Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2016 Sep 2.
Published in final edited form as: Neuron. 2015 Aug 19;87(5):1093–1105. doi: 10.1016/j.neuron.2015.07.012

Neural population evidence of functional heterogeneity along the CA3 transverse axis: Pattern completion vs. pattern separation

Heekyung Lee 1,*, Cheng Wang 1,*, Sachin S Deshmukh 1,3, James J Knierim 1,2
PMCID: PMC4548827  NIHMSID: NIHMS710552  PMID: 26298276

Summary

Classical theories of associative memory model CA3 as a homogeneous attractor network because of its strong recurrent circuitry. However, anatomical gradients suggest a functional diversity along the CA3 transverse axis. We examined the neural population coherence along this axis, when the local and global spatial reference frames were put in conflict with each other. Proximal CA3 (near the dentate gyrus), where the recurrent collaterals are the weakest, showed degraded representations, similar to the pattern separation shown by the dentate gyrus. Distal CA3 (near CA2), where the recurrent collaterals are the strongest, maintained coherent representations in the conflict situation, resembling the classic attractor network system. CA2 also maintained coherent representations. This dissociation between proximal and distal CA3 provides strong evidence that the recurrent collateral system underlies the associative network functions of CA3, with a separate role of proximal CA3 in pattern separation.

Introduction

The hippocampus is viewed as an associative memory system supporting the formation, storage, and retrieval of memories. Computational theories suggest that in order to minimize interference and maximize the storage and recall of memories, the associative memory network performs two competing processes: pattern separation and pattern completion (Guzowski et al., 2004; Marr, 1971; McClelland and Goddard, 1996; McNaughton and Morris, 1987; Rolls and Kesner, 2006; Rolls and Treves, 1998). The dentate gyrus (DG) is modeled as a processing stage for pattern separation, the ability of the network to orthogonalize overlapping input patterns before they are stored (McNaughton and Nadel, 1990). In contrast, CA3, due to its extensive network of recurrent collaterals, is modeled as a processing stage for pattern completion, the ability of the network to retrieve stored output patterns when presented with partial or degraded input patterns (Marr, 1971; McClelland and Goddard, 1996; McNaughton and Morris, 1987; Treves and Rolls, 1992; Hasselmo et al., 1995). Numerous studies have generated support for these proposed functions of the DG and CA3, but most of them were either indirect behavioral studies or physiological studies that did not directly measure the input-output correlations that are required for a true test of these hypotheses (Santoro, 2013; Yassa and Stark, 2011).

Recently, direct physiological evidence of pattern separation in the DG and pattern completion/generalization in CA3 was shown in a study in which CA3 produced an output pattern closer to the originally stored representation when presented with degraded input patterns from the DG (Neunuebel and Knierim, 2014). In that study, the CA3 recordings were confined to the intermediate and distal regions of CA3 (i.e., the regions of CA3 along the hippocampal transverse axis that are outside the blades of the DG; Figure 1A). There are a number of anatomical gradients along this axis that are likely to have important functional consequences. First, the proximal CA3 neurons receive mossy fiber inputs from both the infrapyramidal and the suprapyramidal blades of the DG; the intermediate and distal CA3 neurons receive mossy fiber inputs from the suprapyramidal blade only (Claiborne et al., 1986; Witter, 2007). Second, the proximal CA3 neurons, particularly those constrained by the blades of the DG, have weak perforant path inputs from the entorhinal cortex (EC) (Ishizuka et al., 1995; Ishizuka et al., 1990; Witter, 2007). Third, the proximal CA3 neurons do not contribute recurrent collaterals to intermediate and distal CA3, but rather project back to the hilus and themselves; the intermediate and distal CA3 neurons provide recurrent collaterals to the entire extent of the transverse axis, but the projections to proximal CA3 are the weakest (Ishizuka et al., 1990; Li et al., 1994; Witter, 2007).

Figure 1.

Figure 1

Open in a new tab

CA3 circuitry and experimental paradigm. (A) Schematic of the intrinsic and extrinsic inputs to CA3 along the transverse axis. Solid brown arrows indicate the entorhinal inputs, solid purple arrows indicate the DG inputs, and the dashed arrows indicate the recurrent collaterals in proximal (blue), intermediate (red), and distal (green) CA3. (B) Recording sessions consisted of three standard (STD) sessions interleaved with two mismatch (MIS) sessions. In the STD sessions, the local cues of the track (denoted by the inner ring of 4 textures) and the global cues along the curtain at the periphery (denoted by the black outer ring) were arranged in the familiar configuration that the rat had experienced in all preceding training trials. In the MIS sessions, changes to the sensory inputs were produced by rotating the global cues clockwise and the local cues counterclockwise by the same amount, for a net cue mismatch of 45°, 90°, 135°, or 180°. In this example, session 2 is a 180° mismatch and session 4 is a 45° mismatch. Over the course of 4 recording days, the rats experienced each mismatch amount twice. (C) Tetrodes were distributed across CA3 such that the recording positions covered the entire transverse axis of CA3.

A strong prediction of the computational theories of CA3 is that the network properties that result from the recurrent collateral system (e.g., pattern completion, error correction, and generalization) should be stronger in the distal part of CA3, with the strongest recurrent connections, than in the proximal part of CA3, with a much weaker recurrent collateral system. We tested this critical prediction by analyzing the population responses to a local-global cue mismatch (double rotation) manipulation used previously to show pattern completion/generalization properties in distal CA3 (Knierim, 2002; Lee et al., 2004b; Neunuebel and Knierim, 2014). We show that the population responses to the corrupted inputs were much more coherent in intermediate and distal CA3 than in proximal CA3, which instead resembled strongly prior recordings from the DG (Neunuebel and Knierim, 2014).

Results

Histology and experimental paradigm

Multielectrode recordings were made from neurons along the dorsal CA3 transverse axis from 22 rats as they ran clockwise (CW) on a circular track in a cue-controlled environment (Figure 1B). Four different local textures tiled the surface of the track, and multiple global cues were placed along the walls of the room (Knierim, 2002). In alternating recording sessions, the local and global spatial frameworks were placed in conflict by rotating the track and the global cues in opposite directions. Across animals, recording sites covered the entire transverse axis of CA3 (Figure 1C; Figure S1). Data from a subset of the rats were reported in prior publications (Lee et al., 2004b; Roth et al., 2012; Neunuebel and Knierim, 2014), but these data were not analyzed in terms of the specific hypotheses of the current report.

Individual cells along CA3 transverse axis

There was a gradient in firing rate and spatial tuning along the CA3 transverse axis. The mean firing rates of CA3 cells increased from proximal to distal CA3 (Figure 2A: r = 0.32, p = 0.004), but there was no significant relationship between the peak firing rates and location along the transverse axis (Figure 2B: r = 0.14, p = 0.24). These results suggest that place cells along the entire transverse axis were similar in their firing rates at the center of their place fields (peak rates), but that the place fields were larger in the more distal parts of CA3 than the more proximal parts, resulting in a larger mean firing rate. In support of this notion, the spatial information score decreased from proximal to distal CA3 (Figure 2C: r = −0.43, p = 0.0001) while the place field size increased from proximal to distal CA3 (Figure 2D: r = 0.46, p < 0.0001). When the spikes of the distal CA3 cells were down-sampled to match the mean firing rates of the proximal CA3 cells, the spatial information scores of the distal CA3 cells (median 1.58, interquartile range 1.36 – 1.83) were still significantly lower than the proximal CA3 cells (median 2.32, interquartile range 1.84 – 2.60; Wilcoxon Rank Sum test: z = 3.90, p < 0.0001; see Supplemental Experimental Procedures).

Figure 2.

Figure 2

Open in a new tab

Spatial properties of CA3 principal neurons are nonuniformly represented along the transverse axis. Each point of the scatter plot represents a tetrode recorded in a location along the CA3 transverse axis, starting at proximal CA3 (adjacent to the DG) and extending to distal CA3 (adjacent to CA2). The average value of all the cells recorded across all the sessions and days for each tetrode was plotted. Well-isolated, active principal cells that passed the inclusion criteria (> 20 spikes on the circular track during behavior, spatial information score > 0.5 with significance p < 0.01, and mean firing rate < 10 Hz) were included in the analyses. From proximal to distal CA3, (A) mean firing rates increased (r = 0.32, p = 0.004), (B) peak firing rates were not different (r = 0.14, p = 0.24), (C) spatial information scores decreased (r = −0.43, p = 0.0001), and (D) place field size increased (r = 0.46, p < 0.0001). n = 74, r, Pearson’s linear correlation coefficient.

