Characteristics of CA1 place fields in a complex maze with multiple choice points

2.1 Animals

Four male Long-Evans rats weighing 400–500 g during the recordings were used. During training and recording, the animals were individually housed and on a restricted diet (max 10% weight reduction) on weekdays to ensure motivation. Water was freely available all times. All procedures were approved by the Committee on Animal Care at Massachusetts Institute of Technology and followed U.S. National Institutes of Health guidelines.

2.2 Environment and behavioral tasks

Recordings took place in a complex symmetric “Townmaze” (180 cm × 140 cm) with three chambers equipped with slide doors and winding alleys separated with 20-cm high walls (Figure 1a). The maze was made of black plastic with matt surface. Three parallel alleys in the Townmaze center converged on a T-junction at the far end of the maze. This point is referred to as the Decision Point. Each trial started in the Goal chamber where the rats could consume the earned reward and rest for 1 min. Then the rat was led to the left or right Start chamber in a pseudorandom order and left there for 30 s. After a winding entrance alley, the rat had three alternate paths leading to the Decision Point, where the rat had to decide between a left and a right turn. If the choice was correct, the rat would find its way to the Goal and be rewarded with 4 minipellets (20 mg), while a wrong choice led to a blind alley, which forced the rat to backtrack to the Decision Point and choose the opposite lateral alley to reach the Goal. In this case, only one minipellet was awarded. Only fecal bullets or urine puddles were cleaned during the session, but between the sessions, the maze was cleaned carefully with 70% ethanol to remove all scent marks.

Two task rules were applied, which both required the rat to remember its starting location when it came to the Decision Point and make the choice accordingly. All rats were first trained in a Nonmatch task. If the trial started in the left Start chamber, the correct response was to turn right, while when the trial began from the right Start chamber, a left turn was required (Figure 2). The course of a typical Nonmatch task session is given in Table 1. The animals were first trained to perform the task before the electrode implantation to a criterion of 80% correct. After an initial drop in the performance close to chance level during early recordings, the rats gradually improved and reached a performance level up to 90% correct. To add difficulty to the task, occasionally (in 20% of trials on average), one or two of the paths in the maze mid-section were blocked with an opaque barrier (height 20 cm) to force the rat to take a nonpreferred route. These trials are later referred to as barrier trials in contrast to free trials. Two of the rats still attained a performance level where they rarely made mistakes. It has been demonstrated that upon repetition rats tend to switch from hippocampal-dependent spatial navigation to striatal-dependent response strategy in an attempt to solve a two-choice maze task (Packard & McGaugh, 1996). To encourage the rats' use of hippocampal spatial strategy, we introduced a second, Position task. Here the rat had to ignore its start location and repeatedly choose either left or right (predetermined by the experimenter) lateral alley (Figure 2). The course of a typical Position task session is given in Table 2. Once the rats had learned this new rule, we introduced yet another modification for the last recording sessions. Now, in the middle of the session the rule was unexpectedly changed, so that the opposite lateral alley now led to the Goal. In the Position task, the trial introducing a new rule was not counted as correct or error, only the following trials.

Independent of the task rule, the rats showed a strong preference to choose either the leftmost or middle path when starting from the left, and correspondingly, the rightmost of the middle path when staring from the right. Further, this preference could vary between days. However, as expected, the rat never spontaneously chose the longest path through the opposite third middle alley. When using the barriers, we first ran at least three trials without any manipulations to identify the preferred path of a rat on that particular day.

