Recognition Memory Correlates of Hippocampal Theta Cells

Behavior

Animals were trained on the two-odor simultaneous DNMS task, illustrated schematically in Figure1A, as described inWiebe and Stäubli (1999). Illumination of a cue light above the sample port marked the start of the trial. The first sample nosepoke commenced a 2 sec minimum pre-odor baseline period, after which a nosepoke turned on the odor (either odor A or odor B). After a minimum of 10 sec of odor exposure, a nosepoke terminated the odor and initiated a variable 1–50 sec delay period. After the delay, a sample poke extinguished the cue light, illuminated the ceiling house light, and prompted the delivery of the two odors in the test arms, one odor per arm. The rat exited the sample arm and then simultaneously discriminated the two odors in the center region of the Y maze, just outside the entry point of the test arms. On discrimination of the two odors, the rat selected one of the odors by entering an arm and committing a nosepoke. The test odors (one in each arm) emanated continually from both test arms and could be detected within and just outside the entry point of the arms, regardless of whether a nosepoke was made in the arm. The distance between the test-arm entry point and the infrared beam in the nosepoke device protruding from the end of the test arm was short (24 cm, approximately the body length of the rat) and was traversed rapidly by the water-deprived rats. The time between arm entry and the nosepoke was an accurate measure of locomotion in the test arms because the time span was short (mean, 0.920 sec, on average for each rat) and varied little across trials (SEM, 0.009 sec). A nosepoke in the nonmatch arm resulted in a 0.05 ml water reward. A match response triggered a high-intensity flash discharge. By definition, then, all nonmatch trials were correct trials, and all match trials were error trials. Three seconds after both correct and error responses, odor delivery to the test ports was terminated, the house light was extinguished, and an intertrial interval of 20 sec was imposed. The identity of the sample odor and location of the match and nonmatch test odors were randomized across trials. Performance diminished linearly as a function of delay from 86% correct at 0–5 sec delays to 74% correct at 45–50 sec delays. Figure1B shows the mean performance curve per session summed over all rats, with 100–250 trials per session.

Although the rat could correct initial odor/arm selections in the test phase by stopping after entering one arm, backing up, and then entering the other arm, this postselection corrective behavior was observed on a relatively small percentage of trials (6 ± 1% of the trials per session; mean ± SEM averaged across all rats). This implies that on most trials, the rats made the match/nonmatch decision in the test phase while sampling the two odors simultaneously at the junction between the two test arms, as opposed to selecting one particular odor or arm in the test phase by default and then making a corrective response if necessary. Therefore, the task was performed as “delayed-nonmatch-to-sample” and not using alternative response strategies such as “win–stay/lose–shift” involving corrective responses after initial default choices.

The delayed-nonmatch-to-sample task by definition (Suzuki, 1996) is comprised of three phases: (1) sample item presentation, followed by (2) a delay interval, followed by (3) a recognition phase in which one or more test items are presented, with reward resulting from selection of the nonmatch item. Many rat hippocampal recording studies of recognition memory implementing the DNMS paradigm have used the continuous protocol (Otto and Eichenbaum, 1992; Wood et al., 1999) in which only one item is presented in the test phase. Individual stimuli are presented sequentially, one after another, with the animal rewarded for recognizing an item as different from its antecedent. In such a paradigm, stimuli are at once both samples for the following item and test cues for the preceding item, thereby potentially obscuring detection of any phase-sensitive coding. In the simultaneous two-item DNMS task used here and by others (Mumby and Pinel, 1994; Deadwyler et al., 1996; Murray and Mishkin, 1998), two items (the match cue and the nonmatch cue) are simultaneously presented to the animal in the test phase and continue to be present until the final behavioral response is executed. In the Deadwyler and Mumby studies, spatial cues and nonspatial, visual cues were used as discriminative stimuli, respectively, with the selection response involving lever presses and displacement of three-dimensional objects. In the present study, odors were used as discriminative stimuli, and the behavioral response was a nosepoke. An advantage of the simultaneous DNMS paradigm that was used here, however, with spatially and temporally distinct sample and test phases, was that comparisons could be made between the response properties of cells to stimuli presented in the acquisition versus recall phase of the task.