The individual cell responses to the double rotation manipulations were categorized into 5 groups, similar to previously published reports using the same task (Lee et al., 2004b; Neunuebel and Knierim, 2014; Neunuebel et al., 2013) (Figure 3A). The cell classifications were based on responses to each cue-mismatch (MIS) session compared to the cells’ firing properties in the immediately preceding standard (STD) session. Cells were classified as “Remap” if they met the place-field inclusion criteria in only the MIS session (i.e., place fields APPEARED) or the STD session (i.e., place fields DISAPPEARED). Cells were classified as “Rotate” if the cells met the inclusion criteria in both the STD and the MIS sessions. To determine whether these place fields rotated CCW (to follow the local cues), rotated CW (to follow the global cues), or had an ambiguous (AMB) relationship to the cues, the STD rate map was correlated with the MIS rate map multiple times as the MIS rate map was rotated relative to the STD. The rotation angle producing the greatest correlation was considered the rotation angle of the place field (see Experimental Procedures). Because these classifications often make arbitrary distinctions among cell responses that fall along a continuum, and because we made no attempt to exclude cells recorded over multiple sessions, we did not perform statistical analyses on these categories. Nonetheless, these classifications provide a useful description of the types of single-unit responses that underlie the quantitative, population analyses presented below.

Figure 3.

Figure 3

Open in a new tab

Differences along the transverse axis in remapping vs. rotation of place fields. (A) Examples of spatial firing rate maps showing 5 different response types observed in CA3. Categories included counterclockwise (CCW) rotation with the local cues, clockwise (CW) rotation with the global cues, ambiguous (AMB) rotation responses, APPEAR in the MIS sessions, and DISAPPEAR in the MIS sessions. The rotation correlation analysis between STD1 and MIS session (red line) is shown to the right of the rate maps. These graphs show the correlations between STD and MIS rate maps as the MIS rate map is rotated in 1° increments rel ative to the STD rate map. Peak correlations above a threshold of 0.6 (Neunuebel and Knierim, 2014) located in the dark or light grey box indicated that the fields rotated CW or CCW, respectively. Peak correlations below 0.6 were considered ambiguous responses. Examples of ambiguous responses could be cells that split their place fields in two (as shown here) or cells that markedly changed their place field sizes or specificity between the STD and MIS sessions. In the rate maps, blue indicates 0 firing and red indicates the maximum firing rate for that cell. (B) Categorical classification of responses in different CA3 subregions. “Remap” comprises APPEAR and DISAPPEAR cells; “Rotate” comprises CCW, CW, and AMB cells. A higher proportion of cells in distal and intermediate CA3 rotate with the cues than remap, whereas a higher proportion of cells in proximal CA3 remap than rotate. Of the cells that rotate, a higher proportion of cells rotate with the local cues (CCW) in all the CA3 subregions, although in different proportions.

There were notable differences across the CA3 subregions in the proportion of the response types (Figure 3B). We divided CA3 into 3 regions, such that the numbers of tetrodes sampling each region were approximately equal: the proximal region was defined as 0 – ~40%, the intermediate region as ~40% – ~60%, and the distal region as ~60% −100% along the CA3 transverse axis. This partition closely matched the classic partition of CA3 into a, b, and c subregions (corresponding to distal, intermediate, and proximal subregions, respectively) (Lorente de No, 1934) (see Supplemental Experimental Procedures). The majority (65%) of cells in proximal CA3 remapped (n = 136/210), compared to 43% of cells in intermediate CA3 (n = 140/325) and 42% of cells in distal CA3 (n = 141/335). The higher proportion of remapping responses indicates that the proximal CA3 cells are more likely than the distal CA3 cells to create different representations of the STD and MIS environments (i.e., they perform pattern separation). In contrast, the higher proportion of rotating responses suggests that the intermediate and the distal CA3 cells are more likely than the proximal CA3 cells to maintain the same representation of the STD and MIS environments. The higher proportion of “Remap” cells in proximal CA3 was not due to their lower mean firing rates (Figure S2). These trends held up when we considered separately the experiments in which we recorded, within a region, ensembles ≥ 4 cells that had place fields in either the standard or mismatch session. We calculated the ratio of rotating responses to the overall number for each ensemble. The median ratio for proximal CA3 was less than the medians for intermediate and distal CA3, in agreement with the pooled data in Fig. 3B (proximal: n = 19 ensembles, median = 0.14, interquartile range (IQR) = 0–0.42, range = 0–0.71; intermediate: n = 19 ensembles, median = 0.5, IQR = 0.37–0.71, range = 0.18–1.0; distal: n = 25 ensembles, median = 0.5, IQR = 0.24–0.60, range = 0.12–1.0).

Comparing the subset of cells that rotate CCW or CW, the intermediate and the distal CA3 cells were 4 times more likely to rotate CCW with the local cues than CW with the global cues (Figure 3B), in agreement with two prior reports showing that CA3 displayed a mostly coherent response to the local-global reference frame conflict (Lee et al., 2004b; Neunuebel and Knierim, 2014). In contrast, the proximal CA3 cells were only 1.5 times more likely to rotate with the local than with the global cues. Although there was variability across rats, especially those with small numbers of recorded cells, these patterns were apparent across individuals (Table S1). In combination with the higher proportion of remapping cells in proximal CA3, these data suggest a gradient along the proximal-distal axis from pattern separation in proximal CA3 to pattern completion/generalization/error correction in distal CA3.

Population responses along the CA3 transverse axis

To quantitatively analyze the population responses to the double rotation manipulations, spatial correlation matrices from the population firing rate vectors at each location on the circular track were created (Figure S3A) (Lee et al., 2004b; Neunuebel and Knierim, 2014). For all STD1 × STD2 matrices (columns 1, 3, and 5), there was a strong band of high correlation on the main diagonal for all the CA3 subregions (Figure 4). This result indicates that regardless of the magnitude of the local-global cue mismatch in the intervening MIS session, the representations in the 2 STD sessions were spatially stable. In the STD1 × MIS matrices, bands of high correlation were maintained in intermediate and distal CA3 (columns 4 and 6). The correlation bands were shifted increasingly downward from the main diagonal (dashed line) with increasing mismatch amounts, indicating that the representations were controlled coherently by the local cues in the MIS sessions. However, in proximal CA3, the high correlation band was maintained strongly only in the 45° MIS session; in the 90°, 135°, and 180° MIS sessions, the matrices showed less str ucture in the diagonal bands, indicating that the representations between the STD and MIS sessions were poorly correlated (column 2).

Figure 4.

Figure 4

Open in a new tab

Proximal CA3 shows decorrelated representations between the standard and mismatch sessions while intermediate and distal CA3 show correlated representations. The firing rate of each cell was calculated for each 1° bin on the track and then normalized to its peak rate. The firing rate maps of all n cells in the sample were stacked to create a 360 × n matrix, in which each column of the matrix represents the population firing rate vector for a specific angle of the track (Fig. S3A). The firing rate vectors at each angle of a STD session (STD1) were correlated with (a) the firing rate vectors at each angle of the next STD session (STD2) to create a STD1 × STD2 correlation matrix and (b) the next MIS session to create a STD1 × MIS correlation matrix. The population representation maintained coherence in intermediate and distal CA3 in all mismatch sessions (columns 4 & 6, respectively), indicated by the bands of high correlation. The representation in proximal CA3 was coherent in the 45° MIS session but degraded in MIS sessions > 45° (column 2), indicated by the dispersed and weak bands of correlation. In these normalized matrices, the blue color indicates 0 and the red color indicates 1.

