Abstract
Distinct computations are performed at multiple brain regions during the encoding of spatial environments. Neural representations in the hippocampal, entorhinal, and head direction (HD) networks during spatial navigation have been clearly documented, while the representational properties of the subicular complex (SC) are relatively underexplored, although it has extensive anatomic connections with various brain regions involved in spatial information processing. We simultaneously recorded single units from different subregions of the SC in male rats while they ran clockwise on a centrally placed textured circular track (four different textures, each covering a quadrant), surrounded by six distal cues. The neural activity was monitored in standard sessions by maintaining the same configuration between the cues, while in cue manipulation sessions, the distal and local cues were either rotated in opposite directions to create a mismatch between them or the distal cues were removed. We report a highly coherent neural representation of the environment and a robust coupling between the HD cells and the spatial cells in the SC, strikingly different from previous reports of coupling between cells from co-recorded sites. Neural representations were (1) originally governed by the distal cues under local–distal cue-conflict conditions, (2) controlled by the local cues in the absence of distal cues, and (3) governed by the cues that are perceived to be stable. We propose that such attractor-like dynamics in the SC might play a critical role in the orientation of spatial representations, thus providing a “reference map” of the environment for further processing by other networks.
SIGNIFICANCE STATEMENT The subicular complex (SC) receives major inputs from the entorhinal cortex and the hippocampus, and head direction (HD) information directly from the HD system. Using cue-conflict experiments, we studied the hierarchical representation of the local and distal cues in the SC to understand its role in the cognitive map, and report a highly coherent neural representation with robust coupling between the HD cells and the spatial cells in different subregions of the SC exhibiting attractor-like dynamics unaffected by the cue manipulations, strikingly different from previous reports of coupling between cells from co-recorded sites. This unique feature may allow the SC to function as a single computational unit during the representation of space, which may serve as a reference map of the environment.
- attractor dynamics
- head direction cell
- neural representation
- place cell
- spatial navigation
- subicular complex
Introduction
The hippocampal and parahippocampal networks in the medial temporal lobe play a critical role during spatial navigation by encoding the salient features of the environment. Various studies have highlighted the role of these brain regions in spatial memory (Morris et al., 1982; Moser et al., 1995; Steffenach et al., 2005; Wilson et al., 2013; Jacob et al., 2017; Robinson et al., 2020; Vandrey et al., 2020). Observations of distinct neural representations in the hippocampus and the entorhinal cortex (EC) in dynamically changing environments (i.e., the standard cue configuration of a familiar environment is altered by rotating the local cues and the distal cues in opposite directions to create a cue-conflict environment) (Knierim, 2002) have clearly implied a specific role for these brain regions during spatial navigation (Knierim, 2002; Lee et al., 2004; Deshmukh and Knierim, 2011; Neunuebel et al., 2013; Neunuebel and Knierim, 2014; Wang et al., 2018). The medial EC (MEC) represents the context-based spatial information (where) in an allocentric framework, and the lateral EC (LEC) represents the content based spatial information (what is out there) in an egocentric framework (Deshmukh and Knierim, 2011; Neunuebel et al., 2013; Wang et al., 2018). By combining these allocentric and egocentric spatial information, the hippocampus is suggested to create a conjunctive representation of the environment through internal computations at DG-CA3-CA1 subfields (Knierim, 2002; Lee et al., 2004; Neunuebel and Knierim, 2014), thus resulting in an environment-specific cognitive map. Another major area of the medial temporal lobe, the subicular complex (SC) consisting of the subiculum (SUB) (Matsumoto et al., 2019), the presubiculum (PrS) (Dalton and Maguire, 2017; Simonnet and Fricker, 2018), and the parasubiculum (PaS) (Dalton and Maguire, 2017), is located between the hippocampus and EC, and has extensive anatomic connections with various cortical and subcortical areas of the brain (Amaral and Witter, 1989; O'Mara et al., 2001; Ding, 2013).
Spatial and directional correlates of neuronal firing are identified in the SC, such as the place cells (Barnes et al., 1990; Sharp and Green, 1994; Taube, 1995; Sharp, 1996), head direction (HD) cells (Taube et al., 1990; Boccara et al., 2010), border cells (Lever et al., 2009; Boccara et al., 2010), grid cells (Boccara et al., 2010), and axis-tuned cells (Olson et al., 2017). The SUB place cells maintain similar spatial firing patterns across different environments (Sharp, 1997), expand or contract spatial firing patterns in an open field to fit the size of the environment (Sharp, 1999a), and exhibit stable locational firing fields across light-dark transitions (Brotons-Mas et al., 2010). The dorsal PrS (referred as the postsubiculum [PoS]) receives HD inputs from the anterior dorsal thalamic nuclei (ADN) (Ding, 2013), playing a critical role in associating the HD signals with the distal landmarks to maintain the signal oriented with the external environment (Goodridge and Taube, 1997), and projects forward these HD signal to the hippocampus via the EC (Preston-Ferrer et al., 2016; Simonnet and Fricker, 2018). Moreover, the projections from the PrS and PaS form major inputs to the MEC (van Groen and Wyss, 1990a; Caballero-Bleda and Witter, 1993, 1994; Dalton and Maguire, 2017). The PaS HD cells and border cells are important in providing orientation, stability, and anchorage to the downstream MEC grid cells (Bonnevie et al., 2013; Hardcastle et al., 2015; Krupic et al., 2015; Stensola et al., 2015; Winter et al., 2015).
Since the SUB, PrS, and PaS regions receive diverse information from different anatomic inputs (van Groen and Wyss, 1990a; Matsumoto et al., 2019), the SC may show dissimilar spatial representations under dynamically changing environment. On the contrary, the intrinsic reciprocal connectivity within the SC (Ding, 2013) may lead to a functionally coherent spatial representation because of attractor-like dynamics. In the present study, we tested this hypothesis using a well-established local–distal cue-conflict experimental paradigm (Knierim, 2002; Lee et al., 2004; Yoganarasimha et al., 2006; Neunuebel et al., 2013; Neunuebel and Knierim, 2014) by simultaneously recording the neural activity from different subregions of the SC in male rats; first, by assessing the nature of neural representations in the SC, and then, by quantifying the interaction between the directional and the spatial representations to further enhance the current understanding of functional role of SC during spatial navigation. We report a highly coherent spatial representation of the environment in the SC and a robust coupling between the HD cells and spatial cells exhibiting attractor-like dynamics, very similar to the attractor dynamics reported in the ADN HD cells (Yoganarasimha et al., 2006).
Materials and Methods
Subjects and surgical procedures
Long-Evans rats (n = 5, male, 5-6 months old) were housed individually on reversed light-dark (12:12 h) cycle, and the experiments were conducted during the dark phase of the cycle. All the procedures (animal care, surgical procedures, and euthanasia) were performed in accordance with National Institutes of Health guidelines and were approved by the Institutional Animal Ethics Committee of National Brain Research Centre (Manesar, Haryana), constituted by the Committee for the Purpose of Control and Supervision of Experiments on Animals, Government of India. The surgical procedures were performed under aseptic conditions. Initially, the rats were anesthetized using ketamine (60 mg/kg body weight) and xylazine (8 mg/kg body weight) and then shifted to isoflurane gaseous anesthesia for the entire duration of the surgery. A custom-built recording device (Microdrive) containing 20 independently movable tetrodes (18 tetrodes and 2 references, held inside a single bundle or dual bundle) was surgically implanted over the right hemisphere in a way to simultaneously access different regions of the SC at 5.5-8.5 mm posterior to bregma and 2.5-4.5 mm lateral to the midline. Meloxicam analgesic was administered intramuscularly (1 mg/kg body weight) after surgery and also given orally (1 mg/kg of body weight) during the recovery period (7 d). Postsurgical care was provided on surgical implantation of the microdrive, and the rats were allowed to recover for 7 d. Following postsurgical recovery, the tetrodes were lowered into the brain targeting different regions of the SC and the rats were trained to run clockwise (CW) on a circular track. The rats were maintained at 85% of their free feeding weights during training and subsequent experimental recordings.
Electrophysiology and recording
Tetrodes were made from 17 μm platinum-iridium wire (California Fine Wire), and the tips of individual wires were electroplated with platinum black solution (Neuralynx) to 100-150 kΩ with 0.2 μA current. Electrophysiological recordings were conducted using a 96-channel data acquisition system (Digital Lynx 10S, Neuralynx) by amplifying the signals through headstage preamplifier (Neuralynx). The custom-made microdrive was fitted with EIB-27 board, which connected to the headstage preamplifier HS-27. The headstage was then attached to the commutator using HS-27 tethers. The commutator was connected with the recording cables that connected to the 96-channel Digital Lynx 10S data acquisition system (Neuralynx). The units were recorded against a reference electrode present in a cell-free zone (corpus callosum above the SC) in the brain by filtering the signal between 600 Hz and 6 kHz. Spike waveforms above the threshold were sampled for 1 ms at 32 kHz. Local field potentials were recorded against a ground screw anchored to the skull above the frontal cortex, filtered between 0.1 Hz and 1 kHz, and continuously sampled at 4 kHz. Only the spike data were used, and the LFP data were not used, in the current study. The position and the HD of the animal were tracked with the red and green LEDs attached to the headstages, and captured through ceiling mounted color CCD camera (CV-S3200, JAI) at 25 Hz.
