(C) PLOS One [1]. This unaltered content originally appeared in journals.plosone.org.
Licensed under Creative Commons Attribution (CC BY) license.
url:https://journals.plos.org/plosone/s/licenses-and-copyright

------------



Conjunctive spatial and self-motion codes are topographically organized in the GABAergic cells of the lateral septum

['Suzanne Van Der Veldt', 'Mcgill University', 'Douglas Mental Health University Institute', 'Montreal', 'Guillaume Etter', 'Coralie-Anne Mosser', 'Frédéric Manseau', 'Sylvain Williams']

Date: 2021-09

The hippocampal spatial code’s relevance for downstream neuronal populations—particularly its major subcortical output the lateral septum (LS)—is still poorly understood. Here, using calcium imaging combined with unbiased analytical methods, we functionally characterized and compared the spatial tuning of LS GABAergic cells to those of dorsal CA3 and CA1 cells. We identified a significant number of LS cells that are modulated by place, speed, acceleration, and direction, as well as conjunctions of these properties, directly comparable to hippocampal CA1 and CA3 spatially modulated cells. Interestingly, Bayesian decoding of position based on LS spatial cells reflected the animal’s location as accurately as decoding using the activity of hippocampal pyramidal cells. A portion of LS cells showed stable spatial codes over the course of multiple days, potentially reflecting long-term episodic memory. The distributions of cells exhibiting these properties formed gradients along the anterior–posterior and dorsal–ventral axes of the LS, directly reflecting the topographical organization of hippocampal inputs to the LS. Finally, we show using transsynaptic tracing that LS neurons receiving CA3 and CA1 excitatory input send projections to the hypothalamus and medial septum, regions that are not targeted directly by principal cells of the dorsal hippocampus. Together, our findings demonstrate that the LS accurately and robustly represents spatial, directional as well as self-motion information and is uniquely positioned to relay this information from the hippocampus to its downstream regions, thus occupying a key position within a distributed spatial memory network.

To resolve these outstanding questions, we functionally characterize hippocampal connectivity of LS and asked how information-rich LS GABAergic neurons are in relation to their main pyramidal inputs in dorsal CA3 and CA1. To this end, we used head-mounted miniaturized microscopes to record approximately 2,000 LS neurons across the anterior–posterior and dorsal–ventral axes of the LS, and compared their firing characteristics to pyramidal cells recorded in CA1 and CA3, as animals navigated linear track and open field environments. Results demonstrate that the LS accurately and robustly represents spatial, speed, and directional information in an anatomically organized fashion. While the information of LS neurons is generally comparable to CA3 and CA1, our tracing studies suggest that LS information is sent to downstream brain regions such as the MS and hypothalamus that are not directly connected by the principal cells of the dorsal hippocampus.

Beyond their spatial tuning properties, hippocampal place cells are also noted for the rapid reorganization of the place fields across time (on the order of days) and navigational contexts, a phenomenon termed “remapping” [ 34 – 38 ]. If the LS spatial code is directly inherited from its hippocampal inputs, one would expect comparable dynamics over a similar timescale. While the LS, similar to the hippocampus, has been implicated in behaviors that may require a stable representation of space over time, including context-reward [ 39 , 40 ], contextual fear conditioning [ 16 , 41 – 44 ], spatial learning and memory [ 32 , 45 , 46 ] remapping dynamics in the LS remain largely understudied. Thus, whether and how the LS spatial code changes across time is crucial to clarifying its relationship to the upstream hippocampal spatial code, with important implications for its involvement in this broader spatial memory network and how it supports behaviors across longer timescales.

To date, functional characterizations of the LS have often focused on the spatial coding properties of these neurons, primarily due to the extensive excitatory input from positionally tuned hippocampal pyramidal neurons, commonly referred to as “place cells” [ 4 , 24 – 26 ]. Yet, this body of work has not yet yielded a consensus. Several studies have reported place-like cells in the LS, but the number, information content, and stability vary widely from study to study [ 20 , 27 – 31 ], while others described a lack of canonical place cells altogether [ 4 ]. As previous studies often depended on variable criteria of what defines a place cell, estimates range from 5.3% [ 4 ] to 56.0% [ 20 ] of LS cells classified as spatially modulated. Strikingly, even estimates by the same authors on data acquired from the same subjects engaging the same task ranged between 26.5% [ 32 ] to 56.0% [ 20 ]. When reported, LS place cells are typically described as of lesser quality than classic hippocampal place cells, with lower within-session stability and larger place fields [ 30 , 31 ]. Adding to this complexity, the LS is a large structure that is cytoarchitecturally subdivided into dorsal, intermediate, and ventral subregions that spread across its anterior–posterior axis [ 5 , 26 , 33 ]. This suggests that, in addition to subjective criteria, the striking differences in LS spatial coding characteristics between studies may be in part a product of variations in recording locations within the region. In addition to positional information, LS neurons may also encode direction and self-motion information, including acceleration and velocity [ 20 , 30 , 31 ].

