons in the PaS. In addition, we find that the PaS contains a heterogeneous population of interneurons, with different molecular subtypes showing different distributions along the superficial to deep axis. Electrophysiological data support these findings, with neurons falling into three clusters based on their passive and active properties. Two of these clusters represent putative inhibitory neurons, while the third, largest cluster represents the principal cell population of the PaS.
Materials and Methods
Experimental design and statistical analysis.
To obtain both an anatomical and a physiological overview of the PaS, we performed immunohistochemistry and in vitro slice electrophysiology. For our immunohistochemistry experiments, we used adult vesicular GABA transporter-Venus (VGAT-Venus) transgenic mice (Wang et al., 2009) of both sexes (postnatal day 150 or older, n = 4 animals). For our electrophysiology experiments, we prepared slices from adult mice of both sexes (postnatal day 150 or older, n = 98 animals) from several different genetic mouse lines (Table 1) (Hippenmeyer et al., 2005; Gong et al., 2007; Wang et al., 2009; Chao et al., 2010; Madisen et al., 2010; Kitamura et al., 2014). Further details of sample sizes for individual experiments are reported in the figure legends. All animal maintenance and experiments were performed in accordance with institutional guidelines, guidelines of the local state government (Berlin state government, T0100/03; O0413/12), and the European Union Council Directive 2010/63/EU. The study was not preregistered. Statistics were performed in Python using Scipy and Statsmodels packages. Data were first checked for normality using a Shapiro–Wilk test. Since data were not normally distributed, nonparametric Kruskal–Wallis tests were subsequently performed to test for differences in parameter values across groups, and post hoc Dunn's test performed to determine specific group differences. Significance levels were set to 0.05, for multiple-comparisons p values were adjusted using the false discovery rate (Benjamini/Hochberg) correction method (Benjamini and Hochberg, 1995). All values are reported as median, interquartile range (IQR) unless otherwise stated.
Immunohistochemistry.
Immunohistochemistry was performed on horizontal sections from VGAT-Venus transgenic mice (Wang et al., 2009), to enable cell density counts and labeling of different inhibitory populations. Mice were initially anesthetized with isoflurane, followed by an intraperitoneal injection of ketamine/xylazine (100 mg/15 mg per kg) and transcardially perfused with 0.1 m PBS, pH 7.4, followed by 4% PFA. Brains were harvested and stored in 4% PFA overnight. The following day, brains were rinsed in PBS and sectioned using a vibrating microtome slicer (Leica Microsystems, VT 1200S). Sections of 100 μm were cut and dropped into PBS after slicing. Free-floating sections were washed 3 times in PBS (5 min each) and incubated in a blocking solution composed of 5% NGS (Biozol), 1% Triton-X (Sigma-Aldrich), and PBS, for 3 h at room temperature with gentle agitation. Primary antibodies (Table 2) were diluted in blocking solution (2.5% NGS, 1% Triton-X, PBS), and sections were incubated for 72 h at 4°C. Following this, sections were washed 3 times (20 min each) with PBS and secondary antibodies (goat anti-rabbit AlexaFluor-647 and goat anti-mouse AlexaFluor-555, Invitrogen; 1:500 dilution in PBS) were applied for 3 h at room temperature. Finally, slices were washed 4 times (15 min) with PBS, before being mounted on glass slides in mounting medium (Mowiol). We found antibodies to have good tissue penetration in slices of 100 μm thickness. Any slices where clear antibody signal could not be detected throughout at least a continuous 50 μm stack in the slice were discarded and not used for cell counts.
Slices that had been prepared for electrophysiology and contained recorded and filled cells were fixed overnight in 4% PFA. Sections (400 μm thickness) were then washed 3 times (5 min each) in PBS and stained using the same protocol. To label the biocytin, a conjugated streptavidin marker was used (conjugated to AlexaFluor-488 or AlexaFluor-647). Up to three different primary antibodies were used in addition to the biocytin labeling. Secondary antibodies used were goat anti-rabbit AlexaFluor-405, goat anti-rabbit AlexaFluor-647, goat anti-mouse AlexaFluor-555 or goat anti-guinea pig AlexaFluor-555, and goat anti-chicken AlexaFluor-488 (Invitrogen).
Confocal microscopy and cell counting.
Mounted sections were imaged on an upright TCS SP5 confocal microscope (Leica Microsystems). The following laser lines were used to illuminate fluorescent samples through a 20× immersion objective (0.7 NA; Leica Microsystems): 405 nm (diode), 488 nm (Argon laser), 568 nm (solid state), and 633 nm (Helium, Neon). For cells that had been filled with biocytin, a further image of the cell soma was obtained through a 63× objective (1.4 NA; Leica Microsystems) to identify any post hoc immunostaining. All images were acquired at 1024 × 1024 pixels (pixel size, 0.72 and 0.23 μm for 20× and 63× objectives, respectively). For cell counting, 50 μm stacks were taken with a 0.49 μm step size. Analysis of images was performed in Fiji (https://imagej.nih.gov/ij/). To calculate cell densities, a 100 × 100 μm box was placed over the PaS. The box was first placed as close to the pia as possible, at a position where the distance between the pia and alveus was judged to be the greatest. To avoid potential overlap with the medial-most MEC (mMEC), the box was preferentially placed on the more medial side of the PaS. Cells within this frame were counted according to optical dissector counting rules (West et al., 1991); exclusion lines were drawn on the bottom and left-hand edges of the box, and inclusion lines on the upper and right-hand edges. Cells that touched either exclusion line were excluded, whereas cells touching inclusion lines were included in counts. Only cells coming into focus in the image stack were counted. The grid was moved throughout the depth of the PaS, starting at the pial surface and moving 100 μm away from the pia, toward the alveus, each time, until part of the grid was no longer in the PaS (i.e., superficial to deep axis). Cell densities were calculated by multiplying the cell count by the volume covered by the grid. Shrinkage of sections was not taken into account.
Model fitting.