Experimental design and statistical analyses
Behavior and recording procedures
After 7 d postsurgical recovery, the rat was placed on a pedestal, connected to the recording system and the tetrodes were slowly advanced targeting different regions of the SC over a period of 10-15 d. During this period, the rats were trained in the adjacent behavioral room to run CW seeking chocolate sprinkles placed at random locations on a centrally placed textured circular track (elevated 90 cm from floor level) for 30 min/day for 8-10 d. The path of the rat was blocked if it tried to turn around and go in the counterclockwise (CCW) direction. After repeated training sessions, the rat learned to run in the CW direction on the track. The circular track (56 cm inner diameter; 76 cm outer diameter) with four types of visually distinct textured surfaces, each covering a quarter of the track, served as local cues. Six salient cues (four visually distinct cues of different patterns and shapes made of cardboard hung on the curtain, two cardboard boxes of different sizes placed on the floor at the periphery of the curtain) at the surrounding black curtain reaching from the ceiling to the floor, served as distal cues. The distal cues were purely visual, while the local cues were both visual and somatosensory. The local–distal cue environment ensures uniform sampling of all the directions with relatively constant self-motion information as the rats were trained to run CW on the circular track, and these parameters remain constant across sessions without affecting the results. Further, this task rules out the factor of reward anticipation and/or localization as the animal gets chocolate sprinkles at random locations and at random times. A ceiling-mounted circular LED light at the center of the room provided the required illumination. A centrally placed white noise generator on the floor masked any external sounds during training and subsequent experimental sessions. Once tetrodes reached the region of interest and the animal was trained, the experimental recording sessions were conducted as described below in 5 rats, following the experimental protocol designed by Prof. James J. Knierim (Knierim, 2002). Some tetrodes were adjusted each day looking for well-isolated cell clusters, while other tetrodes were adjusted whenever the cell clusters showed reduced amplitude or did not show better isolation from the background. It is possible that some of the cells might have been recorded for more than 1 d or different cells were recorded every day. Since it is very difficult to establish with certainty the exact number of cells recorded over time, no attempt was made to do so. Once the rat was disconnected from the recording system on completion of that day's experiment, the units recorded on the next recording day were considered as separate units and added to the total number of cells.
Experiment 1: local–distal cue-conflict experiment
Each day of recording consisted of five sessions, in which three standard (STD) sessions were interleaved with two mismatch (MIS) sessions. In STD sessions, the configuration between the local and distal cues was maintained exactly same as that during the training sessions. Local–distal cue-conflict was created in MIS sessions by rotating all the distal cues CW and the textured circular track CCW by either 90° or 45° each to create a mismatch of 180° (MIS 180°) or 90° (MIS 90°) between the cues.
In a subset of these animals (3 of the 5 rats used in the above experiment) where we could still record the neural activity from the SC, we conducted additional experiments to assess the influence of local cues by removing all the distal cues (Experiment 2), and the effect of such exposure on the subsequent neural representations in the SC when encountered with a local–distal cue-conflict environment (Experiment 3).
Experiment 2: local cue rotation in the absence of the distal cues
This set of experiments consisted of either three or five sessions per day, in which STD sessions were interleaved with a session where the textured circular track was rotated by either 180° (Track-180° No Distal) or CCW 135° (Track-135° No Distal) in the absence of all six distal cues. Of the total 5 d of recording in this experiment, the STD sessions had both the local and distal cues in 3 d (three sessions each), while in 2 d (5 sessions each) all the STD sessions were also recorded in the absence of the distal cues (Track-0° No distal).
Experiment 3: local–distal cue-conflict experiment post distal cue removal
This set of experiments was exactly the same as Experiment 1 but was conducted on the next day of Experiment 2 to assess the effect of exposure to no distal cue (Experiment 2) on subsequent neural representations in the SC. Each day of recording consisted of three STD sessions interleaved with two local–distal cue-MIS sessions: MIS 180° and MIS 90°.
Experimental procedure
The rat was first placed inside a covered box for 30 s, and carried by the experimenter on a brief walk in the computer room before entering the adjacent behavior room. After closing the door and curtains, the experimenter walked around the track 3 times to disrupt the rat's sense of direction between the external environment and the behavioral room. This method of disorientation was followed to be consistent with the previous studies and to make legitimate comparison of the neural representation in the SC with that of the hippocampus and the EC (Knierim, 2002; Lee et al., 2004; Yoganarasimha et al., 2006; Neunuebel et al., 2013; Neunuebel and Knierim, 2014). A pedestal was placed in the center of the textured track, the rat was then transferred onto the pedestal, recording tethers with headstages were connected to the electrode interface board of the microdrive, the rat was released onto the track at a random starting point in every session, and the pedestal was removed. Upon completion of 20 laps, the pedestal was returned to the center of the track, the rat was placed on the pedestal to disconnect the tethers, transferred to the box, and carried around the track and into the computer room as mentioned above. This procedure was repeated for each of the experimental sessions. In order to classify different types of SC cells being recorded, each day after the last session on the track, a circular platform (76 cm diameter) was placed on top of the textured track, and the neural activity was recorded for 10 min duration while the rat foraged in the arena. On completing all the experimental sessions, the track was wiped clean with 70% ethanol to clear off any traces that could act as potential cues for the next day of recording. No attempt was made to dissociate the local cues and self-deposited odors between different sessions within a day, to be consistent with the experimental procedure followed in previous reports (Knierim, 2002; Lee et al., 2004; Yoganarasimha et al., 2006; Neunuebel et al., 2013; Neunuebel and Knierim, 2014) for legitimate comparison of the neural representations. However, the rat droppings on the circular track, if any, were removed using a tissue paper after the completion of each session.
Data and statistical analyses
The analysis of data and statistical tests were performed using custom-written software, Microsoft Excel, MATLAB (The MathWorks R2013a), Circular Statistics (Oriana, Kovach Computing Services), and Graph Pad Prism, as described below.
Isolation of single units
Isolation of single units was performed manually with custom-written spike-sorting software Winclust, developed by Prof. James Knierim (Johns Hopkins University, Baltimore). Cells were isolated based on the peak amplitude and energy of the waveforms recorded on four wires of the tetrode. Based on their isolation quality (distance from the background and separation from other clusters), the units were rated on a scale ranging from 1 to 5 (1, very good; 2, good; 3, fair; 4, marginal; 5, poor) and the units rated “fair” and above were used for further analysis. The waveform shape was not used for any inclusion or exclusion criteria or for cell classification.
Cell type identification
For each cell, the border score, grid score, Rayleigh's vector, spatial information score, and directional information score were calculated from the recordings on a circular platform at the end of track sessions. Based on the parameter that most strongly modulated a cell's firing, a cell was categorized as a directional cell or a spatial cell, as follows:
In the directional cell category, a cell was classified as a HD cell if its Rayleigh's vector score was more than the threshold value calculated from the pooled Rayleigh's vector shuffled distribution.
In the spatial cell category, a cell was included if it was classified as a border cell or a grid cell or a place cell or a conjunctive cell (i.e., border × direction cell, grid × direction cell, or place × direction cell). A cell with a border score above the threshold value calculated from the pooled border score shuffled distribution was classified as a border cell. A cell with grid score above the threshold value calculated from the grid score shuffled distribution was classified as a grid cell. A cell was classified as a place cell if it had a significant spatial information score (Skaggs et al., 1993) provided its border score, grid score, and Rayleigh's vector score were less than the threshold value for each of the pooled shuffled distribution parameters. The conjunctive cells were classified as follows: (1) border × direction cell, a cell with both its border score and Rayleigh's vector score more than the threshold values for both border score and Rayleigh's vector score calculated based on pooled cell shuffle distribution; (2) grid × direction cell, a cell with both its grid score and Rayleigh's vector score more than the threshold values for both grid score and Rayleigh's vector score calculated based on pooled cell shuffled distribution; and (3) place × direction cell, a cell with significant spatial information score as well as significant directional information score, based on the method described by Peyrache et al. (2017). The spatial information score and directional information score for all cells were plotted in a log-log scale scatter plot, wherein the data clustered into two groups (i.e., directional and spatial clusters). Based on this plot, the spatial cells clustered around the directional cell cluster were classified as probable place × direction cells. These probable cells were further visually inspected and checked if they had a significant spatial information score and directional information score based on their cell-specific shuffle distributions. The cells that qualified the above criteria were classified as putative place × direction cells.