To analyze statistically the differences between the regions in the strength of the correlation bands, for each correlation matrix we calculated the mean correlation of the bins along each of the 360 diagonals of the matrix and recorded the maximum mean correlation (called the Peak Correlation; Figure S3B). We were primarily interested in comparing the STD1 × MIS matrices across CA3 subregions. However, inspection of the STD1 × STD2 matrices indicated that there may be differences among these regions in the reproducibility of the spatial firing patterns even in the standard sessions; thus, any discrepancies between the subregions in the MIS sessions might be due to these inherent variations rather than to distinct responses to the double rotation manipulation specifically. To control for these differences, we created for each mismatch angle and each subregion a Peak Correlation Difference Index (PCDI), defined as (Peak CorrelationSTD − Peak CorrelationMIS) / (Peak CorrelationSTD + Peak CorrelationMIS). This measure reflected the magnitude by which the peak correlation band in the STD1 × STD2 matrix was greater than the peak correlation band in the STD1 × MIS matrix. To directly compare the subregions, we then subtracted the PCDI of one subregion from that of the others (Figure 5A). The statistical significance of these measures was calculated by a shuffling procedure in which each cell was randomly reassigned to one of the regions under comparison and the PCDI difference was calculated from this shuffled data set. This procedure was repeated 1000 times to create a simulated distribution based on the random shuffling of the data. Three findings are apparent. First, when the mismatch angle was small (45°, top row), there were no significant differences among the 3 CA3 subregions (i.e., in all 3 pairwise comparisons, the real data point (thick, black line) was within the 95% confidence interval of the simulated distributions). Second, there were no significant differences between intermediate and distal CA3 (third column) for any of the 4 mismatch angles. Third, significant differences between proximal and intermediate CA3 and between proximal and distal CA3 were shown in the 90°, 135°, and 180 ° mismatch sessions (first and second columns, rows 2–4).

Figure 5.

Figure 5

Open in a new tab

Shuffling analyses were performed to statistically quantify the population correlation differences between the CA3 subregions. The Peak Correlation Difference Index (PCDI) and the Euclidean distance of the peak correlation bands were calculated for each CA3 subregion. Comparisons between the subregions were then calculated for each measure as proximal CA3 - distal CA3, proximal CA3 - intermediate CA3, and intermediate CA3 - distal CA3. The experimental value (thick, black line) was compared to the distribution produced by 1000 random shufflings of the data. The PCDI difference (A) and the Euclidean distance difference (B) were not significantly different between the subregions in the 45° MIS session, but proximal CA3 was significantly different from intermediate and distal CA3 in MIS sessions > 45°. Because the 12 distributions of each measure are not independent of each other, standard multiple comparison tests (even modified tests such as Holm-Bonferroni or false discovery rate) are not appropriate (see Supplemental Experimental Procedures). With Holm-Bonferroni correction, only the 180° mismatch sessions would b e considered statistically significant. However, no particular comparison is the critical test for these data. Rather, the important results lie in the pattern of low p values for all Prox - Dist and Prox - Int comparisons for mismatch angles > 45° and the high p values for all Int – Dist comparisons.

As a second measure to corroborate these findings, we calculated the Euclidean distance between the peak correlation bands of the STD1 × STD2 and the STD1 × MIS matrices to compare the similarity of the correlations along those diagonal bands. The results of this measure (Figure 5B) are almost identical to the PCDI difference analysis (Figure 5A), although the Euclidean distance measure may be more sensitive to picking up small (but not statistically significant) differences in the 45° MIS sessions. To determine whether the results were affected by any potential noise introduced by small bin sizes (1°), we also calculated the PCDI and Euc lidean distance measures on correlation matrices with 5° bins, and the results were consistent with the smaller bin sizes (data not shown).

Individual rotation amounts

To determine if the local cue controlled patterns in the population matrices were observed in the firing properties of the individual cells, we examined the coherence of cue control on the subset of individual cells that met the inclusion criteria in both the STD and the MIS sessions (i.e., the “Rotating” cells) (Fig. 6). The mean vectors (black lines) for the distal and the intermediate CA3 cells were significant for all the mismatch angles, indicating significant clustering of the cell responses at an angle corresponding to the local cue. In contrast, for the proximal CA3 cells, only the 45° MIS session produced a significant mean vector. This negative result must be viewed with caution, however, as the p values are close to significance for the larger MIS sessions. Because the majority of the proximal CA3 cells remapped rather than rotated their place fields (Figure 3B), there were only a small number of cells in this analysis for each mismatch angle. Thus, it is likely that the lack of significant mean vectors in proximal CA3 arose from a lack of statistical power due to the stronger remapping (i.e., pattern separation) in this region compared to intermediate and distal CA3. The minority of cells that did not remap, however, may be more likely to be controlled by the local cues than the global cues, but with a lower probability than the intermediate and the distal CA3 cells (Figure 3B).

Figure 6.

Figure 6

Open in a new tab

Individual cell rotation amounts show strong local preference in intermediate and distal CA3. Each dot indicates the amount of rotation of a single place field between the standard and mismatch sessions. The line at the center of the polar plots denotes the mean vector; the black and grey tick marks corresponds to the local and global cue rotations, respectively. The mean vector lengths for intermediate and distal CA3 were significant for all mismatch angles, whereas the mean vector length for proximal CA3 was significant only in the 45° MIS se ssion (Rayleigh test, see below). There was no significant clustering for MIS session > 45° (90°: n = 19, z = 2.41, p = 0.09; 135°: n = 12, z = 2.85, p = 0.05; and 180°: n = 15, z = 2.78, p = 0.06), although the lack of significance is likely the result of the decreased statistical power due to the large number of cells that remapped in proximal CA3. Regardless of statistical significance, intermediate and distal CA3 followed the local cues for all mismatch angles and proximal CA3 followed the local cues for 3 out of the 4 mismatch angles (45°, 135°, and 180°). ***p < 0.0001. (proximal: 45°: n = 28, z = 1 5.38, p < 0.0001; intermediate: 45°: n = 64, z = 43.11, p < 0.0001; 90°: n = 41, z = 18.45 , p < 0.0001; 135°: n = 44, z = 18.88, p < 0.0001; 180°: n = 36, z = 20.58, p < 0.0001; di stal: 45°: n = 63, z = 43.06, p < 0.0001; 90°: n = 50, z = 21.26, p < 0.0001; 135°: n = 38, z = 17.26, p < 0.0001; 180°: n = 43, z = 24.44, p < 0.0001).

CA2 population coherence

Adjacent to distal CA3 is the CA2 region. Historically, this small area was considered a simple transition zone between CA3 and CA1, with recurrent connectivity and large soma (like CA3) but lacking mossy fiber input (like CA1). Recent work has suggested that CA2 plays unique roles in hippocampal processing (Caruana et al., 2012; Chevaleyre and Siegelbaum, 2010; Hitti and Siegelbaum, 2014; Jones and McHugh, 2011; Mankin et al., 2015). We thus extended our analysis to include 7 tetrodes that recorded cells in CA2 from 6 rats (Figure S4). CA2 maintained a coherent representation of the STD and the MIS environments (Figure 7A). In the MIS sessions, the coherent representation was controlled by the local cues, apparent by the downward shifts in the correlation bands. Classification of individual cell responses to the cue-mismatch manipulations showed a higher proportion of cells that rotate than remap (Figure 7B). If only the subset of experiments in which ensembles ≥ 4 cells with place fields were recorded, the ratio of rotating responses to the overall number of cells was somewhat higher than in proximal and intermediate CA3 (n = 19 ensembles, median = 0.75, IQR = 0.46 – 0.82, range = 0 – 1.0). The degree of rotation in a subset of individual CA2 cells that passed the inclusion criteria in both the STD and the MIS session showed significant clustering of cell responses, in the direction of the local cues (Figure 7C). Statistical comparisons between CA2 and the CA3 subregions showed that the PCDI difference (Figure 7D) and the Euclidean distance (Figure 7E) measures were not different between CA2 and intermediate and distal CA3 in all mismatch angles with the exception of the 180° MIS session (Dist CA3 – CA2). However, CA2 was significantly different from proximal CA3 in the 90° and 180° mismatch angles, and just missed significance at the 135° mismatch angle. Th ese results demonstrate that CA2 maintained a coherent representation of the environments controlled by the local cues, similar to intermediate and distal CA3.

Figure 7.