GLMs were fit in python using the statsmodels package (www.statsmodels.org). Models were generated with a negative binomial distribution and a log-link function. Several models were created, and the most appropriate was decided upon by comparison of the Akaike information criteria (Akaike, 1974) and maximum likelihood tests between nested models. Model validation was performed using the leave-one-out cross-validation method. Differences between cell counts at different depths and for WFS1 and VGAT-Venus neurons, or different interneuron markers were concluded by nonoverlapping 95% CIs for predicted cell counts, using data predicted by the final chosen GLM.
Morphological reconstructions.
For morphological reconstructions, tiled image stacks of the biocytin-filled neurons were taken using a 63× objective. Images were taken at 1024 × 1024 pixels (pixel size 0.23 μm) and with a z step size of 0.49 μm. Multiple image tiles were stitched together in Fiji using the grid stitching plugin (Preibisch et al., 2009). Following acquisition, images were loaded into NeuTube (Feng et al., 2015) and reconstructed semiautomatically using the software's tracing tools.
Whole-cell patch-clamp recording.
For whole-cell patch-clamp recordings acute brain slices were prepared from adult (postnatal day 50 and over, both sexes) WT mice (C57BL/6n, n = 42) or adult transgenic mice (Table 2). Animals were anesthetized under isoflurane and decapitated. Brains were removed and transferred to an ice-cold, sucrose-based dissection ACSF containing the following (in mm): 87 NaCl, 26 NaHCO3, 50 sucrose, 10 glucose, 2.5 KCl, 1.25 NaH2PO4, 0.5 CaCl2, and 3 MgCl2, and saturated with 95% O2 and 5% CO2. Horizontal slices 400 μm thick were cut across the whole span of the PaS (∼6 slices per hemisphere) using a vibrating microtome slicer (Leica Microsystems, VT1200S). Once cut, slices were transferred to an interface chamber, which was continuously perfused with ACSF containing the following (in mm): 119 NaCl, 26 NaHCO3, 10 glucose, 2.5 KCl, 1.3 MgCl2, 1 NaH2PO4, and 2.5 CaCl2, saturated with 95% O2 and 5% CO2, and maintained at 32°C–34°C. Slices were allowed to recover for at least 1 h before being transferred to a recording chamber. Recordings were performed in a submerged chamber, kept at 32°C–34°C, and perfused with the same ACSF as used for interface storage. Somatic whole-cell recordings were performed using glass pipettes pulled from borosilicate glass capillaries with a tip resistance of 3–6 mΩ. Pipettes were pulled using a horizontal DMZ Universal Puller (Carl Zeiss). Pipettes were filled with an internal solution containing the following (in mm): 120 K-Gluconate, 10 HEPES, 10 KCl, 5 EGTA, 2 MgSO4·7H2O, 3 MgATP, 3 NaGTP, and 5 phosphocreatine Na. Internal solution also contained 0.2% biocytin, to allow for later recovery and identification of recorded neurons. To identify cells for recording, slices were visualized using infrared differential interference contrast microscopy through a digital camera (XM10-IR, Olympus). In some cases, Venus-expressing inhibitory neurons were targeted using epifluorescence illumination by a mercury arc lamp (Olympus). Recordings were performed using Multiclamp 700A/B amplifiers (Molecular Devices). Signals were filtered at 10 kHz and sampled at 20 kHz and either digitized using Digidata 1550 and recorded in pClamp10 (Molecular Devices) or using a BNC-2090 interface board (PCI 6035E A/D Board, National Instruments) and recorded in IGOR Pro 6.12 (WaveMetrics). Series resistance was monitored throughout recordings, and cells were discarded if the series resistance exceeded 25 mΩ. Pipette capacitance neutralization and bridge balance were applied and adjusted as appropriate. Liquid junction potential was not corrected for. Upon obtaining a whole-cell patch, neurons were characterized, in current-clamp configuration, by injecting increasing steps of negative and positive current. Current injections were applied for 1 s, in increasing steps of 40 pA. The voltage responses of neurons to these current injections were used to calculate intrinsic properties and action potential (AP) parameters.
Electrophysiological analysis.
Electrophysiological features were extracted from recordings using custom-written routines in IGOR Pro. Resting membrane potential (Vm) was taken as the mean value of the baseline before current injections were performed. Input resistance was taken as the slope of the current-voltage fit between all negative and subthreshold positive current steps. Rheobase was estimated by fitting a line through the relationship between injected current and AP number and extrapolating this backwards. AP threshold was defined as the membrane potential at the point where dV/dt reached 5% of the maximum dV/dt. AP latency was calculated as the time between current injection and the first AP onset. AP height was taken as the difference between AP threshold and maximal voltage of first AP. The rising and falling slopes of APs were taken as the maximal and minimal differential values, respectively, and these two values used to calculate the slope ratio. AP half-width was taken as the time between the half-height of the upward and downward phases. The interval between AP2 and AP3 was calculated as the time between onset of each of the second and third APs. The same was calculated for AP9 and AP10, using the first current injection step that elicited at least 10 APs. The adaptation index was then calculated by dividing the interval of AP9-AP10 by the interval of AP1-AP2. If a cell did not produce 10 APs with maximal current injection, then the maximum number of APs were used to calculate the last interval, instead of AP9-AP10. A relative firing frequency was calculated by taking the firing frequency two current steps above the rheobase value. Cells were considered as inactivated when current injection no longer produced APs that reversed >0 mV. Afterhyperpolarization (AHP) was defined as the difference between AP threshold and the minimum voltage seen immediately after the AP (within 2 ms). The medium AHP was taken as the difference between baseline Vm and the minimum voltage seen up to 200 ms after the end of current injection. A sag ratio was calculated using 1 s steps of negative current injection and dividing the voltage difference between baseline and the steady-state hyperpolarization by the maximal voltage deflection immediately after the current injection onset. Theta power was calculated by applying a fast Fourier transform to the trace of the highest current injection not producing any APs. For each cell, the maximum power and corresponding frequency in the theta range (3–12 Hz) was taken.
Cell classification and cluster analysis.