We also assessed the topographical distribution of direction and self-motion tuned cells within the LS. We observed a relatively strong dorsal–ventral gradient in the distribution of stable (within-session stability > 0.5) directionally modulated cells (linear regression, R 2 = 0.3837, p = 0.0061, n = 17 mice; Fig 7C ), with most directionally modulated cells located at the more dorsal regions of the LS, but we did not observe such a gradient along the anterior–posterior axis (linear regression, R 2 = 0.0544, p = 0.0615, n = 24 mice; Fig 7C ). Velocity-modulated cells were primarily found at the ventral pole of the LS (linear regression, R 2 = 0.1836, p = 0.0367, n = 24 mice), but no such gradient was found along the anterior–posterior axis (linear regression, R 2 = 0.00348 p = 0.0874, n = 24 mice). Acceleration-modulated cells were found evenly distributed throughout the LS (DV: R 2 = 0.0133, p = 0.2928; AP: R 2 = 0.01332, p = 0.2929, n = 24 mice).

(A) Strategy to record from different anterior–posterior levels in LS (left) and implantation sites covered. (B) Strategy to approximate cell location. Background: maximal projection of representative recording; red: outline of approximated GRIN lens position; white: distance from the center of the GRIN lens to the cells of interest. (C) Top: for 0.2 mm bins along dorsal–ventral axis, proportion of stable (within-session stability > 0.5) spatially modulated cells (n = 24 mice), directionally modulated cells (n = 17 mice), velocity cells (n = 24 mice), and acceleration cells (n = 24 mice), respectively. Bottom: same as top for anterior–posterior axis. Each red dot represents the average per animal per bin along the anterior–posterior axis (left) and along the dorsal–ventral axis. Black lines indicate 95% confidence intervals. *, p < 0.05, **, p < 0.01. Test used in C: linear regression. The underlying data can be found in S1 Data . GRIN, gradient refractive index; LS, lateral septum; ns, not significant.

Next, we examined whether spatially modulated cells recorded in LS had spatial properties that were dependent on the localization of their recording. To measure this, we classified the location of gradient refractive index (GRIN) lens implants along the dorsal–ventral and anterior–posterior axis ( Fig 7A ). Postmortem histological verification of GRIN lens implant combined with within-animal analysis of the location of recorded cells enabled us to approximate the location of each cell ( Fig 7B ). While we observed no differences in cell activity characteristics, such as bursting index or activity probability along the anterior–posterior axis, medial–lateral axis, or dorsal–ventral axis ( S21 Fig ), we found a pronounced increase in the portion of stable spatially modulated cells (within-session stability > 0.5) along the anterior–posterior axis (linear regression, R 2 = 0.1071, p = 0.0022, the proportion of stable cells per 0.2 mm bin per animal, n = 24 mice; Fig 7C ). We observed a similar gradient along the dorsal–ventral axis, with a larger proportion of stable spatial cells found at the more dorsal regions of the LS, although this trend failed to reach significance (linear regression, R 2 = 0.1399, p = 0.0718, n = 24 mice; Fig 7C ). Taken together, these data strongly suggest that regions of the LS that receive stronger innervation from the dorsal hippocampus have a larger proportion of cells that reliably encode space. This suggests that this information could be directly inherited from the hippocampus, which should be tested using targeted inactivation of the hippocampal pyramidal inputs to LS.

To map out the monosynaptic connections between hippocampal and LS neurons, we injected the anterograde transsynaptic Cre-expressing AAV1 viral vector [ 65 ], in either dCA1 or dCA3 ( S19 and S20A Figs). We subsequently injected a Cre-dependent eYFP expressing virus in the LS to visualize cells that receive direct hippocampal input ( S20B Fig ). Importantly, this approach allowed us to quantify the cells along the LS anterior–posterior and dorsal–ventral axis receiving direct hippocampal inputs and assess their distribution density within the anterior regions of the LS (Bregma +1.0 to 0.0 anterior–posterior). Because we can only observe secondary labeled cell bodies at locations where there was infusion of the secondary, Cre-dependent virus, we used a relatively large injection volume (total 800 nL, targeted at the intermediate LS). Both dorsal dCA1 and dCA3 injections resulted in a majority of labeled LS cell bodies in dorsal regions compared to intermediate and ventral LS ( S20D–S20F Fig ). Hence, LS neurons receiving direct inputs from dorsal CA1 and CA3 appear to be localized at more dorsal–posterior regions as compared to more ventral–anterior regions of LS.

(A) Diagram of dual-viral injection strategy for anterograde transsynaptic tracing. The diagram is based on dorsal CA1 targeted injection; dCA3 is also used for injections, together with coronal sections showing primary injection sites in dorsal CA1 (top) or dorsal CA3 (bottom). Red: tdTom expression; blue: DAPI, with zoomed images showing tdTom-positive cell bodies are predominantly located in the pyramidal layer (bottom). For additional images of spread of injection, see S19 Fig . (B) eYFP-positive cell bodies at the anterior dorsal LS and fibers at the level of the MS following dCA1 injection (left) and dCA3 injection (right). Inset left: eYFP-positive fibers at the level of MS. (C) eYFP-positive axons are seen bilaterally at the level of the LH following dCA1 injection (left) and dCA3 injection (right). (D) Injection strategy for Cre-dependent AAV. Synaptag mediated tracing in the LS. (E) Coronal section of dorsal LS, with synaptophysin-bound eYFP at the MS. Top right: zoomed images showing transduction at the injection site. Bottom right: eYFP-positive fibers. (F) Schematic with proposed connections of the LS within the hippocampal network. Scale bars: B, left and right: 800 μm. C, left and right: 500 μm. The underlying data can be found in S1 Data . DBB, diagonal band of Broca; DG, dentate gyrus; DM, dorsomedial hypothalamic nucleus; HPC, hippocampus; LH, lateral hypothalamus; LS, lateral septum; LSd, dorsal lateral septum; LSi, intermediate lateral septum; LSv, ventral lateral septum; MS, medial septum.

