1. Neuroscience
Download icon

Differences in reward biased spatial representations in the lateral septum and hippocampus

  1. Hannah S Wirtshafter  Is a corresponding author
  2. Matthew A Wilson
  1. Department of Biology, MassachusettsInstitute of Technology, United States
  2. Picower Institute for Learning andMemory, Massachusetts Institute of Technology, United States
  3. Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, United States
Research Article
  • Cited 0
  • Views 1,059
  • Annotations
Cite this article as: eLife 2020;9:e55252 doi: 10.7554/eLife.55252

Abstract

The lateral septum (LS), which is innervated by the hippocampus, is known to represent spatial information. However, the details of place representation in the LS, and whether this place information is combined with reward signaling, remains unknown. We simultaneously recorded from rat CA1 and caudodorsal lateral septum in rat during a rewarded navigation task and compared spatial firing in the two areas. While LS place cells are less numerous than in hippocampus, they are similar to the hippocampus in field size and number of fields per cell, but with field shape and center distributions that are more skewed toward reward. Spike cross-correlations between the hippocampus and LS are greatest for cells that have reward-proximate place fields, suggesting a role for the LS in relaying task-relevant hippocampal spatial information to downstream areas, such as the VTA.

Introduction

The lateral septum (LS), which is innervated by all CA fields of the hippocampus, contains place cells, and processes spatial information received from the hippocampus (Bezzi, 2005; Olton et al., 1978; Risold and Swanson, 1997; Wirtshafter and Wilson, 2019; Zhou et al., 1999). It has been hypothesized that this spatial information is sent to downstream areas, such as the hypothalamus and ventral tegmental area (VTA) for use in reinforcement and reward seeking (Jiang et al., 2018; Luo et al., 2011; Swanson et al., 1981; Sweeney and Yang, 2015). The LS has also been implicated in contextual reward seeking and the formation of conditioned place preferences (Jiang et al., 2018; Luo et al., 2011). However, little is known about the septal representation of space during reward seeking and what, if any, transformation place information undergoes from its representation in the hippocampus to the lateral septum remains unknown.

The CA fields of the hippocampus contain place cells that show preferential firing when an animal traverses a specific location, the ‘place field’ (O'Keefe and Dostrovsky, 1971; O'Keefe and Nadel, 1978). Extensive work has been done characterizing these fields, including their size (Fenton et al., 2008; Lyttle et al., 2013), skew (Mehta et al., 2000), dependence on experience (Frank et al., 2004; Lee et al., 2004; Mehta et al., 1997; Mehta et al., 2000; Skaggs and McNaughton, 1998), responsiveness to environment and context (Hasselmo and Eichenbaum, 2005; Knierim, 2002; Knierim and Hamilton, 2011), and their distributions around goal locations (Dupret et al., 2010; Hollup et al., 2001; Kobayashi et al., 1997; Kobayashi et al., 2003; Lee et al., 2006). Spatially specific firing has also been well documented and characterized in other brain areas, including in the entorhinal cortex (Fyhn et al., 2004; Hafting et al., 2005; Quirk et al., 1992), where this firing is believed to play a role in path integration (Fyhn et al., 2007; Knierim et al., 2014; Monaco et al., 2011; Moser et al., 2008), and in the lateral septum (Bezzi et al., 2002; Kita et al., 1995; Leutgeb and Mizumori, 2002; Monaco et al., 2019; Takamura et al., 2006; Wirtshafter and Wilson, 2019; Zhou et al., 1999), though much less is known about the potential role of LS place fields in spatial navigation.

Given the anatomical location of the LS between areas involved in spatial navigation such as the hippocampus, and reward/reinforcement related areas such as the VTA (Luo et al., 2011; Risold and Swanson, 1997), and the established role of the LS in reinforcement seeking (Jiang et al., 2018; Mathieu-Kia et al., 1998; McGlinchey and Aston-Jones, 2018; Olds and Milner, 1954; Oshima and Katayama, 2010; Sotomayor et al., 2005), we hypothesized that LS place fields would be preferentially located at reward locations to a greater extent than hippocampal place fields and contain more reward related information.

In the following experiment, we simultaneously recorded cells from the hippocampal CA1 field and the caudodorsal lateral septum during a spatial navigation task. The caudodorsal LS was chosen as it is the area of the LS most heavily innervated by the hippocampus and known to send the most projections to the VTA (Luo et al., 2011; Risold and Swanson, 1997). We characterized lateral septal place field firing and compared it to that seen in CA1. We found that, although not as common as in the hippocampus, the lateral septum contains a large number of place cells with fields that have properties similar to the place cells of the hippocampus, including similar field size and similar number of fields per cell. However, in contrast to HPC place cells, LS cells are slightly less accurately spatially tuned. Additionally, within a firing field, LS cells ramped up their firing in the direction of reward, and place fields in the lateral septum were preferentially biased toward reward locations compared to the hippocampus. We found that HPC and LS cells with reward proximate fields had significantly higher correlated activity than cells with fields located further from reward. We suggest that reward-related spatial information from the hippocampus is preferentially represented in the LS, and we provide three models by which this could occur. We suggest that this relayed information is used downstream to direct the animal to significant or rewarded locations.

Results

In order to characterize the role of the LS in reward-driven spatial navigation, we recorded the activity of 452 caudodorsal LS units and 178 CA1 hippocampus units in six male Long Evans rats (Figure 1A–B). We specifically targeted the caudodorsal LS as it receives the a large innervation from the hippocampus and is the primary source of LS projections to the VTA (Luo et al., 2011; Risold and Swanson, 1997). Units were recorded when the animals performed a spatial working memory task on a double-sided T maze, used previously to characterize spatial activity in multiple regions, including in the hippocampus (Gomperts et al., 2015; Jones and Wilson, 2005a; Jones and Wilson, 2005b; Siegle and Wilson, 2014), prefrontal cortex (Jones and Wilson, 2005a; Jones and Wilson, 2005b) VTA (Gomperts et al., 2015), and lateral septum (Wirtshafter and Wilson, 2019). In this task, animals are forced to a randomly chosen side of one of the Ts (the forced arm), and must run down a center stem and choose the same side of the opposite T (the choice arm) to be rewarded (Figure 1C–D). The structure of this task results in mirrored turns and track traversals which help control for behavioral variability, allowing us to look specifically as the effect of reward approach without confounding variables such as turn approach.

Rats were implanted with a tetrode array in the HPC and LS and run on a double sided T maze.

(A) Brain section from implanted rat, showing the lateral septum and electrolytic lesions made after recording. The red arrows mark lesions at the tetrode tips. Red shaded area indicates area where about 95% of lesions were seen. (B) Illustration of the maze task, which consisted of two phases: in the forced choice phase, animals were randomly forced with a block (represented by a red rectangle in a schematic) to either side of the T. In the free choice phase, animals had to choose, at the opposite end of the maze, the same side to which they were forced. If they made the correct choice, animals were rewarded with a sucrose and chocolate mixture. (C) Tracked position during one 30 min session.

Place fields are less abundant in the LS as compared to the hippocampus CA1

We first sought to determine the prevalence of place fields in the cdLS, as estimates have varied wildly from one study reporting no fields (Tingley and Buzsáki, 2018), to estimates of about a third to half of all LS cells (Takamura et al., 2006; Zhou et al., 1999). In all recorded LS cells and CA1 principle cells, we evaluated the spatial information content in cell firing using an information measure of bits per spike. We defined a cell with place information to be a cell with a bits per spike cutoff of 0.8 bits/spike, as this cutoff has been used previously for determining spatial firing of non-hippocampal cells (Ji and Wilson, 2007; Markus et al., 1994). We did not include cells with a mean spiking rate of less than 0.05 hz, and also eliminated cells during trials where the animal did not cover the entire track at a speed of at least 12 cm/s. We found that 75.2% (124/165) of CA1 cells (Figure 2A) and 33.6% (127/378) of LS cells (Figure 2B) met or exceeded the 0.8bits/spike threshold, with the average bits/spike of a CA1 cell significantly higher than the average bits/spike of an LS cell (HPC mean 1.54+−1.2 bits/spike, LS mean 0.73+−0.7 bits/spike, two-tailed two sample t-test t(541)=9.42, p<0.001) (Figure 2C) (We also computed mutual information, see Figure 2—figure supplement 1). To ensure that the representation of space was different than would be expected from random Poisson firing, we created 454 artificial LS units using Poisson firing and the mean firing rates of the recorded LS units (Figure 2B, inset). The distribution of the bits/spike for the artificial units was highly significantly different than the distribution of bits/spike for actual units (KS test, p<10−15), and only 7.96% of artificial units had bits/spike measurements of greater than 0.8. The average bits/spike for the artificial units was 0.43, compared to an average value of 0.73 for actual units (two-tailed two sample t-test, t(818)=6.72, p<10−10).

Figure 2 with 1 supplement see all
Hippocampal units have a higher average bits per spike and bits per second than septal units.

(A) Bits per spike of all recorded hippocampal units with an average spike rate >0.05 hz and full track coverage. A bits per spike cutoff of 0.8bits/spike is marked with a dotted line. (B) Larger graph: same as A for units in the lateral septum. Inset: bits per spike for artificially created LS units. The distribution of the bits/spike for the artificial units was highly significantly different than the distribution of bits/spike for actual units (KS test, p<10−15), and only 7.96% of artificial units had bits/spike measurements of greater than 0.8. The average bits/spike for the artificial units was 0.43, compared to an average value of 0.73 for actual units (two-tailed two sample t-test, t(818)=6.72, p<10−10). (C) Comparison of bits per spike for hippocampal and septal units. The average bits/spike of a CA1 cell is significantly higher than the average bits/spike of an LS cell. HPC mean 1.54+−1.2 bits/spike, LS mean 0.73+−0.7 bits/spike, two-tailed two sample t-test t(541)=9.42, p<0.001). Error bars represent standard error. (D) Bits per second of all recorded hippocampal units with an average spike rate >0.05 hz and full track coverage. (E) Same as D for units in the lateral septum. (F) Comparison of bits per second for hippocampal and septal units. CA1 units had a mean of 1.34+−1.4bits/sec, and LS cells had a mean 0.82+−1.1bits/sec. Units in the hippocampus had a significantly greater mean bits/sec than in the LS (two-tailed two sample t-test t(541)=4.57, p<0.001). Error bars represent standard error.