To calculate border score and grid score, spatial firing rate maps were created for all the cells recorded by binning the recording area into 64 × 48 bins (bin size ∼2 × 2 cm). The number of spikes in each bin was divided by the amount of time spent in that bin to obtain a bin-wise firing rate matrix of individual cells, which was then smoothed using a Gaussian kernel of 5 × 5 bins with a σ of 2. To calculate Rayleigh's mean vector length, head directional firing rate tuning curves were created for all the cells recorded by dividing the number of spikes fired when the rat's head was facing a particular direction (bin size 5°) by the total amount of time spent facing that particular direction.
Border score
The MATLAB codes for calculating the border score were obtained through personal communications from Prof. Edward Moser's laboratory, and the border score was computed based on the formula as described by Solstad et al. (2008) as follows:
Grid score
Grid score was calculated by using the MATLAB codes available online (https://github.com/derdikman/Ismakov-et-al.-Matlab-code) (Ismakov et al., 2017). 2D autocorrelograms were generated from spatial firing rate maps of each cell by calculating Pearson's product moment correlation between the original rate map of a cell with its shifted version having spatial lags of τx and τy as follows:
Gridness of a cell was defined as the periodicity of the peaks in the spatial 2D autocorrelogram. The autocorrelogram was correlated with its rotated version (excluding the central peak) for every angle centered around its central peak and Pearson's correlation value for 30°, 60°, 90°, 120°, and 150° rotation was obtained. Grid cells have a repeated pattern of peaks to give high correlation values at 60° and 120° and lower correlation values at 30°, 90°, and 150°. Gridness score is the difference between the minimum correlation value at 60° or 120° and the maximum correlation value at 30° or 90° or 150° (Sargolini et al., 2006). A cell was defined as a grid cell if its grid score was more than the 99th percentile of the grid score values obtained by data shuffling (Langston et al., 2010) (see below for the data shuffling procedure).
Rayleigh's mean vector length
Rayleigh's mean vector length is the measure of directional modulation of the firing rate of a cell across all head directions. The mean vector length for a cell with higher firing rates along narrow range of head directions is higher than a cell with firing rates spread along all head directions. The Rayleigh's vector was calculated in MATLAB by following circular statistical analysis (Fisher, 1995), where the mean firing rate in every angular bini was considered as a point on a unit circle with Xj and Yj coordinates, using the following formula:
A cell was classified as a HD cell if its Rayleigh's vector score value was more than the 99th percentile of Rayleigh's vector score values obtained through data shuffling (see Data shuffling), provided its border/grid score value was less than those corresponding threshold values.
Data shuffling
Data shuffling was performed to rule out the possibility of spuriously classifying the cells by chance into different types. Shuffling was performed separately for each cell where a cell's entire spike sequence recorded on the platform session was time-shifted by a random interval between 20th second and 20 s before termination of the session. Spikes exceeding the total time of the session were wrapped around to be assigned to the beginning of the session to generate a new spike time sequence for the cell. This procedure was repeated 100 times for each cell, and the Border score, Grid score, and Rayleigh's mean vector length were calculated on these shuffled spike time series. Individual shuffle distributions from separate cells were combined into one distribution from which a single 99th percentile value for each score was taken as the threshold value for that cell type (Langston et al., 2010).
Spatial information score
The spatial information value indicates the amount of information about the rat's position, conveyed by the firing of a single spike from a cell (Skaggs et al., 1993), and was calculated as follows:
Directional information score
The directional information value indicates the amount of information about the rat's head direction, conveyed by the firing of a single spike from a cell (Skaggs et al., 1993; Peyrache et al., 2017), and was calculated as follows:
Rotational correlation
In order to assess the change in the preferred firing direction or firing field of the HD cells and the spatial cells (place, border, grid, and conjunctive cells), we performed the rotational correlation analysis by following the methods described by Neunuebel et al. (2013). To begin with, 2D circular track data were linearized using 5° binning to obtain one-dimensional firing rate array (of 72 bins) for each cell. Directional tuning curve was calculated for the HD cells by dividing the total number of spikes fired when the animal's head faced a particular direction by the total amount of time spent facing that particular direction on the circular track. In the case of spatial cells, one-dimensional firing rate arrays were created by dividing the number of spikes fired when the animal was in a particular spatial bin by the total amount of time spent in that particular bin on the circular track. A speed filter of 1 cm/s was used for all spatial cells to avoid the accumulation of spikes fired by a cell when the animal was too slow or stationary on the circular track. To quantify the amount of rotation of cells between STD sessions or between STD and MIS sessions, we have measured the Pearson's product-moment correlation of linearized firing rate arrays of a cell between those two sessions. Then, the firing rate bins of the session being correlated were shifted by one bin (5° shift) to obtain shifted correlations, and this procedure was repeated 71 times. The shift angle at which the maximum correlation value was obtained was considered as the amount of rotation of a cell between the two sessions being compared. We performed circular statistical tests (Zar, 1999) to calculate the angle of the mean vector and the length of the mean vector, using circular statistics software (Oriana, Kovach Computing Services). The mean angle of rotation of the cells was represented by the angle of the vector, while the length of the vector was inversely proportional to the variance of the distribution around that mean. The angle of the mean vector for all CW rotations was between 0° and 180°, while the same for CCW rotations was between 0° and −180°.
Population coherence
To understand the SC activity, the population responses were measured by creating 2D spatial correlation matrices from peak-normalized population firing rate vectors at each location on the track. Population correlation matrices for STD versus STD sessions and STD versus MIS sessions were created for all the cells recorded. Firing rate vectors for each session were created by pooling firing rate arrays (bin size 1°) of all the cells from a particular session yielding an N × 360 matrix where N is the number of cells. The STD session firing rate vector at each bin was correlated (using Pearson's product-moment correlation) with subsequent STD or MIS session bin to produce a 360 × 360 correlation coefficient matrix. The population correlation matrix consists of the correlation values obtained from the bin-wise correlation of the one-dimensional firing rate vectors. For STD versus STD correlation, higher correlation values would be expected at ∼0° as the preferred firing direction or the firing field did not change its location compared with the previous STD session and thus would accumulate along the central diagonal of the population correlation matrix. For STD versus MIS correlation, higher correlation values would be expected at the angle equivalent to the rotation of the preferred firing direction or the firing fields; and thus, they would accumulate above or below the central diagonal of the population correlation matrix as illustrated in Figure 8A.
In order to quantify the population response, the 2D spatial correlation matrices were reduced to 1D polar plots by calculating the mean correlation of pixels in each of the 360 diagonals in the population correlation matrices. The angle at which the mean correlation was maximum was considered as the amount of rotation of the cell population between the two sessions (Gothard et al., 1996; Lee et al., 2004; Neunuebel et al., 2013).
Spatial cross-correlation (SXC)
To assess the possibility of coupling between HD cells and spatial cells in the SC, the spatial offset between the co-recorded cell pairs across sessions on the circular track was compared by following the procedure of Bassett et al. (2018), with the exception that the SXCs were performed on the spike data for the whole session. The procedure for calculation of spatial offset between the co-recorded cell pairs is illustrated in Figure 11A.
One-dimensional firing rate array (of 72 bins) for each cell was created as described above under rotational correlation analysis and was normalized to its peak firing rate. The SXC values were calculated through Pearson's product-moment correlation of normalized firing rate arrays of a cell pair in a session, and the normalized firing rate array of the cell being compared was shifted by one bin (5° shift) to obtain shifted correlation value, and the procedure was repeated 71 times. The angle at which maximum SXC was observed for a cell pair was defined as the peak correlation angle value. SXC matrices were created by pooling the cell pairs from all days of recordings, session-wise, and rank ordered based on their ascending peak correlation angle value in STD 1 session for visualizing the alignment of SXC peak values across various sessions. For quantitative analysis, polar plots of SXCs were created for each of the cell pairs and the mean direction of the polar plot (ranging from −180° to 180°) was calculated using the circular statistics toolbox (Berens, 2009) in MATLAB. The mean direction value represented the spatial offset between co-recorded cells within a session on the circular track. The coupling between the cells was quantified by comparing the spatial offset values of cell pairs across sessions (after correcting for −180/180° transition values) using correlation analysis, and the r2 values (indicating goodness of fit of the linear equation) and its significance were calculated. As the HD system is proposed to exhibit attractor dynamics (Skaggs et al., 1995; Redish et al., 1996; Zhang, 1996), we first assessed the coupling between co-recorded HD–HD cell pairs across sessions in datasets having a minimum of two HD cells. Upon confirmation of an attractor-like coupling among the HD cells in the SC, we quantified the coupling between co-recorded HD–spatial cell pairs across sessions.