Hierarchical clustering and principal component analysis were performed using the following electrophysiological parameters: Vm, input resistance, sag ratio, rheobase, input–output slope, AP latency, AP2-AP3 interval, AP9-AP10 interval, minimum interspike interval, AP threshold, AP amplitude, AP half-width, adaptation index, medium AHP, AHP, relative firing frequency, maximum depolarization rate, maximum repolarization rate, and the ratio of maximum and minimum dV/dt. Data were first standardized by centering (subtracting the mean of a parameter from each data point of the given parameter) and scaling (dividing by the SD). Clustering was based on Ward's method (Ward, 1963). Clustering was implemented using the prcomp function in the stats package in R (http://www.r-project.org/). Euclidean distance was calculated using the scaled and centered data and the distance matrix used to compute the hierarchal clustering. To estimate the optimal number of clusters, we used the gap statistic method, whereby the within-cluster dispersion was compared with that from a reference distribution over 500 trials, for k = 1 to 10 clusters. The reference distribution was randomly assigned within the same range as our dataset. The mean and SD of the gap between the intracluster variation from our dataset, and that from the reference distribution was calculated across the 500 trials and the optimal number of clusters defined as the first peak where Gap(k) was greater or equal to Gap(k + 1) − SD(k + 1). The Calinski–Harabasz (CH) index was calculated to examine the ratio of intercluster and intracluster variability. CH index was calculated using the calinski_harabaz_score function from the sklearn package in Python. Cells were divided into clusters according to distance. Clusters containing <3 cells were merged with the most correlated cluster (Gouwens et al., 2019).
Results
Delineation of the PaS/MEC border using coexpression of markers
The PaS is situated between the PrS and the MEC in the parahippocampal region (Fig. 1A). As previously reported (Luuk et al., 2008; Ramsden et al., 2015; Ray et al., 2017), we found the protein marker WFS1 to be highly expressed in the PaS (Fig. 1B). On the medial side, a clear border can be distinguished between the PrS and PaS, with the PrS distinctly lacking WFS1 expression. On the lateral side, the distinction between the PaS and the MEC is less obvious, with pyramidal neurons in L2 of the MEC also expressing WFS1 and seemingly merging with the lateral edge of the PaS (Fig. 1B,C). However, we find several molecular distinctions between the two regions that may be used to delineate the PaS from the MEC. We made use of a transgenic mouse line, VGAT-Venus (Wang et al., 2009), to identify the inhibitory population of neurons. The most striking difference between the PaS and the MEC is the proportion of reelin cells that colocalize with VGAT-Venus across the two regions (Fig. 1D). In the PaS, we find 88% of reelin-labeled neurons overlap with VGAT-Venus expression. In contrast, in the MEC we find the inverse, with only 17% of reelin cells colocalizing with VGAT-Venus (Fig. 1E). A second clear distinction between the two regions can be seen in the colocalization of calbindin with WFS1 and VGAT-Venus (Fig. 1F). In the MEC, calbindin is often used alongside WFS1 as a marker for L2 pyramidal neurons (Ray et al., 2014; Kitamura et al., 2015; Sun et al., 2015). We checked for overlap between calbindin, WFS1, and VGAT-Venus expression in neurons in both the MEC and PaS (Fig. 1G,H). We find that, in the PaS, calbindin is exclusively inhibitory (100% of calbindin cells were VGAT-Venus positive; Fig. 1G). In the MEC, calbindin colocalizes with both WFS1 and VGAT-Venus in a mutually exclusive manner (calbindin-WFS1 overlap: 35%; calbindin-VGAT-Venus overlap: 37%; Fig. 1G). Meanwhile, in the PaS, there is no overlap between WFS1 and calbindin staining, compared with in the MEC where 74% of WFS1 cells were colabeled with calbindin (Fig. 1H).
We also made two observations suggestive of a transitional zone between the PaS and MEC, which we believe corresponds to the previously reported mMEC (Fujimaru and Kosaka, 1996; Ray et al., 2017). While in the MEC WFS1 is expressed only in L2 neurons, at the interface between PaS and MEC, WFS1 can be seen across all superficial layers. Two properties appear to separate these WFS1 neurons from the WFS1 neurons seen in the PaS. First, in this region, there is colocalization of WFS1 and calbindin (Fig. 1Fii). Second, the orientation of the apical dendrite of WFS1 cells in this region appear orthogonal to those expressing WFS1 in the PaS (Fig. 1C). We observed this phenomenon in at least 19 slices where multiple cells had been filled with biocytin in the PaS and at the interface with the MEC. Combining these observations, we consider this region to be more alike to the MEC than the PaS; therefore, in the present study, we exclude this region from our analysis of the PaS.
Cellular composition of the PaS
We next looked at the neuronal composition of the PaS. To this end, horizontal sections containing PaS cut from the perfused brains of VGAT-Venus mice were immunostained with antibodies for NeuN and WFS1 (Fig. 2A). This enabled us to look at the total proportion of neurons expressing either WFS1 or VGAT-Venus. The PaS has been previously referred to as both a three-layered (Mulders et al., 1997; Burgalossi et al., 2011; Tang et al., 2016) and a six-layered structure (Funahashi and Stewart, 1997; Glasgow and Chapman, 2007; Boccara et al., 2010). Instead of assigning layers directly, we used a binning approach to look at cell densities at different distances from the pial surface to determine whether this gave indications of the laminar structure of the PaS. Neurons were counted at 100 μm increments from the pial surface (Fig. 2A), up to 500 μm from the pia. The total number of neurons (identified by NeuN label), as well as the number of WFS1 and VGAT-Venus-expressing neurons were counted (Fig. 2B). To allow for valid statistical comparisons across depth and cell type, we modeled our cell count data using a negative binomial GLM and found that both immunomarker (referred to in models as “stain”) and depth from the pia, as well as their interaction, were significant factors contributing to cell count. We tested several models and compared them using the Akaike Information Criteria (AIC) and deviance (Fig. 2C). Our data were best fit by a cubic expression that included staining, depth, and the interaction of these two terms, indicating that both molecular marker and location along the superficial-deep axis influence cell count (cell count ∼ stain*depth + stain*depth2 + depth3; where * denotes the inclusion of each individual factor as well as their interactions; Table 3). We used the CIs of estimated cell density, predicted from our model, to compare WFS1 and VGAT-Venus distributions with each other and within themselves across the depth of the PaS (Fig. 2D). Using this approach, we found that the distribution of WFS1 labeling followed a nonuniform distribution throughout the superficial-deep axis of the PaS. We found WFS1 density to be at its lowest in the most superficial and deepest portions of the PaS, whereas the highest proportion of WFS1 was seen across the center of the PaS, ∼100–400 μm from the pial surface (mean ± SEM WFS1 density at 0–100 μm: 5516 ± 818 cells/mm3; 100–200 μm: 16,368 ± 689 cells/mm3; 200–300 μm: 16,776 ± 574 cells/mm3; 300–400 μm: 12,400 ± 737 cells/mm3; 400–500 μm: 5188 ± 515 cells/mm3). The WFS1 density distribution followed closely the distribution of NeuN-labeled neurons, reflecting that the majority of the PaS neuronal population is made up of putatively excitatory, WFS1-expressing neurons (Fig. 2E). In the central region of the PaS (200–300 μm from pia), WFS1-positive cells made up 80% of all neurons. VGAT-Venus exhibited a different distribution, showing a relatively constant density across the middle body of the PaS, with the lowest density seen in the most superficial region and the highest density seen at the furthest away region from the pia (Fig. 2D). Despite having the lowest VGAT-Venus density in the most superficial region, inhibitory neurons accounted for 31% of all neurons in this sparsely populated area (Fig. 2E). In the deepest region (400–500 μm from the pia), where VGAT-Venus was at its highest density, inhibitory neurons accounted for 38% of all neurons. Expression of WFS1 and VGAT-Venus was entirely nonoverlapping, suggesting that the WFS1 population, which makes up the majority of neurons in the PaS, is putatively excitatory.