2.3 Electrophysiology

Microdrive arrays carrying 18 independently adjustable gold-plated tetrodes aimed at area CA1 of the right dorsal hippocampus (2.4 mm lateral and 4.0 mm posterior, relative to bregma) were implanted under general anesthesia (induction: ketamine 50 mg/kg i.p. + xylazine 6 mg/kg i.p + isoflurane 3%; maintenance: isoflurane 1–2%). Postoperatively the rat received ketoprofen 5 mg/kg i.p. for pain relief. Tetrodes (Nguyen et al., 2009) and microdrives (Kloosterman et al., 2009) were constructed as described. Each tetrode consisted of a twisted bundle of four polyimide-insulated microwires (13 µm in diameter, RediOhm-800, Kanthal, Palm Coast, FL), fused and cut to create a blunt tip. Electrodes were slowly lowered into the CA1 pyramidal cell layer over the course of 1–2 weeks. Differential recordings of extracellular action potentials (sampled at 31.25 kHz per channel, filtered between 600 Hz and 6 kHz) and continuous LFP (sampled at 2 kHz per channel, filtered between 0.1 and 475 Hz) were made against a local reference electrode in overlying white matter, using patch boxes (Cheetah system) and amplifiers by Neuralynx (Boseman, MT). A screw in the skull overlying cerebellum served as ground. Position was acquired at 30 Hz by an array of diodes located above the electrode drive. Individual units were isolated by manual clustering on peak spike waveform amplitudes and spike widths across all four tetrode channels using custom software (Xclust; M.A.W.). No spike overlap was allowed between selected clusters. Spike stability was verified by plotting the peak amplitude of the tagged cluster in the tetrode with the highest amplitude across the entire 1–2 h recording session. If the amplitude dropped 25% or more during the session, the unit was considered unstable and discarded. Cluster boundaries calculated on session N were applied to session N + 1, and were never found to fully overlap even for a single tetrode, suggesting some movement with regard to the tissue. Therefore, isolated units in each session were considered separate. At the end of the recordings, the tetrode sites were marked by passing direct current through one wire in a tetrode. The rats were deeply anesthetized, perfused transcardially with 4% PFA, and the brains were cut into 50 µm coronal sections. Lesion marks were detected in cresyl violet stained sections. All tetrodes with reported units were confirmed to locate in the CA1 pyramidal cell layer.

2.4 Data analysis

Animal location and head direction were captured at 30 Hz by video tracking of two head-mounted LEDs flickering out of phase, using an overhead camera. The velocity was smoothed with a Gaussian kernel (SD = 0.25 s). Unit data were analyzed only for epochs when the movement velocity was 5 cm/s or higher. To create a firing rate map, the maze was divided into 5 cm x 5 cm bins, and the firing rate per bin was smoothed with a Gaussian filter (2 SD). Well-isolated units with a spike width > 350 ms, maximum firing rate > 1.0 Hz, mean firing rate > 0.05 Hz, and spatial information content (Markus, Barnes, McNaughton, Gladden, & Skaggs, 1994) > 1.0 displaying 1–5 identified place fields were included in the analysis as place cells. Place fields were defined as 5 or more adjacent pixels (shared at least one side with another pixel in the place field) with a firing rate ≥ 3 × unit mean rate. Further, firing rate in the place field should have only one peak. In practice, this meant that long fields in one maze alley or round fields in two adjacent alleys with continuous chain of adjacent pixels above 3 × unit mean firing rate were considered two separate fields if the firing rate distribution demonstrated two peaks. Well-isolated units with a spike width < 350 ms, mean rate ≥ 3.0 Hz displaying theta modulation were included in the analysis as interneurons.

We first calculated the mean number of place fields per pyramidal neurons considering all maze sections. However, neurons were included in this and further analyses only if they had at least one place field outside the Goal and Start compartment where the mice were mainly stationary.

Next, place field responses to barriers were categorically analyzed by subtracting the firing field map during barrier trials from the map during free trials. If the differential map showed no field after applying the “5 or more adjacent pixels with a firing rate ≥ 3 × unit mean rate” rule or showed < 25% per pixel in-field firing rate in the original place, the field was classified as Stable. If ≥ 25% of the original firing rate was present in the differential map, the field response was classified as (firing rate) Decrease. Conversely, if the rate increased by the same amount, the response was classified as (firing rate) Increase. If the in-field firing rate was stable but 20–50% of pixels did not overlap with the original field, the response was classified as field (location) Shift. If the place field in the barrier trials did not meet the “5 or more adjacent pixels with a firing rate ≥ 3 x unit mean rate” criterion any more or showed than 50% with the original field, the response was classified as field Disappearance. In contrast, if the differential map showed a “negative” field according to the “5 or more adjacent pixels with a firing rate ≥ 3 × unit mean rate” in a location with no field during free trials, the response was classified as New field. To avoid the potential confound of differential reward (4 pellets for a correct choice, 1 for incorrect), the Goal and Start chambers were excluded from the field analysis.