Figure 7

Open in a new tab

Population coherence in CA2 is similar to intermediate and distal CA3 but different from proximal CA3. (A) The population representation maintained coherence in CA2 in all mismatch angles, as indicated by the bands of high correlation (column 2). The correlation bands shifted below the main diagonal, demonstrating control by the local cues. In these normalized matrices, the blue color indicates 0 and the red color indicates 1. (B) Categorical classification of cell responses in CA2 to cue-mismatch manipulations. A higher proportion of cells in CA2 “Rotate” than “Remap”. Of the cells that rotate with the cues, a higher proportion of cells rotate with the local cues (CCW). (C) Individual cell rotation amounts between the standard and mismatch sessions in CA2. The mean vector length was significant for all mismatch angles, following the local cues (Rayleigh test, 45°: n = 30, z = 13.52, p < 0.0001; 90°: n = 32, z = 12.00, p < 0.0001; 135°: n = 24, z = 7.07, p < 0.0001; 180°: n = 28, z = 11.17, p < 0.0001). (D) PCDI difference and (E) Euclidean distance difference between CA2 and CA3 subregions. The real experimental value (thick, black line) was compared to the shuffled distributions. CA2 was different from proximal CA3 in MIS sessions, with p values < 0.05 or close to 0.05. The peak correlation difference in the 180° MIS session was also significantly different between CA2 and distal CA3. As with the comparisons among the CA3 regions (Fig. 5), the overall patterns of p values are the important result, not whether any particular comparison reaches a particular statistical significance value. ***p < 0.0001.

The STD1 × STD2 matrices of the CA2 data (Figure 7A) appeared less coherent than the corresponding matrices of the 3 CA3 subregions (Figure 4), with higher correlations in scattered pixels off the main diagonal. This result suggests that CA2 place fields may not have been as stable as CA3 place fields across standard sessions. To test this, we analyzed a subset of cells that passed the inclusion criteria in all three STD sessions (session 1, session 3, session 5). Place cells were ordered by the peak position of their linearized rate maps in the first STD session (session 1). For all CA3 subregions, place field ordering remained stable in all STD sessions (Figure 8A), showing that the place fields returned to the same firing locations in the STD session after the MIS session. In CA2, however, place fields started to misalign in the second STD session (session 3) and by the third STD session (session 5), many place fields fired in locations out of order compared to the first STD session. To quantify the stability changes, we calculated the rotation angle between the place fields in the first and last STD sessions. The dispersion of the rotation angles was different among the 4 regions, with the CA2 dispersion being different from each of the 3 CA3 subregions (Figure 8B).

Figure 8.

Figure 8

Open in a new tab

Place fields are unstable in CA2 compared to CA3 subregions. (A) From all four recording days, place cells that passed the inclusion criteria in all three STD sessions were ordered by the peak positions of their linearized rate maps in the first STD session (session 1). The positions of most place fields remained stable across session 1, session 3, and session 5 in all CA3 subregions. In CA2, the positions of the place fields started to misalign in session 3 and many were out of order in session 5. . In these plots, the firing rates of each place cell were normalized across the 3 sessions. The blue color indicates the minimum normalized firing rate (0) and the red color indicates maximum firing rate (1). (B) For a subset of place cells that passed the inclusion criteria in all three STD sessions, the rotation amounts of the place fields between the first STD session (session 1) and the last STD session (session 5) were calculated. To avoid counting the same cells over multiple days, we analyzed for each tetrode only the single day of recording that produced the largest number of well-isolated units. Comparisons of the rotation amounts in the CA2 and CA3 subregions showed significant region difference (Kruskal-Wallis test; χ2(3) = 21.82, p < 0.0001. Post-hoc Mann-Whitney U test showed that CA2 was different from proximal (z = 3.74, p < 0.0001), intermediate (z = 3.85, p < 0.0001) and distal (z = 4.33, p < 0.0001) CA3 but the CA3 subregions were not different from each other (Dist vs. Int: z = −0.95, p = 0.34; Dist vs. Prox: z = −0.24, p = 0.81; Int vs. Prox: z = 0.62, p = 0.53).

Discussion

Anatomical differences along the CA3 transverse axis suggest that the proximal and distal parts of CA3 may serve different computational roles in hippocampal processing. In the present study, place cells had different spatial tuning properties along the CA3 transverse axis. Cells located in the proximal end of CA3, near the DG, had lower mean firing rates, carried more spatial information, and had smaller place fields than cells located in the distal end. In the local-global, cue-mismatch manipulations, proximal CA3 cells tended to remap (i.e., become silent or gain a new place field), whereas intermediate and distal CA3 cells tended to rotate their place fields to follow the local cues. These differences among the subregions were reflected in the population correlation matrices, as proximal CA3 showed decorrelated representations of the mismatch environments for mismatch angles > 45°, but intermediate and distal CA3 maintained coherent representations. CA2 place fields were similar to intermediate and distal CA3, in that they maintained a high correlation between the STD and MIS sessions and tended to follow the local cues.

CA3 recurrent collaterals and autoassociative memory

Classic theories and models of memory studied recurrently connected, associative networks of neurons in which memories were stored as distributed patterns of connection weights (Hopfield, 1982; Marr, 1971; McClelland and Goddard, 1996; McNaughton and Morris, 1987; Treves and Rolls, 1992). Simple learning rules, operating locally, endowed these networks with robust storage capabilities that displayed such properties as content-addressability, error correction, pattern completion, generalization, and graceful degradation. With its extensive recurrent collateral system, the CA3 region of the hippocampus has been modeled as an autoassociative network. Consistent with these models and with previous recordings from intermediate and distal CA3 (Lee et al., 2004b; Neunuebel and Knierim, 2014), we have shown that CA3 can retrieve coherent representations of an environment when the spatial cues in that environment are altered. As an important, further test of these models, we demonstrated that the proximal region of CA3, which has the least extensive recurrent collateral system (Ishizuka et al., 1990; Li et al., 1994; Witter, 2007), showed much less population coherence than distal and intermediate CA3. Moreover, the proximal CA3 population response closely resembled previous results from the DG under the same experimental conditions (Neunuebel and Knierim, 2014). These results strongly support suggestions that proximal CA3 should be considered as a component of a computational unit with the DG, and less as a component of the autoassociative network of intermediate and distal CA3 (Hunsaker et al., 2008; Scharfman, 2007).

Differences in both external connectivity and recurrent collaterals along the proximal-distal axis likely interact to determine the observed variations in CA3 firing responses to the mismatches between local and distal cues. During learning of an environment, the DG is thought to impose a pattern of activity on CA3 through the powerful mossy fiber inputs (McNaughton and Morris, 1987). In addition to the greater density of recurrent collaterals compared to proximal CA3, intermediate and distal CA3 receive stronger perforant path inputs from the entorhinal cortex (Ishizuka et al., 1990; Witter, 2007) (Figure 1A), which are believed to act as memory retrieval cues (Treves and Rolls, 1992). The CA3 activity patterns become associated with the EC inputs (heteroassociation) and with themselves through the recurrent collaterals (autoassociation). In a very well-learned environment, the associative network can create strong attractors—stable network states that are resistant to perturbation, can be self-sustaining, and can be reactivated from an initial state by weak inputs (Amit, 1989; Hasselmo et al., 1995; Knierim and Zhang, 2012; Lu et al., in press; Samsonovich and McNaughton, 1997; Treves and Rolls, 1992; Zhang, 1996). During memory retrieval, the putative CA3 attractors receive external input from both EC and DG. Contrary to the general notion that the DG is the first component of the hippocampal trisynaptic loop, the CA3 population can produce spiking activity as a response to perforant path stimulation earlier than the DG population (Derrick, 2007; Derrick et al., 2000; Yeckel and Berger, 1990). Thus, one plausible model for intermediate/distal CA3 is that in the cue-mismatch sessions, input from the EC causes the learned, recurrent-collateral-mediated attractor bump to form at a location corresponding to the local cues on the track. By the time the slower input from the DG arrives, the strong attractor dynamics may override the DG input and prevent the network from remapping. In contrast, proximal CA3 has weaker retrieval-cue input from the EC, and its attractor basins may be weaker than those of intermediate/distal CA3 (because of its less extensive, localized recurrent collaterals). Furthermore, proximal CA3 receives input from both blades of the DG, at apical and basal dendrites, which suggests stronger DG input to this region than to intermediate/distal CA3. This combination of anatomical connection patterns may thus cause proximal CA3 to follow the DG input and remap the mismatch environment.