We also looked to determine whether the DV axis had an influence on cell count, since it is known that other structures within the parahippocampal region show differences across this axis. Slices were compared with a reference map of horizontal slices, and, using landmark features including the shape and location of the PaS were assigned to either the top, central, or bottom third of the PaS structure, corresponding to dorsal, intermediate, and ventral portions, respectively. Subsequently, we added DV location as a new factor in our GLM and found that it was also a significant factor (cell count ∼ stain*depth*DV location + stain*depth2 + depth3; Fig. 2F; Table 4). Thus, we modeled estimated cell densities for WFS1 and VGAT-Venus at our three locations. In the superficial PaS (0–100 μm from the pia), we found higher densities of both WFS1 and VGAT-Venus (Fig. 2G,H) in the ventral portion compared with intermediate and dorsal portions. WFS1 expression reached its peak density between 100 and 200 μm from the pia in the ventral PaS, compared with 200–300 μm in the intermediate and dorsal portions. WFS1 density in the ventral portion then decreased and became significantly lower than in intermediate and dorsal portions at deeper depths. Furthermore, the shape of the PaS changes along the DV axis; and in ventral slices, we were unable to unequivocally distinguish the PaS beyond 400 μm from the pia. Thus, no counts were made in ventral slices at the 400–500 μm bin. VGAT-Venus remained higher in the ventral PaS than intermediate and dorsal PaS across all depths. To examine what these results might mean for the balance of putative excitatory and inhibitory populations, we next looked at the ratio of WFS1 to VGAT-Venus cells in slices where both markers were counted (Fig. 2I). For each counting box, the number of WFS1 cells was divided by the number of VGAT-Venus cells, giving us an indication of the putative excitatory to inhibitory cell ratio. In all regions, the ratio was >1, indicating that WFS1 cells outnumber VGAT-Venus cells throughout the PaS. In the most superficial region, the ventral WFS1:VGAT-Venus ratio was higher than intermediate or dorsal ratios. However, at all other depths, the ventral PaS showed a lower ratio of WFS1:VGAT-Venus than intermediate or dorsal PaS. Moreover, the peak ratio was much higher in dorsal and intermediate portions than in the ventral PaS, again indicating that the ventral PaS has a greater inhibitory contribution.
Different inhibitory markers exhibit differential expression profiles in the PaS
To examine the inhibitory population of neurons in the PaS in more detail, further immunostaining was performed using several different inhibitory markers (Fig. 3A), and cell counts were made for each marker (Fig. 3B). By using the VGAT-Venus transgenic mouse line, we were able to determine the relative proportion of each marker across the interneuron population (Fig. 3C). We applied the same approach as with WFS1 and VGAT-Venus density measurements and modeled our count data with a GLM (Fig. 3D; Table 5). Again, we found that both stain and depth and their interaction were good predictors of cell density, which could be described with the same expression. We observed several distinct distributions in the inhibitory markers tested (Fig. 3E). Notably, we found that, in the most superficial region of the PaS (0–100 μm from pial surface), reelin showed a higher density above all other markers, with the exception of vasoactive intestinal peptide (VIP), where the difference was not significant (median, 25th and 75th percentile density, at 0–100 μm from the pia for reelin, 1600, 1200–2000; calretinin, 200, 0–400; parvalbumin (PV), 100, 0–350; somatostatin (SOM), 0, 0–100; VIP, 400, 300–600; calbindin, 0, 0–200; cholecystokinin (CCK), 0, 0–0; neuropeptide Y (NPY), 0, 0–0 cells/mm3). Reelin was also at its highest density in this region, before showing relatively constant, lower density levels throughout the rest of the PaS. In contrast, several other markers, including calbindin, PV, and SOM, showed the inverse pattern, starting with very low levels in the superficial PaS and having the highest density counts throughout the middle and deepest portions of the PaS. Meanwhile, NPY, VIP, calretinin, and CCK did not show such distinct distributions, with relatively uniform, low expression levels throughout the superficial to deep axis of the PaS. However, VIP and calretinin did have significantly higher counts than CCK and NPY in the area spanning 200–300 μm from the pial surface.