Because bits/spike is sensitive to differences in firing rate, we also calculated bits per second for the CA1 and LS units. We found a mean of 1.34+−1.4bits/sec for CA1 cells, and 0.82+−1.1bits/sec for LS cells (Figure 2D–E). Units in the hippocampus had a significantly greater mean bits/sec than in the LS (two-tailed two sample t-test t(541)=4.57, p<0.001) (Figure 2F).

We next determined how many cells with bits/spike greater than 0.8 had definable place fields. Place fields boundaries have previously been defined in a multitude of ways, including with a flat firing rate threshold (Rich et al., 2014). Because firing rate for LS place fields may not follow the same criteria as hippocampal place fields and may be contingent on the LS’s innate firing properties, we opted to use a threshold derived from the units’ average firing rate. We identified a place field as connected area at least 15 cm long with a peak firing rate of at least two standard deviations above the unit’s mean firing rate, and the boundaries of the place field to be when firing drops below one standard deviation above mean firing rate (see Methods).

Of the 164 HPC units with bits/spike greater than or equal to 0.8, 104 units had at least one place field, with 63 of these cells with one field, 33 with two fields, seven with three fields, and a single cell with four fields (Figure 3A,C). Of the 127 LS cells with bits/spike greater than 0.8, 100 had at least one place field, with 54 with one field, 33 with two fields, six with three fields, five with four fields, and two with five fields (Figure 3B,D). The distributions of field numbers were not significantly different between CA1 and LS cells (two sample Kolmogorov-Smirnov (KS) test, p>0.5). We have previously shown that LS cells are modulated by speed and acceleration (Wirtshafter and Wilson, 2019). To determine the potential contribution of spatially biased speed and acceleration to our place field results, we performed a multiple linear regression of each spatial bin’s firing rate against the bin’s average speed and acceleration. Only two cells with place fields had an r2 of >0.5, with the median r2 value of 0.06 (Figure 3—figure supplement 1), so very little spatial firing could be explained by correlations with speed or acceleration.

Figure 3 with 2 supplements see all
CA1 and LS place cells have a comparable number of place fields.

The distribution of field numbers was not significantly different between HPC and CA1 cells (two sample Kolmogorov-Smirnov (KS) test, p>0.5). (A) Number of fields in CA1 cells with a bits per spike greater than or equal to 0.8 bits/spike. (B) Same as A for lateral septum. (C) Example units in the CA1. Top: Example of a unit with a single place field. Bottom: example of a unit with three place fields. (D) Same as C but for the LS.

In summary, out of 165 recorded CA1 cells that met spike rate and track coverage criteria, 104 (63.0%) met or exceeded a bits per spike cut off of 0.8 and had at least one place field that met the described criteria, while out of 378 recorded LS cells meeting spike rate and track coverage criteria, 100 (26.5%) met the same criteria (see Figure 3—figure supplement 2 for additional place cell examples and representations of firing rates). Place cells in the CA1 are therefore significantly more abundant than in the cdLS (two-tailed two sample t-test t(541)=8.6, p<0.001).

CA1 and LS place fields have comparable average sizes and are shaped by experience

We next determined if there were differences in size between hippocampal and LS place fields. In order to qualify as a place field, we applied a minimum length standard of at least 15 cm. The average length of a CA1 place field was 29.1+−14.7 cm (Figure 4A,C), while the average length of an LS place field was 28.8+−16.6 cm (Figure 4B,D). The average field lengths were not significantly different (two-tailed two sample t-test t(320)=-0.16, p>0.05).

CA1 and LS place fields have comparable lengths and are modified with experience.

The average field lengths in CA and LS were not significantly different (two-tailed two sample t-test t(320)=-0.16, p>0.05). (A) Distribution of field lengths in the CA1. The average length of a CA1 place field was 29.1+-14.7cm. (B) Distribution of field lengths in the LS. The average length of an LS place field was 28.8+-16.6cm. (C) Example fields in the CA1. Top: example of a shorter field with a length of 22cm. Bottom: example of a longer field with a length of 78cm. (D) Example fields in the LS. Top: example of a shorter field with a length of 20cm. Bottom: example of a longer field with a length of 76cm. (E) Distance of highest HPC firing location from mean place field center during each lap. Average distance across all laps is marked with a dotted line The average distance from the field center in lap one is slightly but significantly different than the average field distance from the center in lap 30 (two-tailed two sample t-test t(177)=-2.7, p<0.01). (F) Distance of highest LS firing location from mean place field center during each lap. The average distance from the field center in lap one was not significantly different than the average field distance from the center in lap 30 (two-tailed two sample t-test t(207)=0.2, p>0.05). On average, compared to the LS, the HPC is slightly more accurate on the first lap as well as across all laps. (On the first lap, HPC has a mean distance of 2.7cm versus 4.9cm for the LS, two-tailed two sample t-test t(319)=-3.1, p<0.005. Across all laps, mean distance of 4.70cm for the HPC versus 5.12cm for the LS, two-tailed two sample t-test t(6792)=-2.5, p<0.05). (G) HPC spiking frequency per cm as a function of location around the place field center. The average spiking rate/cm of the first five runs through the field is marked in red, and the last five runs in black. Error bars represent standard error. The difference between the first and last run averages was significant for all cm values except +18cm and +20cm (all two-tailed two sample t-test). (H) Same as G but for the LS. The difference between the first and last run averages was significant for all cm values except -18cm (all two-tailed two sample t-test).

Previous work has demonstrated that hippocampal place fields stabilize and become more tuned to position with experience (Mehta et al., 1997). We compared the time periods over which HPC and LS place fields become stable. For each pass through a place field, we determined how far the center (determined by maximum spiking) of the place field was from the average place field center. Both the hippocampus (Figure 4E) and LS (Figure 4F) had highly accurate place fields starting with the first pass of the place field, though, on average the HPC is slightly more accurate on the first lap as well as across all laps. (On the first lap, HPC has a mean distance of 2.7 cm versus 4.9 cm for the LS, two-tailed two sample t-test t(319)=-3.1, p<0.005. Across all laps, mean distance of 4.70 cm for the HPC versus 5.12 cm for the LS, two-tailed two sample t-test t(6792) = −2.5, p<0.05). The slight but significant decrease in accuracy from the first to later laps in the HPC can be explained by a slight shift of the place field peak toward the direction of travel (Figure 4G). Fields in both the HPC and the LS significantly increase firing in their place field with experience on the track (Figure 4G–H).

CA1 and LS place fields have different skew and location distributions

We also wondered if bias toward reward location could also be seen within a place field. To determine this, looked at place field skew as a function of travel direction, computing skew for a place field using total within-field firing (for skew during individual laps, see Figure 5—figure supplement 1). We found that, when travelling toward a reward site, cells in the hippocampus were skewed positively toward the direction of travel, while, when traveling away from a reward site, cells were skewed negatively away from the direction of travel, and this difference was significant (two-tailed two sample t-test t(207)=-2.1, p<0.05) (Figure 5A,B, see also Figure 5—figure supplement 23). Conversely, when traveling toward a reward site, LS place fields were skewed negatively away from the direction of travel, and when traveling away from a reward site, the fields were skewed positively toward the direction of travel, though this difference is not significant (two-tailed two sample t-test t(246)=-1.7, p=0.09) (Figure 5A,C). Although the HPC and LS both contains uni- and bi- directional place fields, there was no significant difference for values of skew based on whether a place cell was uni or bi directional (both two-tailed two sample t-tests, p>0.05, see Figure 5—figure supplement 4). Because skew can be sensitive to changes in speed, we also computed the firing rate asymmetry index (FRAI, see methods) (Mehta et al., 2000) for HPC and LS cells, and found a highly significant relationship between skew and FRAI values (Figure 5—figure supplement 5), showing that skew cannot be explained by the animal’s speed.

Figure 5 with 5 supplements see all
HPC and LS place cells have opposite directions of skew based on travel direction.

(A) Comparison of mean skew values within immediate reward proximity. Stars indicate significant differences between multiple populations. Plus signs indicate the mean is significantly different than 0. Xs indicate significant differences from a shuffled distribution, see Figure 5—figure supplement 3. Error bars represent standard error. In the hippocampus, there is a significant difference in mean skews based on direction of travel (two-tailed two sample t-test t(207)=-2.1, p<0.05). (B) Graph of distributions of skews in the hippocampus based on direction of travel. (C) Graph of distributions of skews in the septum based on direction of travel. (D) Schematic showing arms of the maze examined during immediate reward proximity. Toward reward direction is marked with a blue arrow, away from reward with a red arrow. (E) Comparison of mean skew values within immediate reward proximity. Stars indicate significant differences between multiple populations. Plus signs indicate the mean is significantly different than 0. Xs indicate significant differences from a shuffled distribution, see Figure 5—figure supplement 3. Left: Error bars represent standard error. Skew values were not significantly different for hippocampus traveling to and from reward (two sample two sided t-test, t(74)=-0.59 p>0.05), but were significantly different for LS cells to and from reward (two sample two sided t-test, t(108)=-2.10, p<0.05). (Skew for HPC away from reward is significantly different from a shuffled sample, two sample two sided t-test, t(301)=-2.05, p<0.05. Skew for LS away from reward is significantly different from zero, one sample two sided t-test, t(58)=-2.0, p=0.05). Right: Schematic for clarification of skew relative to reward location for both directions of travel. Arrow represents direction of travel, ‘R’ represents reward location. Note that while HPC skew away from reward for reward proximal cells appears to have a different direction than for all HPC cells when traveling away from reward, the two means are not significantly different (two sample two sided t-test, t(131)=1.62, p>0.05).