To understand the function of this downstream copy of the hippocampal spatial code, we assessed the downstream targets of the dCA1/dCA3 to LS projection. First, we used a Cre-dependent, anterograde tracing approach to confirm LS projections to the hypothalamus and ventral tegmental area (VTA) in the mouse ( S18 Fig ), as it was initially described in the rat [ 3 , 5 , 25 ]. Next, we leveraged the anterograde transsynaptic properties of Cre-expressing AAV1 viral vector [ 65 ] and assessed the targets of the dorsal CA1-lateral septum (dCA1-LS) as well as dorsal CA3-LS (dCA3-LS) projections ( Fig 6 ). For either dCA1 or dCA3 injections (Figs 6A and S19 ), we observed a direct pathway from the principal cells of the hippocampus, via the LS, leading to dense innervation of the MS ( Fig 6B ) and the hypothalamic regions ( Fig 6C ). We observed that dCA1-LS preferentially targets the LH, and dCA3-LS targets the hypothalamus more broadly as well as the nuclei of the medial zone. Whether the GABAergic cells of the LS form functional synapses at the level of the MS/diagonal band complex is debated [ 5 , 26 , 66 – 68 ]. Thus, in order to determine whether the projections observed at the level of the MS with both our anterograde tracing approaches constituted passing fibers or synaptic connections, we used an AAV-Flex-Synaptag [ 69 ], a Cre-dependent viral construct to express mRuby in the cytosol and eGFP bound to synaptophysin, a protein primarily expressed in the presynaptic endings of cells participating in synaptic transmission ( Fig 6D ). For this, we used a CaMKIIα-Cre mouse line, expressing Cre only in CAMKII-positive cells, a marker that is abundant in the GABAergic cells of the LS but absent in the MS, thereby preventing any unspecific labeling. After expressing AAV-Flex-Synaptag in the intermediate LS, we observed significant synaptophysin-bound projections in the MS ( Fig 6E ), confirming the existence of a synaptic interface between the LS and the MS. Together, our work shows that LS receives direct inputs from dorsal CA1 and CA3 and, in turn, projects to regions that are not directly receiving inputs from the hippocampus itself ( Fig 6F ).

We compared the proportion of LS cells significantly modulated by each of these variables with the proportion of those cell types found in dCA1 and dCA3. The proportion of directionally modulated cells was significantly different across regions (LS: 27.89 ± 2.16, dCA1: 46.11 ± 4.98, dCA3: 28.92 ± 4.52; Kruskal–Wallis, H(3) = 6.586, p = 0.0371, Fig 5B ) with a higher proportion of directionally modulated cells in CA1 as compared to LS (Dunn’s multiple comparisons test, p = 0.0340; Fig 5B ). The proportion of velocity encoding cells was not significantly different between LS (20.43 ± 2.598), CA1 (24.38 ± 7.73), or CA3 (20.04 ± 3.62; Kruskal–Wallis, H(3) = 0.1016, p = 0.9504, Fig 5D ). Similarly, no difference was found in the proportion of acceleration encoding cells between LS (25.11 ± 2.34), CA1 (24.14 ± 4.13), or CA3 (22.60 ± 2.98, Kruskal–Wallis, H(3) = 0.3337, p = 0.8463; Fig 5F ). We found a proportion of cells that were significantly tuned to more than one of these variables in both LS as well as dCA1 and dCA3 (Figs 5G , S15A–S15D and S17 ), with 22.03% of LS cells (271/1,230 cells) encoding more than one modality ( Fig 5H ). Together, this conjunctive coding for location, directionality, and velocity indicates that LS cells fire in response to more complex navigational features, similarly to the hippocampus.