Surgery

After criterion behavioral DNMS performance was reached, animals were surgically implanted with an 8 × 2 microwire (diameter, ∼45 μm) electrode array (NB Labs, Denison, TX). To help eliminate low-frequency movement artifacts, the recording signals were subtracted from that of a low-impedance–tip reference microwire positioned just posterior and at the same depth as the array. The rat was anesthetized with sodium pentobarbital (50 mg/kg, i.p.) and given atropine (0.1 mg/kg, i.p.) to reduce mucous secretions. The center pair of electrodes was positioned at coordinates 4.0 mm posterior to bregma, 3.3 and 2.8 mm lateral to midline for CA3 and DG-CA1 placement, respectively. The longitudinal axis of the array was rotated to a 30° angle from midline, with the anterior end more medial and the posterior end more lateral, to follow the contour of the hippocampus. The array was driven slowly (∼25 μm/min) through the brain to a depth of 3.4–4.0 mm for CA3 and 2.2–3.2 mm for CA1. Neural activity from the microwire electrodes was monitored throughout surgery to ensure placement near the hippocampal cell layers. After array placement, the cranium was sealed with bone wax and dental cement. The animal was allowed to recover for 6–7 d before DNMS retraining and recording commenced. The scalp wound was treated periodically with Neosporin antibiotic to prevent infection.

Histology

At the conclusion of recording, the location of each recording electrode was histologically verified (Fig. 1C) by passing 40 μA current through each microwire and reacting with potassium ferrocyanide to form a Prussian blue stain. A Nissl stain was used to aid the visualization of the cell body layers. All animal care and experimental procedures conformed to National Institutes of Health and Society for Neuroscience guidelines for care and use of experimental animals.

Unit recording technique

Extracellular action potentials (“spikes”) and behavioral responses were digitized and time-stamped for computer processing in relation to successive behavioral events within each DNMS trial. Neuronal activity was digitized at 40 kHz and isolated by time–amplitude window discrimination and template matching using a Multichannel Acquisition Processor system (Plexon, Dallas, TX). Up to four single units could be isolated per microwire (Fig.1D). Identified spikes were tracked from session to session by waveform and firing characteristics within the task (perievent histograms). To maximize the likelihood that single units were recorded, only waveforms with zero spike counts in the first 1 msec time bin of their interspike interval histogram were included in the analysis. Also, to help ensure that the same neurons were recorded continuously over time, waveforms were required to have stable perievent firing rates across sessions. Although it is possible that the neuronal spikes that were discriminated on a given microwire may not have isolated single units or consistently identified the same neuron over time (McNaughton et al., 1983b), selecting only waveforms with absolute 1 msec refractory periods, and with constant firing rates and behavioral correlates across recording sessions greatly reduced the probability that different neurons were mistaken as single units (Deadwyler et al., 1996).

Identification of hippocampal principal cells and theta interneurons

Theta interneurons can be distinguished from principal cells in the hippocampus on the basis of their physiological characteristics. Interneurons in the dentate gyrus (Mizumori et al., 1989; Jung and McNaughton, 1993) and CA3/1 fields (Ranck, 1973; Fox and Ranck, 1975,1981) exhibit high (>5 Hz) overall mean firing rates modulated at theta rhythm (5–12 Hz) and have narrow spike widths (most with <250 μsec negative-going spike widths); in contrast, granule and pyramidal cells discharge in complex-spike bursts of two to seven spikes at 3–5 msec interspike intervals with low (<2 Hz) overall mean firing rates and possess large spike widths (>250 μsec negative-going spike widths). Although mean firing rate alone or spike width alone is insufficient to clearly distinguish the principal cell and theta interneuron populations because of partially overlapping distributions, the two cell groups can be effectively separated if the two parameters are considered together in firing rate–spike width space (Csicsvari et al., 1999). In the present study, up to four individual units could be isolated on each microwire, including at times both principal cells and interneurons on the same wire. A linear cutoff in firing rate–spike width space was used in conjunction with the autocorrelation function to classify the two cell types. Cells with mean firing rates r (in Hertz) and negative-going spike widths w (in microseconds) such that r < 0.04 · w − 3.5 and a peak at 3–5 msec in the autocorrelogram followed by a fast exponential decay (i.e., complex-spike activity) were classified as principal cells. Cells with mean firing rates and negative-going spike widths such thatr > 0.04 · w − 3.5 and a prominent peak in the 80–200 msec range (5–12 Hz) of the autocorrelogram were classified as theta interneurons. The two populations of cells (Fig.1D) were separated reasonably well in firing rate–spike width space by the linear cutoff method, as measured by multivariate ANOVA (F_(2,1237) = 817.29; p < 10⁻¹⁰). The firing correlates of the theta interneurons (firing rate, 17.4 ± 1.13 Hz, mean ± SEM; negative-going spike width, 206 ± 6 μsec; n = 139) were compared with that of the complex-spike principal cells (firing rate, 1.1 ± 0.05 Hz; negative-going spike width, 305 ± 3 μsec; n = 1101) reported in Wiebe and Stäubli (1999).