We then examined the effect of reward proximity on skew for place cells that were in the immediate reward proximity (in the rewarded arms, Figure 5D). Comparing the mean skew values in immediate reward proximity (Figure 5E, see also Figure 5—figure supplements 23), skew values were not significantly different for hippocampus traveling to and from reward (two sample two sided t-test, t(74)=-0.59 p>0.05), but were significantly different for LS cells to and from reward (two sample two sided t-test, t(108)=-2.10, p<0.05).

We then looked at field location for HPC and LS place fields to determine which arms of the maze were most represented in the hippocampus and LS (Figure 6A). Place field locations were based on maximum spiking (field center) in the place field. In the hippocampus, 49/154 of place fields occurred on the forced sides, 36/154 on the center stem, and 45/154 on the end of the choice sides (Figure 6B,D). (Force and choice points were excluded as they were difficult to assign to an arm, but 24 place fields occurred at the forced and choice point. For data further subdivided by location see Figure 6—figure supplement 1.) In the LS, 37/168 of place fields occurred on the forced sides, 40/168 on the center stem, and 69/168 on the end of the choice sides (22 occurred at choice points) (Figure 6C,E). The distribution of place fields is significantly different between the hippocampus and the lateral septum (Pearson’s chi2 test, X2 = 6.03, p<0.05). The LS has about 1.4 times, proportionally, significantly more place fields in the choice side than the HPC (41.1% of total fields in the LS, versus 29.2% of total fields in the HPC, two-tailed two sample t-test t(320)=2.23, p<0.05). In the hippocampus, the number of fields in the choice arms versus the forced arms was not significantly different (two-tailed two sample t-test t(364) = 2.34, p>0.05), while, in the LS, there were significantly more fields in the choice arms compared to the forced arms (two-tailed two sample t-test t(334)=-3.8, p<0.001).

Figure 6 with 2 supplements see all
CA1 and LS fields have different location distributions, with LS place fields more biased toward reward locations.

The distribution of place fields is significantly different between the hippocampus and the lateral septum (Pearson’s chi2 test, X2=12.7, p<0.05). (A) Schematic showing where the maze is split to identify place field location. (B) Scatter plot of HPC field centers (C) Scatter plot of LS field centers D. Distribution of HPC field centers. In CA1, place fields are no more likely to be on the choice side of the maze than the forced side (two-tailed two sample t-test t(364) = -2.34, p>0.05) (E) Distribution of LS field centers. There were significantly more fields in the choice arms compared to the forced arms (two-tailed two sample t-test t(334)=-3.8, p<0.001). The LS also has about 1.4 times, proportionally, more place fields in the choice side than the HPC does (41.1% of fields in the LS, versus 29.2% of fields in the HPC, two-tailed two sample t-test t(320) = -2.23, p<0.05). (F) Distribution of HPC place field centers by direction of travel. Top: traveling to reward. Bottom: traveling away from reward. (G) Distribution of HPC place field centers by direction of travel. Top: traveling to reward. Bottom: traveling away from reward. The LS had significantly more place fields on the choice side than the forced side in both travel directions, with the difference highly pronounced travelling away from reward (travelling toward reward two-tailed two sample t-test t(254)=1.82 p<0.10, travelling away from reward two-tailed two sample t-test t(218)=3.8 p<0.001). The distribution of place fields in the choice side was also significantly different based on the animal’s travel (one-tailed two sample t-test t(102)=-2.15, p<0.05). (H) The probability of finding a hippocampal place field as a function of distance from rewarded locations. As above, red represents the forced side of the maze, blue the middle arm, and purple the choice side. Note that the divisions between the three segments of the maze are not exactly represented in the histogram due to binning. (I) Same as H but for the LS.

We wondered if place field location depending on direction of travel; for instance, if it was more likely to see a place field by a reward site after the site had been visited. To determine this, we split fields by direction, based on whether the animal was traveling to or from reward (if a field existed in both directions, we analyzed its parameters in both directions). This resulted in a total of 209 hippocampal place fields (115 toward reward and 94 away from reward, with, out of the total, 133 being unidirectional and 76 being bi directional) and 248 LS place fields (138 toward reward, 110 away from reward, with, out of the total, 177 unidirectional and 71 bidirectional. There was no significant difference of numbers of uni- or bi- directional HPC or LS cells, two-tailed two sample t-test, t(461)=1.377, p>0.05), and no difference in the distribution of HPC or LS place cell locations depending on if a cell was uni- or bi- directional both (KS test, p>0.05). There was no significant different between field sizes based on direction within the LS, within the HPC, or across the LS and HPC (all two-tailed two sample t-test p>0.05).

In the LS, the difference seen in the number of fields on the forced side versus choice side was most stark when the animal was traveling away from reward (two-tailed two sample t-test t(218)=3.8 p<0.001), although there was a clear trend while the animal traveled toward reward (two-tailed two sample t-test t(254)=1.82 p<0.1) (Figure 6G). In the hippocampus, there were never significantly more fields on the choice side (Figure 6F). While LS place fields were more concentrated around reward sites, average place field firing rate did not scale with distance to reward in either the LS or the hippocampus (Figure 6—figure supplement 2).

We also computed the probability of finding a place field as a function of distance from reward (Figure 6H–I). In the hippocampus, there was an increase in the probability of a spatial firing field in the last 60 cm of reward approach (Figure 6H). However, the largest peeks in HPC place field probability were around the forced and choice points of the maze, approximately 200–220 cm and 80–60 cm away from reward, respectively. In the LS, the entire forced arms were highly overrepresented, and the probability of a place field also increased upon reward approach.

In order to determine if HPC place cells were preferentially innervating LS cells, we computed spike train cross correlations between HPC and LS place cells with place fields in similar locations (centers less than or equal to 20 cm apart) (Wilson and McNaughton, 1994). (To adjust for spiking variance, we also computed the cross correlation with a shuffled LS spike train, and subtracted the mean of this cross correlation from the computed pairwise cross correlation.) We found that place cell pairs with fields in the choice side of the maze had a significantly higher mean average correlation (over +−100 ms) than place cell pairs with fields in the middle or forced arm (Figure 7A) (forced versus choice: two-tailed two sample t-test t(67)=-2.2, p<0.05, middle versus choice: two-tailed two sample t-test t(54)=-2.3, p<0.05). For pairs in the choice arm, the maximum value of the cross correlation happened at a mean lag of 20 ms with the HPC leading (Figure 7B, for all cross correlations, see Figure 7—figure supplement 1), within approximately the same time course seen for sharp wave ripple propagation from the HPC to the LS (Wirtshafter and Wilson, 2019). It does not appear that the higher cross correlations for cells on the choice side of the track were due to higher firing rates of cells proximal to reward, as there was no significant difference between the mean and maximum firing rates of LS place cells in all three locations (see Figure 7—figure supplement 2). The higher cross correlations were also not the result of greater proximity of field centers between LS-HPC place cell pairs in the choice arm (see Figure 7—figure supplement 3).

Figure 7 with 3 supplements see all
Reward proximate LS cells are more synced with hippocampal activity than not proximate cells.

(A) Averaged cross correlation across a lag of −100 to 100 ms between coupled pairs of HPC and LS cells, based on place field location. Error bars show standard error. Forced side versus choice side: two-tailed two sample t-test t(67)=-2.2, p<0.05, middle arm versus choice side: two-tailed two sample t-test t(54)=-2.3, p<0.05). (B) Average of cross correlation traces at all lags between coupled pairs of HPC and LS cells, based on place field location. Error bars show standard error. Mean correlation peak for forced arm correlations was at 20 ms. Results from unpaired two tailed t-tests between forced and choice arms are indicated with stars: * indicates a p value of < 0.05 on an unpaired two tailed t-test, ** indicates a p value of < 0.01, and *** indicates a p value of < 0.005. Analogous results are shown between choice arm and middle arm using +. (C) Differences in cross correlations for HPC-LS pairs during the first, middle, and last third of a single trial. Error bars show standard error. * indicates a p value of < 0.05 on an unpaired two tailed t-test, *** indicates a p value of < 0.005. (D) Three models of LS innervation by HPC that may explain the overrepresentation of LS place fields by rewarded locations. Top: model 1. HPC cells with reward-proximate place fields selectively innervate more LS place cells. Middle: model 2. HPC cells with reward proximate place fields, and cells with other fields innervate the same number of LS cells. Hippocampal cells with non-proximate fields innervate overlapping cells, causing interference and resulting in fewer place fields in the LS that are not reward proximate. Bottom: model 3. HPC cells with reward proximate place fields, and cells with other fields innervate the same number of LS cells. Hippocampal cells with non-proximate fields innervate cells that are also innervated by other inputs, causing interference and resulting in fewer place fields in the LS that are not reward proximate.

Finally, we sought to determine if correlations between HPC and LS spiking became stronger throughout a single day’s training (Figure 7C). Comparing pairs of cells within the first third, middle third, and last third of training, there were no significant differences within group (e.g. cross correlation of reward proximate cells in the first third versus the last third of training) (all paired double sided t tests, p>0.05). However, across all three periods of training, cross correlations for reward proximate cells were consistently higher than for more reward-distal cells. It therefore appears that activity correlations between HPC and LS reward-proximate cells are either developed early in training before the task is learned and not lost between training sessions, or they are not developed within training sessions.

Discussion

In this study, we have completed a comprehensive characterization and analysis of LS place cells and their properties using parameters identical to those used to characterize HPC place cells. We found that, while the percentage of place cells is lower in the LS than in CA1, the observed LS place cells are similar to those in the CA1 in terms of field numbers and field size. We additionally determined that LS do not converge on a place field center with as much precision as CA1 cells. Finally, we found that LS place fields are distributed more toward reward locations than place fields in CA1, and that this distribution is even more biased toward rewarded locations if the animal is travelling away from the reward site. We suggest that these characteristics are the result of selective LS innervation by HPC place cells, which then allows reward related information to be sent to areas downstream of the LS.