Histologic procedures and identification of recording sites
Upon completion of the electrophysiological experiments, marker lesions were performed on a few selected tetrodes by passing current (10 μA for 10 s) while the rat rested quietly on the pedestal, without anesthesia. On the following day, rats were perfused transcardially with 4% formalin solution, the brain was extracted, and stored in 30% sucrose-formalin until it sank in the solution. Brains were sectioned in the coronal plane (40 μm thick), mounted, and processed for Nissl staining using 0.1% cresyl violet. Images of serial sections were captured on Leica DFC265 digital camera attached to Leica M165-C stereo microscope and saved as TIFF files. The distance from the midline to the tetrode track markings was measured from these serial sections, plotted in an Excel spreadsheet to visualize the configuration of tetrode tracks. The tetrodes were identified by comparing this configuration with the arrangement of tetrodes in the microdrive, and cross verifying with the marker lesions. We performed depth reconstruction of the tetrode track to identify the brain region at which the cells were recorded each day, based on the distance from the bottom tip of the tetrode by taking into account a 15% shrinkage of tissue because of histologic processing.
Data and code accessibility
The data that support the findings are available from the corresponding author on request.
Results
In vivo neurophysiological recordings from the SC
We conducted in vivo neurophysiological studies by recording the neural activity from all subregions of the SC simultaneously using multitetrode electrophysiology technique, to understand the representational properties of the SC while the animals navigated in a dynamically changing environment (see Materials and Methods) in three experiments. Figure 1A shows the extent of the recording sites from different subregions of the SC in 5 rats. Figure 1B shows the number of experimental days and sessions in which the SC cells were co-recorded for each of the experiments. On most of the experimental days, the cells were recorded simultaneously from a minimum of two regions of the SC. Figure 2 shows the representative examples of the tetrode localizations in SUB, PrS, and PaS, respectively. We have recorded a total of 456 cells from 5 rats in the present study, while they ran CW on a textured circular track in the presence (Experiments 1 and 3) or absence (Experiment 2) of the distal cues. After the last session on the circular track each day, a circular platform was placed on top of the track and a single recording session of 10 min duration was conducted to identify the cell type. Figure 3 shows the classification of the SC cells based on the observed Rayleigh's mean vector length, grid score, and border score values for each of the cells (Fig. 3A) and the values obtained after data shuffling (Fig. 3B). Figure 3C–I shows an example each for the cells classified as a place cell, a border cell, a grid cell, an HD cell and conjunctive cells (place × direction, border × direction, and grid × direction). Figure 4 shows the number of various cell types recorded from different subregions of the SC in each experiment (Fig. 4A), the mean vector length, border score, and grid score of the cells classified as place cells, border cells, grid cells, or HD cells (mean ± SEM) (Fig. 4B), the directional and spatial information scores of the different cell types of the SC (Fig. 4C), and a comparison of the mean firing rates versus the spike widths of all the HD cells and the spatial cells (Fig. 4D) based on the recordings from the circular platform. The border cells, the grid cells, and the HD cells had their category specific scores higher than their respective 99th percentile threshold value calculated based on the shuffled data, while these scores in case of place cells were lower than the threshold values of the mean vector length, the border score, and the grid score. The data consisted of regular spiking and high rate cells as has been shown previously (Barnes et al., 1990; Sharp and Green, 1994). Further, the mean firing rate, peak firing rate, spike width, spike amplitude, and the directional/spatial information scores of SC cells showed a significantly high correlation between the STD and cue manipulation session on the circular track in all three experiments (respective r2 values: 0.95, 0.95, 0.93, 0.99, and 0.63 [HD cells]; 0.98, 0.86, 0.94, 0.99, and 0.91 [spatial cells]; p < 0.0001 for all comparisons), suggesting that these parameters remained constant in different experimental conditions.
Recording sites in the SC and experimental details. A, Schematic representation of the extent of in vivo neurophysiological recordings from different subregions of the SC. Tetrode locations of individual rats are color-coded and shown in the right hemisphere, while the details of different subregions of the SC are mentioned in the left hemisphere (adapted from Paxinos and Watson, 2007). B, Pie charts represent details of simultaneous recordings from different subregions of the SC (days and sessions) in different rats and experiments.
Representative examples of tetrode tracks in the SC. A–C, Representative examples of Nissl-stained coronal sections of the rat brain show tetrode tracks (indicated by arrow) in the SUB (A), the PrS (B), and the PaS (C) regions of the SC. The boundaries of different subregions of the SC at corresponding plates from the rat brain atlas (adapted from Paxinos and Watson, 2007) are shown below. Scale bar, 1 mm.
Identification of cell types in the SC. A, B, Distribution of the mean vector length, the grid score, and the border score values of all the cells recorded (observed) from different regions of the SC on the circular platform and distribution of these values after random shuffling of spike sequence (shuffled). Red line indicates threshold values (99th percentile of the shuffled data). C–I, Representative examples each for different types of cells recorded from the SC region. Each row shows the trajectory of the rat (gray lines) superimposed with the spikes (red dots), firing rate map, spatial autocorrelogram, firing rate as a function of head direction, followed by their mean vector length (r), grid score (g), border score (b), directional information score (Dir info), and spatial information score (Spatial info) values based on the recordings on the circular platform. Scores in red represent above the threshold value. Corresponding firing rate maps or HD tuning curves of these cells recorded on the circular track session are shown at the end, along with peak firing rate. The rate maps are color coded (red represents >90% of the peak firing rate; blue represents no firing), and the successive decrements in peak firing rates are shown with intervening colors of the spectrum.
Description of the cells recorded from the SC. A, Table represents the number of cells recorded in each experiment from different subregions of the SC. *Number of conjunctive cells (border × direction, grid × direction, and place × direction) recorded under each category of spatial cells. B, Bar graphs represent the average mean vector lengths, border scores, and grid scores (mean ± SEM) of different cell types in the SC (based on the recordings from the platform session). C, Scatter plot in log-log scale represents the distribution of different cell types based on their directional information score and spatial information score. Red, green, black, and gray open circles represent HD cells, border cells, grid cells, and place cells, respectively. Green, black, and gray filled circles represent border × direction cells, grid × direction cells, and place × direction cells, respectively. D, Scatter plot comparing the mean firing rate versus spike width of all the HD cells and the spatial cells from the SC recorded on the circular platform.
Experiment 1: local–distal cue-conflict experiment
In this experiment, we have recorded a total of 338 cells in different regions of the SC (obtained over a period of 41 recording days comprising of 193 sessions) from 5 rats (Fig. 1B) while they ran CW on a textured circular track (the local cues) in the presence of the distal cues following a well-established experimental paradigm (Knierim, 2002; Lee et al., 2004; Yoganarasimha et al., 2006; Neunuebel et al., 2013; Neunuebel and Knierim, 2014) (Fig. 5A). The firing activity patterns of the cells recorded from different subregions of the SC in the local–distal cue-conflict environment may exhibit binding to either the local cues or the distal cues or both the cues simultaneously, as has been reported for the cells in different subfields of the hippocampus or the EC or the ADN, recorded under the same experimental conditions (Knierim, 2002; Lee et al., 2004; Yoganarasimha et al., 2006; Neunuebel et al., 2013; Neunuebel and Knierim, 2014). Each day of recording consisted of three STD sessions interleaved with two MIS sessions, in which the local cues and the distal cues were rotated in opposite directions by either 90° or 45° each to create a mismatch of 180° (MIS 180°) or 90° (MIS 90°) between the cues. Representative examples of the HD cell tuning curves, firing patterns of place, grid, and border cell (spatial cells) recorded from different rats across five sessions of a day are shown in Figure 5B. Both the HD cells and the spatial cells maintained their preferred firing direction and firing field in STD sessions on the circular track and rotated CW following the distal cues in MIS sessions.
Local–distal cue-conflict experiment. A, Schematic representation of the Local–distal cue-conflict experimental paradigm. B, Representative examples of the HD cells and the spatial cells recorded on textured circular track from different rats across STD and MIS sessions within 1 d. The axes for the HD cells (1-6) are scaled for their maximum firing rates (112.6, 85.5, 17.0, 164.8, 22.4, and 17.2 Hz, respectively). The numbers inside the firing rate maps indicate the peak firing rate in Hz. The rate maps are color coded, as described in Figure 3. Each day, on completion of track sessions, a circular platform was placed on the track and the neural activity was recorded. HD cell tuning curve (superimposed with directional occupancy in gray for circular platform session) and firing rate map of the spatial cell are shown for each example cell. Values indicate the peak firing rate, Rayleigh's mean vector length (r), peak occupancy in seconds for HD cells, and peak firing rate for spatial cells. The grid and border score are mentioned for the grid cell (cell 11) and border cell (cell 12), respectively.