( A) Left: activity of an example LS neuron during free exploration in an open field; top: HD (red); middle: binarized activity (yellow); bottom: raw calcium activity (blue). Right: corresponding polar plot indicating the probability of the cell being active as a function of the animals’ HD with (black, p(active | direction); red lines indicate 95% upper and lower percentile. Blue line indicates the normalized time spent for each direction. MI calculated using 40 bins of 9° ( S15 Fig ). (B) Proportion of significantly directionally modulated cells in LS, dCA1, and dCA3. (C) Left: activity of an example LS neuron during free exploration modulated by velocity (top, red); middle: binarized activity (yellow); bottom: raw calcium activity (blue). Right: Tuning curve for velocity for the example cell (red) with 95% upper and lower percentile in gray. MI calculated using 20 bins. (D) Proportion of significantly speed modulated cells in LS, dCA1, and dCA3. (E) Left: activity of an example LS neuron during free exploration modulated by acceleration (top, red); middle: binarized activity (yellow); bottom: raw calcium activity (blue). Right: Tuning curve for acceleration for the example cell (red) with 95% upper and lower percentile in gray. MI calculated using 20 bins. (F) Comparison of proportions of significantly modulated cells for each region. (G) Example of an LS cell that is both significantly head direction modulated (left), as well as spatially modulated (right). (H) Left: proportion of cells that are significantly modulated by only one modality (gray), 2 modalities (yellow), 3 (red) or all 4 of the investigated variables (black). The number above the bars indicates the absolute number of cells found to be modulated in the total population (n = 1,230 cells, n = 19 mice). Right: absolute proportion of cells modulated by any combination of variables. Test used in B, D, F: Kruskal–Wallis test with Dunn’s multiple comparisons test. *, p < 0.05. The underlying data can be found in S1 Data . A, acceleration; D, direction; HD, head direction; LS, lateral septum; MI, mutual information; ns, not significant; S, spatial coding; V, velocity.

In CA1, place cells have been found to carry directional [ 59 – 61 ] and speed-related information [ 61 – 64 ]. Previous studies found that subsets of LS neurons show some degree of modulation by direction of travel [ 31 ] as well as velocity and acceleration [ 20 ] through correlative measures in spatial alternation or T-maze navigation tasks, where the relationship between cell activity patterns and movement-associated variables may be confounded with task-dependent variables. Here, we again employed an information-theoretic approach to compute the MI between calcium activity and each of these self-motion correlates in a subset of mice recorded during a 15-minute free foraging task in the open field. We found that 28.13% of LS cells are significantly modulated by head-direction (346/1,230 cells, n = 19 mice; Figs 5A and S15 ), 18.70% by velocity (230/1,230 cells; Fig 5C ) and 24.63% by acceleration (303/1,230 cells; Fig 5E ). We assessed the stability of LS directional and self-motion tuning over short (3 days) and longer timescales (8 days; S16 Fig ), and we observed a subset of LS directionally modulated cells that were stable over time ( S16A Fig ). Mean tuning vectors correlation of aligned LS cell pairs significantly increased over days (two-way ANOVA, F(2,774) = 3.682, p = 0.0256, main effect of time) with LS cells being statistically more stable than shuffled surrogates (F(1,774) = 54,42, p < 0.0001, main effect of shuffling; S16C Fig ). For velocity or acceleration tuning, we did not observe such stable tuning over time (two-way ANOVA, velocity: F(1,506) = 2.609, p = 0.1069, acceleration: F(1,320) = 0.1986, p = 0.6561, main effect of shuffle; S16D and S16E Fig ).

We next assessed the significant tuning map correlations on subsequent day pairs for those cells that were significantly spatially modulated on day 1 and were subsequently found on all recording days ( Fig 4D ), as well as those that passed our criteria for spatial modulation on all days ( Fig 4E ). We then compared the place field correlation for different day pairs (F(5,363) = 2.625, p = 0.0239, interaction effect), and we found that on shorter timescales, tuning maps for LS cells (0.316 ± 0.0507) are significantly less correlated than CA1 (0.497 ± 0.0411; Fisher’s LSD day 1 to 2, p = 0.0153; Fig 4F ). On the other hand, the correlation for day pairs 3 to 8 is higher for the LS (0.369 ± 0.0513) compared to dorsal CA1 (0.174 ± 0.0669; Fisher’s LSD, p = 0.0098; Fig 4F ). Together, this suggests that a subset of LS spatial cells encodes spatial information over longer periods of time.

( A) Experimental setup (top) with a representative example of an animal implanted in the LS and aligned spatial footprints of cells recorded over days (bottom). (B) Tuning maps for 2 sets of stable cells recorded over all days, with each row being one aligned cell. Within-session correlation indicated in blue. Tuning map correlation indicated at the bottom in red. (C) Significant tuning map correlation for aligned cell pairs (black) vs. shuffled pairs (red) for progressive days for LS (day 1–2, n = 157 cells; day 2–3, n = 122 cells; day 3–8, n = 129 cells; n = 5 mice) and dorsal CA1 (day 1–2, n = 165 cells; day 2–3, n = 122 cells; day 3–8, n = 99 cells; n = 3 mice). (D) Significant tuning map correlations for all cells found on all days for LS (n = 84) and CA1 (n = 86). (E) Matrix of mean tuning map correlation for all aligned cells that were significantly spatially modulated on each day, for LS (n = 23–37 cells, n = 5 mice) and CA1 (n = 23–41 cells, n = 3 mice). (F) Mean tuning map correlation for data shown in E. *, p < 0.05, **, p < 0.01, ****, p < 0.0001. Test used in C, two-way ANOVA, with Sidak’s multiple comparisons test, F, two-way ANOVA, with Fisher’s LSD post hoc test. The underlying data can be found in S1 Data . LS, lateral septum.