Previous reports have varied wildly on the prevalence of place cells in the LS, with estimates ranging from none (Tingley and Buzsáki, 2018), to a third (Leutgeb and Mizumori, 2002; Zhou et al., 1999) to almost half or more of all cells (Takamura et al., 2006). We found that 26.5% of LS cells could be characterized as place fields based on both a bits per spike cutoff and spatial firing characteristics. The difference in estimates can be accounted for by several variables: first, using a stringent bits/spike and field size criteria eliminates many cells that may fit one criteria or the other, or that may have very small fields. Additionally, as we found that LS place fields tended to be clustered around reward locations, the presence, absence, and location of reward could have affected the total number of place fields. Finally, the caudodorsal LS receives the densest innervation from the hippocampus (Swanson, 1977), so more ventral areas of the LS may have fewer place fields, a result that has been previously observed (although place fields have been found throughout the entirety of the LS) (Takamura et al., 2006).

LS place field sizes are comparable to HPC field sizes, and both HPC and LS cells increase place field spiking with experience. However, LS place cells are slightly but significantly less accurate than HPC place fields (Figure 4E–H). This may be the result of LS cells incorporating other parameters beyond current spatial location. We have previously shown that LS cells’ firing rate may be affected by an animal’s speed or acceleration (Wirtshafter and Wilson, 2019), which may decrease the precision of place field firing. The incorporation of proximity to reward location, discussed below, may also serve to modulate LS place cell firing beyond the animal’s current location.

We observed that the distribution of place fields in the lateral septum was more biased toward the rewarded locations of the maze than the distribution of place fields in CA1 (Figure 6), and that, unlike HPC place fields, LS place fields tended to skew toward reward direction regardless of the direction of travel, particularly when close to rewarded locations (Figure 5). Previous work has also established that the LS can represent both reward delivery and reward locations (Nerad and McNaughton, 2006; Wirtshafter and Wilson, 2019). There are many possible ways this result could be accounted for at the circuit level: First, reward information may be introduced into the lateral septum from an area outside of the hippocampus. The LS has connections, often reciprocal, with many brain areas that respond to reinforcing stimuli, including the VTA (Gomperts et al., 2015; Jiang et al., 2018; Luo et al., 2011; Roesch et al., 2007), hypothalamus, which is involved with the LS in a circuit that regulates feeding (Sweeney and Yang, 2016), ventral hippocampus (Zhou et al., 2019), and amygdala (Krettek and Price, 1978; Sheehan et al., 2004). It is possible that information about reward location is being sent from, for example, the VTA to the LS, rather than vice versa. However, we think it unlikely that this circuit accounts for the all, or the majority, of reward proximate place firing in the LS for several reasons: First, reward related cells in the VTA are specifically reactivated during hippocampal replay and sharp wave ripples at a time course consistent with the signal first travelling through the LS from the hippocampus (Gomperts et al., 2015; Wirtshafter and Wilson, 2019). Additionally, cells in the LS that change firing during a reward period tend to show signatures of hippocampal interactions, such as theta coherence and increased firing during hippocampal sharp-wave ripples (Wirtshafter and Wilson, 2019). Finally, we found that LS cells with reward proximate firing are more highly cross correlated with hippocampal firing than cells without reward proximate place fields (Figure 7). This suggests that a predominant input to these LS cells that are biased to reward locations is from the hippocampus.

In the present task, the probability of finding a HPC place field increases as the animal approaches final goal location (Figure 6H). However, the representation of the goal is decreased relative to locations proximate to the forced and choice points of the maze. Past work has found that the HPC uniquely over-represents salient or goal locations on tasks demanding increased spatial memory (Dupret et al., 2010). It is possible, therefore, since the important spatial memory components of this task occurred at the forced and choice points, that these locations came to be even more over-represented in the HPC that the goal locations.

Reward-related spatial information may be transmitted from the HPC to the LS in multiple ways. First, there may be a fixed population of HPC cells that consistently represent reward, which selectively innervate the LS. There is direct evidence for such a population (Gauthier and Tank, 2018), but it is unknown if it comprises a large portion of LS afferents. Alternatively, different HPC place cells may represent reward in different spatial environments, but the LS cells receiving place cell input are differentially modulated by external inputs depending on the specific task.

So how are reward proximate place cells more common in the LS than in the hippocampus? First, it is possible that hippocampal place cells that have fields proximate to reward locations selectively, either anatomically or functionally, innervate LS cells (Figure 7D, Model 1). Additionally, it is possible that non-reward related place cell inputs from the hippocampus are diluted in the LS if they converge with non-reward related input. This input may be from other place cells without reward proximate fields (Figure 7D, Model 2), or from other areas, such as speed or acceleration related information from the brainstem (Figure 7D, model 3; Wirtshafter and Wilson, 2019). In this model, reward proximate place cells would innervate a separate group of LS cells which would account for the over-representation of reward proximate place fields in the LS (Figure 7D, Model 2). Our finding that LS and HPC place cells with fields proximate to reward have increased cross correlations compared to cells with more distal place fields is consistent with any of these models, as the higher cross correlation may be due to a selective innervation of the LS by reward-proximate HPC cells, or the ‘dilution’ of non-reward related place cells by competing inputs (Figure 7D, Model 3). However, our previous work (Wirtshafter and Wilson, 2019) has shown that LS cells containing place information are less likely to contain speed or acceleration information than cells without place fields, suggesting that speed or acceleration information may be competing with spatial information sent from the HPC, a result most easily accommodated by model 3.

We also found differences in place cell activity based on whether the animal was moving toward or away from the reward site. We first found that the reward site itself was overrepresented in LS place fields right after the reward site had been visited (Figure 6G). We also found that during reward approach, within a place field, LS place cells, on average ramped up firing (Figure 5E). It is possible that both this overrepresentation and difference in skew may serve memory or planning functions during navigation. We have previously found that the HPC-LS coherence increased during working memory tasks (Wirtshafter and Wilson, 2019), and increased place field activity at a site of significance could be evidence of hippocampal and LS coordination during encoding of significant task locations. There is increased VTA reactivation of reward related cells during ripples and VTA cells also ramp up firing during reward approach (Gomperts et al., 2015), showing that the significance of reward related locations has been established downstream of the LS. That information may be communicated to the VTA from the LS by the over-representation of just-visited reward related locations in place cell firing, as well as by increased within-field firing upon reward approach.

The distribution of field numbers was not significantly different between HPC and CA1 cells (two sample Kolmogorov-Smirnov (KS) test, p>0.5).

Top panel is traveling to reward, bottom panel is traveling away. Because of the non-linearity of the track, different place fields were visited different numbers of times, and therefore each field contributes a different number of bins to these histograms.

Materials and methods

Subject details

Request a detailed protocol

All procedures were performed within MIT Committee on Animal Care and NIH guidelines. Six male Long-Evans rats (275 g to 325 g) were sourced from Charles River and implanted with tetrode arrays and run on a double-sided T maze (see Figure 1). Different data from the same subjects were previously published in Wirtshafter and Wilson, 2019. Animals were individually housed in an animal facility with a 12 hr light dark cycle.

Method details

Tetrode implementation and electrophysiology

Request a detailed protocol

Rats were implanted under isoflourine anesthesia (induction 4%, maintenance 1–2%) with two multi-tetrode arrays, each containing 16 independently moveable tetrodes (see Jones and Wilson, 2005b). One tetrode array was directed toward the dorsal CA1 hippocampus (stereotaxic coordinates Bregma −3.7, midline −3.2), while the other was directed toward the caudo-dorsal lateral septum (stereotaxic coordinates Bregma +.05, midline −0.5). Animals were grounded with a skull screw posterior to Lambda. Over several days tetrodes were individually lowered to their goal location. A CA1 reference tetrode was placed in the corpus collosum white matter tract above CA1. A lateral septum reference tetrode was placed in white matter above the LS, or at a quiet site in the lateral septum.

Electrical signals were passed through two 16 channel headstage preamplifiers to a custom patchbox which was then used to select a reference channel. The signal was then fed to Neurolynx amplifiers. Extracellular action potentials were acquired at 31 kHz, 0.3–6 kHz filtering. Data were collected using custom lab software ARTE (Hale and Wirtshaftee, 2019, June 30). The animal’s position during trial runs was collected at 30 Hz via overhead cameras. Position was collected and later extracted using OAT (Newman et al., 2017, December 10).

After data collection, CA1 and LS cells were manually isolated using a custom software package (Xclust) using spike amplitude on each of the four channels. Septal and hippocampal cells with large amounts of drift and/or unstable waveforms were excluded in the analysis. After completion of the study, animals were lesioned with 15μA of current for 10 s to mark tetrode location. Animals were then perfused at least one week post lesioning, and tetrode locations were then verified with histology.

Behavioral training

Request a detailed protocol

During training, animals were food deprived to 85% body weight. Implanted animals were trained for 2–4 weeks to run a spatial choice task (Gomperts et al., 2015; Jones and Wilson, 2005b) on an end-to-end T maze (Figure 1). Animals were free running on this task and were only handled to be placed on the maze and removed from the maze. The maze had two phases: in the forced choice phase, animals were randomly forced to either side of the T into the forced arms. (Animals were not directed to one side more than three consecutive times.) In the free choice phase, animals had to choose, at the opposite end of the maze, the same side that they were forced to in the free choice arm. If animals made the correct choice, animals were rewarded with 0.2 mL of 20% sucrose 10% chocolate milk powder dispensed remotely from a syringe pump. After a trial, animals self-initiated a new trial by returning to the forced arm of a maze. Tail tip in an arm was used as the criteria for arm entrance. Animals were trained to a criterion of 75% correct choices. There was wide variability among all the animals for the length of time it took to learn the task, as well as their ability to maintain their performance at criterion. Animals were each run for 30 min a day.