To assess responses to barrier manipulation more objectively, we first calculated pixel-by-pixel cross-correlations across free and barrier trial maps using Pearson rho. Then we converted the rho values to z values using Fisher transformation: z = 0.5 × ln((1 + r)/(1 - r)). The z-transformation was also done to the early vs. late trial correlations.

Further, we determined whether introduction of barriers introduced a net change in the unit firing rate. To separate a change due to the manipulation from general drift in the firing rate during the 1–2 h recording session, we calculated a rate score (R) as follows:

The value of R tells whether there was a net decrease or increase in the firing rate between free and barrier trials.

To compare the effect of barriers in their vicinity vs. more distal section of the Townmaze, we divided all alleys in the maze to two roughly equally large zones: a proximal zone and a distal zone (Figure 1). All barrier manipulations took place in the proximal zone. The Goal and Start chambers were not included in the comparison, since those are wide, open spaces that cannot be directly compared with long alleys. The similarity and rate scores were compared between proximal and a distal zones using paired t-tests.

Finally, to assess whether task performance affects the place field responses to the barriers, we split the sessions of each rat in two categories. Performance was consider impaired in a session if the rat made more errors in the presence of barriers than without, while unimpaired sessions included trials with similar or higher accuracy in the presence of barriers than during free trials. The performance accuracy varied between 33% and 100% correct, and there was a roughly equal number of units with acceptable place fields during both kind of trials. Then we applied ANOVA for repeated measures with maze zone and within-unit and task performance and between-unit factor.

3.1 Most place cells have many fields in the Townmaze and code for a position on a route

The place field properties were assessed on 353 well-isolated and stable CA1 pyramidal neurons that met the inclusion criteria for place cells (see Methods). Four rats provided cells for the study, and their individual contribution to the data is summarized in Table 3. On average, the cells displayed 2.44 ± 0.05 (mean ± sem) fields. To further assess the pattern of field distribution, we divided the Townmaze into 11 logical segments separated by choice points, plus the Start and Goal compartments. In addition, the Decision Point (DP) and the DP and Goal corners were considered separate segments, since many fields extended on both sides of the corner (Figure 1b). The fields were unevenly distributed, so that the left and right initial segments (1L, 1R), the left and right middle segments (3L, 3R) and the left lateral segment (5L) were overrepresented, while the corners were underrepresented (Tables 1-3). The fields were almost exclusively confined to a single alley (path) and had a mean length of 34.3 ± 0.7 cm, except on the lateral segments that were about twice as long as the remaining ones. There the mean field length was 64.0 ± 0.7 cm, suggesting scaling to the alley length (Table 4).