Immunolabeling for different inhibitory markers was always performed using pairs of antibodies, in slices from VGAT-Venus transgenic mice. Thus, we were also able to determine overlap between certain pairs of inhibitory markers (Fig. 3F). We found that both PV and SOM overlapped partially with calbindin, explaining, in part, the similarities in density distribution patterns seen across these markers. We also observed some coexpression between reelin and both SOM and calbindin. Furthermore, we found that a small percentage of reelin-positive neurons were not VGAT-Venus-positive. These neurons were not included when calculating the total percentage of reelin in the overall VGAT-Venus population. We did not find any overlap between WFS1 and any of the inhibitory markers with which it was coimmunostained for (calretinin, PV, calbindin, or reelin). This work reveals that there is a small population of noninhibitory reelin-positive neurons, and that these cells are, at least molecularly, distinct from the main pyramidal class in the PaS.
Parasubicular neurons can be classified into three clusters based on their electrophysiological properties
To characterize the electrophysiological features of neurons in the PaS, we made whole-cell recordings in acute brain slices from 391 neurons within this brain region. Following recordings, slices were immunostained to enable visualization of the recorded neurons, which were filled with biocytin during recording, as well as to label for potential molecular markers. We performed unsupervised cluster analysis and principal component analysis on our electrophysiological data to determine whether we could identify different cell groups based on their intrinsic and AP firing properties. In addition to electrophysiological parameters, for a subset of cells, we were able to recover post hoc immunohistochemical labeling of molecular markers, which we matched up to cells after clustering (Fig. 4A). To assess our clustering, we calculated the gap statistic (Tibshirani et al., 2001) to look at intercluster variance, and the CH index to examine the ratio of intercluster to intracluster variability. Both measures were calculated for up to 10 clusters (Fig. 4B). The CH index decreased as the number of clusters increased, indicating that clusters became less compact the more clusters there were. Our gap plot indicated four to be the optimal number of clusters. Cutting our dendrogram at the level of four clusters produced one cluster containing two cells. Thus, we merged this cluster with its closest neighbor cluster and looked at the resulting three clusters (Fig. 4C). The first cluster (55 of 391 cells) contained many inhibitory labeled neurons. In 10 cells, PV immunostaining was recovered; and in a further 7 cells, VGAT-Venus labeling was confirmed. Two cells were identified as reelin-positive. Cluster 2 (52 of 391 cells) contained 17 reelin-positive cells, as well as a further 3 VGAT-Venus-positive neurons and 2 WFS1-labeled neurons. Combined with our previous finding that, in the PaS, reelin is predominantly an inhibitory marker, we propose that Cluster 2 may also represent a putatively inhibitory cluster of neurons. Cluster 3 represented more than two-thirds of all recorded neurons (284 of 391 cells). More than half of Cluster 3 cells (155 of 284) were found to be WFS1-positive. Given the lack of colocalization between WFS1 and VGAT-Venus seen earlier, we propose Cluster 3 to be a putatively excitatory group of neurons.
In 301 of our recorded cells, we were able to successfully recover the biocytin labeling and identify the recorded neuron. For these cells, we measured the minimal distance to the pial surface from the cell soma (Fig. 4A,D) and binned them into 100 μm intervals matching those used in our immunohistochemistry experiments (Fig. 4E). We looked at the proportion of putative inhibitory and excitatory neurons recorded at different depths of the PaS by combining Cluster 1 and Cluster 2 cells to represent putative inhibitory neurons. We found the greatest proportion of putative inhibitory neurons to be in the range of 300–400 μm from the pia (38% putative inhibitory, 62% putative excitatory), followed by the most superficial region, 0–100 μm from the pia (33% putative inhibitory, 67% putative excitatory). In the deepest portion (>400 μm from pia), we found only Cluster 3, or putative excitatory neurons. However, overall only 5 cells were recorded in this region due to difficulties in targeting accurately the deep region of this structure. In the central regions (100–300 μm), we saw high levels of putative excitatory neurons (100–200 μm: 81% putative excitatory; 200–300 μm: 70% putative excitatory) reflecting similarities with our immunohistochemistry data. We next looked at the distribution of each cluster across the measured depths (Fig. 4G–I). We found that Cluster 1 cells were predominantly found in the central region of the PaS (37% of all Cluster 1 cells in each 100–200 μm and 200–300 μm bins). Meanwhile, the majority of Cluster 2 cells were found in the most superficial region (0–100 μm: 57% all measured Cluster 2 cells). Cluster 3 cells were found to be most abundant in the central portions of the PaS (100–200 μm: 43% of all Cluster 3 cells; 200–300 μm: 25% of all Cluster 3 cells).