To test whether the LS and the hippocampus exhibit comparable evolutions of their spatial code over time, we recorded from animals implanted in LS and dorsal CA1 in a novel open field over both short (3 days) and longer timescales (8 days; Figs 4 and S14 ). We assessed the stability of the spatial map across sessions and observed a subset of LS spatially modulated cells that were stable over time ( Fig 4B ). Correlating the tuning maps of aligned LS cell pairs over days lead to a significant increase in mean pairwise correlation value over days (two-way ANOVA, F(2,808) = 3.259, p = 0.0389, main effect of time) with LS cells being statistically more stable than shuffled comparisons (F(2,808) = 277.0, p < 0.0001, main effect of shuffling; Figs 4C and S14C ). We observed the opposite pattern for pyramidal cells recorded from dorsal CA1, with a significant decrease in mean pairwise correlation value over days (two-way ANOVA, F(2,766) = 33.47, p < 0.0001 main effect of time, F(1,766) = 319.8, p < 0.0001 main effect of shuffle; Fig 4C ). For cells recorded in the LS, we observed the strongest within-session stability on day 3 (Kruskal–Wallis, H(4) = 19.16, p = 0.0003; S14E Fig ), as well as an increase in spatial information (Kruskal–Wallis, H(4) = 13.27, p = 0.0041; S14B Fig ). There was no significant increase in the proportion of stable cell pairs over subsequent days (one-way ANOVA, F(3) = 0.400, p = 0.9537; S14D Fig ), or the proportion of spatially modulated cells per animal for each days (one-way ANOVA, F(3,16) = 1.618, p = 0.2247; S14E Fig ).

When decoding the location of the animals using bootstrap samples of 80 cells from each region ( Fig 3F ), decoding error was significantly lower in LS (20.16 ± 1.48 cm) compared to shuffled surrogates (26.03 ± 0.84 cm; two-way RM ANOVA, F(1,13) = 89.09, p < 0.0001 for main effect of shuffle), even when decreasing the number of neurons used to only 40 neurons ( S13A and S13B Fig ). Omission of temporal filtering recapitulated these results ( S13E Fig ). The same held true for decoders trained using CA1 or CA3 data, respectively ( S13C and S13D Fig ). Strikingly, a decoder trained on data from LS does not perform significantly worse than a decoder trained on CA1 or CA3 data (F(2,13) = 2.186, p = 0.2384 for main effect of region; Fig 3H ). Similarly to recordings on the linear track, we used an activity cutoff of P(A) > 0.001 to confirm that the differences observed in decoding error between regions are not due to the inclusion of a large number of pseudo-silent cells in the bootstrapped sample ( S13F Fig ).

( A) Example tuning maps of spatially modulated cells recorded from LS, dorsal CA1, and dorsal CA3 in a 45 × 45 cm open field (3 × 3 cm bins). (B) Proportion of spatial cells per animal (LS: n = 28 mice; dCA1: n = 6 mice; dCA3: n = 8 mice). (C) MI (bits) per binarized event for all cells recorded from each region (LS: n = 1,899 cells from n = 28 mice; dCA1: n = 1,031 cells, n = 6 mice; dCA3: n = 534 cells, n = 8 mice). (D) Within-session stability for spatial cells (LS: n = 734 spatial cells from n = 28 mice; dCA1: n = 424 spatial cells, n = 6 mice; dCA3: n = 138 spatial cells, n = 7 mice). (E) Left: mean dispersion computation. Right: mean dispersion for all spatial cells recorded from each region. (F) Method for computing the mean decoding error. (G) Bootstrapping approach using 80 randomly selected cells, for 30 bootstrapped samples. (H) Mean decoding error for LS (n = 8 mice), CA1 (n = 4 mice), and CA3 (n = 4 mice). *1, p < 0.05, **, p < 0.01, ****, p < 0.0001. Test used in B, one-way ANOVA, C–E, Kruskal–Wallis, Dunn’s multiple comparison test, H, two-way ANOVA, Sidak’s multiple comparisons test. The underlying data can be found in S1 Data . LS, lateral septum; MI, mutual information.

We compared the place field properties of cells recorded in LS to those of dorsal CA1 and CA3 ( Fig 3 ). For each animal, we computed the portion of significantly positionally modulated cells per region, which was not significantly different across LS (36.71 ± 2.16%), dCA1 (38.32 ± 9.36%), and dCA3 (28.39 ± 6.73%; one-way ANOVA, F(2,39) = 1.105, p = 0.3414; Fig 3B ). We computed the information for each binarized event and observed that the LS (1.020 × 10 4 ± 7.56 × 10 7 bits/binarized event) carries significantly less information per binarized event than CA1 (1.200 × 10 4 ± 1.067 × 10 6 bits/binarized event; Kruskal–Wallis, H(3) = 264.6, p < 0.0001; Dunn’s multiple comparisons test, p < 0.0001) and CA3 (1.199 × 10 4 ± 1.605 × 10 6 bits/binarized event; p < 0.0001). As expected on the basis of prior results [ 4 , 28 , 30 , 31 ], within-session stability of LS cells (0.31 ± 0.011) was significantly lower than that of spatially modulated cells recorded in CA1 (0.44 ± 0.017; Kruskal–Wallis, H(3) = 89.52, p < 0.0001; Dunn’s multiple comparisons test, p < 0.0001) and CA3 (0.56 ± 0.027; p < 0.0001; Fig 3D ), and place fields were more dispersed (LS: 14.47 ± 0.13 cm; CA1: 12.00 ± 0.207 cm; CA3: 10.61 ± 0.35 cm; Kruskal–Wallis, H(3) = 155.7, p < 0.0001; Fig 3E ). Similar to recordings in a 1D environment, we observed large differences in P(A) between spatially modulated cells recorded from LS and those from dorsal CA1 and CA3 ( S12A Fig ). Therefore, we also assessed for each region the proportion of spatially modulated cells, split-half stability and mean-dispersion using an activity cutoff of P(A) > 0.001, and confirmed our results ( S12B Fig ).