Distal-cue-controlled highly coherent representations in the SC
Through rotational correlation analysis of firing rate arrays, we quantified the change in preferred firing direction/firing field of the HD cells and the spatial cells separately in STD versus STD and STD versus MIS sessions in Experiment 1 and calculated the angle of the mean vector and length of the mean vector using circular statistics (see Materials and Methods). The angle of the mean vector for the population of cells represents the circular mean of the angle of rotation of the cells, while the length of the mean vector signifies the compactness of the distribution of the angle of rotation around their circular mean and is inversely proportional to the variance of the distribution around that mean. The amount of rotation of the HD cells and the spatial cells between the STD sessions and between the STD versus MIS sessions is shown in Figure 6. The polar plots show the distribution of data from all the rats for each of these comparisons, categorized by the different subregions of the SC (Fig. 6A) and with data from all the subregions combined (Fig. 6B). In all STD versus STD comparisons, the angle of the mean vector did not deviate significantly from zero between sessions, and the length of the mean vector (r) signified that the angles were significantly clustered (Fig. 6A: SUB spatial: STD 1 vs STD 2 r = 0.99, STD 2 vs STD 3 r = 0.99, PrS HD: STD 1 vs STD 2 r = 0.99, STD 2 vs STD 3 r = 0.99, PrS spatial: STD 1 vs STD 2 r = 0.99, STD 2 vs STD 3 r = 0.99, PaS HD: STD 1 vs STD 2 r = 0.99, STD 2 vs STD 3 r = 0.99; PaS spatial: STD 1 vs STD 2 r = 0.98, STD 2 vs STD 3 r = 0.99, p < 0.0001 for all comparisons. Fig. 6B: SC HD cells: STD 1 vs STD 2 r = 0.99, STD 2 vs STD 3 r = 0.99, SC spatial cells: STD 1 vs STD 2 r = 0.99, STD 2 vs STD 3 r = 0.99; p < 0.0001 for all comparisons), suggesting that the preferred firing direction or firing fields of both the HD cells and the spatial cells did not change between the sessions, thus exhibiting stable representation. In STD versus MIS comparisons, the angle of the mean vector deviated significantly from zero to an angle equal to the amount of rotation of the distal cues in the MIS sessions, indicating that both the HD cells and the spatial cells rotated CW in the cue-conflict environment. Even in this comparison, the length of the mean vector (r) signified that the angles were significantly clustered (Fig. 6A: SUB spatial: STD 1 vs MIS 180° r = 0.98, STD 2 vs MIS 90° r = 0.99, PrS HD: STD 1 vs MIS 180° r = 0.99, STD 2 vs MIS 90° r = 0.99, PrS spatial: STD 1 vs MIS 180° r = 0.98, STD 2 vs MIS 90° r = 0.99, PaS HD: STD 1 vs MIS 180° r = 0.99, STD 2 vs MIS 90° r = 0.99, PaS spatial: STD 1 vs MIS 180° r = 0.99, STD 2 vs MIS 90° r = 0.99; p < 0.0001 for all comparisons. Fig. 6B: SC HD cells: STD 1 vs MIS 180° r = 0.99, STD 2 vs MIS 90° r = 0.99, SC spatial cells: STD 1 vs MIS 180° r = 0.98, STD 2 vs MIS 90° r = 0.99; p < 0.0001 for all comparisons), indicating that the representations were similar and highly coherent across all subregions of the SC (Fig. 6A) and the cell types (Fig. 6B).
Distal-cue-controlled highly coherent representation in the SC. A, The amount of rotation of preferred firing direction of the HD cells (green triangle) and firing fields of the spatial cells (blue circle) recorded from different subregions of the SC between STD sessions and between STD versus MIS sessions, represented around the circle (triangle/circle; open = 1 cell, filled = 2 cells). B, The amount of rotation of preferred firing direction of the HD cells and firing fields of the spatial cells, from all subregions of the SC combined (triangle/circle, open = 1 cell, filled = 3 cells). The direction of the arrow inside the circle represents the mean angle of rotation of the population, and the length of the arrow is inversely proportional to the variance of the distribution around the mean angle. Red lines indicate the rotation angle of the local (L) and distal (D) cues in MIS sessions. Values next to each plot indicate the angle of the mean vector, 95% CI, length of the mean vector (r), and corresponding significance level (p).
Ensemble coherence in the SC
As the above analysis was performed on the combined data from all days of recording, we quantified the nature of representation in each of the ensemble recordings through rotational correlational analysis (Fig. 7). On average, we have recorded ∼8 cells per ensemble (min 2, max 16) over a period of 41 d of recording (Fig. 7A). The change in preferred firing direction/firing field of the cells in STD versus STD and STD versus MIS sessions in individual ensemble recordings was quantified using circular statistics. The averaged angle of the mean vector and the length of the mean vector for each of the comparisons are shown in Figure 7B and Figure 7C, respectively. Even in individual ensemble recordings, all the cells maintained their preferred firing direction or firing fields across different STD sessions (mean vector; STD 1 vs STD 2: −2.1° ± 3.3°, STD 2 vs STD 3: −0.5° ± 2.9°, mean ± SD), and rotated as an ensemble to an angle equal to the amount of rotation of the distal cues in MIS sessions (mean vector: STD 1 vs MIS 180°: 86.0° ± 4.9°, STD 2 vs MIS 90°: 42.9° ± 3.9°, mean ± SD) (Fig. 7B). The length of the mean vector was similar across different comparisons (STD 1 vs STD 2: 0.99 ± 0.006, STD 1 vs MIS 180°: 0.99 ± 0.006, STD 2 vs STD 3: 0.99 ± 0.005, STD 2 vs MIS 90°: 0.99 ± 0.005, mean ± SD) (Fig. 7C), indicating that the angles were clustered around the mean vector because of highly coherent representation in each of the ensemble recordings.
Ensemble coherence in the SC. A, Histogram represents the number of cells (HD and spatial cells, all subregions of the SC) in each of the ensemble recordings. B, Averaged angle of the mean vector (mean ± SD) calculated for each of the ensemble recordings, showing the change in preferred firing direction/firing fields of cells between STD sessions and between STD versus MIS sessions. C, Averaged length of the mean vector (mean ± SD) calculated for each of the ensemble recordings across STD versus STD and STD versus MIS comparisons.
Along with the robust coherency, the stability of representations in the SC under dynamically changing environments is unlike the hippocampal and EC representations. The firing fields of co-recorded neurons in the hippocampus and EC followed distal or local cue rotations, appeared, disappeared, or developed ambiguous firing patterns (Knierim, 2002; Lee et al., 2004; Yoganarasimha et al., 2006; Neunuebel et al., 2013). On the contrary, the SC neurons show stable firing across all track sessions (both STD and MIS sessions) regardless of whether the firing fields rotate CW or CCW during cue manipulation sessions. This stability of the SC neurons shows compelling similarity to the representations observed in the ADN-HD system (for like-to-like comparison between the hippocampus, EC, ADN, and SC representations, see Table 1).
Comparison of neuronal responses across different brain regions to cue-conflict conditionsa
Population response in the SC
We further supplemented this categorical analysis by measuring the population response through bin-wise correlation of firing rate vectors of all the cells at each location on the track between STD versus STD and STD versus MIS sessions, as illustrated in Figure 8A. A band of high correlation was observed on the central diagonal in STD versus STD correlation matrices indicating no shift in the firing direction/fields, while in STD versus MIS correlation matrices, the band of high correlation shifted CW (following rotation of the distal cues) in SUB, PrS, and PaS, when assessed individually (Fig. 8B–D) or all subregions of the SC combined (Fig. 8E). This observation suggests that the population response of the SC was stable across different STD sessions and rotated as an ensemble in MIS sessions, thus exhibiting a highly coherent activity. In order to quantify the population response, 1D polar plots were created by calculating the mean correlation of pixels in each of the 360 diagonals from the corresponding 2D correlation matrices (Fig. 8F–I). The angle of peak correlation (angle at which the mean correlation was maximum) and the length of the mean vector (Fig. 8J–M) were calculated for each of these polar plots. The angle of peak correlation (representing the amount of rotation of the cell population between two sessions) was around zero in all STD versus STD comparisons (SUB: STD 1 vs STD 2 = −3°, STD 2 vs STD 3 = 0°; PrS: STD 1 vs STD 2 = −3°, STD 2 vs STD 3 = −2°; PaS: STD 1 vs STD 2 = −4°, STD 2 vs STD 3 = −2°; all regions of the SC: STD 1 vs STD 2 = −3°, STD 2 vs STD 3 = 0°) and was almost equivalent to the amount of rotation of the distal cues in all STD versus MIS comparisons (SUB: STD 1 vs MIS 180° = 88°, STD 2 vs MIS 90° = 44°; PrS: STD 1 vs MIS 180° = 88°, STD 2 vs MIS 90° = 40°; PaS: STD 1 vs MIS 180° = 85°, STD 2 vs MIS 90° = 41°; all regions of the SC: STD 1 vs MIS 180° = 88°, STD 2 vs MIS 90° = 40°), when the population response of different subregions of the SC was quantified individually (Fig. 8J–L) or all subregions combined (Fig. 8M), suggesting that the neural representation in each of the SC regions is strongly governed by the distal cues in local–distal cue-conflict condition. Further, there was no significant difference (two-way ANOVA) in the length of the mean vectors between SUB, PrS, and PaS regions (p = 0.1045) or across different comparisons (p = 0.7005), indicating that the coherency of the population response was similar across different subregions of the SC (Fig. 8J–M).