Viewed in isolation, the current results might be interpreted as showing simply that distal CA3 receives primarily local cue inputs, whereas proximal CA3 receives both global cue and local cue inputs. However, previous experiments under identical conditions have shown that the medial entorhinal cortex (MEC) is likely to be controlled by the global cues and the lateral entorhinal cortex (LEC) is likely to be controlled by the local cues (Neunuebel et al., 2013). The LEC and MEC appear to project equally at each location along the transverse axis of CA3 (although the total EC input changes along this axis), with the LEC projecting to the outer parts of the CA3 dendrites and the MEC projecting to the middle parts of the CA3 dendrites (Witter, 2007) (Figure 1A). Furthermore, in many prior experiments in which global cues were rotated in the absence of strong local cues, intermediate and distal CA3 cells were strongly controlled by the global cues, showing that global-cue inputs can exert strong control over place cells in these experiments (see Knierim & Hamilton, 2011, for review). Indeed, the local cue signal from LEC appears to be rather weak (Neunuebel et al., 2013), yet the CA3 network follows this input. Attractor network models show that a weak external input is sufficient to seed an “activity bump” to form at a predicted location in the neural state space (e.g., Zhang, 1996). The weak LEC input may seed the putative CA3 attractor bump if the LEC inputs are initially stronger than MEC, perhaps amplified by preferential connectivity to newborn dentate gyrus granule cells (Vivar et al., 2014; Neunuebel et al., 2013; Neunuebel and Knierim, 2014). Thus, in conjunction with complementary results along the CA3 transverse axis recently reported by Lu et al. (in press), the present results appear to reflect a competition between inputs that is resolved in a winner-take-all manner suggestive of attractor dynamics in the network.

These considerations suggest that the computational output of proximal CA3 is very different from that of intermediate/distal CA3. What are the functional implications of this difference? Proximal CA3 projects preferentially to distal CA1, which receives input preferentially from the LEC over the MEC (Steward, 1976; Witter, 2007) Distal CA3 projects preferentially to proximal CA1, which receives input preferentially from the MEC over the LEC. The putatively weaker attractor dynamics of proximal CA3 suggests that the LEC-recipient zone of CA1 does not require the nonlinear dynamics of these types of networks in order to perform its computational processing on the LEC inputs. This part of CA1, which may be more involved in object or item processing (Burke et al., 2011; Deshmukh and Knierim, 2011; Henriksen et al., 2010; Knierim et al., 2014; Nakamura et al., 2013) , may require the DG-mediated pattern separation processing without the nonlinear attractor dynamics of intermediate/distal CA3 and the resulting competition between pattern completion and pattern separation (Guzowski et al., 2004; Lee et al., 2004b; Leutgeb et al., 2004; O'Reilly and McClelland, 1994; Vazdarjanova and Guzowski, 2004). In contrast, the MEC-recipient zone of CA1, involved in spatial context representation (Hafting et al., 2005; Knierim et al., 2014), may require these classic attractor functions of CA3. Further theoretical and experimental work is required to elucidate the reasons why stronger attractor dynamics of CA3 would benefit the MEC stream and why a strong bias for pattern separation would benefit the LEC stream.

Relationship to prior results of CA3

The current results support recent behavioral and gene expression studies that have shown a functional dissociation along the CA3 transverse axis. Hunsaker et al. (2008) reported that proximal CA3 lesions led to a deficit in the ability to discriminate small changes in the distance between two objects, as measured by differential exploration. This deficit was similar to lesions of the DG. Lesions to intermediate and distal CA3, however, did not produce a deficit. Although the overall pattern of results from this and other tests were complex in their study, Hunsaker et al. (2008) suggested that the proximal CA3 region should be considered as interacting with the DG in performing spatial pattern separation. Nakamura et al. (2013) showed that proximal CA3 was more likely than distal CA3 to express the immediate early gene Arc when the animal was performing an odor discrimination task. Marrone et al. (2014) used Arc expression to argue that proximal CA3 ensembles showed stronger pattern separation than intermediate or distal CA3. Although Marrone et al. (2014) did not directly compare these regions against each other statistically (Nieuwenhuis et al., 2011), the trends in the data matched the increased remapping seen in our proximal CA3 data compared to distal CA3. The Arc studies are useful indicators of gross changes in the populations of active cells (Guzowski et al., 1999), but Arc expression is not always a faithful indicator of neural activity per se (Guzowski et al., 2006) and it lacks the resolution of single-unit recordings to uncover more specific coding mechanisms. Thus, our direct measurements of the spatial representations along the CA3 transverse axis and our analyses of the population coherence of the responses to cue altered environments, in the context of our previous studies of DG, MEC, and LEC with the same manipulations (Neunuebel and Knierim, 2014; Neunuebel et al., 2013), provide compelling evidence that the proximal CA3 region acts in concert with the DG region and functions differently from the intermediate/distal CA3.

Relationship between CA3 and CA2

CA2 occupies a location in the transverse axis in between the distal part of CA3 and the proximal part of CA1. It was thus important to determine whether this long-neglected, but newly appreciated, subregion of the hippocampus responded similarly to the distal CA3 region. Population coherence in CA2 was well maintained in the mismatch sessions, and the population representation strongly followed the local cues, similar to intermediate and distal CA3. Because our prior results from MEC and LEC do not show a strong, local-cue-dominated input that could easily explain the CA2 response (Neunuebel et al., 2013), it is likely that the CA3 inputs drove CA2 in our experiments. This suggestion may appear inconsistent with earlier in vitro work (Chevaleyre and Siegelbaum, 2010), which suggested that entorhinal input is stronger than CA3 in exciting CA2. However, that study demonstrated the presence of a strong excitatory drive from CA3 to CA2 when inhibition was blocked pharmacologically. On a reasonable assumption that the balance between excitation and inhibition is dynamically modulated in vivo, under some conditions the CA3-CA2 drive is likely to be dominated by excitation (i.e., such as in the present experiment) and under others by inhibition. Direct projections from the DG to CA2 have also been shown recently in mice, but the connections from the DG to CA2 were weaker than to CA3 (Kohara et al., 2014). In rats, the DG to CA2 projection, if present at all, appears extremely weak (Amaral and Witter, 1989). Moreover, CA2 has local axonal collaterals at a similar density as that seen in the distal part of CA3, indicative of substantial recurrent connections (Tamamaki et al., 1998; Cui et al., 2013; Ishizuka et al., 1990). Thus, the presence of recurrent collaterals in CA2, similar to intermediate and distal CA3, and the strong drive from CA3 may have likely caused coherence in CA2.

Our finding that CA2 place fields are not as stable as CA3 place fields across standard sessions is consistent with CA2 place field instability over time in the same environment (Mankin et al., 2015). Thus, the coherent response of CA2 in the mismatch session appears to ride on top of a time-dependent decorrelation of the CA2 spatial representation. The STD1 and STD2 sessions were separated by a greater time interval than the STD1 and Mismatch sessions, but the STD1 and Mismatch sessions had greater differences in spatial input than the STD1 and STD2 sessions. Thus, the correlation matrices of Figure 7A should reflect both sources of instability of the CA2 representation. The mechanisms generating the CA2 temporal instability are currently unknown (see Mankin et al., 2015, for a discussion of possible mechanisms). CA2 representational plasticity has been studied with IEG expression (Wintzer et al., 2014), and interesting differences with CA3 have been reported. It will be important for future studies to concentrate on this long-neglected component of the hippocampal pyramidal layer to understand its precise computational roles in comparison with the differentiated roles of CA3 and CA1 that have begun to be unraveled in the past decade (Guzowski et al., 2004; Lee et al., 2004a; Lee et al., 2004b; Leutgeb et al., 2004).

Summary

The present study demonstrates a functional gradient along the CA3 transverse axis, extending to the CA2 region, in terms of putative pattern separation and pattern completion/generalization processes at the level of neuronal population coding. Computational models of CA3 as an autoassociative network have assumed, for simplicity, a homogeneity of function along this axis, as have past neurophysiological recording studies. In reality, the striking gradients show that the CA3 functional topography is much more complex than traditionally thought. Future theoretical and experimental work will need to account for the functional and anatomical heterogeneity in understanding the computational role(s) performed by CA3 in the mnemonic processing performed by the hippocampus.

Experimental Procedures

Subjects and Surgery

Male, Long-Evans rats (Charles River Laboratories), 4–6 months old, were individually housed with ad libitum access to water during a 12 hr light/dark circadian cycle (lights off at 12:00 noon). A custom-built recording drive that contained 15 or 18 independently moveable tetrodes was surgically implanted over the right hemisphere. To optimize the drive placement, recordings were performed during the surgery to find the lateral edge of CA3, which served as a landmark for the mediolateral placement of the drives; the most lateral tetrode ranged from 3.7–4.1 mm lateral to bregma and 3.3–3.6 mm posterior to bregma. All animal care and housing procedures conformed to the National Institute of Health standards using protocols approved by the Institutional Animal Care and Use Committee at Johns Hopkins University or the University of Texas Health Science Center at Houston. (See Supplemental Experimental Procedures for more details.)