Thus, in a large and complex environment, such as the Townmaze, most place cells display more than one well-defined field. In this set, 149 pyramidal neurons had 3–5 fields, which allowed us to investigate larger field constellations than in conventional smaller arenas. The most striking feature of these multiple fields was that they seemed to be located along the most typical paths that the rats used to traverse through the maze. The fields were dispersed along the path like scent marks that a rat would leave to mark its trail in the wild. Figure 3 illustrates some examples or these “path segment” cells. Usually these included one element in the Start compartment or initial alley (1L, 1R), one element in the mid-section (3L, 3M, 3R) and one after the Decision Point (Figure 1b). Although it appears logical to have such a representation of the path, it is essential to rule out the possibility that this can happen only by chance. To evaluate this possibility, we can consider the most typical cases (n = 22 out of 94) of fields representing a single path that is a correct choice in the Nonmatch task during free trials. A correct path from the left Start compartment until the Decision Point would include the following elements: Start compartment, 1L, left Choice Point (CPL), 2L–CPM–3M, if the rat follows the middle-mid path, or 3L directly after the CPL if the rat chooses the lateral path. (The rat never took the longest path if not forced to do so by barriers.) In both cases, the path would continue as DP, 4R, DP corner R, 5R, Goal corner R, 6R, and the Goal. This means either 13 or 11 elements in total (Figure 1b). For simplicity, we can take the average, 12 elements. Correspondingly, a correct path from the right Start would comprise the elements: Start, 1R, CPR, 2R–CPM–3M, or 3R directly from CPR, and further DP, 4L, DP corner L, 5L, Goal corner L, 6L, and the Goal (Figure 1b), again on average 12 elements. The nature of the task (advancement only in one direction, balanced number of left and right starts) ensures practically even occupancy time per segment (except for massive overrepresentation of Start and Goal compartment and 3 middle vs. 2 lateral paths, which bias is balanced by higher running speed in lateral alleys). This is further confirmed by the occupancy maps in Figure 3. Therefore, one can assume an even occupancy for all maze segments for simplicity. Since the total number of path elements in the maze is 24, the likelihood for three fields to be situated in the correct path from the left Start is thus 12/24 for the first field, 11/24 for the second field and 10/24 for the third field, because we consider only one field per element. The likelihood that all three match with the path is their product (p = 0.095). Similarly, the likelihood of having four fields in the correct path starting from left is p = 0.036 and for five fields p = 0.012. The same probabilities apply for trials starting from the right. The combined probability would then be the sum of left and right trials. Thus, the combined probability of three fields to be located on the correct path that begins either from left or right will be 2 × 0.095 = 0.191, for four fields 2 × 0.036 = 0.072, and for five fields 2 × 0.012 = 0.024. As summarized in Table 5, it appears that the observed number of 3 to 5 fields falling on a correct path was above the expected value by chance (X²(2) = 9.3, × < 0.01).

In turn, the theoretical probability of 3–5 fields to fall on the path of an erroneous choice can be calculated as follows. For instance, a path starting from left and ending with an erroneous choice would follow the correct path until the DP. Then it will contain only 3 elements (4L, DPcorL, 5L), since segments 6L and Goal are not included in the erroneous path (Figure 1b). This will leave 10 maze elements on average for trials starting from left and the same number for trials starting from right. The likelihood for three fields to be situated in the erroneous path from the left Start is thus 10/24 for the 1st field and 6/24 for the 5th field. The combined probability for three fields to be located on the erroneous path that begins either from left or right will be 2 × 0.052 = 0.104. The corresponding probabilities for four fields will be 2 × 0.015 = 0.030 and for five fields 2 × 0.004 = 0.008. In contrast to correct paths, the observed number of multiple fields that matched with the erroneous choice (n = 3 out of 94) was not different from the chance level (X²(2) = 2.1, p > 0.10; Table 6). In addition to fields that were compatible with either a correct or erroneous choice, 8 cells had a field constellation that matched with a single path up to the Decision Point, but could not be classified as either correct or erroneous paths with not fields representing the choice of the rat. The remaining 75 neurons with multiple fields had a field constellation that would only match with representation of two paths at the same time, either trials starting from both left and right or trials starting from one position but with both correct and erroneous choices.