Population response in the SC. A, Schematic representation of population correlation analysis for STD versus STD and STD versus MIS sessions. B–E, Correlation matrices from population firing rate vectors at each location on the track between STD versus STD and STD versus MIS sessions, created separately for the cells recorded from SUB (B), PrS (C), PaS (D), and by pooling all the cells recorded from different regions of the SC (E). The central diagonal is indicated by a dashed white line in STD versus MIS matrices. Correlation matrices are color coded between 0 and 1 (Pearson's r values) to show the gradation from the lowest correlation value 0 (blue) to the highest correlation value 1 (brown), and a single color bar is shown for all the correlation matrices at the bottom. F–I, The population activity between STD versus STD sessions (blue) and STD versus MIS sessions (red) are represented as polar plots (created from 2D spatial correlation matrices). The values next to the polar plot indicate the angle of peak correlation (r) in the case of SUB (F), PrS (G), PaS (H), and all subregions of the SC combined (I). J–M, Length of the mean vector in STD versus STD sessions (blue) and STD versus MIS sessions (red) for each of the corresponding polar plot shown in F–I.
As all the cells recorded were always controlled by the rotations of the distal cues only, we conducted two more experiments in a subset of these animals where we could still record the neural activity, to assess the SC representations in the absence of the distal cues (Experiment 2), and then the effect of such exposure on subsequent representations in the SC during the local–distal cue-conflict condition (Experiment 3), as described below.
Experiment 2: local cue control of the SC representations in the absence of the distal cues
In this experiment, we recorded a total of 57 cells in different regions of the SC (obtained over a period of 5 recording days comprising of 19 sessions) from 3 rats (Fig. 1B) while they ran CW on a textured circular track (used in Experiment 1 in the same room) (the local cues) to assess whether the local cues had any influence on the representations in the SC by removing all the distal cues for a session (Fig. 9A) or all five sessions of a day (Fig. 9B). One rat has experienced the protocol shown in both Figure 9A and Figure 9B, but on different days. Other rats have experienced the protocol of either Figure 9A or Figure 9B only once, not both. In between the STD sessions, the local cues were rotated either 180° or CCW 135°. Figure 9C, D shows representative examples of HD cells (cells 1-3; 7-9) and spatial cells (cells 4-6; 10-12) recorded under these experimental conditions. In both the cases, all the cells maintained their preferred firing direction and firing field in STD sessions and rotated CCW to an amount equal to the amount of rotation of the local cues in sessions where the track was rotated either 180° or CCW 135°. The amount of rotation of the preferred firing direction/firing field of the cells was quantified by pooling together all the cells recorded, but separately for the distal cues removed for a session or all five sessions of a day (Fig. 9E,F), as all the cells from the SC region recorded in our main experiment were found to be highly coherent. Consistent with the results of Experiment 1, the angle of the mean vector did not deviate significantly from zero between STD sessions, suggesting no change in the preferred firing direction or firing fields. In the STD versus Track rotation sessions, the angle of the mean vector deviated significantly from zero to an angle equal to the amount of rotation of the local cues (track rotated either 180° or CCW 135°). The angles were significantly clustered around the mean angle in each of these comparisons, as signified by the length of the mean vector (r) (Fig. 9E: STD 1 vs STD 2 r = 0.99, STD 1 vs Track 180° r = 0.98; and p < 0.0001 for all comparisons; Fig. 9F: STD 1 vs STD 2 r = 0.95, STD 2 vs STD 3 r = 0.99, STD 1 vs Track 180° r = 0.97, STD 2 vs Track 135° r = 0.98; and p < 0.0001 for all comparisons), indicating a highly coherent representation in the SC even under this condition. These observations suggest that, in the absence of the distal cues, the local cues govern the SC representation. However, the representations are always coherent. Interestingly, in one of the sessions (Fig. 9G) where all the cells recorded (both the HD and the spatial) rotated to an arbitrary angle, the representations were still found to be highly coherent (STD 1 vs STD 2 r = 0.99, STD 1 vs Track 180° r = 0.98; and p < 0.0001 for both comparisons).
Local-cue-control of the SC representations in the absence of the distal cues. A, B, Schematic representation of the Local cue rotation in the absence of the distal cue experimental paradigm. C, D, Representative examples of the HD cell (cells 1-3 and 7-9) and the spatial cell (cells 4-6 and 10-12) response to rotations of the local cue in the absence of the distal cues in one session (C) or all five sessions (D) within a day. The axes for the HD cells are scaled for their maximum firing rates (cells 1-3 and 7-9; 121.7, 60.9, 17.1, 112.1, 24.9, and 11.8 Hz, respectively). The numbers inside the firing rate maps indicate peak firing rates in Hz. The rate maps are color coded as described in Figure 3. Each day, on completion of track sessions, a circular platform was placed on the track and the neural activity was recorded. HD cell tuning curve (superimposed with directional occupancy in gray for circular platform session) and firing rate map of the spatial cell are shown for each example cell. Values indicate the peak firing rate, Rayleigh's mean vector length (r), peak occupancy in seconds for HD cells, and peak firing rate for spatial cells. E, F, The amount of rotation of the preferred firing direction or firing field of all the cells (green triangles represent HD cells; blue circles represent spatial cells) recorded from different subregions of the SC between STD sessions and between STD versus Cue manipulation sessions in the absence of the distal cues in one session (E) or all five sessions (F) within a day, represented around the circle. The direction and length of the arrow inside the circle are as described in Figure 6. Red line indicates the rotation amount of the local cues (L). G, Amount of rotation of preferred firing direction of one HD cell (from PrS) and firing fields of six spatial cells (from SUB) recorded as an ensemble on 1 d (Rat 06, d 04). E–G, Values below each plot indicate the angle of the mean vector, 95% CI, length of the mean vector (r), and corresponding significance level (p).
Experiment 3: the SC representations are governed by the cues that are perceived to be stable
In this experiment, we recorded a total of 61 cells in different regions of the SC (obtained over a period of 6 recording days comprising of 28 sessions) from 3 rats (Fig. 1B) exactly as per the experimental protocol (Fig. 10A) followed in Experiment 1. This experiment was performed to assess whether the SC cells still exhibit coherent representation and coupling between the HD cells and the spatial cells in the subsequent local–distal cue-conflict environment, following the exposure to a no-distal-cue-environment (single or all the sessions) in Experiment 2. It is possible that, because of their exposure to a no-distal-cue-environment in Experiment 2, the animal may completely rely on the local cues of the environment and show local-cue-control of the SC cells even after the distal cues are reintroduced to the environment, or switch back to following the distal cues immediately because of their dominance, or the HD cells and the spatial cells may decouple from each other and exhibit a heterogeneous representation.
The SC representations are governed by the cues that are perceived to be stable. A, Schematic representation of the Local–distal cue-conflict post distal cue-removal experimental paradigm. B, D, Representative examples of the HD cell (cells 1, 2, 5, and 6) and the spatial cell (cells 3, 4, 7, and 8) recorded across three STD sessions interleaved with two MIS sessions within a day. The axes for the HD cells are scaled for their maximum firing rates (cells 1, 2, 5, and 6; 135.2, 21.7, 50.1, and 20.5 Hz, respectively). The numbers inside the firing rate maps indicate peak firing rates in Hz. The rate maps are color coded as described in Figure 3. Each day, on completion of track sessions, a circular platform was placed on the track and the neural activity was recorded. HD cell tuning curve (superimposed with directional occupancy in gray for circular platform session) and firing rate map of the spatial cell are shown for each example cell. Values indicate the peak firing rate, Rayleigh's mean vector length (r), peak occupancy in seconds for HD cells, and peak firing rate for spatial cells. C, E, The amount of rotation of the preferred firing direction or firing field of all the cells (green triangles represent HD cells; blue circles represent spatial cells) recorded from different subregions of the SC between STD sessions and STD versus MIS sessions, represented around the circle. The direction and length of the arrow inside the circle are as described in Figure 6. Red lines indicate the rotation amounts of the local cues (L) and distal cues (D) in MIS sessions. Values below each plot (C,E) indicate the angle of the mean vector, 95% CI, length of the mean vector (r), and corresponding significance level (p). Sessions in which the preferred firing directions or firing fields of cells recorded in an ensemble rotated CCW in MIS 180° and CW in MIS 90° (B,C) or CW in MIS 180° and CCW in MIS 90° (D,E) were grouped separately and analyzed. C, 28 HD cells (PrS-17 and PaS-11) and 21 spatial cells (SUB-10, PrS-4 and PaS-7) in MIS 180° session; 25 HD cells (PrS-14 and PaS-11) and 11 spatial cells (SUB-2, PrS-2 and PaS-7) in MIS 90° session. E, 4 HD cells (PrS) and 8 spatial cells (SUB-4 and PrS-4) in MIS 180° session; 6 HD cells (PrS) and 13 spatial cells (SUB-8 and PrS-5) in MIS 90° session.