With theta-rhythmic cells in the LS having higher firing rates than theta-rhythm-independent cells [ 56 ], and the suggestion that theta-rhythmic cells could receive direct hippocampal inputs (though see [ 57 , 58 ]), we asked whether LS cells with higher firing rates were also more likely to be spatially modulated. Indeed, spatially modulated cells were significantly more active than nonmodulated cells (Mann–Whitney test, U = 342,219, p < 0.0001; S9G and S9H Fig ) and displayed a higher bursting index, defined as the probability of a cell being active, given it was already in an active state, or P(A t |A t-1 ), than nonspatially modulated cells (Mann–Whitney test, U = 368,276, p < 0.0001; S9H Fig ).

In order to assess spatial coding during free exploration in a 2D environment, we recorded Ca 2+ activity of 1,899 GABAergic neurons in 28 mice implanted in distinct subregions of the LS, while the animals were freely exploring a novel open field. Out of 1,899 recorded cells, 37.80% (718/1,899 cells, n = 28 mice) of LS cells displayed significant spatial information (p < 0.01; Figs 3A and 3B and S9 ), as compared to 31.33% (323/1,031 cells, n = 6 mice) in CA1 and 25.84% (138/534 cells, n = 8 mice) CA3 cells. Despite the LS’ known involvement in feeding-related behaviors [ 11 – 14 ], we did not observe any increased number of cells with firing fields around food zones, nor did we observe overrepresentation of objects or walls ( S10 Fig ). To assess reward-modulation in the context of increased task demand, we assessed spatial firing of LS GABAergic neurons recorded while animals performed a delayed nonmatching to place task in a T-maze but did not observe any significant reward code compared to shuffled surrogates ( S11 Fig ).

In order to compare decoding accuracy for LS, CA1, and CA3, a decoding score was computed ( Fig 2E ; see Methods ), and we found that LS significantly outperformed CA3 (one-way ANOVA, F(2,13) = 6.277, p = 0.0124; Holm–Sidak’s multiple comparisons test, p < 0.05), but not CA1. Overall, spatially modulated cells recorded in LS were significantly more active than those recorded in the hippocampus (H(3) = 155.1, p < 0.0001; S5A and S5B Fig ), which could be a contributing factor to the overall higher decoding accuracy using LS cells. To confirm that the differences observed in decoding error are not due to the inclusion of a large number of pseudo-silent cells in the dCA1 or dCA3 bootstrapped samples, we confirmed our findings using an activity cutoff of P(A) > 0.001 ( S8E and S8F Fig ). Similarly, to confirm that the observed differences in decoding error are not due to the use of a temporal smoothing window, we replicated our results using a decoder that omitted temporal filtering ( S8G and S8H Fig ).

Previous studies have used decoding methods to predict behavioral variables from calcium imaging data recorded in the hippocampus [ 34 , 38 , 55 ], yielding insights into the amount of information encoded by a neuronal assembly. Here, we asked whether we could reliably estimate the mouse location solely from LS neuronal activity patterns ( Fig 2 ). Using 30 bootstrap samples of 60 cells ( Fig 2B and 2C ), decoding location using LS neuronal activity significantly outperformed a decoder that was trained using shuffled data (paired t test, t(6) = 13.56, p < 0.0001; Fig 2D ), even when decreasing the number of neurons used ( S8A and S8B Fig ). Similarly, a decoder trained using CA1 or CA3 data also yielded significantly less error than one using shuffled surrogates (CA1: t(4) = 7.558, p = 0.0016; CA3: t(3) = 3.233, p = 0.0481; Figs 2D and S8C and S8D ). While using a small bootstrap sample size allowed a fair comparison between recording regions, it also induced higher error rates. To ensure the proper functioning of our decoder, we confirmed that the decoding error decreased significantly when increasing bootstrapped sample size ( S8C and S8D Fig ).