Data and code availability

Request a detailed protocol

All analysis code was custom written and anaylysis was performed using Matlab (MathWorks, Natick, Massachusetts). Code is public on https://github.com/hsw28/data_analysis/ (Wirtshafter, 2020; copy archived at https://github.com/elifesciences-publications/data_analysis).

Statistical analysis

Position and velocity sampling

Request a detailed protocol

Position was sampled by overhead cameras at about 30 Hz. Due to occlusion, sampling rates of position were often at approximately 15 Hz. Speed was determined by taking the hypotenuse of the coordinates of the points immediately before and after the time point of interest. Speed was then smoothed using a Gaussian kernel of 1 s standard deviation and was then converted from pixels/s to cm/s.

Bits/spike and bits/second

Request a detailed protocol

All animal occupancies and spikes were found when the animal was travelling at or over 12 cm/second. Firing per occupancy found in 1 cm2 increments, smoothed with 10 cm standard deviation Gaussian kernel.

A bits per spike measurement was calculated as follows:

i=spatial bin number
Pi=occupancy probability for bin i
Ri=mean firing rate at bin i
R=overall mean firing rate
bits per spike=iPi RiRlog2RiR

A bits per second measurement was calculated as follows:

bits per second=iPi Rilog2RiR

Cells were discarded if mean rate was below 0.05 hz during the time period above velocity threshold. Mutual information was found as in Ego-Stengel and Wilson, 2007.

Place field analysis

Request a detailed protocol

Velocity was measured as determined above, and threshold was set at 12 cm/second. All animal occupancies and spikes were found when the animal was travelling at or over 12 cm/second. Firing per occupancy was found in 2 cm increments, smoothed with 10 cm standard deviation Gaussian kernel.

A place field was identified as a region of connected bins (connected on at least one of four sides, not diagonally) where firing rate was equal to or less than one standard deviation greater than the mean firing rate. In order to be classified as a place field, firing in at least one location in this region must be greater than two standard deviations greater than the mean firing rate. Place field area must be at least 15 cm long. We analyzed place field selection in our HPC data using these criteria and criteria previously established in Rich et al., 2014. We found that out criteria identified place fields in the hippocampus consistent with the previously established criteria (Rich et al., 2014).

Cells with a max rate of less than 0.05 hz were discarded from analysis.

Place field centers were determined by the coordinates within the place field that had the highest spiking rate.

Directionality was determined by computing place fields in both directions. If a unit had fields in both directions with centers separated less than 20 cm, the field was considered bidirectional. For bidirectional place cells, skew was computed in both directions.

In lap by lap analysis, place fields were included if the field was passed through at least 15 times, and laps were included if at least 20 place fields had that many laps.

Skew was determined as the ratio of the third moment of the place field firing rate spatial distribution found in the direction of travel, divided by the cube of the standard deviation of this distribution (Spiegel, 1961). Because skew was determined lengthwise along the track, firing rates along the width of the place field were averaged.

  • FRAI (Mehta et al., 2000) was computed as

    • F1 = mean firing rate for first half of spikes in field

    • F2 = mean firing rate for second half of spikes in field

      • FRAI = F1 F2 / (F1+F2)

Cross correlation analysis

Request a detailed protocol

Place cells from HPC were matched with place cells from the LS recorded during the same session based on field firing center (if a cell had multiple fields it was matched multiple times). In order to be matched, firing field centers must have been within 20 cm of each other (Wilson and McNaughton, 1994). If there were multiple matches, the match with the most similar average firing rate was chosen. If no matches could be found, the cell was excluded from analysis.

Cross correlations of HPC spike train x LS spike train were determined between all track spiking for matched cell pairs. Spike trains for cells were found in 10 ms bins and cross correlations were taken over +−100 ms. For control, an additional cross correlation was determined with HPC spike train x shuffled LS spike train. This control was subtracted from the previously determined cross correlation to get the final cross correlation for analysis.

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
  54. 54
    Schaum's Outline of Theory and Problems of Statistics
    1. MR Spiegel
    (1961)
    New York: Schaum Pub Co.
  55. 55
    The anatomical organization of septo-hippocampal projections
    1. LW Swanson
    (1977)
    Ciba Foundation Symposium. pp. 25–48.
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
  61. 61
  62. 62
  63. 63
  64. 64
  65. 65

Decision letter

  1. Laura L Colgin
    Senior and Reviewing Editor; University of Texas at Austin, United States
  2. Alexey Ponomarenko
    Reviewer

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

There are still so few studies about the lateral septum, disproportional to its potential significance. This is solid work with a novel finding that increases our understanding of mechanisms underlying motivated behaviors and neural representations of space.

Decision letter after peer review:

Thank you for submitting your article "Differences in reward biased spatial representations in the lateral septum and hippocampus" for consideration by eLife. Your article has been reviewed by Laura Colgin as the Senior Editor and Reviewing Editor and three reviewers. The following individual involved in review of your submission has agreed to reveal their identity: Alexey Ponomarenko (Reviewer #1).

The reviewers have discussed the reviews with one another and the Senior Editor has drafted this decision to help you prepare a revised submission.

Wirtshafter and Wilson investigated spatial firing of lateral septal (LS) and CA1 neurons using parallel recordings from the two regions during a rewarded navigation task. They show that a large number of LS cells display place-selective firing, a finding the authors have previously reported. Here, they perform more detailed analysis of place field characteristics of LS neurons, showing that LS place fields are similar in many ways to HPC fields and that LS place fields are more likely to occur near reward/choice locations. They found that one-dimensional place fields of LS neurons are in many respects similar to CA1 place fields, yet they display different skewness in relation to reward location. Further, firing of LS and CA1 cells is more correlated for those cells with place fields close to the reward location. Because neural representations in LS are poorly understood, studies of circuits including LS can provide new insights into functions of hippocampus and other interconnected regions. The present high-quality data contribute to the understanding of information processing in LS and substantially extend earlier reports (Takamura et al., 2016) showing the influence of reward location on spatial firing of LS neurons. The LS has been given somewhat short shrift by the broader hippocampal community, and although the work is somewhat exploratory, this was viewed as appropriate for the relatively early days of investigating the LS.

However, the study was viewed as having some shortcomings in its quantifications that leave open important questions. The paper needs to show a more transparent presentation of the results. The results should also be better situated in the existing literature.

Essential revisions:

1) There was a considerable variability of firing in individual runs, and LS place fields maxima did not converge to a single location. Reviewers had questions about the variability in LS firing patterns:

a) Was this variability in the different arms of the maze, and other features of spatial representations in general, related to behavioral performance or behavioral variations in the task?

b) Related to the above point, an underlying assumption is that LS neurons are selectively encoding information about reward proximity or recent reward receipt (supported by the data in Figure 6). The authors observe a higher probability of LS place fields on the Choice Side when the rat is leaving vs. approaching a reward well. An open question which is important to interpret these results is whether LS neurons care only about location relative to *possible* reward, or rather care about location relative to *actual* reward. Given that rat performance is ~75% correct according to the methods, do the authors observe an effect of success or failure on the task? Do LS neurons differentiate between runs toward or away from the same location when that location is actually rewarded vs. when it was not? On a similar note, given the increase in LS place fields upon leaving the reward site, do the authors observe larger LS place fields after correct vs. incorrect trials? Alternatively do the authors observe LS place fields which do not distinguish between the two arms, but instead only represent correct vs. incorrect trials?

c) There was confusion about the results described in Figure 4E-F. The fact that the curve for hippocampal neurons eventually reaches zero indicates that spatial representation stabilizes over time. However, while the curve is downward for LS cells (indicating some level of stabilization), the reason why it never reaches zero was unclear. Do these data simply mean that each lap is highly variable for LS cells even after a small amount of initial stabilization? If so, and if the final distance of lap place field center to average place field center is 20 cm, this indicates that the 'place field' on each lap moves up to 40 cm (+/- 20 cm from average center). Considering that the track center is ~120 cm long and each arm is ~70 cm long, movement of 40 cm from lap-to-lap seems quite substantial. Is it possible that LS place fields shift slowly across time, making it near-impossible for any given lap to align with the average across all laps? Given the bias for LS place fields to be located near the reward well, are near-reward place fields more or less likely to stabilize than reward-distant fields?

d) LS cells are quite heterogeneous in their firing rate and the regularity of discharge. Was there any systematic relationship between spatial firing properties and firing rate, coefficient of variation and the recording location? This information may also help explain apparent discordances with other studies of spatial firing in LS (Tingley and Buzsaki, 2018).

2) Regarding the data in Figure 2, the distribution of HPC unit bits/spike appears to be comprised of at least two (possibly three) clusters: one with very low bits/spike (near zero), one with moderate bits/spike (centered around 1.25), and possibly a third with very high bits/spike (centered around 2.9). This is what would be expected in a population of neurons in which most, but not all, encode spatial information in any given environment. Appropriately, the division between cells that encode spatial information and those that do not (bits/spike = 0.8) does seem to correctly divide those populations. However, for the LS neurons, the distribution appears to be composed of a single population, and the separation that the authors make between cells that encode spatial information and those that do not seems rather arbitrary for LS cells. In other words, if you only looked at the distribution of bits/spike for LS cells, where would you draw the red dashed line? Reviewers raised a concern that the distribution of bits/spike for LS cells is not reflective of a true spatial encoding, but rather, what one might expect from random firing patterns. If the authors shuffle cell IDs for each spike, do they still observe a similar distribution of bits/spike across the shuffled units? If the authors create 452 artificial units using Poisson firing, do they observe a similar distribution of bits/spike for the artificial units? In other words, is the place representation of some neurons really different than one would expect from chance, given that the authors are sampling 452 neurons and at least some of those neurons are likely to fire in spatially restricted locations by chance?