This experiment was conducted on the next day in the same rats after Experiment 2. Figure 10B, D shows representative examples of the HD cells (cells 1-2; 5-6) and the spatial cells (cells 3-4; 7-8) recorded across three STD and two MIS sessions in a day. All the cells maintained their preferred firing direction and firing field in STD sessions. In 5 (2 d in Rat 06 and 3 d in Rat 11) of a total of 6 d of recordings from 3 rats, the preferred firing directions or firing fields of all the cells rotated CCW to an amount equal to the rotation of the local cues in the first cue-conflict session (MIS 180°) even in the presence of all the distal cues, and the same cells rotated CW to an amount equal to the amount of rotation of the distal cues in the second cue-conflict session (MIS 90°) (Fig. 10B). However, in a single day of recording from Rat 45, the preferred firing directions or firing fields of cells rotated CW to an amount equal to the amount of rotation of the distal cues in MIS 180° session, and rotated CCW to an amount equal to the amount of rotation of the local cues in MIS 90° session in the presence of the distal cues (Fig. 10D). To quantify the amount of rotation and coherency, all the cells recorded in an ensemble from various sessions exhibiting either CCW or CW rotations were grouped separately, and the angle of the mean vector and the length of the mean vector were calculated (Fig. 10C,E). In STD versus STD comparisons, the angle of the mean vector did not deviate significantly from zero, suggesting no change in the preferred firing direction or firing fields between STD sessions. In STD versus MIS comparisons, the angle of the mean vector deviated significantly from zero CCW or CW in MIS sessions to an angle equal to the amount of rotation of the local cues or the distal cues. The angles were significantly clustered around the mean angle as signified by the length of the mean vector in all the comparisons (Fig. 10C: STD 1 vs STD 2 r = 0.99, STD 2 vs STD 3 r = 0.99, STD 1 vs MIS 180° r = 0.97, STD 2 vs MIS 90° r = 0.99; and p < 0.0001 for all comparisons; Fig. 10E: STD 1 vs STD 2 r = 0.99, STD 2 vs STD 3 r = 0.99, STD 1 vs MIS 180° r = 0.99, STD 2 vs MIS 90° r = 0.99; and p < 0.0001 for all comparisons), indicating a highly coherent representation in the SC in each case. These findings are consistent with the results of our Experiments 1 and 2, that the SC exhibits highly coherent representations even when the preferred firing direction/fields of the cells rotated either CCW or CW as an ensemble in MIS sessions following either the local cues or the distal cues, on prior exposure to a no distal cue condition for a day. The data further indicate that the selection of the frame of reference to orient the spatial representations in the SC may be governed by the cues that are perceived to be stable at any given session because, on reintroduction of distal cues, the SC representation dynamically switches its orientation in MIS sessions to either a local or global frame of reference, possibly because of the competition between already stabilized local cues (as rats experienced only local cues in Experiment 2) and inherently dominant distal cues. However, even during such a switch in orientation governed by either local or distal cues, the SC ensemble representations were found to be stable and highly coherent.
Attractor-like dynamics in the SC
As we observed, a highly coherent representation of the environment in the SC, much similar to the attractor activity reported for the HD system, we conducted further analysis to check for the possibility of coupling between the HD cells and the spatial cells in the SC by comparing the spatial offset between the co-recorded cell pairs across sessions on the circular track (Bassett et al., 2018) (Fig. 11A; see Materials and Methods), by combining the data from all three experiments. SXC values were calculated for each of the co-recorded cell pairs (HD–HD, HD–spatial), and session-wise SXC matrices were created by rank ordering the cell pairs based on their ascending peak correlation angle values in STD 1 session (Fig. 11B,D). We observed the alignment of SXC peak values along the diagonal in all session-wise SXC matrices (Cue manipulation 1, STD 2, Cue manipulation 2 and STD 3) as the spatial offset between co-recorded cell pairs was maintained in all the sessions, indicating a unified response by all the cell pairs across sessions. As expected, the mean direction of SXCs of HD–HD cell pairs (representing the spatial offset between co-recorded HD cells within a session) across sessions (STD vs STD, STD vs Cue manipulations) was significantly correlated (STD 1 vs STD 2 r2 = 0.98, STD 2 vs STD 3 r2 = 0.99, STD 1 vs Cue Manipulation 1 r2 = 0.97, STD 2 vs Cue Manipulation 2 r2 = 0.99; and p < 0.0001 for all comparisons) (Fig. 11C), indicating an attractor-like activity in the HD cells of the SC. Interestingly, the mean direction of SXCs of even the HD–spatial cell pairs was also significantly correlated (STD 1 vs STD 2 r2 = 0.90, STD 2 vs STD 3 r2 = 0.95, STD 1 vs Cue Manipulation 1 r2 = 0.84, STD 2 vs Cue Manipulation 2 r2 = 0.93; and p < 0.0001 for all comparisons) (Fig. 11E), suggesting that even the directional and spatial cells in different subregions of the SC also exhibit robust coupling between each other. This striking result indicates that the SC representations remain consistent despite environmental perturbations (i.e., removal of the dominant distal cues or its reintroduction) and always exhibits a highly coherent representation along with a robust coupling between the HD cells and the spatial cells in different subregions of the SC, whether they rotate CW following the distal cues or rotate CCW following the local cues or even when they rotate to an arbitrary angle ignoring the cue rotation (Fig. 9G), suggesting attractor-like dynamics in the SC.
Attractor-like dynamics in the SC. A, Illustration represents the calculation of spatial offset between the co-recorded cell pair (e.g., cells A and B) in different sessions on the circular track. B, D, SXC matrices of co-recorded HD–HD cell pairs (310 cell pairs) (A) and HD–spatial cell pairs (640 cell pairs) (C) across three STD and two Cue manipulation sessions (all three experiments), sorted based on their ascending peak correlation angle values in STD 1 session. C, E, Scatter plots represent the correlation between the mean direction of SXCs of HD–HD cell pairs (B) and HD–spatial cell pairs (D) in STD versus STD and STD versus Cue manipulation sessions. The r2 values and the significance level (p) are mentioned below the scatter plots for each comparison.
Discussion
We report that different categories of cells (HD, place, border, grid, and conjunctive cells) recorded from the SC exhibit attractor-like dynamics by showing strong coherence in the representation of the environment in standard and cue-conflict conditions. The orientation of neural representation in the SC was originally governed by the distal cues, while in the absence of salient distal cues in the environment, the representation was oriented with reference to available local cues only. In subsequent local–distal cue-conflict conditions, the orientation of the SC representations was governed by either the distal cues or local cues. However, the ensemble representations were always found to be highly coherent and stable. Further, in all the conditions, we observed a strong coupling between the HD cells and spatial cells. The representational coherency between the SUB, the PrS, and the PaS regions of the SC shown in this study is strikingly different from previous reports of spatial correlates between co-recorded brain regions.
The hippocampus is proposed to form the locus of the cognitive map of the environment through the integration of inputs from different neural networks, involving sensory inputs, heading direction, path integration, and behavioral contingencies of the animal (O'Keefe and Nadel, 1978; Jacobs and Schenk, 2003). The heterogeneous nature of spatial representation in the hippocampal circuitry along with pattern completion or pattern separation functions may help distinguish and create distinct cognitive maps for each of the environments based on the behavioral context and episodic events (Knierim, 2002; Lee et al., 2004; Wills et al., 2005; Yoganarasimha et al., 2006; Leutgeb et al., 2007; Neunuebel and Knierim, 2014). Although the SC has reciprocal anatomic connectivity with both the hippocampus and the EC (Ding, 2013), our findings clearly rule out the possibility of simple transfer of spatial information from the hippocampus to other regions of the brain, as the spatial representations in the SC are homogenous and highly coherent compared with the hippocampal representations (Knierim, 2002; Lee et al., 2004; Yoganarasimha et al., 2006; Neunuebel and Knierim, 2014) and appear much stronger than the EC representations (Neunuebel et al., 2013), reported under the same experimental conditions (Table 1). Nearly 72.8% of the CA1 place cells and 40.8% of the CA3 place cells in the hippocampus exhibit diverse responses (appear, disappear, or ambiguous) to cue-conflict conditions, while the rest of the cells followed either the local or distal cues (CA1: 14.3% local, 12.9% distal; CA3: 51.5% local, 7.7% distal) (Lee et al., 2004). When the CA1 place cells exhibited cue-control, 75% of the sessions showed split representation of the cue-conflict environment (i.e., in an ensemble of cells), some followed the rotation of the local cues, while others followed the distal cues and showed coherent representation in only 25% of the sessions (CW 15%, CCW 10%) (Yoganarasimha et al., 2006), compared with highly coherent SC representations in 100% of the sessions (Fig. 6). While the MEC and LEC representations have been shown to be controlled by the global frame (distal cues) and local frame (local cues), respectively, the representations were ambiguous in almost 52% of the MEC cells and 79% of the LEC cells (Neunuebel et al., 2013). This is in stark contrast to our findings from the present study where 100% of the SC neurons showed stable firing patterns along with a highly coherent representation across all track sessions in all three experiments. Also, the population correlation of the MEC cells decreased with an increase in the cue-MIS angle, unlike the population response in the SC (Fig. 8). The cohesiveness of directional and spatial representations has been reported in the ADN, PaS, and the MEC (Hargreaves et al., 2007), indicating that the feedforward HD signals via the PoS may set the orientation of the MEC grid cells. Further, studies have reported attractor dynamics in the HD network (Taube and Burton, 1995; Zhang, 1996; Yoganarasimha et al., 2006; Bassett et al., 2018). Hence, our findings of a robust coupling between the HD cells and the spatial cells of the SC may suggest an attractor-like neural network dynamics.