Each cluster could be distinguished by several electrophysiological features that separated it from the remaining two clusters (Table 6). To identify the features that most strongly typified each class, we calculated a measure of effect size (ES) for each parameter. Since many of our parameters did not follow a normal distribution, we used Cliff's δ to quantify the differences between clusters (Cliff, 1996) (Table 6). Several features were found to be strongly different across all three clusters. These include the ratio of AP rise and fall rates (Fig. 5A; median, IQR dV/dt ratio C1:1.21, 1.1–1.32; C2: 1.94, 1.79–2.13; C3: 2.57, 2.17–3.2; p < 0.001 for all comparison pairs using Dunn's test; ES C1 vs C2: −0.85; C1 vs C3: −0.99; C2 vs C3: −0.61), minimal interspike intervals (Fig. 5B; median, IQR min interspike interval C1: 10, 6.9–14.3 ms; C2: 18.1, 14.9–22.6 ms; C3: 44.7, 34.5–54.9 ms; C1 vs C1, p = 0.04, ES: −0.67; p < 0.001 for all other comparison pairs using Dunn's test; ES C1 vs C3: −0.95; C2 vs C3: −0.89) and input resistance (Fig. 5C; median, IQR input resistance C1: 62.7, 46–76 mΩ; C2: 152.7, 120.6–230.1 mΩ; C3: 116.6, 89.4–149.3 mΩ; p < 0.001 for all comparison pairs using Dunn's test, ES C1 vs C2: −0.83; C1 vs C3: −0.67; C2 vs C3: 0.44). Furthermore, the three clusters all showed distinct input–output curves (Fig. 5Di–Diii; median, IQR input–output slope C1: 0.8, 0.7–1.2; C2: 0.7, 0.5–0.9; C3: 0.2, 0.2–0.3; p < 0.001 for all comparisons using Dunn's test; ES C1 vs C2: 0.36; C1 vs C3: 0.97; C2 vs C3: 0.78). Cluster 1 cells showed later activation, reflected in their high rheobase values, whereas Clusters 2 and 3 showed much earlier activation (median, IQR rheobase C1: 358, 249–440 pA; C2: 98, 46–152 pA; C3: 109, 71–148 pA; C2 vs C3, p = 0.2; p < 0.001 for all other comparison pairs using Dunn's test; ES C1 vs C2: 0.84; C1 vs C3: 0.75). Inactivation of some neurons was seen upon injection of high current values and was characterized by the failure of neurons to fire APs that reversed >0 mV. This inactivation followed the same pattern as activation, occurring early on in Cluster 2 cells and later or not at all in Cluster 1 cells. Other features were highly indicative of specific clusters. Cluster 1 cells showed high firing frequencies (Fig. 5D,E; median, IQR relative firing frequency C1: 223, 181–288 Hz; C2: 94, 71–114 Hz; C3: 56, 42–76 Hz; p < 0.001 for all comparisons using Dunn' test; ES C1 vs C2: 0.94; C1 vs C3: 0.99; C2 vs C3: 0.57) and a short AP half-width (Fig. 5F; median, IQR AP half-width C1: 0.24, 0.18–0.28 ms; C2: 0.56, 0.49–0.67 ms; C3: 0.67, 0.56–0.77 ms; p < 0.001 for all comparison pairs using Dunn's test; ES C1 vs C2: −0.99; C1 vs C3: −1; C2 vs C3: −0.36). Cluster 3 cells were typified by a large AP height (Fig. 5F; median, IQR AP height C1: 61.8, 51–71.7 mV; C2: 56.4, 50.1–67.5.1 mV; C3: 70.4, 63.5–77.3 mV; C1 v C2, p = 0.12; p < 0.001 for all other comparison pairs using Dunn's test; ES C1 vs C3: −0.37; C2 vs C3: −0.55) and the presence of a sag potential in response to negative current injection (Fig. 5G; median, IQR sag ratio C1: 0.9, 0.87–0.93; C2: 0.92, 0.9–0.94; C3: 0.88, 0.83–0.91; C1 v C2, p = 0.025, ES: −0.23; C1 vs C3, p = 0.007, ES: 0.24; C2 v C3, p < 0.001, ES: 0.49). It has been previously reported that PaS neurons exhibit membrane oscillations within the theta range (Glasgow and Chapman, 2008). Thus, we checked subthreshold membrane depolarizations for activity in the theta frequency band. We found prominent subthreshold theta-range membrane activity in Cluster 2 and 3 neurons, but not in Cluster 1 neurons (Fig. 5H–J; median, IQR subthreshold theta power C1: 2 × 105, 1.2 × 105 to 6.3 × 105 mV2; C2: 5.8 × 105, 2.7 × 105 to 2.2 × 106 mV2; C3: 4.6 × 105, 2.3 × 105 to 1.2 × 106 mV2; p < 0.001 for all comparison pairs using Dunn's test; ES C1 vs C2: −0.39; C1 vs C3: −0.32; C2 vs C3: 0.12).
In addition to electrophysiological properties, we also saw morphological differences between cells in the three clusters. Cluster 1 was the only cluster in which PV staining was observed (Fig. 6A). Cells in this cluster tended to show morphologies that covered a large area of the PaS (Fig. 6B,C). In contrast, Cluster 2 cells often showed small morphologies (Fig. 6D,E), also reflected in their low membrane capacitance values (median, IQR Cm C1: 89.9, 70.6–133 pF; C2: 46.8, 33.1–87 pF; C3: 137.7, 103.2–189.8 pF; p < 0.001 for all comparison pairs using Dunn's test; ES C1 vs C2: 0.55; C1 vs C3: −0.35; C2 vs C3: −0.76). Cluster 3 cells showed typical pyramidal morphology with apical dendrites extending to the pial surface. Axons from Cluster 3 cells were seen projecting toward the alveus (Fig. 6F,G) as well as the MEC (Fig. 6H,I).
Our initial cluster analysis suggested the possible presence of further subgroups of cells, in particular within the principal neurons comprising Cluster 3. Therefore, we reran our clustering algorithm on Cluster 3, after first removing any VGAT-positive cells to leave us with a putatively pure excitatory set of cells (Fig. 7A, n = 280 cells). The CH index decreased with increasing numbers of clusters, again indicating the presence of fewer, tighter clusters (Fig. 7B). Our gap plot indicated 3 to be the optimal number of clusters (Fig. 7B). Moreover, the largest decrease in Euclidean distance was also seen at this stage (Fig. 7C). Thus, we cut the dendrogram at the level for three clusters and looked at these groups in more detail. All three groups contained a high proportion of WFS1, and reelin was also seen in all three groups (C3a total cluster size: 115 cells; 62% WFS1, 6% reelin; C3b: 75 cells; 45% WFS1, 1% reelin; C3c total cluster size: 90 cells; 56% WFS1, 3% reelin). While all of these cells are putatively excitatory pyramidal neurons, the three groups differed in a number of parameters (Table 7). Neurons in C3a were characterized by a short AP height and high threshold (Fig. 7D,E; median, IQR AP height in C3a, 62.2, 55.6–67.8 mV; C3b: 71.9, 69.4–77 mV; C3c: 78.5, 73.3–82.8 mV; C3a vs C3b, p < 0.001, ES: −0.73; C3a vs C3c, p < 0.001, ES: −0.81; C3b vs C3c, p = 0.003, ES: 0.4; Dunn's test and Cliff's