A plausible hypothesis for the development of path-segment fields is that through hundreds of repeating passages of the rat through the maze following the same path, the route becomes represented even at the level of a single CA1 pyramidal neuron through associative learning. One way to test this hypothesis is to compare the frequency of path-segment cells during the overlearned Nonmatch task to the Position task while the rat was still learning the new rule. In total, we found 55 neurons with 3 to 5 fields during the Position task. The theoretical probability of 3 to 5 fields to fall on the path of a correct choice in this task is the same as in the Nonmatch task, since the path lengths remain the same. As summarized in Table 7, it appears that 10 neurons had 3 to 4 fields falling on a correct path (none had 5 fields), which is not different from chance distribution (X²(2) = 1.1, p > 0.10). Notably, 5 of these neurons represent the same route as in the Nonmatch task (e.g., left Start–right choice in the Go Right task) and only 5 the newly learned route. In addition, 7 cells had a field constellation that matched with a single path up to the Decision Point (again common between the two tasks). Only three neurons had three fields that would correspond specifically to an erroneous choice. These data lend tentative support to the contention that the path-segment feature of single place cells is a result of sequence learning.

3.2 Some place cells with multiple fields code for general geometric features

A second conspicuous constellation of multiple place fields were “axis-aligned” fields. These were all located either on the longitudinal or transverse alleys of the Townmaze. Unfortunately, the nature of the memory task made the rats to run through most of the maze segments in only one direction. Therefore, we could not systematically test the direction specificity of these fields. Nevertheless, in several cases the aligned fields of a single cell fell on segments where the movement direction was the opposite, speaking against head-direction like coding. Figure 4 illustrates some clear examples among 32 cells of this type. If we assume that a field ∼30 cm in length can locate in three different ways—straight longitudinally, straight transversally or in an angle - there are 11 possible longitudinal segments (two counted in the long lateral alleys and two parallel in the initial segment = 1La, 1Lc, 1Ra, 1Rc, 3L, 3M, 3R, 5La, 5Lb, 5Ra, 5Rb), 10 possible transverse segments (1Lb, 1Rb, 2L, 2R, 3Lb, 3Rb, 4L, 4R, 6L, 6R) and 10 corners (1Lab, 1Lbc, 1Rab, 1Rbc, 3Lab, 3Rab, DPcorL, DPcorR, GcorL, GcorR) in the maze (Figure 1b). The likelihood of having three fields all in a longitudinal segment is thus (11/31) × (10/31) × (9/31) = 0.033 and for having them in a transverse segment correspondingly (10/31) × (9/31) × (8/31) = 0.024. As summarized in Tables 8 and 9 the observed number of either longitudinally (X²(2) = 262.0, p < 0.001) or transversally (X²(2) = 356.1, p < 0.001) aligned fields was way above the chance level.

A third notable pattern of place field arrangement were “symmetric fields” that were located symmetrically on both left and right halves of the maze. Considering the left vs. right emphasis of the applied tasks, one would expect a substantial portion of cells to express such symmetry. In total, 26 pyramidal neurons out of 353 displayed at least two symmetric fields, while only two neurons had two pairs of symmetric fields. Assuming that we have the above mentioned 11 longitudinal, 10 transverse, and 10 corner segments (31 in total) in the maze, the likelihood of having two fields in symmetric segments is (31/31) × (1/31) = 0.032, and the likelihood of having two pairs of symmetric field correspondingly (31/31) × (1/31) × (29/31) × (1/31) = 0.001. Considering that 276 out of 353 analyzed pyramidal cells had two or more fields, the observed 26 cells with symmetric pairs is twice the number expected by chance (0.032 × 276 = 8.8) and the observed two cells with two symmetric field pairs 6-fold higher than by chance (0.001 × 276 = 0.3). This distribution is also significantly above chance (X²(1) = 44.1, p < 0.001).

The presence of the aforementioned three types of field constellations suggests that the CA1 place cells encode both task-relevant features and geometric aspects of the environment.