3) Could increased cross-correlations computed for a broad range of lags of 100 ms in the choice arm be due to higher firing rates proximal to rewards? Information about firing rates (peak rates, average rates in field) in different arms is difficult to find in the manuscript.

4) Given the importance of LS place fields to this study, reviewers would like to see more than four examples. Reviewers suggested a Supplemental Figure that presents a large number of LS place fields, selected in an unbiased way.

5) Reviewers would also like to see examples of spike cross-correlations supporting the data in Figure 7.

6) The authors note in the introduction that the hippocampal encoding of goal locations has been characterized, but they do not cite a highly-relevant study: Dupret et al., 2010. These experiments showed that the hippocampal over-representation of goal is not universal, but instead depends on the cognitive demands of the task. It is important to know, therefore, whether goal locations are over-represented in the present study. One straightforward way to address this would be to re-analyze the data from panels 6E-F. If the probability of a field were *per unit distance*, results from the different track segments would be directly comparable. This would reveal whether hippocampal place cells themselves are clustered near reward in this task, and to what degree LS neurons might amplify that clustering.

7) The plot in Figure 7B shows an intriguing correlation. This quantification only shows the effect is present when averaging across all neurons, however. Does every neuron show such a correlation, or just a subpopulation? This information could help to corroborate or reject the authors' models. Related to the previous point, it would also provide an estimate of what fraction of hippocampal neurons might be specialized for encoding reward.

8) The recording location in the caudodorsal LS is very close to the septohippocampal nucleus (SHN) based on Paxinos and Watson's stereotaxic atlas, and indeed, at least one of the recording sites in Figure 1A appears to be in the SHN. I am unaware of any literature quantifying place representation in the SHN, but given the controversy regarding whether LS neurons have place fields, it is important to determine whether the place-selective units in the current study are more likely to be LS or SHN neurons. Do the authors observe a correlation between medial/lateral (or other stereotaxic orientation) recording location and likelihood of observing place-specific firing? Perhaps in a supplemental analysis, the authors could replicate several of their core findings using only units recorded on the lateral-most tetrodes, which would be the least likely to be in the SHN.

9) In Figure 5, the authors quantify skewness and compare runs to reward vs. runs away from reward, reporting a significant difference for hippocampal fields, but not for LS fields. To facilitate interpretation of these results, reviewers would like to know if the skewness of HPC or LS neurons during runs to reward were different from zero (or different from a shuffle distribution). The same point was raised for runs away from reward.

10) Do LS cells have direction-specific place fields, and how does direction-selective firing (in HPC and LS neurons) impact skewness? Is the skewness driven entirely by uni-directional fields, or is it observed for both uni- and bi-directional fields?

11) The datapoints in Figure 5 and Figure 5—figure supplement 2 don't seem to line up, and there was confusion about these figures. Is it correct that each dot represents a cell with its average place field peak at that location (relative to reward) and with that skewness? If that is correct, shouldn't the data in Figure 5F-G be a subset of the data in Figure 5—figure supplement 2? However, those points aren't the same. The difference should be made clear in the text or in the figure legend.

[Editors’ note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your article "Differences in reward biased spatial representations in the lateral septum and hippocampus" for consideration by eLife. Your article has been reviewed by Laura Colgin as the Senior Editor and Reviewing Editor, and three reviewers. The following individuals involved in review of your submission have agreed to reveal their identity: Alexey Ponomarenko (Reviewer #1).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

Reviewers agree that authors have satisfactorily addressed most concerns. Reviewers are grateful that the authors identified the coding error that led to many of the reviewer questions and also appreciate the additional work the authors have performed on analyses and reworking the text of the manuscript. Reviewers agree that this is a considerably stronger manuscript after the revisions. There are only a few comments remaining:

Essential revisions:

1) It is still possible that the higher cross-correlations over many tens of milliseconds between HPC and LS cells with place fields in the choice side (Figure 7) can be secondary to a higher number of place fields in both regions in this part of the maze (Figure 6). This might be addressed, for instance, by comparing cross-correlations for forced, middle and choice sides for subsampled spike trains selected to ensure a matched degree of overlap of place fields in HPC and LS. However, the paper would still be interesting and worthy of publication even if it turns out that cross-correlations are due to concentration of Hip and LS place fields on the choice side close to rewards.

2) Figure 5 seems to be consistent with the lack of influence of hippocampal inputs on the skewness of LS place fields. The latter are similarly skewed at different locations. However, HPC place fields change their directional skewness depending on the proximity of the reward. It would be helpful to integrate this finding in the models proposed.

https://doi.org/10.7554/eLife.55252.sa1

Author response

Essential revisions:

1) There was a considerable variability of firing in individual runs, and LS place fields maxima did not converge to a single location. Reviewers had questions about the variability in LS firing patterns:

The high amounts of field center variability/maxima originally seen in Figure 4E-F were reevaluated and we discovered a minor computational error involving cells with multiple fields. That mistake has been fixed and all results in Figure 4E-F have been redone, with the addition of Figure 4G-H. We discuss this more in response to question 1C below, but, in brief, the average distance from a place field center for peak LS firing has now been found to be 5.12cm, with a standard deviation of +-7.1cm (as opposed to the >20cm of distance from field center previous reported). This correction does not influence our overall conclusions regarding LS place fields.

a) Was this variability in the different arms of the maze, and other features of spatial representations in general, related to behavioral performance or behavioral variations in the task?

As explained above, the variability is much smaller than originally reported, and does not appear to differ across arms of the maze. Because LS cells are known to incorporate speed and acceleration information in spike rate, we attempted to take into account this behavioral variation by computing a linear regression of spike rate against speed and acceleration for all LS cells with a bits/spike over 0.8. However, very few cells had large r2 values (Figure 3—figure supplement 1). This is consistent with our previously published work which shows that cells that code for spatial location tend not to also code for acceleration or speed. We have made this point more clear in the text.

b) Related to the above point, an underlying assumption is that LS neurons are selectively encoding information about reward proximity or recent reward receipt (supported by the data in Figure 6). The authors observe a higher probability of LS place fields on the Choice Side when the rat is leaving vs. approaching a reward well. An open question which is important to interpret these results is whether LS neurons care only about location relative to *possible* reward, or rather care about location relative to *actual* reward. Given that rat performance is ~75% correct according to the methods, do the authors observe an effect of success or failure on the task? Do LS neurons differentiate between runs toward or away from the same location when that location is actually rewarded vs. when it was not? On a similar note, given the increase in LS place fields upon leaving the reward site, do the authors observe larger LS place fields after correct vs. incorrect trials? Alternatively do the authors observe LS place fields which do not distinguish between the two arms, but instead only represent correct vs. incorrect trials?

We attempted to do this analysis prior to the original submission, but unfortunately the bounds of the task make it rather hard. With between 20 and 30 trials a day, the animal is likely only to visit an individual rewarded arm between 10 and 15 times. if the animal is performing at above 75%, that often meant the animal passed through a place field two or fewer times per day on incorrect trials on their way to reward. Unfortunately, that did not provide a sufficient sample to determine whether individual cells were responding to the presence or absence of reward versus reward expectation, or the effect of reward receipt on place cell characteristics.

We did examine whether some cells appeared to represent rewarded locations preferentially. To do this, we looked at all cells that had more than one place field, and determined if the odds of having two place fields in the rewarded arms were higher than having a place field at a rewarded arm and a choice arm or the middle stem.

Out of the total number of cells with multiple place fields, 23 of these cells had exactly two place fields that occurred on one of the stems (as opposed to the choice points). Out of the corresponding 46 fields, 11 (23.9%) occurred in the forced arm, 12 (26.1%) occurred in the middle stem, and 23 (50.0%) occurred in the choice side, with 11 (23.9%) on one side of the arm and 12 (26.1) on the other. In any pair of fields, therefore, expect a cell with a place field on each end of the forced side about 12.48% of the time. Out of the 23 place field pairs, we see this combination 5 times, or 21.74% of the time. While this is suggestive, the low number of pairs makes the result not significantly different from expected so we did not report it in our results.

c) There was confusion about the results described in Figure 4E-F. The fact that the curve for hippocampal neurons eventually reaches zero indicates that spatial representation stabilizes over time. However, while the curve is downward for LS cells (indicating some level of stabilization), the reason why it never reaches zero was unclear. Do these data simply mean that each lap is highly variable for LS cells even after a small amount of initial stabilization? If so, and if the final distance of lap place field center to average place field center is 20 cm, this indicates that the 'place field' on each lap moves up to 40 cm (+/- 20 cm from average center). Considering that the track center is ~120 cm long and each arm is ~70 cm long, movement of 40 cm from lap-to-lap seems quite substantial. Is it possible that LS place fields shift slowly across time, making it near-impossible for any given lap to align with the average across all laps? Given the bias for LS place fields to be located near the reward well, are near-reward place fields more or less likely to stabilize than reward-distant fields?

As a consequence of the error reported above, we re-computed the results for place cells with multiple fields. These new results give us a much clearer picture of highly accurate place fields in the LS and HPC, with increased firing at the place field center with experience. These results do not effect our overall conclusion that the LS is less tuned to space than the HPC. We have redone Figure 4E-F and also added Figure 4G-H. We have explained them in the text as follows:

“Previous work has demonstrated that hippocampal place fields stabilize and become more tuned to position with experience (Mehta, Barnes and McNaughton, 1997). We compared the time periods over which HPC and LS place fields become stable. For each pass through a place field, we determined how far the center (determined by maximum spiking) of the place field was from the average place field center. Both the hippocampus (Figure 4E) and LS (Figure 4F) had highly accurate place fields starting with the first pass of the place field, though, on average the HPC is slightly more accurate on the first lap as well as across all laps. (On the first lap, HPC has a mean distance of 2.7cm versus 4.9cm for the LS, two-tailed two sample t-test t(319)=-3.1, p<0.005. Across all laps, mean distance of 4.70cm for the HPC versus 5.12cm for the LS, two-tailed two sample ttest t(6792)=-2.5, p<0.05). The slight but significant decrease in accuracy from the first to later laps in the HPC can be explained by a slight shift of the place field peak towards the direction of travel (Figure 4G). Fields in both the HPC and the LS significantly increase firing in their place field with experience on the track (Figure 4G-H).”

d) LS cells are quite heterogeneous in their firing rate and the regularity of discharge. Was there any systematic relationship between spatial firing properties and firing rate, coefficient of variation and the recording location? This information may also help explain apparent discordances with other studies of spatial firing in LS (Tingley and Buzsaki, 2018).