( A) Retrograde rabies tracing injection site, blue: DAPI staining, green: TVA.oG coupled to an eGFP, red: Rabies coupled to mCherry, orange: starter cells expressing both TVA.oG.eGFP and Rabies.mCherry. (B) Retrograde labeling in the dorsal hippocampus, coronal section. Bottom left: example of mCherry-positive CA1 pyramidal cell, bottom right: mCherry-positive CA3 pyramidal cells, blue: DAPI staining, red: Rabies coupled to mCherry. (For additional images, see S1 Fig ). (C) Top: diagram of one-photon calcium recording setup in LS, CA1, and CA3 in freely behaving mice, with GCaMP6f expression restricted to GABAergic cells in LS and restricted to pyramidal cells in dorsal CA1 and CA3, middle: histological verification of implantation site (see also S2 and S3 Figs), and bottom: extracted calcium transients for LS, CA1, and CA3. (D) Top: linear track paradigm, with sucrose rewards on either end. Bottom: probability of an example cell to be active given the location on the linear track (red) with 95% upper and lower percentile (gray). Examples shown are from 2 different LS mice and a representative spatially modulated cell from one CA1 and CA3 animal each. (E) Activity of cells sorted along location in the maze for each region (blue, low; yellow, high). (F) Percent of significantly spatially modulated cells for each animal for each recording region. LS: N = 15 mice. CA1: N = 5 mice, N = 6 mice. (G) Within-session stability of spatially modulated cells in each region active (LS: n = 475 cells, CA1: n = 336 cells, CA3: n = 143 cells). (H) MI per binarized event for all cells (LS: n = 1,030 cells, n = 15 mice. CA1: n = 1251 cells, n = 5 mice, CA3: n = 464 cells, n = 6 mice). *, p < 0.05, **, p < 0.01, ****, p < 0.0001. Test used in F–H: Kruskal–Wallis with Dunn’s multiple comparisons test. The underlying data can be found in S1 Data . eGFP, enhanced green fluorescent protein; LS, lateral septum; LSd, dorsal lateral septum; LSi, intermediate lateral septum; MI, mutual information; SFP, spatial footprint.

Discussion

While spatial coding has been extensively characterized in the hippocampal formation, how downstream regions integrate this information has only recently begun to receive attention. The LS has reemerged as a critical region implicated in a space encoding network, although electrophysiological recordings of the LS in rats have led to disparate estimates of the quantity and quality of place cells in the region [4,20,28,32]: LS place cells recorded on a circular track were described to be almost absent [4] but were found to be abundant in reward-seeking tasks [20,32]. Using a large calcium imaging dataset and unbiased information metric and decoding approaches, we found that 37.80% of GABAergic LS cells robustly code for space during free exploration in an open field and 43.90% of cells during linear track alternation. One of the major features of the linear track is the induction of directional place fields in LS, which may underlie some of the differences we observed in 1D versus 2D spatial coding characteristics.

In addition to spatial information, we found that the LS also reliably encodes velocity, acceleration, and directional information, suggesting that the LS encodes more complex navigational features than previously thought. Neuronal representations in the hippocampus have been found to change over time, with place code ensembles changing rapidly: An estimated approximately 40% [34] to approximately 75% to 85% [38] of CA1 neurons were found to remap over days. The stability of the LS spatial code has received little attention to date. In the current study, we found a subpopulation of cells in the posterior LS that display stable place fields over 8 days. The functional role of such stability remains to be elucidated. One possibility is that this ensemble may mediate the stable encoding of contexts over longer durations of time, which could account for the critical importance of the LS for a number of spatial behaviors, including contextual fear conditioning [16,41–44], context-reward associations [39,40], and radial-maze navigation tasks [45,46].

Similar to the change in properties of hippocampal place cells along the dorsal–ventral axis [70,71], a previous work has hinted at a dorsal to ventral organization of the LS place code: Electrophysiology studies recorded a slightly larger proportion of LS place cells in the more dorsal regions [30]. A complementary anatomical microstructure has been previously described for the rate-independent phase code in the LS, where the strength of the phase code increased as a function of recording depth along the dorsal–ventral axis [4]. Here, we find that lateral septal spatially modulated cells are arranged along an anterior–posterior gradient similar to the gradient of hippocampal inputs into the LS—a finding that helps to reconcile the variation in previously reported estimates of spatial cells in the LS, which ranged between 5.3% of cells recorded along the dorsal–ventral axis [4] up to 56% recorded at the most dorsal level of the LS [20]. This pattern suggests that the inhibitory neurons of the LS may inherit a place code from the hippocampus, with subregions receiving dorsal hippocampal inputs being most similar to the classical hippocampal place cells. One potential caveat of using miniaturized microscope when comparing hippocampus and LS coding properties is that of circuit damage due to GRIN lens implantation, which may be particularly the case for CA3 implants due to the occasional damage to CA1. Interestingly, the LS place code described here is not the only case of GABAergic neurons inheriting some spatial properties from hippocampal pyramidal cells: Interneurons within CA1 have been shown to exhibit significant spatial modulation and comparable information as pyramidal place cells, although with a higher firing rate and greater spatial dispersion [72]. It should be considered that, although the hippocampus seems a likely source of the positional information encoded by the LS, this information could also arise from incoming projections from any other region projecting to LS, such as the entorhinal cortex [5,25], which could account for part of head direction and velocity tuning.