3.3 Place field responses to barriers

To assess changes in the place fields upon introduction of the barriers, we compared the place field map of all free trials to that of all barrier trials for each cell. First, to get a qualitative estimate of these changes, we divided the responses (i.e., comparison free map—barrier map) in the following categories (see Methods for definitions). A field could remain stable both in terms of firing rate and location (Stable), show a pure firing rate change (Decrease, Increase) without a change in location, or display a shift of the peak firing or field expansion with no overall change in rate (Shift). In addition, a field present during free trials could disappear (Dis) during barrier trials, or conversely, a field appears only during barrier trials (New). Figure 5 illustrates typical examples of these place field changes. One neuron with two fields shows Decrease in both fields (Figure 5a), one neuron with a single field shows Disappearance of the field together with a New field (Figure 5b), while one neurons with two fields shows Decrease of one field, Increase of a second field, and in addition, a New field (Figure 5c). As illustrated in Figures 5a and 6c, neurons with multiple fields could show discordant responses to the barriers, i.e. individual fields respond in a different way. However, this was not common, since among 211 neurons with two or more place fields, 70.9 ± 1.7% (mean ± sem) of the fields responded the same way as the majority of fields.

Notably, the barrier trials were intermingled between the free trials (Table 1), so that a simple drift in firing properties of the recorded cell across time should not lead to major differences between the two maps. On the other hand, it is possible that the introduction of barriers may alter the maze representation during subsequent trials without barriers. To compare the magnitude of barrier-induced alterations in the field properties vs. changes in the firing rate during the long recording session, we divided the free trials between early and late ones, and compared the changes between free and barrier trials to those between early and late trials without barriers. Notably, this analysis could not be done on sessions where two rules were applied, like Go Left in early trials vs. Go Right in late trials. Typically, the fields remained stable across all free trials and changes were only observed during the barrier trials, as illustrated in Figure 6a. Here the neuron has two place fields, which remain stable across all free trials. However, during the barrier trial, one field remains in its original location, one field disappears while a new field appears. In addition, we sometimes also found a response type that has not been described in barrier manipulations before: a field present only during free trials substantially intensified after the barrier manipulation was introduced (Figure 6b), or in the extreme case, the field was present only during the latter half of the free trials (Figure 6c). Notably, in both cases, the response was specific to the free trials, since the corresponding field location did not show any firing during the barrier trials. Of a total of 368 place fields included in this analysis, 8 (2%) fields showed specific firing during late free trials. In the analysis of field responses to barriers, these kind of fields were categorized as Decrease or Disappear type, using the late free trials as a comparison. Table 10 summarizes the extent of field changes between free vs. barrier trials and those between early free and late free trials. Most notably, the percentage of stable fields was significantly higher between the early and late free trials than between free and barrier trials, whereas firing rate decrease was significantly more often associated with the presence of barriers.

Recorded interneurons also showed firing rate changes in response to barrier manipulations. The response could be a chance in the firing rate all over the maze (Figure 7a) or a more local change in one or several maze sections (Figure 7b).

3.4 Barrier-induced new fields are largely limited to the proximal zone while firing rate changes are zone independent

To assess whether the introduction of barriers induces a global remapping of the Townmaze or more local changes near the barriers, we divided the maze into a proximal and a distal zone of approximately equal size (Figure 1a). These sections contain all the maze alleys, while the open spaces (Goal and Start compartments) were not included. Figure 8a summarizes the field changes in both zones. Overall, only 1/3 of the fields remained stable during the barrier trials. Moreover, even in the distal zone, more than half showed changes. The percentage of fields with rate changes did not differ between the distal and proximal zones, but the proportion of new fields was significantly higher in the proximal zone, while correspondingly, the proportion of stable fields was higher in the distal zone (Figure 8a). Further, in the comparison between free vs. barrier trials and early free vs. late free trials limited to the Nonmatch task, the proportion of new fields was significantly dependent on the zone while other field changes were not (Table 10).

Of the total 56 interneurons, 29 decreased and 8 increased their firing rate by more than 25% in the proximal zone during the barrier trials compared to free trials. The corresponding numbers in the distal zone were 34 and 8, respectively. In contrast to the pyramidal neurons, the responses of interneurons were largely zone independent, since the correlation between firing rate changes between these two zones was 0.96 (Spearman's rho).