For demonstration of the rate variability, we have added Figure 3—figure supplement 2 which shows more examples of place fields and also shows the distribution of firing rates. In answering review concern #3 we also compared mean and maximum firing rate across different track segments and found no significant difference (Figure 7—figure supplement 2). There thus appeared to be no significant relationship between firing rate and the rat’s location in the maze.

As far as tetrode location could be mapped, we saw no difference in firing rate and place cell probability from more lateral to more medial locations in the LS. We recorded the vast majority of cells in the most dorsal area of the LS so, similarly, the vast majority of place fields were located there. We found that cells appeared to be sparser, as judged by number of cells picked up by a tetrode, in more ventral LS areas.

Previous studies that recorded from the entire LS (dorsal to ventral) reported place cells along the entire depth of the LS. (For example, see Takamura et al., 2006, Figure 6:)

Examples of studies finding LS place cells in areas other than the most dorsal LS include:

- Bezzi et al., 2002

- Kita et al., 1995

- Leutgeb and Mizumori, 2002

- Monaco et al., 2019

- Nishijo et al., 1997

- Takamura et al., 2006

- Zhou et al., 1999 among others (all of which are referenced in text), so we do not think tetrode location can explain the discrepancies with the unique results found in Tingley and Buzsaki, 2018.

2) Regarding the data in Figure 2, the distribution of HPC unit bits/spike appears to be comprised of at least two (possibly three) clusters: one with very low bits/spike (near zero), one with moderate bits/spike (centered around 1.25), and possibly a third with very high bits/spike (centered around 2.9). This is what would be expected in a population of neurons in which most, but not all, encode spatial information in any given environment. Appropriately, the division between cells that encode spatial information and those that do not (bits/spike = 0.8) does seem to correctly divide those populations. However, for the LS neurons, the distribution appears to be composed of a single population, and the separation that the authors make between cells that encode spatial information and those that do not seems rather arbitrary for LS cells. In other words, if you only looked at the distribution of bits/spike for LS cells, where would you draw the red dashed line? Reviewers raised a concern that the distribution of bits/spike for LS cells is not reflective of a true spatial encoding, but rather, what one might expect from random firing patterns. If the authors shuffle cell IDs for each spike, do they still observe a similar distribution of bits/spike across the shuffled units? If the authors create 452 artificial units using Poisson firing, do they observe a similar distribution of bits/spike for the artificial units? In other words, is the place representation of some neurons really different than one would expect from chance, given that the authors are sampling 452 neurons and at least some of those neurons are likely to fire in spatially restricted locations by chance?

We chose to use a 0.8bits/spike cutoff for several reasons: first, we wanted to use the same criterion to judge the spatial firing properties of LS cells as to judge the spatial firing properties of HPC cells. Secondly, previously literature examining spatial firing in nonhippocampal areas, such as in the visual cortex, used a 0.8bits/spike cutoff for spatial firing (see Ji and Wilson 2007), and we have added a mention of this to the text. Thirdly, we wished to predefine the criterion for place cell inclusion prior to data analysis to obtain the most objective results.

To address the concerns of the reviewers, we created artificial units with Poisson firing, matching the units average firing rate to actual LS units. We added this data to figure 2B, and added the following in the text:

“To ensure that the representation of space was different than would be expected from random Poisson firing, we created 454 artificial LS units using Poisson firing and the mean firing rates of the recorded LS units (Figure 2B, inset). The distribution of the bits/spike for the artificial units was highly significantly different than the distribution of bits/spike for actual units (KS test, p<10-15), and only 7.96% of artificial units had bits/spike measurements of greater than 0.8. The average bits/spike for the artificial units was 0.43, compared to an average value of 0.73 for actual units (two-tailed two sample ttest, t(818)=6.72, p<10-10).”

3) Could increased cross-correlations computed for a broad range of lags of 100 ms in the choice arm be due to higher firing rates proximal to rewards? Information about firing rates (peak rates, average rates in field) in different arms is difficult to find in the manuscript.

This seems unlikely, as the middle arm actually had the highest mean and maximum firing rate per place field and the lowest cross correlation, and the mean and maximum rates for the three locations are not significantly different (both one-way anovas). We have added Figure 7—figure supplement 2 showing the mean and maximum in field firing rates for LS cells in each location, and have added the following lines to the text:

“It does not appear that the higher cross correlations for cells on the choice side of the track were due to higher firing rates of cells proximal to reward, as there was no significant difference between the mean and maximum firing rates of LS place cells in all three locations (see Figure 7—figure supplement 2).”

The addition of Figure 3—figure supplement 2 also gives a clearer picture of firing rates across LS place cells, and we have added a histogram to this figure as well to represent all firing rates of LS cells within place fields.

4) Given the importance of LS place fields to this study, reviewers would like to see more than four examples. Reviewers suggested a Supplemental Figure that presents a large number of LS place fields, selected in an unbiased way.

Figure 3—figure supplement 2A was added which shows spatial firing rate heat maps of 30 randomly chosen LS cells.

5) Reviewers would also like to see examples of spike cross-correlations supporting the data in Figure 7.

Figure 7—figure supplement 1 was added which includes all cross correlations for data in Figure 7, as well as data for shuffled spike trains.

6) The authors note in the introduction that the hippocampal encoding of goal locations has been characterized, but they do not cite a highly-relevant study: Dupret et al., 2010. These experiments showed that the hippocampal over-representation of goal is not universal, but instead depends on the cognitive demands of the task. It is important to know, therefore, whether goal locations are over-represented in the present study. One straightforward way to address this would be to re-analyze the data from panels 6E-F. If the probability of a field were *per unit distance*, results from the different track segments would be directly comparable. This would reveal whether hippocampal place cells themselves are clustered near reward in this task, and to what degree LS neurons might amplify that clustering.

Great suggestion. We ran the data as suggested and added this data as panels on Figure 6. We found, to our positive surprise, that the LS, steadily increased the number of place fields as the animal approached the reward on the final arm. The hippocampus, in the final 60cm of award approach, appeared to drastically increase the number of place fields as well. However, this was offset somewhat by the large increase in fields 60-80cm from reward, which was mirrored by a large increase 200-220cm from reward. Both of these locations are proximal or overlapping the forced/choice point in the maze.

To explain this, we added the following text in the Results section of the paper:

“We also computed the probability of finding a place field as a function of distance from reward (Figure 6H-I). In the hippocampus, there was an increase in the probability of a spatial firing field in the last 60cm of reward approach (Figure 6H). However, the largest peeks in HPC place field probability were around the forced and choice points of the maze, approximately 200-220cm and 80-60cm away from reward, respectively. In the LS, the entire forced arms were highly overrepresented, and the probability of a place field also increased upon reward approach.”

We also added the following in the Discussion section:

“In the present task, the probability of finding a HPC place field increases as the animal approaches final goal location (Figure 6H). However, the representation of the goal is decreased relative to locations proximate to the forced and choice points of the maze. Past work has found that the HPC uniquely over-represents salient or goal locations on tasks demanding increased spatial memory (Dupret et al., 2010). It is possible, therefore, since the important spatial memory components of this task occurred at the forced and choice points, that these locations came to be even more over-represented in the HPC that the goal locations.”

7) The plot in Figure 7B shows an intriguing correlation. This quantification only shows the effect is present when averaging across all neurons, however. Does every neuron show such a correlation, or just a subpopulation? This information could help to corroborate or reject the authors' models. Related to the previous point, it would also provide an estimate of what fraction of hippocampal neurons might be specialized for encoding reward.

To answer this question, we shuffled the spike trains of all the HPC and LS pairs on the forced arm (Figure 7—figure supplement 1) and found an average correlation of 7.71e-05, with a 95% confidence interval of [-8.89e-05, 2.43e04]. Out of 36 unit pairs on the forced arm, the average of 26 of these pairs (72%) fell above the 95% confidence interval for shuffled data. Therefore, it appears that about 72% of pairs at the choice arm have a significantly higher correlation than would be expected by chance (and this number would clearly be higher if using a one-sided measurement). As pairing of neurons is not exact, I would also expect an even higher number of significant correlations with additional pairing options. Because the neurons pairs in the choice arm set are, by definition, proximal to reward, this finding appears to suggest that by virtue of being reward proximate the vast majority of HPC neurons are reward-encoding.

8) The recording location in the caudodorsal LS is very close to the septohippocampal nucleus (SHN) based on Paxinos and Watson's stereotaxic atlas, and indeed, at least one of the recording sites in Figure 1aAappears to be in the SHN. I am unaware of any literature quantifying place representation in the SHN, but given the controversy regarding whether LS neurons have place fields, it is important to determine whether the place-selective units in the current study are more likely to be LS or SHN neurons. Do the authors observe a correlation between medial/lateral (or other stereotaxic orientation) recording location and likelihood of observing place-specific firing? Perhaps in a supplemental analysis, the authors could replicate several of their core findings using only units recorded on the lateral-most tetrodes, which would be the least likely to be in the SHN.

Based on implant location and histology results, the number of tetrodes even potentially in the SHN is extremely small (<5), and we have recording data from over 100 tetrodes. Thus, it is impossible for tetrode placement in the SHN to explain the presence of place cells.

Additionally, based on the atlas, we are not even confident that these aforementioned 5 tetrodes are actually in the SHN versus the LS.