Analysis

Event-responsive activity. Perievent histograms were generated for each cell around each DNMS event. The associated event, bin width, histogram boundary limits, and grouping factors for each perievent histogram are shown in Table1. The responsivity of cells to events in the sample, delay, and test phase was determined by comparing the firing rate within each perievent histogram with its appropriate baseline period. For perievent histograms in the test phase, in which the rat's position was different for each event, the baseline used was mean firing rate over the entire trial. A neuron was classified as a “test cell” if there was a significant (ANOVA; p < 0.01) firing rate increase in one or more perievent histogram bins relative to baseline. Only increases in firing were considered for events in the test phase, which did not all occur at a common location to avoid classifying low, out-of-field firing of a spatially selective cell as a response (e.g., to avoid classifying a place cell with a receptive field in the sample arm and no activity in the test arms as “responsive” to events in the test phase). For perievent histograms in the sample and delay phases in which the location of the rat was fixed at the sample port, the pre-odor interval at the sample port was used as the baseline reference period. A cell was classified as a “sample cell” or “delay cell” if there was a significant change (either enhancement or suppression) in the firing rate of one or more perievent histogram bins relative to pre-odor baseline. Further discriminant analyses were performed then on each event-responsive cell.

Determination of firing selectivity. The test-phase perievent histograms constructed around the test-arm entry and poke response consisted of three factors: position [left/right (L/R)], odor (A/B), and trial type [correct/error (C/E)]. The sample phase and delay phase perievent histograms constructed around odor onset and offset consisted of two factors: odor (A/B) and trial type (correct/error). Note that correct/error in the sample and delay phases refers to trials in which a nonmatch or match response, respectively, was made in the test phase.

Discriminant analysis (Rencher, 1995) is a mathematical technique to project data in high-dimensional space onto a reduced number of dimensions (i.e., a set of basis vectors) that maximally separate one or more groups. For linear discriminant analysis of perievent activity, this translates to finding weights for each perievent histogram time bin such that the weighted linear sum maximally separates one or more of the groups. This process is equivalent to plotting the perievent firing rate of a cell for each trial in n-dimensional space, where n is the number of time bins, and computing directions in n-dimensional space along which the groups are maximally separated [see Wiebe and Stäubli (1999) for schematic illustration]. For example, if odor and correct/error are the two factors and the four groups are correct odor A trials, correct odor B trials, error odor A trials, and error odor B trials, and if the perievent firing rate of the cell, X, for trial jin time bin i is x_ij, then the first direction (first discriminant function) D¹ = (d ${}_1^1$, d ${}_2^1$,… , d ${}_{\text{n}}^1$) is computed such that the distribution of

across the j trials maximally separates one or more of the groups. The first discriminant function is chosen therefore such that the data projection onto it accounts for the maximal separation between the groups (maximal variance). Mathematically, this is equivalent to finding the eigenvector with the largest eigenvalue λ₁ of the matrix B ·W⁻¹, where B is the between-group covariance matrix and W is the within-group covariance matrix for the perievent histogram. The second discriminant function D² = (d ${}_1^2$, d ${}_2^2$,… , d ${}_{\text{n}}^2$) is then selected such that it maximizes the variance between the groups and is uncorrelated (linearly independent) with D¹. A total of minimum (number of time bins, number of groups) discriminant functions is calculated in this fashion for each perievent histogram such that successive functions account for maximal variance of the perievent activity of the cell across groups that are linearly independent from previously calculated functions. The proportion of perievent firing variance across trials accounted for by the discriminant function projections D¹ · X, D²· X, … is ordered, therefore, from highest to lowest. The ability of a discriminant function projection to separate one or more of the groups was determined using the χ² approximation of the Wilks Λ statistic. Significant discriminant function projections (p < 0.01; χ²) were analyzed then using two- and three-way ANOVAs to determine which groups were separated. Cells with discriminant function projections that had a significant main effect (ANOVA; p < 0.01) were deemed selective for that factor (e.g., odor) and to discriminate the two groups in that factor (e.g., odor A versus odor B) in its perievent firing rate.