δ; median, IQR AP threshold C3a: −33.6, −34.8 to −30.9 mV; C3b: −35.8, −37.5 to −34.5 mV; C3c: −36.3, −38.5 to −34.3; C3a vs C3b, p < 0.001, ES: 0.51; C3a vs C3c, p < 0.001, ES: 0.54; C3b vs C3c, p = 0.489). C3b neurons were typified by a short AP half-width (Fig. 7F; median, IQR AP half-width C3a, 0.73, 0.63–0.83 ms; C3b, 0.55, 0.5–0.58 ms; C3c, 0.73, 0.65–0.81 ms; C3a vs C3b, p < 0.001, ES: 0.8; C3a vs C3c, p = 0.567; C3b vs C3c, p < 0.001, ES: 0.84), a large AHP (median, IQR AHP C3a: 15.7, 13.9–18.4 mV; C3b:18.2, 16.4–20.5 mV; C3c: 14.6, 12.6–17.4 mV; C3a vs C3b, p < 0.001, ES: −0.39; C3a vs C3c, p = 0.025, ES: 0.2; C3b vs C3c, p < 0.001, ES: −0.53), and a low input resistance (median, IQR input resistance C3a: 131, 110–165 mΩ; C3b: 91, 76–107 mΩ; C3c: 124, 94–159 mΩ; C3a vs C3b, p < 0.001, ES: 0.63; C3a vs C3c, p = 0.152; C3b vs C3c, p < 0.001, ES: 0.46). Cells in C3c showed the lowest firing frequencies (median, IQR relative firing frequency C3a: 60, 49–76 Hz; C3b, 75, 63–86 Hz; C3c, 41, 29–49 Hz; C3a vs C3b, p = 0.002, ES: −0.31; C3a vs C3c, p < 0.001, ES: 0.62; C3b vs C3c, p < 0.001, ES: −0.81), the shortest onset latency to first AP (median, IQR onset C3a: 141, 97–191 ms; C3b: 160, 116–210 ms; C3c: 95, 61–135 ms; C3a vs C3b, p = 0.06; C3a vs C3c, p < 0.001, ES: 0.34; C3b vs C3c, p < 0.001, ES: −0.5), and the highest dV/dt ratio (median, IQR dV/dt ratio C3a: 2.3, 2.0–2.6; C3b: 2.4, 2.1–2.6; C3c: 3.6, 3.1–4.1; C3a vs C3b, p = 0.286; C3a vs C3c, p < 0.001, ES: −0.84; C3b vs C3c, p < 0.001, ES: 0.81). We checked the location of cells from each of the three groups to determine whether they might be differentially distributed throughout the PaS (Fig. 7F). However, cells from each group were found in each of our binned distances from the pia, and we saw no statistical difference between groups (Kruskal–Wallis test, p = 0.316). Thus, these subgroups of pyramidal neurons do not appear to occupy distinct regions but rather intermingle throughout the PaS.
Discussion
The PaS is home to several functionally specialized cell types important for the integration of spatial information, including head-direction, border, and grid cells (Taube, 1995; Boccara et al., 2010). How these functional cell types interact with one another will be key in detangling how the spatial network operates. Such connectivity studies have been performed in the MEC (Couey et al., 2013; Fuchs et al., 2016; Winterer et al., 2017) and the PrS (Peng et al., 2017). However, our knowledge of the basic cell types in the PaS is poor compared with other parahippocampal structures. We performed immunohistochemistry and electrophysiology to determine the basic neuronal composition of the PaS. Cell counts revealed differential expression of molecularly distinct neurons across both superficial to deep and DV axes. Unsupervised clustering of electrophysiological parameters identified three major groups of neurons, which showed a number of parallels with our histological data.
Lying at the interface between the three-layered hippocampus and six-layered cortex, the PaS has been varyingly described as a three- and six-layered structure (Funahashi and Stewart, 1997; Mulders et al., 1997; Glasgow and Chapman, 2007; Boccara et al., 2010; Burgalossi et al., 2011; Tang et al., 2016). We analyzed neuronal markers along the superficial to deep axis to shed more light on the issue. Overall cell density was lowest in the most superficial region, consistent with L1 of the neocortex across numerous species (Gabbott and Somogyi, 1986; Amaral et al., 1987; Meyer et al., 2010). This region was enriched with inhibitory neurons, especially reelin-expressing, in line with reports that L1 is interneuron rich, and specifically the 5HT3-receptor interneuron subclass of which reelin is a member (Tremblay et al., 2016; Keller et al., 2018). Cell density reached its peak in the central PaS, before decreasing again in the deepest region. The MEC exhibits a cell sparse lamina dessicans, which dissects the deep and superficial layers (Canto et al., 2008). In the majority of our slices, the PaS was no longer discernible from its neighboring structures beyond 500 μm, thus suggesting that the low-density region is the deepest layer of the PaS. In this deepest measured region, we saw the peak density of VGAT-Venus, and in particular PV and SOM. Similarly, in the cortex, these markers exhibit their peak densities in the deepest layers (Keller et al., 2018). Thus, it appears that the cell sparse layer and PV/SOM rich deep layers merge in the PaS. It is, however, possible that the size of our measuring bins precludes us from definitively distinguishing between different deep layers.
The hippocampus, subiculum, and MEC have all been reported to show differences across the DV axis, from gene expression and intrinsic properties to projection targets and functional separations (Moser et al., 1995; Risold and Swanson, 1996; Brun et al., 2008; Giocomo et al., 2011; Kheirbek et al., 2013; Cembrowski et al., 2016). To determine whether the PaS also exhibits DV differences, we looked at the ratio of putative excitatory to inhibitory neuron markers across this axis. We found a higher proportion of inhibitory neurons in the ventral PaS compared with the dorsal PaS, with the exception of the most superficial region. This contrasts with work in the MEC showing a greater inhibitory drive onto L2 stellate cells in dorsal MEC compared with ventral (Beed et al., 2013). It is important to note that our findings relate only to the numbers of inhibitory neurons and may not reflect functional levels of inhibition in dorsal and ventral PaS. Thus, further experiments are required to determine the consequence of this observed difference.