In addition to spatial modulation of LS cells, we observed a significant number of velocity- and acceleration-modulated cells in the LS. These cells have been previously described by others on the basis of correlative measures, although estimated proportions vary widely: For velocity tuning in reward-seeking tasks, estimates range from almost absent [4] to almost 60% of cells [20]. For acceleration, approximately 45% of LS cells were found to show some degree of correlation, with almost 30% of recorded cells correlated with both speed and acceleration [20]. Here, using information metrics and a free exploration task in which firing characteristics are not confounded with task parameters, we find that 18.70% of cells were modulated by velocity and 24.63% by acceleration. Cells reliably coding for both velocity and acceleration were relatively rare (2%). Velocity and acceleration cells were distributed dissimilarly in the LS, with velocity cells being more abundant toward the ventral pole, whereas acceleration-modulated cells were found to be distributed equally. This raises the question whether velocity and acceleration information are inherited from different areas upstream of the LS. Interestingly, cells coding for direction were more abundant toward the dorsal portion of the LS but distributed equally along the anterior–posterior axis. Previous studies have identified classical head direction cells in multiple, well-described circuits, including the postsubiculum [73,74], anterodorsal thalamic nucleus [75], mammillary bodies [76], entorhinal cortex [77], retrosplenial cortex [78,79], and parasubiculum [80,81]. A likely source of directional information to the LS is the entorhinal cortex [82]. Our work supports the perspective that LS neurons combine a wide range of modalities and may form a complex representation of context over longer timescales than previously reported. In addition to space, velocity, and reward information, hippocampal pyramidal cells have been shown to encode variables such as time [83,84], odor [85,86], and sound frequency [87]. Whether this information is also relayed to the LS remains to be elucidated. Despite this interconnectedness with the LH and previous reports of LS place cells being skewed toward reward in a spatial navigation task [32], we did not observe any overrepresentation of LS spatially modulated cells around food zones during free exploration or around the reward zone in a T-maze non-match-to-place paradigm.

The question arises of what the function could be of having a region downstream of the hippocampus that expresses such seemingly similar coding characteristics. Previous work estimated the convergence of hippocampal efferents onto LS neurons to be 20 to 800 times denser than to any of its cortical targets [4]. Understanding how individual LS neurons integrate the thousands of synaptic inputs they receive from these hippocampal pyramidal neurons will be critical to understanding how the hippocampal map is processed downstream. Through the process of synaptic integration, a target neuron can fire an action potential upon receiving sufficient temporally coincidental excitatory input on its dendrites. The activation of an LS neuron could thus require multiple hippocampal pyramidal cells to spike simultaneously or in close temporal proximity. Hippocampal pyramidal cells will fire concomitantly when their place fields are in overlapping regions or in close vicinity of another. Due to the high interconnectivity of LS GABAergic neurons [88,89], the activation of one LS cell could subsequently lead to the inhibition of other neurons in the nucleus [24,90], thereby reducing noise. Thus, through this process of coincidence detection and recurrent inhibition, the spatial map could converge to be represented by a much lower number of cells. As such, the LS could accurately convey hippocampal information to downstream regions using a much more distributed code and thereby support more effective information transmission and associative learning [91,92].

Finally, our tracing work shows LS cells receiving hippocampal inputs from principal cells in dorsal CA1 and CA3 in turn project directly to the LH and the MS. Although the existence of an LS–MS projection has long been debated [5,26,66,68,93], we observed both dense innervation at the level of the MS after transduction of dCA1-input and dCA3-input receiving LS cells, as well as synaptic densities between the LS and MS. This suggests that principal cells of the hippocampus send indirect projections to the MS through LS neurons. This circuit is complementary to the well-described projections of hippocampal GABAergic interneurons directly to the MS, which may play a key role in theta-rhythm generation [94,95]. The MS plays a role in generating and propagating theta-rhythms throughout the hippocampal formation [96–98], which, in turn, organizes hippocampal place cell activity [99], as well as a behavioral role in the initiation and velocity of locomotion [100,101]. One hypothesis is that the information on locomotion-related information reported in specific MS neurons [101] may originate from the LS. Moreover, we observed a direct projection from CA1 and CA3, via the LS, to the hypothalamus. This pathway was previously shown by previously using nonspecific monosynaptic anterograde tracing [3], with similar unilateral projection patterns for CA1 and bilaterally for CA3 as described here. Here, we show for the first time that the same LS neurons that receive inputs from dorsal CA1 and CA3 project directly to the hypothalamus. The hypothalamus is a highly connected region known for its role in regulating feeding behaviors [102], arousal and motivation [103], and, more recently, learning and memory [104–107]. Additionally, the LH is crucial for the control of locomotion [108–111] and thought to mediate motivational and goal-directed processes underlying feeding [12,102,112,113].

Together, our findings show that GABAergic cells of the LS may provide the hypothalamus and MS with information about location, direction, and speed, and therefore constitute a core node within a distributed spatial memory network. Within this network, the LS may be necessary for the translation of spatial information to goal-directed or environmentally appropriate actions necessary for survival.

[END]

[1] Url: https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.3001383

(C) Plos One. "Accelerating the publication of peer-reviewed science."
Licensed under Creative Commons Attribution (CC BY 4.0)
URL: https://creativecommons.org/licenses/by/4.0/


via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/plosone/

You are viewing proxied material from magical.fish. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.