Local-global cue mismatch (double rotation) manipulation

The double rotation experiments were conducted for 4 days, typically during the dark portion of the light/dark cycle. Baseline sessions (each 15–20 min) in which the rat rested in the holding dish were recorded prior to the start and at the end of the experiment, in order to compare recording stability before and after the experiment. During behavior, rats ran 5–6 track sessions. After an initial baseline STD session in some rats (not analyzed in this paper), track sessions consisted of three STD sessions (local and global cue relationships remained constant) interleaved with two MIS sessions (local and global cues were rotated by equal increments, but in opposite directions, producing mismatch angles of 45°, 90°, 135°, and 180°). For example, a 180° mismatch represents a 90° CCW local cue rotati on plus a 90° CW global cue rotation. Mismatch angles were chosen in pseudorandom order such that each angle was chosen once during the first 2 days of recording and once again during the second 2 days.

Electrophysiological Recordings

Tetrodes were made from 12.5-mm nichrome wires or 17-mm platinum-iridium wires (California Fine Wire Co., Grover Beach, CA). Impedance of the nichrome wires was reduced to ~200 kOhms by electroplating them with gold. Platinum-iridium wires were not plated and had impedance of ~700 kOhms. Neural signals for 8 rats were recorded using a 64-channel wireless transmitter system (Triangle Biosystems International, Durham, NC). For all other rats, the signals were passed through a unity-gain preamplifier headstage (Neuralynx, Bozeman, MT). The signals from both groups of rats were transmitted to a Cheetah Data Acquisition System (Neuralynx, Bozeman, MT). The signals were amplified 1,000- to 5,000 times and filtered between 600 Hz and 6 kHz (for units) or 1 and 475 Hz (for LFP). The spike waveforms above a threshold of 40–70 mV were sampled for 1 ms at 32 kHz, and LFPs were continuously sampled at 1 kHz. The rat’s position was tracked with an overhead camera recording light emitting diodes (LEDs) positioned over the head of the rat (red LEDs in front and green LEDs behind) at 30 Hz.

Data Analysis

Unit Isolation

Multiple waveform characteristics (i.e., spike amplitude peak and area under the waveform) were used to isolate single units using custom-written, manual cluster-cutting software. Cells were assigned to subjective isolation quality scores from 1 (very good) to 5 (poor), depending on the distance each cluster was separated from the other clusters and from the background noise. Cluster isolation was judged independent of the behavioral firing correlates of the cells. Only well-isolated cells (scores 1–3) were included in the analysis. Putative interneurons with mean firing rates > 10 Hz were excluded from the analysis.

Rate maps and place fields

The position and the head direction of the rat were based on tracking the LEDs on the headstage connected to the hyperdrive. Analysis was performed on data restricted to times when the animal’s head was within the boundaries of the circular track and speed-filtered. For 2D rate maps that were used for display purposes, the x and y dimensions of the camera image (640 by 480 pixels) were divided into bin sizes of 10 pixels. The number of spikes of a single neuron in each bin was divided by the amount of time the rat spent in that bin, in order to create a firing rate map of the cell. For quantitative analysis, the circular 2D rate maps were transformed into linear rate maps by converting the rat’s Cartesian position into units of degrees on the track and used to calculate the spatial information score (Skaggs et al., 1996), the mean and the peak rates, and the place field size. Linearized rate maps were velocity filtered > 3 cm/s, divided into 360 bins (1°/bin), and smoothed with a Gaussian filter (σ = 3°). Place cells were identified as neurons with spatial information scores > 0.5 bits/spike, spatial information significance p < 0.01, peak firing rates > 1.5 Hz, and mean firing rates < 10 Hz. Spatial information significance was calculated with a shuffling procedure, in which the spike train and the position train were shifted relative to each other by a random time (minimum of 30 s), the rate map was recalculated, and a spatial information score was calculated. This procedure was performed 1000 times, and the cell was considered significant at p < 0.01 if the true information score exceeded the values of more than 990 scores from the shuffling procedure. Place field boundaries were determined using a floor threshold set to 10% of the peak firing rate. Bins with firing rates greater than the floor threshold were tracked in each direction from the peak bin of the place field until two consecutive subthreshold bins were reached. The size of each place field was measured across those above-threshold bins. The peak firing rate was the maximum firing rate in the rate map, and the mean firing rate was calculated by dividing the number of running spikes by the duration of the session.

Population correlation matrices

Population correlation matrices were created by forming normalized firing rate vectors for the sample of cells at each 1° bin of t he track to create 360 firing rate vectors (Figure S3A). These vectors were correlated with the vectors for every location in a comparison session (Lee et al., 2004b; Neunuebel and Knierim, 2014). The correlation matrix contains the Pearson product moment correlation values for each pair of the firing rate vectors. The mean correlation of all bins in a diagonal was calculated for each diagonal band in the matrix; the maximum mean correlation of the diagonal bands was defined as the peak correlation and was used to calculate the peak correlation difference index (PCDI) and the Euclidean distance measures (Figure S3B; see Supplemental Experimental Procedures).

Rotational analysis

The rotation angle of rate maps between the sessions was determined for each cell that passed the inclusion criteria in both the STD and the MIS sessions. The linearized rate map in the STD session was correlated with the linearized rate map in the MIS session. The linearized rate map in the MIS session was circularly shifted in 1° increments and correlated with the STD session rate map. The shift producing the maximum correlation was assigned as that cell’s rotation angle.

Stability analysis

Cells that passed the inclusion criteria in all 3 STD sessions were used in this analysis. Place cells were ordered by the peak position of the linearized rate map in the first STD session. For each cell, the rate map was normalized by its respective global peak rate in all three rate maps.

Statistical analysis

Statistical tests were calculated in Matlab (Mathworks, Natick, MA). Every statistical analysis was 2-tailed and considered significant at p < 0.05, unless stated otherwise (see Supplemental Experimental Procedures).

Histological Procedures

Rats were deeply anesthetized and perfused with 4% formaldehyde. Frozen coronal sections (40 mm) were cut and stained with cresyl violet. Images of the brain slices were acquired with IC Capture DFK 41BU02 camera (The Imaging Source, Charlotte) attached to a Motic SMZ-168 stereoscope. All tetrode tracks were identified, and the lowest point of the track was used to determine the recording location (see Supplemental Experimental Procedures).

Supplementary Material

Highlights.

  • Neural population analyses reveal functional dissociation along CA3 transverse axis

  • Proximal CA3 population activity demonstrates computational pattern separation

  • Distal CA3 population activity demonstrates computational pattern completion

  • CA2 population activity is similar to distal CA3 in a cue-mismatch environment

Acknowledgments

We thank I. Lee, E. D. Roth, and J. P. Neunuebel for use of their experimental data; J. D. Monaco for assistance with data analysis; K. Bandeen-Roche for advice on statistics; and J. L. Johnson and A. Smolinsky for technical assistance. This work was supported by NIH grants R01 NS039456 and R01 MH094146, and a grant from the Johns Hopkins University Brain Sciences Institute.

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Author Contributions

Conceptualization, H.L. and J.J.K.; Investigation, H.L and S.S.D.; Software, C.W. and S.S.D.; Formal Analysis, C.W. and H.L.; Writing – Original Draft, H.L., C.W., and J.J.K; Writing – Review & Editing, H.L., C.W., S.S.D, and J.J.K.; Visualization, H.L. and C.W.; Supervision, J.J.K.; Funding Acquisition, J.J.K.