Additionally, previous work has found LS place fields spread throughout the entirety of the LS, including the more lateral and ventral regions (see our response to 1D above) (Bezzi et al., 2002; Kita et al., 1995; Leutgeb and Mizumori, 2002; Monaco et al.,, 2019; Nishijo et al., 1997; Takamura et al., 2006; Zhou et al., 1999).

Some anatomists, including Swanson, who published the seminal works on LS anatomy and chemoarchitecture , also consider that the SHN may be a part of the LS, and that the projections to and from the SHN (including the organization of these projects) are not markedly different from those seen in the remainder of the LS (Risold and Swanson 1997a,b). They also state “Furthermore, additional evidence now suggests that the septofimbrial nucleus, which lies caudomedially adjacent to the LS, as well as the tiny septohippocampal nucleus dorsomedially adjacent., may well constitute additional parts of what might be called the lateral septal complex, or at least should be considered with the LS.”

These considerations argue against redoing the analysis with a small subset of cells, particularly as including only the most lateral tetrodes would greatly diminish our statistical power.

9) In Figure 5, the authors quantify skewness and compare runs to reward vs. runs away from reward, reporting a significant difference for hippocampal fields, but not for LS fields. To facilitate interpretation of these results, reviewers would like to know if the skewness of HPC or LS neurons during runs to reward were different from zero (or different from a shuffle distribution). The same point was raised for runs away from reward.

A random distribution was obtained by shuffling firing rates within a firing field and then computing skew for the shuffled data. Because of a large variance in averages for shuffled data, averages for 500 shuffled distributions were computed. We added these shuffled averages as Figure 5—figure supplement 3. For hippocampal skew, the average skew for traveling to reward, traveling away from reward, and traveling away from reward for reward proximal cells all occurred less than 5% of the time in shuffled data. The skew result for skew traveling to reward for reward proximal cells did not fall outside 95% of shuffled data. These results have been added to Figure 5.

For the LS, the skew results for traveling away from reward, and reward proximal in both directions occurred less than 5% of the time in shuffled data. Traveling to reward occurred more than 5% of the time. These results have also been added to Figure 5.

We also used a one sample two-sided t-test to determine if any of the skew distributions had a mean significantly different than zero. The only significant result was for LS reward proximal cells while traveling away from reward (mean skew for LS away from reward is significantly different from zero, one sample two-sided t-test, t(58)=-2.0, p=0.05). This result has also been added to Figure 5.

10) Do LS cells have direction-specific place fields, and how does direction-selective firing (in HPC and LS neurons) impact skewness? Is the skewness driven entirely by uni-directional fields, or is it observed for both uni- and bi-directional fields?

The LS has both unidirectional and bidirectional place fields. A paragraph was added to Materials and methods section to describe how it was determined whether a field was uni or bi directional:

“Directionality was determined by computing place fields in both directions. If a unit had fields in both directions with centers separated less than 20cm, the field was considered bidirectional. For bidirectional place cells, skew was computed in both directions.”

We also added the following to the text to emphasize there are both uni and bi directional HPC and LS place fields:

“We wondered if place field location depending on direction of travel; for instance, if it was more likely to see a place field by a reward site after the site had been visited. To determine this, we split fields by direction, based on whether the animal was traveling to or from reward (if a field existed in both directions, we analyzed its parameters in both directions). This resulted in a total of 209 hippocampal place fields (115 towards reward and 94 away from reward, with, out of the total, 133 being unidirectional and 76 being bi directional) and 248 LS place fields (138 towards reward, 110 away from reward, with, out of the total, 177 unidirectional and 71 bidirectional. There was no significant difference of numbers of uni- or bi- directional HPC or LS cells, two-tailed two sample ttest, t(461)=1.377, p>0.05).”

The average LS skew is not significantly different for uni and bi-directional fields directional (two-tailed two sample t-test, t(246)=-0.19, p=0.85) and there are a wide range of skew values for both uni and bi-directional LS place fields. We also computed the values for uni and bi directional place cells based on direction of travel, and there was also no significant difference across the values (one way anova, F(3,244) = 1.05, p>0.05). We have added Figure 5— supplemental figure 4 to display these results.

We completed analogous analysis for HPC place cells and found analogous results: the mean skew was not significantly different between uni- and bi- directional place cells (two-tailed two sample t-test, t(207)=-0.59, p=0.55), nor did directionality matter for skew measured based on direction travelled relative to reward (one way anova, F(3,205) = 0.29, p>0.05). We have also added this data to Figure 5—figure supplement 4.

11) The datapoints in Figure 5 and Figure 5—figure supplement 2 don't seem to line up, and there was confusion about these figures. Is it correct that each dot represents a cell with its average place field peak at that location (relative to reward) and with that skewness? If that is correct, shouldn't the data in Figure 5F-G be a subset of the data in Figure 5—figure supplement 2? However, those points aren't the same. The difference should be made clear in the text or in the figure legend.

We thank the reviewers for catching this slip, which is also a result of the computational error reported above (Figure 5—figure supplement 1 is correct). After re-running the data, although there is a trend in the direction seen in originally in Figure 5F, it is not significant and the Figure 5F was moved to the supplement (now Figure 5—figure supplement 2). However, with this correction made, the result in Figure 5E is more striking than previously.

The finding has been discussed in the text as follows:

“We observed that the distribution of place fields in the lateral septum was more biased towards the rewarded locations of the maze than the distribution of place fields in CA1 (Figure 6), and that, unlike HPC place fields, LS place fields tended to skew towards reward direction regardless of the direction of travel, particularly when close to rewarded locations (Figure 5).”

[Editors’ note: further revisions were suggested prior to acceptance, as described below.]

Essential revisions:

1) It is still possible that the higher cross-correlations over many tens of milliseconds between HPC and LS cells with place fields in the choice side (Figure 7) can be secondary to a higher number of place fields in both regions in this part of the maze (Figure 6). This might be addressed, for instance, by comparing cross-correlations for forced, middle and choice sides for subsampled spike trains selected to ensure a matched degree of overlap of place fields in HPC and LS. However, the paper would still be interesting and worthy of publication even if it turns out that cross-correlations are due to concentration of Hip and LS place fields on the choice side close to rewards.

We understand and appreciate this concern. We first decided to determine if there was a difference in the average distance between LS and HPC place field pairs for the forced side, central stem, and choice side. When we completed this analysis, we found that there was no significant difference between the average distance between LS and HPC pairs in the forced arm versus choice arm, so the proximity of HPC-LS pairs did not account for the difference in cross correlation values for forced vs. choice sides. There was a small (3cm) but significant difference in distances when comparing the choice side to the middle stem.

We then followed the suggestion to subsample and found that eliminating the very closest pairs (pairs that had centers within 3cm of each other) was more than sufficient to result in an insignificant difference between forced, choice, and middle pair distances. However, the average cross correlation for pairs on the choice side was still significantly higher than the average in either the forced side or middle stem.

Therefore, it does not appear that differences in cross correlations are due to differences in the concentration of place fields in the forced, middle, and choice sides. We have added this to the text and created Figure 7—figure supplement 2.

2) Figure 5 seems to be consistent with the lack of influence of hippocampal inputs on the skewness of LS place fields. The latter are similarly skewed at different locations. However, HPC place fields change their directional skewness depending on the proximity of the reward. It would be helpful to integrate this finding in the models proposed.

The only difference in HPC skewness direction based on proximity occurs while traveling away from reward, where, looking at all HPC cells, the average skewness is negative, but for reward proximal cells, the average skewness is positive. However, the values for skew are not significantly different from one another (double sided t test t(131) = 1.62, p>0.05). We have added this clarification into the text for Figure 5: Note that while HPC skew away from reward for reward proximal cells appears to have a different direction than for all HPC cells when traveling away from reward, the two means are not significantly different (two sample two sided t-test, t(131)=1.62, p>0.05).

https://doi.org/10.7554/eLife.55252.sa2

Article and author information

Author details

  1. Hannah S Wirtshafter

    1. Department of Biology, MassachusettsInstitute of Technology, Cambridge, United States
    2. Picower Institute for Learning andMemory, Massachusetts Institute of Technology, Cambridge, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Funding acquisition, Validation, Investigation, Methodology, Writing - original draft, Writing - review and editing
    For correspondence
    hsw@mit.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-4684-7074
  2. Matthew A Wilson

    1. Department of Biology, MassachusettsInstitute of Technology, Cambridge, United States
    2. Picower Institute for Learning andMemory, Massachusetts Institute of Technology, Cambridge, United States
    3. Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, Cambridge, United States
    Contribution
    Supervision, Funding acquisition, Project administration, Writing - review and editing
    Competing interests
    No competing interests declared

Funding

U.S. Department of Defense (NDSEG Fellowship)

  • Hannah S Wirtshafter

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

HSW was supported by the Department of Defense (DoD) through the National Defense Science and Engineering Graduate Fellowship (NDSEG) Program. We thank Israel Donato Ridgley and Molly Quan for support with analysis and experiments.

Ethics

Animal experimentation: All procedures were performed within MIT Committee on Animal Care and NIH guidelines under Wilson protocol 0417-037-20. All surgeries were done under isoflourine anesthesia (induction 4%, maintenance 1-2%) and every effort was made to minimize suffering.

Senior and Reviewing Editor

  1. Laura L Colgin, University of Texas at Austin, United States

Reviewer

  1. Alexey Ponomarenko

Publication history

  1. Received: January 20, 2020
  2. Accepted: May 24, 2020
  3. Accepted Manuscript published: May 26, 2020 (version 1)
  4. Version of Record published: June 5, 2020 (version 2)

Copyright

© 2020, Wirtshafter and Wilson

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 1,059
    Page views
  • 221
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Neuroscience
    2. Physics of Living Systems
    Joanne C Gordon et al.
    Research Article Updated
    1. Neuroscience
    Julia Erb et al.
    Research Article