Distinct dendritic morphology across the blades of the rodent dentate gyrus

Abstract

The dentate gyrus (DG) is a hippocampal region that has long been characterized as a critical mediator of enduring memory formation and retrieval. As such, there is a wealth of studies investigating this area. Most of these studies have either treated the DG as a homogeneous structure, or examined differences in neurons along the septal-temporal axis. Recent data, however, have indicated that a functional distinction exists between the suprapyramidal and infrapyramidal blades of the DG, with the former showing more robust responses during spatial tasks. To date, few anatomical studies have addressed this functional gradient in rats, and no study has done so in the mouse. To address this, we investigated dendritic morphology and spine density in hippocampal granule cells of rats and mice using the Golgi-Cox technique. We find that granule cells from the suprapyramidal blade of the DG contains greater dendritic material in the region receiving spatial information from the medial perforant path. This provides a potential anatomical substrate for the asymmetric response of the DG to spatial input.

Graphical Abstract

Introduction

A wealth of evidence links the hippocampus with episodic memory (reviewed in Davachi, 2006; Manns and Eichenbaum, 2006; Squire and Wixted, 2011). Consistent with the complexity of memory phenomena, individual regions of the hippocampal formation are thought to make unique contributions to memory encoding and retrieval. Most theories on the contribution of the dentate gyrus (DG) emphasize pattern separation. That is, the DG is thought to extract unique information from events to differentiate their internal representation from similar memories (Aimone et al., 2011; Schmidt et al., 2012).

Consistent with the critical role the DG plays in memory, a great deal of research has investigated the structure and function of this region. The majority of this work, however, has either treated the DG as a homogeneous structure (e.g., Treves et al., 2008), or has examined the functional gradient from the septal to temporal poles of the hippocampus, emphasizing spatial information in the former and non-spatial associations like those that drive fear conditioning in the latter (e.g., Kesner, 2013). Although this is an important functional gradient within the hippocampus, recent data show that an additional gradient may exist along the transverse axis of the DG. Along this axis, the DG may be divided into the suprapyramidal (DG_sp) and infrapyramidal (DG_ip) blades (Fig. 1a). Although unit recordings have been performed almost exclusively in the DG_sp, immediate-early gene markers suggest that granule cells in the DG_sp are an order of magnitude more active during spatial tasks than the neighboring DG_ip (Chawla et al., 2005; Ramirez-Amaya et al., 2006; Marrone et al., 2012a, 2012b; Satvat et al., 2012). Moreover, the blunted response of the DG_ip occurs much later, typically several hours following behavior (Ramirez-Amaya et al., 2005; 2013). These observations have renewed interest in the factors mediating this functional gradient, including anatomical differences between the DG_sp and DG_ip.

Figure 1.

Overview of the methods used in the current experiment. (a) Following Golgi-Cox staining, granule cells were selected from the suprapyramidal (DG_sp, red overlay) and infrapyramidal (DG_ip, blue overlay) blades of the dentate gyrus (shown in a, scale bar = 200 μm). (b) Schematic of a granule cell demonstrating each of the parameters measured. The size of the dendritic arbor was measured using the Sholl method, which overlays concentric circles spaced 20 μm apart and counts the number of intersections (left). In addition, the complexity of the dendritic tree was measured by counting the number of intersections and assigning each branch a consecutively higher branch order number (right). Finally, dendritic spines were counted (inset, scale bar = 10μm) under 100x magnification under oil in each of the inner (IML, grey area nearest the soma), middle (MML), and outer (OML, grey area furthest from soma) third of the molecular layer.

Unfortunately, only a handful of studies have compared the DG_sp to the DG_ip (Seress and Pokorny, 1981; Desmond and Levy, 1982, 1985; Green and Juraska, 1985; Claiborne et al., 1990). These studies indicate that granule cells in the DG_sp have greater dendritic length and spine density. However, few of these studies have systematically examined the extent of both dendritic branches and spines across the depth of the dendritic tree. This is an important issue since the arbor of granule cells can be split into 3 parts, each with unique inputs (Amaral et al., 2007). The inner third, known as the inner molecular layer (IML, closest to the soma) contains predominantly commissural and associational fibers as well as recurrent fibers from mossy cells (e.g., Buckmaster et al., 1992), while the middle third, termed the middle molecular layer (MML) receives input from the medial perforant path, and the outer third, termed the outer molecular layer (OML) receives lateral perforant path input (e.g., Steward, 1976). Moreover, these pathways carry different kinds of information. The MML input, for instance, carries much more spatial information relative to the OML (e.g., Knierim et al., 2013). Thus, knowing the molecular layer in which the arbor of granule cells of the DG_sp and DG_ip differ may provide insight into the mechanism for the unique physiological responses of these cells during behavior. To the authors’ knowledge, no study has examined this anatomical gradient in mice, despite their widespread use in studies of DG function and the fact that the mouse DG differs from the rat in several subtle but potentially impactful ways. For instance, adult-born granule cells are greater in number and have a greater impact on the behavioral response of the rat DG relative to the mouse (Snyder et al., 2009). Moreover, the pattern of projections from the entorhinal cortex shows topographical distinctions in the two species (van Groen et al., 2002). These subtle differences may lead to considerable divergence in DG morphology or function. To address this issue, the dendritic arborization and spine density across this gradient is examined in both rats and mice using the Golgi-Cox technique.