References

  1. Amaral DG, Witter MP. The three-dimensional organization of the hippocampal formation: a review of anatomical data. Neuroscience. 1989;31:571–591. doi: 10.1016/0306-4522(89)90424-7. [DOI] [PubMed] [Google Scholar]
  2. Amit DJ. Modeling Brain Function: The World of Attractor Neural Networks. Cambridge, UK/New York: Cambridge Univ. Press; 1989. [Google Scholar]
  3. Burke SN, Maurer AP, Nematollahi S, Uprety AR, Wallace JL, Barnes CA. The influence of objects on place field expression and size in distal hippocampal CA1. Hippocampus. 2011;21:783–801. doi: 10.1002/hipo.20929. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Caruana DA, Alexander GM, Dudek SM. New insights into the regulation of synaptic plasticity from an unexpected place: hippocampal area CA2. Learn Mem. 2012;19:391–400. doi: 10.1101/lm.025304.111. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Chevaleyre V, Siegelbaum SA. Strong CA2 pyramidal neuron synapses define a powerful disynaptic cortico-hippocampal loop. Neuron. 2010;66:560–572. doi: 10.1016/j.neuron.2010.04.013. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Claiborne BJ, Amaral DG, Cowan WM. A light and electron microscopic analysis of the mossy fibers of the rat dentate gyrus. J Comp Neurol. 1986;246:435–458. doi: 10.1002/cne.902460403. [DOI] [PubMed] [Google Scholar]
  7. Cui Z, Gerfen CR, Young WS., 3rd Hypothalamic and other connections with dorsal CA2 area of the mouse hippocampus. J Comp Neurol. 2013;521:1844–1866. doi: 10.1002/cne.23263. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Derrick BE. Plastic processes in the dentate gyrus: a computational perspective. Prog Brain Res. 2007;163:417–451. doi: 10.1016/S0079-6123(07)63024-6. [DOI] [PubMed] [Google Scholar]
  9. Derrick BE, York AD, Martinez JL., Jr Increased granule cell neurogenesis in the adult dentate gyrus following mossy fiber stimulation sufficient to induce long-term potentiation. Brain Res. 2000;857:300–307. doi: 10.1016/s0006-8993(99)02464-6. [DOI] [PubMed] [Google Scholar]
  10. Deshmukh SS, Knierim JJ. Representation of non-spatial and spatial information in the lateral entorhinal cortex. Front Behav Neurosci. 2011;5:69. doi: 10.3389/fnbeh.2011.00069. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Guzowski JF, Knierim JJ, Moser EI. Ensemble dynamics of hippocampal regions CA3 and CA1. Neuron. 2004;44:581–584. doi: 10.1016/j.neuron.2004.11.003. [DOI] [PubMed] [Google Scholar]
  12. Guzowski JF, McNaughton BL, Barnes CA, Worley PF. Environment-specific expression of the immediate-early gene Arc in hippocampal neuronal ensembles. Nat Neurosci. 1999;2:1120–1124. doi: 10.1038/16046. [DOI] [PubMed] [Google Scholar]
  13. Guzowski JF, Miyashita T, Chawla MK, Sanderson J, Maes LI, Houston FP, Lipa P, McNaughton BL, Worley PF, Barnes CA. Recent behavioral history modifies coupling between cell activity and Arc gene transcription in hippocampal CA1 neurons. Proc Natl Acad Sci U S A. 2006;103:1077–1082. doi: 10.1073/pnas.0505519103. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Hafting T, Fyhn M, Molden S, Moser MB, Moser EI. Microstructure of a spatial map in the entorhinal cortex. Nature. 2005;436:801–806. doi: 10.1038/nature03721. [DOI] [PubMed] [Google Scholar]
  15. Hasselmo ME, Schnell E, Barkai E. Dynamics of learning and recall at excitatory recurrent synapses and cholinergic modulation in rat hippocampal region CA3. J Neurosci. 1995;15:5249–5262. doi: 10.1523/JNEUROSCI.15-07-05249.1995. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Henriksen EJ, Colgin LL, Barnes CA, Witter MP, Moser MB, Moser EI. Spatial representation along the proximodistal axis of CA1. Neuron. 2010;68:127–137. doi: 10.1016/j.neuron.2010.08.042. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Hitti FL, Siegelbaum SA. The hippocampal CA2 region is essential for social memory. Nature. 2014;508:88–92. doi: 10.1038/nature13028. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Hopfield JJ. Neural networks and physical systems with emergent collective computational abilities. Proc Natl Acad Sci U S A. 1982;79:2554–2558. doi: 10.1073/pnas.79.8.2554. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Hunsaker MR, Rosenberg JS, Kesner RP. The role of the dentate gyrus, CA3a,b, and CA3c for detecting spatial and environmental novelty. Hippocampus. 2008;18:1064–1073. doi: 10.1002/hipo.20464. [DOI] [PubMed] [Google Scholar]
  20. Ishizuka N, Cowan WM, Amaral DG. A quantitative analysis of the dendritic organization of pyramidal cells in the rat hippocampus. J Comp Neurol. 1995;362:17–45. doi: 10.1002/cne.903620103. [DOI] [PubMed] [Google Scholar]
  21. Ishizuka N, Weber J, Amaral DG. Organization of intrahippocampal projections originating from CA3 pyramidal cells in the rat. J Comp Neurol. 1990;295:580–623. doi: 10.1002/cne.902950407. [DOI] [PubMed] [Google Scholar]
  22. Jones MW, McHugh TJ. Updating hippocampal representations: CA2 joins the circuit. Trends Neurosci. 2011;34:526–535. doi: 10.1016/j.tins.2011.07.007. [DOI] [PubMed] [Google Scholar]
  23. Knierim JJ. Dynamic interactions between local surface cues, distal landmarks, and intrinsic circuitry in hippocampal place cells. J Neurosci. 2002;22:6254–6264. doi: 10.1523/JNEUROSCI.22-14-06254.2002. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Knierim JJ, Neunuebel JP, Deshmukh SS. Functional correlates of the lateral and medial entorhinal cortex: objects, path integration and local-global reference frames. Philos Trans R Soc Lond B Biol Sci. 2014;369:20130369. doi: 10.1098/rstb.2013.0369. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Knierim JJ, Hamilton DA. Framing spatial cognition: Neural representations of proximal and distal frames of reference and their roles in navigation. Physiol Rev. 2011;91:1245–1279. doi: 10.1152/physrev.00021.2010. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Knierim JJ, Zhang K. Attractor dynamics of spatially correlated neural activity in the limbic system. Annu Rev Neurosci. 2012;35:267–285. doi: 10.1146/annurev-neuro-062111-150351. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Kohara K, Pignatelli M, Rivest AJ, Jung HY, Katamura T, Suh J, Frank D, Kajikawa K, Mise N, Obata Y, et al. Cell type-specific genetic and optogenetic tools reveal hippocampal CA2 circuits. Nat Neurosci. 2014;17:269–279. doi: 10.1038/nn.3614. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Lee I, Rao G, Knierim JJ. A double dissociation between hippocampal subfields: differential time course of CA3 and CA1 place cells for processing changed environments. Neuron. 2004a;42:803–815. doi: 10.1016/j.neuron.2004.05.010. [DOI] [PubMed] [Google Scholar]
  29. Lee I, Yoganarasimha D, Rao G, Knierim JJ. Comparison of population coherence of place cells in hippocampal subfields CA1 and CA3. Nature. 2004b;430:456–459. doi: 10.1038/nature02739. [DOI] [PubMed] [Google Scholar]
  30. Lein ES, Callaway EM, Albright TD, Gage FH. Redefining the boundaries of the hippocampal CA2 subfield in the mouse using gene expression and 3-dimensional reconstruction. J Comp Neurol. 2005;485:1–10. doi: 10.1002/cne.20426. [DOI] [PubMed] [Google Scholar]
  31. Lein ES, Hawrylycz MJ, Ao N, Ayres M, Bensinger A, Bernard A, Boe AF, Boguski MS, Brockway KS, Byrnes EJ, et al. Genome-wide atlas of gene expression in the adult mouse brain. Nature. 2007;445:168–176. doi: 10.1038/nature05453. [DOI] [PubMed] [Google Scholar]