3.5 Fields with decreased firing rate are more common during trials with impaired performance

Next, we wanted to determine how place field responses to barriers correlated with the overall task performance of the rats. The choice accuracy varied greatly (33%–100%) between sessions due to task modifications and daily fluctuation in performance. Further, in about half of the sessions, the choice accuracy dropped due to the barrier manipulation, but during some days, the rats performed even better during the barrier trials than free trials. Thus, we divided the sessions roughly into those with impaired performance due to barriers and those without such impairment. We composed an ANOVA model with impairment (present vs. absent) as between-session factor and the maze zone (proximal vs. distal) as a within-session factor, and compared the proportion of each place field response types to the barriers. Independent of the choice accuracy, New field responses were far more common in the proximal zone than in the distal zone (F_1,120 = 18.4, p < 0.001; Figure 8), while Stable responses were less common in the proximal zone (F_1,123 = 6.9, p = 0.010; Figure 8). In contrast, the choice accuracy significantly affected the relative proportion of Stable vs. Decrease responses. Independent of the maze zone, Stable responses were far more common during maintained choice accuracy (F_1,123 = 9.1, p = 0.003; Figure 8b and c), whereas Decreased field rates were more common during impaired task performance (F_1,120 = 6.5, p = 0.012; Figure 8b and c). It is not that surprising that stable fields, both in terms of location and firing rate, are associated with good task performance. However, a new finding is that appearance of new place fields (global remapping) is independent of task performance, while decrease in firing rate (rate remapping) associated with the task performance. We also assessed if concordant field responses would associate with better task performance and discordant field responses. However, the concordance was not influenced by the choice accuracy (t₂₀₉ = 0.71, p = 0.48).

3.6 Global measures of field location stability are influenced by the proximity of barriers while overall firing rate reflects performance accuracy

To compare the field responses to barriers between the proximal and distal zones and their correlation with task performance in a more objective way, we calculated z-transformed Pearson correlation coefficients between free and barrier trials separately for proximal and distal maze zones. Further, to assess barrier-induced changes across time (Figure 6) we similarly calculated z-transformed Pearson rhos between early and late free trials. Finally, we composed the ANOVA model described above with impairment as the between-session and maze zone as the within-session factor.

The maze zone proved a highly significant factor for the similarity between free and barrier trials: in both pyramidal neurons (F_1,321 = 52.3, p < 0.001) and interneurons (F_1,54 = 36.3, p < 0.001), the z-score was highly significantly lower in the proximal than distal zone (Figure 9a). In contrast, the choice accuracy did not influence the similarity score in either pyramidal neurons (p = 0.59) or interneurons (p = 0.75). The same pattern was observed when we compared early and late free trials but only with a marginal significance: the z-score was lower in the proximal than distal zone in both pyramidal neurons (F_1,298 = 5.5, p = 0.020) and interneurons (F_1,54 = 4.1, p = 0.047; Figure 9b).

In line with the field analysis, there was an impairment × zone interaction in the firing rate of pyramidal neurons (F_1,338 = 7.0, p = 0.008), such that during poor task performance the firing rate in the proximal zone tended to increase but decrease in the distal zone. In addition, firing rate increase was more robust in the proximal zone (F_1,338 = 12.3, p = 0.001). No such effects were observed in the interneurons (Figure 9c). Also in the comparison between early and late free trials, the zone proved to a significant factor of pyramidal neuron firing: firing rate increase was more robust in the proximal zone (F_1,326 = 17.1, p < 0.001) than in the distal zone, while the impairment factor and impairment x zone interaction were nonsignificant (p = 0.35 vs. p = 0.19). In contrast, the changes in interneuron firing rate between early and late trials were highly dependent on the task performance: the firing rate decrease was significantly smaller during poor task performance (F_1,338 = 7.2, p = 0.009; Figure 9d). Overall, there were no differences in the interneuron firing between the zones.

Taken together, both the field and global analyses revealed a common pattern that robust field changes (appearance of new field and disappearance of old ones) during barrier manipulations occur mostly in the zone proximal to the barriers. However, these changes do not correlate with task performance, whereas changes in overall firing rate or both pyramidal neurons and interneurons do correlate.