Materials and Methods

Subjects

The current study included eight F344 rats (males aged 3-4 months) and six C57/Bl6 mice (males aged 2-6 months). All animals were housed in 12:12 light cycles with ad lib food and water, and all procedures were approved by either the Animal Care Committee of Wilfrid Laurier University (rat procedures) or the Institutional Animal Care and Use Committee of the University of Arizona (mouse). All animal work followed the guidelines of either the National Institutes of Health or Canadian Council on Animal Care.

Histology

The tissue from all animals was prepared using the Golgi-Cox technique, as previously described (Gibb and Kolb, 1998). Briefly, animals were decapitated under isofluorane and the brains were rapidly extracted, rinsed in 0.9% saline and immersed in Golgi-Cox solution for at least 14 days followed by 30% sucrose for at least 3 days. Brains were then sectioned on the coronal plane using a Vibratome at 200 μm and mounted on gelatin-coated slides. Ten fully-impregnated cells were selected from the DG_sp and DG_ip of each animal, and were imaged in a z-stack throughout the thickness of the section using a bright-field microscope (Leica Mcrosystems, Concord, ON) equipped with a digital camera (Pixellink, Ottawa, ON). Granule cells were included in the analysis if (a) the cell could be identified as a granule cell on the basis of morphology, (b) the dendritic tree appeared well impregnated and largely intact in the plane of section, and (c) the cell was not obscured with precipitate, blood vessels, or astrocytes.

Cells were selected from the septal third (rat: −2.28mm to −3.24mm from Bregma, Paxinos & Watson, 2006; mouse: −1.35mm to −2.35mm from Bregma, Allen Mouse Brain Atlas, 2015) of the hippocampus. Granule cells were randomly selected along the longitudinal axis as well as throughout the depth of the granule cell layer. However, depth within the granule cell layer affects morphology (Claiborne et al., 1990, Green and Juraska, 1985). More specifically, significant differences in granule cell morphology can be found between cells that are superficial (i.e., with cell bodies closer to the molecular layer) relative to cells that are deep (i.e., with cell bodies closer to the hilus). Because of this, the sampled granule cells were classified post-hoc (see Table 1) and analyzed to determine if there were any significant differences in the number of superficial vs deep granule cells across regions or species. The classification of cells involved bisecting the thickness granule cell layer in each perpendicular to its long axis. Those cells that contained the majority of their soma with the half closest to the hilus were classified as deep, while those predominantly in the furthest half were classified as superficial.

Table 1.

Characteristics of granule cells sampled

	Rat		Mouse
	DGsp	DGip	DGsp	DGip
Total granule cells	64	64	80	80
Superficial	34	32	46	45
Deep	30	32	34	35

Analysis

Using ImageJ (NIH, Bethesda, MD) and Metamorph (Molecular Devices, Sunnyvale, CA) software, several measures of dendritic arborization were obtained (Fig. 1). The amount of dendritic material was estimated using the Sholl technique (Sholl, 1953), which involves overlaying the cell with a series of concentric circles (20 μm apart) centered on the soma and recording the number of dendritic processes intersecting each circle to a maximal distance of 380 μm. Note that although this analysis was conducted on a 3-dimensional image stack, the measure is inherently 2-dimensional.

In addition, branching order was calculated for all cells. The primary dendrites originating from the soma are assigned a branch order of one, while dendritic processes originating from that dendrite are second order branches, and each subsequent bifurcation is assigned a progressively higher branch order (Uylings et al., 1986). Dendritic spine density was analyzed at 100x under oil immersion. In each section, the IML, MML, and OML were defined by dividing the length of the molecular layer into 3 equal parts. Dendritic spines were counted in a sample of 2 random 20-μm-long segments within each zone (i.e., the IML, MML, and OML) in each included cell. Sholl data were analyzed by repeated-measures ANOVA, while branching order and spine density measures used general factorial ANOVA, with region (i.e., DG_sp vs DG_ip) and species (i.e., rat vs mouse) as between-subject factors.

The statistical models described above were also assessed with the addition of depth (i.e., deep vs superficial granule cells) as an additional factor in order to assess possible differences in morphology based on cell depth.