The degree to which the difference in firing rate between separated groups contributed to the overall perievent firing rate variance across trials was determined by the ratio of the eigenvalue of the corresponding discriminant function to the total firing rate variance in the perievent histogram across trials. Because the discriminant functions are uncorrelated, the eigenvalues additively partition the total firing rate variance across trials, and the percentage of firing rate variance accounted for by the ith discriminant function is given by:

The percentage of variance in firing rate across trials accounted for by discriminant functions that separated sample odor A versus B, test odor A versus B, test position left versus right, and correct versus error trials in the test phase was calculated (see Fig.6A) to compare the relative strength of each correlate. For instance, a discriminant function projection D · X which separated odor A versus odor B trials and contributed to 80% of the perievent firing variance of the cell would indicate that 80% of the variation in perievent activity across trials was attributable to differential activity on odor A versus odor B trials.

Controls for movement

Because the firing rates of some classes of interneurons are known to be correlated to walking speed (McNaughton et al., 1983a;Rivas et al., 1996), we compared velocity distributions between groups separated by the discriminant analysis, as calculated by subtracting the test-arm entry and nosepoke response times. A significance level of p > 0.05 (ANOVA) signified that walking speeds on odor A versus odor B, left versus right, or correct versus error trials were indistinguishable. The distance between the entry point of the test arm and the nosepoke port infrared beam was 24 cm, approximately equal to the body length of the rat. In test arms in which a nosepoke was executed, the rats traversed this length rapidly (mean time, 0.920 sec on average for each rat) and in a highly stereotyped fashion (SEM, 0.009). This effectively rules out the possibility that other variants of overt behavior, such as vigorous head movements, etc., could have been responsible for generating the observed position, odor, or correct/error firing selectivity in the test phase. The 0.920 sec average test-arm traversal time was less than the 1 sec length of the perievent firing rate histograms analyzed around the test-arm entry and in the pre-poke period. Hence, the resolution of the walking speed measurement was of the same order of magnitude as the time resolution of the perievent firing rate analyses. In all other perievent histograms analyzed [post-poke in the test phase; odor-fast, odor-slow, and odor-off in the sample phase; and the delay histogram (Table 1 and Fig. 2)], the body and head position of the rat was stationary, with the nose positioned inside 20-mm-diameter cylindrical nosepoke ports.

Theta cell odor A/B, correct/error, and left/right position selectivity in the test phase not attributable to differential walking speeds between groups

The test phase differed from the sample and delay phases in that the rat was walking while smelling and discriminating the odor stimuli. The majority of the test cells with selective odor, correct/error, or position firing while the rat was moving from the test-arm entrance to the poke port (i.e., in the test-entry and pre-poke periods) did not have significant differences (ANOVA; p > 0.05) in walking speed for odor A versus odor B trials (20 of 23 cells; 87%), correct versus error trials (20 of 25; 80%), and left versus right position (79 of 104; 76%), respectively. The walking speed was calculated using the time between test-arm entry and the end-of-arm nosepoke [0.920 ± 0.009 sec (mean ± SEM) on average for each rat]. This indicates that the selectivity of the theta cell firing observed before the nosepoke in the test phase cannot be attributed to differential rates of locomotion within the different analysis groups.

Correct/error-selective cells with reinforcement-correlated activity after the test nosepoke excluded from subsequent analyses

To determine whether correct/error-selective firing immediately after the test nosepoke was associated with the behavioral execution of the nosepoke or was attributable to differential responding to the water reward and light flash, the reinforcement signal was delayed 1.5 sec on a select number of randomly interspersed trials. Correct/error-selective cells in the post-poke period were classified as either reinforcement-correlated or poke-correlated on the basis of whether their poke-aligned or reinforcement-aligned perievent histogram contained the larger peak firing rate. An example of a reinforcement-correlated correct/error cell is shown in Figure 5A. The elevated activity of the cell after error match, but not correct nonmatch, responses was better correlated with the flash reinforcement than the nosepoke response (i.e., more temporally locked in the flash-centered versus the poke-centered histogram), indicating that the activity could be interpreted as a response to the light flash. An example of a poke-correlated cell is shown in Figure 5B. The increased activity after correct nonmatch responses compared with error match responses was correlated better with the time of the poke than the water reinforcement delivery (i.e., more temporally locked in the poke-centered versus the reinforcement-centered histogram), indicating that the activity of the cell represented the execution of a correct nosepoke and not simply a response to the water delivery. Reinforcement-correlated C/E cells were discarded, and only C/E-selective cells with poke-correlated activity were included from the post-poke period in subsequent analyses (Table 2, Figs. 4,6).