We observe a notably different organization in the PaS compared with its neighboring cortex. Coexpression of specific molecular markers proved an effective way of discriminating between PaS and MEC. We observed two major differences between the two areas, and identified a previously reported intermediate region between them, which we designate to the mMEC (Fujimaru and Kosaka, 1996; Ray et al., 2017). First, in the PaS, calbindin is exclusively expressed in inhibitory neurons; whereas in the MEC, it colocalizes with WFS1 and labels the pyramids of L2 (Ray et al., 2014; Kitamura et al., 2015; Ramsden et al., 2015); and second, in the PaS, reelin is predominantly expressed in inhibitory neurons; whereas in the MEC, it labels the second population of excitatory principals cells, stellate cells (Varga et al., 2010). In contrast to the MEC, where principal cells can be divided into pyramids and stellates based on electrophysiological and morphological features (Alonso and Klink, 1993; Klink and Alonso, 1997), as well as molecular markers (Varga et al., 2010; Kitamura et al., 2014; Ray et al., 2014; Fuchs et al., 2016; Winterer et al., 2017), we find that, while principal cells in the PaS do show some subgrouping with distinct physiological properties, they are largely labeled by a single marker, WFS1. Moreover, these subgroups were not defined by specific locations in the PaS, unlike in the MEC where L2 and L3 pyramidal neurons can be separated by Vm and input resistance, as well as the presence and lack of WFS1 expression, respectively (Winterer et al., 2017). These results again highlight the apparent lack of well-defined layers in the PaS, and indicate a subtler subdivision of principal cells in the PaS compared with the MEC. Earlier studies have referred to pyramidal and stellate populations in the PaS, although reported them to have similar electrophysiological profiles (Funahashi and Stewart, 1997). We saw a small population of putatively excitatory reelin cells, which did not exhibit a distinct profile and could potentially comprise stellate cells. Further definition of principal cell subclasses in the PaS may be established through application of additional techniques, such as morphological analysis, RNAi-seq (Cembrowski et al., 2018), or tracing studies to identify specific input or output targets.
Our cluster classification produced three major groups of neurons in the PaS, comprising of a group of fast-spiking interneurons, a group of nonfast spiking putative inhibitory neurons, and the putative principal cells of the PaS, which could be further subdivided into three groups. Cluster 1 was made up of cells with fast depolarization and repolarization rates, narrow APs, and high firing frequencies. Matching this profile, several Cluster 1 cells were PV-positive, and these neurons were primarily found in the central and deeper region of the PaS, paralleling our PV cell counts. PV interneurons have been shown to be key modulators of grid cell activity in the MEC (Buetfering et al., 2014; Miao et al., 2017) and are also implicated in induction and control of gamma oscillations (Traub et al., 1997; Cardin et al., 2009). Moreover, given the strong theta activity seen in this region (Tang et al., 2016) and the known roles of PV neurons in regulating theta activity (Stark et al., 2013; Amilhon et al., 2015) and controlling precision spike timing (Hu et al., 2014), it is likely that this cluster of neurons plays a key role in synchronizing local oscillatory activity in the PaS. Cluster 2 contained the greatest proportion of reelin-positive neurons of the three clusters. Furthermore, corresponding with our reelin cell counts, Cluster 2 neurons were most frequently recorded in the superficial PaS. Together with our finding that in the PaS reelin is predominantly an inhibitory marker, this indicates that Cluster 2 encompasses a group of superficial interneurons. Cluster 2 cells showed the strongest theta-band activity in their subthreshold membrane potential. Oscillatory activity in the membrane potential of PaS neurons has been previously described and, moreover, was found to be more prominent in superficially located neurons, including interneurons, than deeper cells (Glasgow and Chapman, 2007, 2008). Cluster 3 covered the major putative principal cell population of the PaS and contained many WFS1-positive neurons. A closer look into this cluster of cells revealed three subgroups exhibiting distinct electrophysiological features. All three of these subgroups expressed high proportions of WFS1. Small numbers of reelin cells were also found in each of these subclusters, suggesting that the non-VGAT-Venus reelin-expressing neurons do not form a physiological subclass of neurons. A previous study of PaS neurons in the rat reported subtle differences in electrophysiological properties between deep and superficial principal cells (Funahashi and Stewart, 1997). However, this was based on small sample sizes and may have also included some interneurons. While we are not able to definitively rule out that our Cluster 3 contains only excitatory neurons, the large sample size combined with the abundance of WFS1 labeling provides us with reasonable confidence that our measurements describe well the principal cell population in the PaS. Our unbiased clustering approach did not detect physiological differences between deep and superficial principal cells but rather found that all three of our identified principal subgroups were dispersed throughout the PaS. Thus, this discrepancy may be result of a difference in species, or different experimental and analytical approaches. We saw axons of Cluster 3 cells projecting to both the MEC as well as toward the alveus, consistent with anatomical studies reporting PaS neurons projecting to both the MEC and the contralateral PaS (Köhler, 1985; van Groen and Wyss, 1990; Caballero-Bleda and Witter, 1993, 1994; Agster and Burwell, 2013). It is not known whether the same neurons project both contralaterally and ipsilaterally or whether this may be a potential distinction between different subclasses of these neurons. Cluster 3 cells showed subthreshold theta oscillations in their membrane potential; and although weaker in power than Cluster 2 cells, given the proportion of PaS neurons that this group encompasses, it is likely that these cells contribute significantly to the theta activity observed in this region.
In conclusion, we provide a detailed overview of the molecular and physiological characteristics of neurons in the PaS. We observe several distinguishing features that stand the PaS apart from the neighboring MEC. These clear differences provide future investigations with a well-defined means to avoid grouping of multiple structures or misallocation of measurements, which may potentially mask functional differences between regions. Moreover, our work defines molecular and physiological characteristics of cell types that will be key to our further understanding of the navigational and wider parahippocampal circuitry.
Footnotes
This work was supported by Deutsche Forschungsgemeinschaft Grants SFB 958 and Exc 257, Bundesministerium für Bildung und Forschung Bernstein Center for Computational Neuroscience Berlin Grant 01GQ1001A, and Bernstein Focus Learning Grant 01GQ0972. We thank Susanne Rieckmann and Anke Schönherr for excellent technical assistance; Dr. Atsushi Miyawaki for providing the pCS2-Venus plasmid; Dr. Yuchio Yanagawa for generous donation of VGAT-Venus mice; Prof. Susumu Tonegawa for generous donation of WFS1-Cre mice; and Jörg Breustedt, Constance Holman, and John Tukker for discussions and comments on the manuscript.
The authors declare no competing financial interests.
- Correspondence should be addressed to Dietmar Schmitz at dietmar.schmitz{at}charite.de or Rosanna P. Sammons at rosanna.sammons{at}charite.de.