Results

Consistent with previous reports in rats (Desmond and Levy, 1982; Green and Juraska, 1985), DG_sp granule cells contain more dendritic material than DG_ip granule cells (main effect of region: F_1,24 = 6.03; p = 0.02). Moreover, this increase was consistently observed in both rats and mice (main effect of species: F_1,24 = 2.37; p = 0.14). Post-hoc testing with Tukey's HSD shows that the dendritic arbor is selectively increased in the middle third of the dendritic tree (corresponding to the MML) in both rats and mice, as well as the outer third (corresponding to the OML) in mice (see Fig. 2a,b). This increase occurred without an increase in branching complexity (F_1,24 = 0.31; p = 0.58) or a change in spine density (F_3,22 = 0.37; p = 0.14), as neither parameter shows significant differences on the basis of anatomical location. This pattern of results (i.e., an increase in the number of intersections in Sholl analysis without a change in branch order) suggests an increase in the length of dendritic segments rather than an increase in the number of segments in (and thus the complexity of) the dendritic arbour.

Figure 2.

Quantification of granule cell measurements in rats (a,c,e) and mice (b,d,f). Sholl analysis (a,b) shows an increase in dendritic material in DG_sp (open diamonds) relative to the DG_ip (filled squares) in the MML (i.e., middle third) of both rats and mice, and well as the OML (i.e., outer third) of mice. Analysis of the branch orders (c,d) and the density of spines (e, f) reveals no significant difference between the DG_sp (grey) and DG_ip (white) in either species (*p < 0.05, Tukey's HSD).

The number of deep vs. superficial cells was analyzed, since previous studies have shown that granule cell morphology can vary according to their depth in the granule cell layer (e.g., Claiborne et al., 1990; Green and Juraska, 1985). No significant difference were observed on the basis of granule cell layer depth (p > 0.26 in all cases). These data suggest that differences observed are representative of differences between DG_sp and DG_ip and are not the result of sampling error.

To further determine any potential morphological differences, statistical comparison was conducted between deep and superficial cells in rats and mice in each region of the DG (see Table 2). The addition of cell depth as a factor did not result in any significant differences based on depth, or any significant interactions between depth and region (i.e., DGsp vs DGip) or species (p > 0.18 in all cases). Thus, the current study has failed to replicate previous observations showing significant differences in granule cell morphology based on cell depth. It should be noted, however, that because the study was not originally designed for this comparison, the study may be under-powered for this comparison. As a result, it remains difficult to draw conclusions from this analysis.

Table 2.

Morphology of Deep vs Superficial Cells

	Rat				Mouse
	DGsp		DGip		DGsp		DGip
	super2	deep3	super	deep	super	deep	super	deep
Sholl1
0-60 μm	6.23 ± 0.61	7.38 ± 0.71	6.93 ± 0.92	6.67 ± 1.07	12.84 ± 0.89	11.43 ± 0.85	14.43 ± 0.95	11.68 ± 0.99
61-120 μm	11.13 ± 1.12	11.46 ± 1.71	11.55 ± 1.60	12.17 ± 4.94	20.46 ± 0.91	17.93 ± 1.24	19.38 ± 1.18	18.06 ± 0.53
121-180 μm	16.85 ± 1.30	15.92 ± 0.69	14.70 ± 1.85	14.74 ± 1.90	22.20 ± 1.03	21.69 ± 1.41	19.23 ± 1.03	18.52 ± 1.02
181-240 μm	18.86 ± 1.26	19.06 ± 1.06	18.83 ± 1.68	18.01 ± 1.72	19.36 ± 0.69	20.61 ± 1.44	13.74 ± 1.37	14.55 ± 1.67
241-300 μm	19.42 ± 1.20	20.04 ± 1.29	21.32 ± 1.21	21.11± 0.95	13.72 ± 1.04	14.26 ± 1.50	5.46 ± 0.96	7.40 ± 1.29
300+ μm	54.88 ± 5.62	67.34 ± 17.89	49.86 ± 9.77	54.76 ± 15.17	4.54 ± 0.99	4.48 ± 0.78	1.14 ± 0.36	1.42 ± 0.64
Branch Order
1	2.60 ± 0.32	2.40 ± 0.39	2.85 ± 0.20	2.75 ± 0.31	2.15 ± 0.17	1.90 ± 0.19	2.36 ± 0.13	1.78 ± 0.11
2	5.52 ± 0.70	5.10 ± 0.78	5.94 ± 0.63	6.14± 0.68	4.20 ± 0.28	3.98 ± 1.34	4.24 ± 0.32	3.67 ± 0.21
3	7.67 ± 0.61	7.31 ± 0.64	7.31 ± 0.75	7.08 ± 0.73	5.61 ± 0.36	4.81 ± 0.34	4.85 ± 0.26	5.13 ± 0.39
4	6.00 ± 0.73	6.40 ± 0.63	5.78 ± 0.66	5.54 ± 0.80	4.15 ± 0.37	4.13 ± 0.59	3.86 ± 0.49	3.60 ± 0.57
5	1.73 ± 0.27	2.25 ± 0.41	1.95 ± 0.36	1.31 ± 0.34	1.43 ± 0.32	1.45 ± 0.30	1.71 ± 0.38	1.39 ± 0.29
6+	1.02 ± 0.77	1.84± 0.78	0.86 ± 0.24	0.70 ± 0.30	0.55 ± 0.19	0.67 ± 0.28	0.31 ± 0.16	0.39 ± 0.17
Spines
IML	17.28 ± 1.40	17.65 ± 1.12	19.13 ± 1.09	19.09 ± 0.66	16.76 ± 0.70	17.13 ± 0.64	16.30 ± 0.57	17.21 ± 0.52
MML	18.35 ± 1.06	18.58 ± 1.14	20.26 ± 0.79	19.57 ± 0.69	17.29 ± 0.50	17.39 ± 0.53	17.08 ± 0.34	17.34 ± 0.54
OML	16.40 ± 0.70	16.86 ± 0.87	18.71 ± 0.72	16.95 ± 0.89	16.77 ± 0.47	18.05 ± 0.46	16.55 ± 0.50	17.15 ± 0.44