Correct/error recognition signaling most prominent in CA3/1 subfields, odor-selective in DG/CA3, and integrated with spatial representations

Responsive theta cells with odor, correct/error (including 15 C/E cells in the test-entry period, 15 C/E cells in the pre-poke period, and 28 poke-correlated C/E cells from the post-poke period), and position selectivity were tallied over all three test-phase perievent histograms, as shown in Table 2 and Figure 4A. Position (115 of 126; 91%), odor (43 of 126; 34%), and correct/error selectivity (48 of 126; 38%) in the test phase occurred in a greater proportion of responsive theta cells than principal cells [400 of 571 (70%), 155 of 571 (27%), and 121 of 571 (21%), respectively). Roughly one-third (37 of 115) of the spatial theta cells were selective for position exclusively with no significant (ANOVA; p< 0.01) main or interaction effects with regard to odor or C/E, whereas ∼95% of the C/E (45 of 48) and odor (41 of 43) cells had spatial correlates (Table 2, Fig. 4A).

Correct/error trial-selective activity was seen in twice as many theta cells in CA3/1 (43 of 101 or 43% of responsive neurons) as in DG (5 of 25; 20%) (χ² test; χ²₍₁₎ = 4.33;p < 0.05) (Fig. 4C). The majority of correct/error cells in DG and CA3 were odor selective (18 of 32; 56%), whereas only 2 of 16 (12%) C/E cells in CA1 had conjunctive odor correlates (Fig. 4D). This distribution of odor selectivity of C/E cells in the test phase was more steeply graded (χ² test; χ²₍₂₎ = 12.78;p < 0.01) than that observed for the C/E-selective principal cells [10 of 17 (59%) in DG, 32 of 76 (42%) in CA3, and 10 of 28 (36%) in CA1] and was inverse and distinct (χ² test; χ²₍₂₎ = 58.2;p < 0.01) from the theta cell odor-selectivity distribution in sample phase in Figure 4B. The distribution of odor selectivity of responsive theta cells in the test phase [7 of 25 (28%) in DG, 27 of 65 (42%) in CA3, and 9 of 36 (25%) in CA1], although not inversely distributed across subfields relative to the sample phase, was statistically different (χ² test; χ²₍₂₎ = 21.6;p < 0.01) from the sample-phase distribution.

Equivalent relative strength of odor, position, and correct/error recognition correlates in interneurons and principal cells

For each odor-selective cell in the sample phase or position-, C/E-, or odor-selective cell in the test phase, we calculated the percentage of the total variance in perievent firing across trials accounted for by the associated discriminant functions. A summary of the pooled results over all DNMS events for both interneurons and principal cells is shown in Figure6A. The percentage of total perievent firing rate variance accounted for by differential sample odor A versus B, test odor A versus B, test left versus right position, and test correct versus error trial activity for interneurons did not differ from that of the principal cells for each of the four variables (ANOVA;p > 0.1; pairwise comparisons). Nor was there any significant difference in percentage of total perievent firing variance among the four variables for the interneurons (ANOVA; p> 0.1).

Majority of C/E theta cells exhibited greater activity on correct trials than error trials, but with smaller difference in firing rate compared with principal cells

To compare the magnitude of interneuron correct/error recognition correlates in absolute terms with that of principal cells, we calculated the ratio of the mean perievent activity (averaging across all time bins in each perievent histogram) for correct (C) and error (E) trials, [(C − E)/(C + E)], for all C/E-selective cells. The majority of both C/E-selective principal cells (84 of 121; 70%) and interneurons (38 of 48; 79%) exhibited higher firing rates on correct nonmatch trials than error match trials. The ratio of correct to error activity, however, was considerably smaller in interneurons [45 of 48 (94%) of the theta cells with <20% difference] compared with the principal cell population [55 of 124 (44%) of the cells with >20% difference] (Fig.6B). The inset portion of Figure6B illustrates that the majority of C/E theta cells did exhibit moderately (>0%, <25%) higher firing rates on correct nonmatch trials than error match trials.