Distal-cue-controlled, highly coherent representation has been reported in the ADN HD cells, recorded under the same experimental condition (Yoganarasimha et al., 2006). The ADN-HD system may drive neural activity in the PoS (Peyrache et al., 2015) and shows dependence on the PoS for landmark cue control (Goodridge and Taube, 1997). The PoS HD cells have also been shown to carry spatial information (Peyrache et al., 2017), and studies have reported the emergence of an adult-like HD signal in the PoS and MEC before any spatial activity (Wills et al., 2010; Langston et al., 2010). Further, it has been proposed that the HD cell network may play a critical role in the generation and functioning of the grid cells in both the PaS and the MEC (Winter et al., 2015). Interestingly, the coherency of representation in our data was not limited to the PoS HD cells but was observed in all subregions and across different cell types of the SC. This might be because of the intrinsic connectivity within the SC (Ding, 2013). The SUB projects topographically to the PrS and the PaS (Swanson and Cowan, 1977; Swanson et al., 1978; Köhler, 1985; Van Groen and Wyss, 1990a, 1990b; O'Reilly et al., 2013), the PrS has bilateral connectivity with the SUB and the PaS (Köhler, 1985; Van Groen and Wyss, 1990a; Honda et al., 2011), and the PaS has both ipsilateral and contralateral connectivity with the PrS (Van Groen and Wyss, 1990a) while projecting weakly to the SUB (Köhler, 1985; Witter et al., 1988; Van Groen and Wyss, 1990a). This may allow the SC to exhibit an attractor-like activity, resulting in strong coupling between the HD and spatial cells and a highly coherent representation of the environment.
The attractor dynamics in the navigation system have been associated with maintaining coherent spatial representation (Knierim, 2002; Knierim and Zhang, 2012). These network properties are important for the formation and stabilization of a representation, which over time may be modified in terms of reweighting of various task or environment-based contingencies (Knierim, 2002). In view of the dynamically changing environments (e.g., the local–distal cue-MIS experiments performed in our study), the output of an attractor-like network dynamics would follow certain conditions: (1) the cells would follow either the local cues or the distal cues in the environment and never both; (2) overlapping fields in the STD session would all follow either the local or the distal cues; (3) the cells that fire at different locations on the track during STD session, as part of different attractor states, would not fire simultaneously if they overlapped during a MIS session; and (4) individual cells would follow only one set of cues and, thus, not split its firing field into two because of the rotation of the local–distal cues in opposite directions (Knierim, 2002). The stability and coherency of spatial representation observed in the present study follow all of the above conditions and strongly indicate attractor-like dynamics in the SC. The attractor dynamics have also been attributed to the presence of recurrent connectivity within a system. CA3 shows more coherent activity compared with CA1, DG, MEC, and LEC neurons (Lee et al., 2004; Neunuebel et al., 2013; Neunuebel and Knierim, 2014) because of the presence of recurrent collaterals. The SUB also has recurrent connections within its pyramidal neurons (Böhm et al., 2015; Wee and MacAskill, 2020), thus enabling such network dynamics. Such nested loops in the form of recurrent connectivity allows the SUB to compare multiple versions of a representation created from diverse inputs (Naber et al., 2000). This architectural advantage and the presence of HD network (ADN, PrS, and PaS) may allow the SC to function as an attractor-like network.
The cue-conflict paradigm has been used to assess the extent to which an attractor circuitry may incorporate changes to maintain a coherently stable representation, and shows that an increase in the mismatch between the proximal (local) and distal cues decreases the coherency in the hippocampal and the EC representation (Knierim, 2002; Lee et al., 2004; Yoganarasimha et al., 2006; Neunuebel et al., 2013; Neunuebel and Knierim, 2014). On the contrary, the present study shows that the spatial representation in the SC cells was coherently maintained even during maximum MIS conditions, suggesting that the SC may act as a single computational unit to provide a consistent spatio-directional code for orientation to other interdependent neural circuits to form a specific spatial representation. Such a platform may act as a current state indicator in the form of a “reference map.” This concept was initially proposed only for SUB region of the SC (Sharp, 1999b, 2006), providing a “universal map.” Based on our findings, we further extend this concept that the SUB, PrS, and PaS may act synergistically and, thus, are better equipped to function as a “reference map.”
The “reference map” may be thought of as a template that must constantly update itself from the animal's heading direction to provide correct orientation, failure of which may lead to a dysfunctional spatial representation. This orientation is dependent on the selection of the frame of reference, which may be governed by the cues that are perceived to be stable through rapid reweighting (Jeffery et al., 2016). An attractor-like activity showing strong coupling between the directional and the spatial cells of the SC may contribute to a context invariant spatial representation, allowing the representation to encode changes in the environment (expansion or contraction) similar to the previous reports from SUB (Sharp, 1999a). Although the current data do not conclusively infer this, it may serve as a plausible explanation for why such a striking coherence should exist in the SC. Since the SUB region is considered to be an output structure of the hippocampus that exhibits heterogeneous spatial representation under cue-conflict conditions, it is expected that the SUB may also show a response similar to that of the hippocampus. Instead, the representations were found to be stable and highly coherent because of attractor-like activity in the SC. Further, the proposed reference map in the SC may primarily function as a source providing orientation and locational information on a common spatio-directional platform, which the hippocampus may use to compare changes in the environment to undertake pattern separation and pattern completion. For example, SUB lesions lead to deficits in heading and landmark bearing (Morris et al., 1990), whereas combined lesions of the SUB and the hippocampus impaired pattern separation of visual stimuli during delayed matching-to-place task guided by distal visual cues in a modified radial-arm maze (Potvin et al., 2009). Also, PoS lesions disrupted the stability of the hippocampal place cells in a familiar environment (Calton et al., 2003) and impaired the performance in spatial tasks but not in nonspatial tasks (Taube et al., 1992). Further, combined lesioning of PrS and PaS affected spatial tuning of the place cells (Liu et al., 2004), impaired object recognition and spatial memory (Liu et al., 2001), and impaired performance on a 12-arm maze requiring working memory for spatial location (Kesner and Giles, 1998). Thus, this suggests a close collaboration between the SC and the hippocampus in performing spatial functions. However, recordings from the hippocampus under similar experimental conditions with inactivation of specific subregions of the SC may be required to substantiate the causal role of the SC.
Although the results of the present study and the reasons stated above strongly support attractor-like dynamics in the SC, as a caveat, alternate possibilities for the striking coherency between the different SC cell types despite environmental perturbations cannot be ruled out. It may be possible that these different SC cells may belong to a common attractor network or these different cells belong to separate networks that may not have any attractor properties but that receive the same input from another brain region (probably the directional inputs from the ADN), and that input may impose a common orientation onto the different representations. As the alternate possibility cannot be fully disproved with the present data, future studies are needed to address this issue by targeting afferent regions of the SC, especially the HD inputs.
In conclusion, our data show that characteristic features of the attractor-like dynamics, a highly coherent spatial representation along with robust coupling between the HD cells and the spatial cells in the SC, is the first study suggesting the role of the SC in providing a common spatial platform in the form of a “reference map” of the environment, which might then be relied on by the EC and the hippocampus to form a cognitive map of the environment, along with further processing in terms of behavioral contingencies, context, and episodic events.
Footnotes
This work was supported by Department of Biotechnology, Government of India Grant BT/PR14057/Med/30/352/2010 to D.Y. and National Brain Research Center, India core funding. We thank James J. Knierim, Geeta Rao, and Francesco Savelli for critical inputs on the manuscript; Edward I. Moser for sharing the MATLAB codes for calculating the border score; and Guncha Bhasin and Kuldeep Srivastava for the assistance during behavioral experiments.
The authors declare no competing financial interests.
- Correspondence should be addressed to Doreswamy Yoganarasimha at yoganarasimha.doreswamy{at}gmail.com