¹Mean number of line crossing in Sholl analysis within each third of the molecular layer.

²Superficial: the majority of the cell body was contained within the half of the granule cell layer closest to the hilus

³The majority of the cell body was contained within the half of the granule cell layer closest to the hilus

Discussion

The selective increase in perforant path input to DG_sp may provide an anatomical substrate for the asymmetric response of the DG to spatial information. The medial perforant path contains the output of highly spatially-tuned cell populations such as grid cells (reviewed in Moser et al., 2014) as well as head-direction and border cells (Zhang et al., 2013). In contrast, the lateral perforant path contains little spatial tuning (Knierim et al., 2013). Consistent with these differences in information content, the medial perforant path also disproportionately supports performance of spatial tasks while the lateral perforant path supports non-spatial tasks such as object recognition (Myhrer, 1988; Hunsaker et al., 2007). It is intriguing to note that the expansion of dendritic material in the OML of the mouse DG_sp suggests differences in the DG response to non-spatial information between these species. That is, the rat DG_sp receives more synapses selectively from the MML, which contains highly spatially-tuned input from the medial perforant path, and thus it is intuitive that the rat DG_sp should show a much stronger response than the DG_ip during spatial tasks. In contrast, the mouse DG_sp receives more input from both the spatially-tuned medial perforant path and the non-spatially tuned lateral perforant path. This result is consistent with earlier observations that the input from the medial entorhinal cortex (to the MML) contains robust bilateral projections in rats but the projections are largely unilateral in the mouse (van Groen et al., 2001). This discrepancy suggests that the mouse DG_sp may show greater activity not only during spatial tasks, but also during hippocampus-dependent non-spatial tasks as well. This hypothesis however, needs to be tested experimentally, as the relationship between synaptic density and granule cell activity is necessarily complex.

A prime example of the complex relationship between synapse number and synaptic drive can be observed in the changes that occur in the DG as a result of aging. With progressive age, the number of synapses is reduced (Geinisman et al., 1986; 1992), but the remaining synapses become much more powerful in terms of their ability to drive granule cell responses (Barnes and McNaughton, 1980). As a result, the output of the DG of aged animals remains comparable to that of younger animals under many circumstances, and the changes that do occur are complex (reviewed in Burke and Barnes, 2006; Gray and Barnes, 2015). Thus, it may be too simplistic to assume that a significant increase in spine number will result in a comparable increase in synaptic drive.

Several other observations have been proposed to account for the DG's asymmetric activity. For example, the proliferation and survival of newly-generated granule cells follows a qualitatively different time course in the two blades. In the DG_sp, fewer neurons are generated, but those that are produced are more likely to survive (Snyder et al., 2012). Moreover, entorhinal projections activate the DG_sp more rapidly in response to a stimulus, which is thought to act along with recurrent inhibition to allow the DG_sp to out-compete DG_ip in response to stimulation (Canning and Leung, 1997). In addition, DG_sp receives twice as much supramammillary input (Wyss et al., 1979), although these projections terminate in the IML and thus cannot account for the differences observed here. In addition to the expansion of the MML-recipient zone observed here, these anatomical differences may account for the unique physiological responses of these two parts of the DG, a structure often considered to be homogeneous.

The current observations add to a growing body of literature describing novel functional gradients within the hippocampal formation. As more functional gradients are uncovered, it is likely that many anatomical observations will need to be revisited to build a more accurate model of the hippocampal formation and its contribution to learning and memory.