  32. Leutgeb S, Leutgeb JK, Treves A, Moser MB, Moser EI. Distinct ensemble codes in hippocampal areas CA3 and CA1. Science. 2004;305:1295–1298. doi: 10.1126/science.1100265. [DOI] [PubMed] [Google Scholar]
  33. Lu L, Igarashi KM, Witter MP, Moser EI, Moser M-B. Topography of place maps along the CA3-to-CA2 axis of the hippocampus. Neuron. doi: 10.1016/j.neuron.2015.07.007. (in press). [DOI] [PubMed] [Google Scholar]
  34. Li XG, Somogyi P, Ylinen A, Buzsaki G. The hippocampal CA3 network: an in vivo intracellular labeling study. J Comp Neurol. 1994;339:181–208. doi: 10.1002/cne.903390204. [DOI] [PubMed] [Google Scholar]
  35. Lorente de No R. Studies on the structure of the cerebral cortex II. Continuation of the study of the ammonic system. J Psychol Neurol. 1934;46:113–177. [Google Scholar]
  36. Lu L, Igarashi KM, Witter MP, Moser EI, Moser M-B. Topography of place maps along the CA3-to-CA2 axis of the hippocampus. Neuron. doi: 10.1016/j.neuron.2015.07.007. (in press). [DOI] [PubMed] [Google Scholar]
  37. Mankin EA, Diehl GW, Sparks FT, Leutgeb S, Leutgeb JK. Hippocampal CA2 Activity Patterns Change over Time to a Larger Extent than between Spatial Contexts. Neuron. 2015;85:190–201. doi: 10.1016/j.neuron.2014.12.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Marr D. Simple memory: a theory for archicortex. Philos Trans R Soc Lond B Biol Sci. 1971;262:23–81. doi: 10.1098/rstb.1971.0078. [DOI] [PubMed] [Google Scholar]
  39. Marrone DF, Satvat E, Odintsova IV, Gheidi A. Dissociation of spatial representations within hippocampal region CA3. Hippocampus. 2014;24:1417–1420. doi: 10.1002/hipo.22367. [DOI] [PubMed] [Google Scholar]
  40. McClelland JL, Goddard NH. Considerations arising from a complementary learning systems perspective on hippocampus and neocortex. Hippocampus. 1996;6:654–665. doi: 10.1002/(SICI)1098-1063(1996)6:6<654::AID-HIPO8>3.0.CO;2-G. [DOI] [PubMed] [Google Scholar]
  41. McNaughton BL, Morris RGM. Hippocampal synaptic enhancement and information storage within a distributed memory system. Trends Neurosci. 1987;10:408–415. [Google Scholar]
  42. McNaughton BL, Nadel L. Neuroscience and Connectionist Theory. Hillsdale, NJ: Lawrence Erlbaum; 1990. [Google Scholar]
  43. Nakamura NH, Flasbeck V, Maingret N, Kitsukawa T, Sauvage MM. Proximodistal segregation of nonspatial information in CA3: preferential recruitment of a proximal CA3-distal CA1 network in nonspatial recognition memory. J Neurosci. 2013;33:11506–11514. doi: 10.1523/JNEUROSCI.4480-12.2013. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Neunuebel JP, Knierim JJ. CA3 retrieves coherent representations from degraded input: direct evidence for CA3 pattern completion and dentate gyrus pattern separation. Neuron. 2014;81:416–427. doi: 10.1016/j.neuron.2013.11.017. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Neunuebel JP, Yoganarasimha D, Rao G, Knierim JJ. Conflicts between local and global spatial frameworks dissociate neural representations of the lateral and medial entorhinal cortex. J Neurosci. 2013;33:9246–9258. doi: 10.1523/JNEUROSCI.0946-13.2013. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Nieuwenhuis S, Forstmann BU, Wagenmakers EJ. Erroneous analyses of interactions in neuroscience: a problem of significance. Nat Neurosci. 2011;14:1105–1107. doi: 10.1038/nn.2886. [DOI] [PubMed] [Google Scholar]
  47. O'Reilly RC, McClelland JL. Hippocampal conjunctive encoding, storage, and recall: avoiding a trade-off. Hippocampus. 1994;4:661–682. doi: 10.1002/hipo.450040605. [DOI] [PubMed] [Google Scholar]
  48. Rolls ET, Kesner RP. A computational theory of hippocampal function, and empirical tests of the theory. Prog Neurobiol. 2006;79:1–48. doi: 10.1016/j.pneurobio.2006.04.005. [DOI] [PubMed] [Google Scholar]
  49. Rolls ET, Treves A. Neural Netowrks and Brain Function. Oxford: Oxford University Press; 1998. [Google Scholar]
  50. Roth ED, Yu X, Rao G, Knierim JJ. Functional differences in the backward shifts of CA1 and CA3 place fields in novel and familiar environments. PLoS One. 2012;7:e36035. doi: 10.1371/journal.pone.0036035. [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Samsonovich A, McNaughton BL. Path integration and cognitive mapping in a continuous attractor neural network model. J Neurosci. 1997;17:5900–5920. doi: 10.1523/JNEUROSCI.17-15-05900.1997. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Santoro A. Reassessing pattern separation in the dentate gyrus. Front Behav Neurosci. 2013;7:96. doi: 10.3389/fnbeh.2013.00096. [DOI] [PMC free article] [PubMed] [Google Scholar]
  53. Scharfman HE. The CA3 "backprojection" to the dentate gyrus. Prog Brain Res. 2007;163:627–637. doi: 10.1016/S0079-6123(07)63034-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  54. Skaggs WE, McNaughton BL, Wilson MA, Barnes CA. Theta phase precession in hippocampal neuronal populations and the compression of temporal sequences. Hippocampus. 1996;6:149–172. doi: 10.1002/(SICI)1098-1063(1996)6:2<149::AID-HIPO6>3.0.CO;2-K. [DOI] [PubMed] [Google Scholar]
  55. Steward O. Topographic organization of the projections from the entorhinal area to the hippocampal formation of the rat. J Comp Neurol. 1976;167:285–314. doi: 10.1002/cne.901670303. [DOI] [PubMed] [Google Scholar]
  56. Thompson CL, Pathak SD, Jeromin A, Ng LL, MacPherson CR, Mortrud MT, Cusick A, Riley ZL, Sunkin SM, Bernard A, et al. Genomic anatomy of the hippocampus. Neuron. 2008;60:1010–1021. doi: 10.1016/j.neuron.2008.12.008. [DOI] [PubMed] [Google Scholar]
  57. Treves A, Rolls ET. Computational constraints suggest the need for two distinct input systems to the hippocampal CA3 network. Hippocampus. 1992;2:189–199. doi: 10.1002/hipo.450020209. [DOI] [PubMed] [Google Scholar]
  58. Vazdarjanova A, Guzowski JF. Differences in hippocampal neuronal population responses to modifications of an environmental context: evidence for distinct, yet complementary, functions of CA3 and CA1 ensembles. J Neurosci. 2004;24:6489–6496. doi: 10.1523/JNEUROSCI.0350-04.2004. [DOI] [PMC free article] [PubMed] [Google Scholar]
  59. Vivar C, Potter MC, Choi J, Lee JY, Stringer TP, Callaway EM, Gage FH, Suh H, van Praag H. Monosynaptic inputs to new neurons in the dentate gyrus. Nat Comm. 2012;3:1107. doi: 10.1038/ncomms2101. [DOI] [PMC free article] [PubMed] [Google Scholar]
  60. Wasserman L. All of Statistics. New York: Springer; 2004. [Google Scholar]
  61. Wintzer ME, Boehringer R, Polygalov D, McHugh TJ. The hippocampal CA2 ensemble is sensitive to contextual change. J Neurosci. 2014;34:3056–3066. doi: 10.1523/JNEUROSCI.2563-13.2014. [DOI] [PMC free article] [PubMed] [Google Scholar]
  62. Witter MP. Intrinsic and extrinsic wiring of CA3: indications for connectional heterogeneity. Learn Mem. 2007;14:705–713. doi: 10.1101/lm.725207. [DOI] [PubMed] [Google Scholar]
  63. Yassa MA, Stark CE. Pattern separation in the hippocampus. Trends Neurosci. 2011;34:515–525. doi: 10.1016/j.tins.2011.06.006. [DOI] [PMC free article] [PubMed] [Google Scholar]
  64. Yeckel MF, Berger TW. Feedforward excitation of the hippocampus by afferents from the entorhinal cortex: redefinition of the role of the trisynaptic pathway. Proc Natl Acad Sci U S A. 1990;87:5832–5836. doi: 10.1073/pnas.87.15.5832. [DOI] [PMC free article] [PubMed] [Google Scholar]
  65. Zhang K. Representation of spatial orientation by the intrinsic dynamics of the head-direction cell ensemble: a theory. J Neurosci. 1996;16:2112–2126. doi: 10.1523/JNEUROSCI.16-06-02112.1996. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

NIHMS710552-supplement.pdf (2MB, pdf)

RESOURCES