INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Aging has been shown to produce deficits in the ability to learn hippocampus-dependent forms of learning (Geinisman et al. 1995). Trace eyeblink classical conditioning in the rabbit is a hippocampus-dependent task (Kim et al. 1995; Moyer et al. 1990; Solomon et al. 1986) that is often used to study the neurophysiological mechanisms of aging (Powell et al. 1991; Solomon et al. 1988; Woodruff-Pak and Thompson 1985). In this paradigm, each trial consists of a neutral auditory conditioned stimulus (CS) followed by a stimulus free trace period, then a nonnoxious eyeblink-eliciting airpuff unconditioned stimulus (US) presented to the cornea. After repeated trials, young animals and humans learn to associate the CS and US, whereas a subset of aged animals and humans show poor learning on this task (Finkbiner and Woodruff-Pak 1991; Thompson et al. 1996; Woodruff-Pak et al. 1999). This ability to separate aging-impaired from -unimpaired subjects has made trace eyeblink conditioning a very effective tool in animal and human research on aging.

A great deal of research has been aimed at understanding the hippocampal neurophysiology of trace eyeblink conditioning. Berger and Thompson were the first to demonstrate that hippocampal neurons encode learning-related information during eyeblink conditioning (Berger et al. 1976). They conducted a series of studies examining the multiple- and single-neuron activity of the pyramidal cell layer of the hippocampus during delay eyeblink conditioning (Berger and Thompson 1978a,b; Berger et al. 1983). Although the delay task was not hippocampus dependent, they found that pyramidal cell activity increased during the initial trials of training. These increases in activity paralleled the amplitude time course of the behavioral response and preceded it temporally. Recently, we have expanded on the work of Berger and Thompson using variations of the stereotrode recording technique, which allows a large number of single neurons to be isolated from a multiple-unit record (McNaughton et al. 1983). Our studies have shown similar early learning-related increases in CA1 pyramidal cell activity; however, we have also found that there are a number of different excitatory and inhibitory patterns of pyramidal cell activity that encode learning-related information during trace eyeblink conditioning (McEchron and Disterhoft 1997). These heterogeneous response profiles suggest that CA1 pyramidal neurons may not encode trace eyeblink information in an additive fashion but rather different patterns of single-neuron activity could interact within a network to encode information.

There is ample evidence from several laboratories demonstrating that CA1 neurons are affected by the process of aging (e.g., Clark et al. 1992; Landfield and Pitler 1984; Potier et al. 1992; Shen and Barnes 1996; West et al. 1994). In vitro work from our laboratory suggests that CA1 neurons of aged animals are less excitable, and this altered excitability may play a role in the ability to learn trace eyeblink information (Moyer et al. 2000). A number of studies have examined the activity of CA1 pyramidal cells in vivo and have found that the activity of these cells is altered in aged animals (Mizumori et al. 1996; Shen et al. 1997; Tanila et al. 1997). Although several investigations have described how hippocampal single-neuron activity is altered as the result of aging, there are no studies describing how the interaction of different types of single neurons within local ensembles may play a role in aging-related learning deficits. Therefore the present study had two goals, to describe the different CA1 single-neuron response types that occur during trace eyeblink conditioning and to understand how the encoding of learning-related information within these ensembles is affected in aged animals that are unable to learn the trace eyeblink task. The approach of the present study was to characterize heterogeneous single-neuron response types within ensembles. Each ensemble was recorded from the tip of a single fixed electrode. Thus the present study examined how heterogeneous single-neuron response types with a close physical relationship interact to encode trace eyeblink information. Some of the initial pilot data from this investigation were briefly mentioned in a theoretical review (McEchron and Disterhoft 1999).

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Subjects

Subjects were 11 young (3-6 mo) and 9 aged (30-34 mo) New Zealand albino rabbits, Oryctolagus cuniculus. All rabbits were females obtained from Kuiper rabbitry (Gary, IN) or from Covance Laboratories (Denver, PA). Aged rabbits were retired breeders that had ceased all nursing for at least 3 mo. Previous work from our laboratory (Thompson et al. 1996) has shown that the age range from 30 to 34 mo produces heterogeneous levels of impairment in trace eyeblink conditioning that range from mild to severe. All rabbits were housed individually and maintained on a 14/10 h light/dark cycle with food and water provided ad libitum.

Surgery

All subjects were allowed to remain undisturbed in their cages for 1 wk prior to any handling or surgery. Surgery was carried out under National Institutes of Health and Northwestern University Animal Care and Use Committee approved procedures. Animals were anesthetized with ketamine (60 mg/kg im) and xylazine (10 mg/kg im), and the eyes were kept moist with a thin coat of antibacterial ophthalmic ointment. The skull was positioned in a stereotaxic frame with lambda 1.5 mm below bregma. The skull was then exposed, and a 3-mm-diam hole was drilled above the left CA1 area of the hippocampus. Five self-tapping screws (No. 2 × 1/4-in) were inserted ~2 mm into the skull to anchor the final dental cement-head assembly. In each animal, one or two nonmovable stereotrode recording bundles were stereotaxically lowered into the left CA1 area of the hippocampus (~3 mm ventral to dura) until action potentials with pyramidal cell firing characteristics were recorded (Ranck 1973). This procedure ensured that the electrode tip was located within the pyramidal cell layer of CA1. The coordinates for electrode placement were 5.0-5.2 mm caudal to bregma and 5.2-5.4 mm lateral to midline. Dental cement was then used to secure the electrodes to the skull and close the remaining wound area. Rabbits were given Buprenorphine (0.3 mg/kg sc) to minimize discomfort after recovery from anesthesia.

Behavioral training

Rabbits were allowed 5-7 days of recovery before any handling or testing. For all training sessions, animals were placed in a cloth restraining jacket and Plexiglas restraining box and then placed in a sound-attenuating chamber. The right eye was held open in a comfortable position with eyelid hooks attached to a velcro strap. The ends of rubber tubes (1-cm diam) were placed comfortably in each ear and served to deliver the auditory tone CS (100 ms; 90 dB; 6 kHz; 5-ms rise/fall time) from headphones. An airpuff tube was placed 1 cm from the animal's right eyeball and served to deliver the airpuff US (150 ms; 3 psi). The airpuff was supplied by compressed air and controlled by a regulator and solenoid valve. The airpuff intensity was adequate to reliably evoke an eyeblink that extended the nictitating membrane across the globe of the eyeball. A lightweight infrared sensor was used to transduce extensions of the nictitating membrane (Thompson et al. 1994). The sensor was fastened to the dental acrylic on the animal's head. This enabled the animal's head to move comfortably within the restrainer. Changes in voltage from the infrared sensor were sampled by a computer which also controlled the delivery of CSs and USs (Akase et al. 1994).

Animals were first given one 30-min acclimation session during which no stimuli were presented. One day following acclimation, trace-conditioned rabbits received daily sessions (10 session total) consisting of 80 CS-US trials presented at an intertrial interval of 30-60 s (mean, 45 s). Each trial consisted of the CS, followed by a 500-ms stimulus free trace period, then the US. For the first 4 days of training, these animals also received a US-alone test trial on trials 11, 32, 53, and 74. These test trials were used to determine if the CS information on CS-US trials augmented the unconditioned behavioral response, similar to reflex facilitation described by Gormezano et al. (1983). The reflex facilitation analyses used only the CS-US trials and US-alone trials where the onset of blinks occurred following the US onset.

Pseudoconditioned control rabbits also received acclimation followed by daily sessions consisting of 80 CS-alone trials and 80 US-alone trials (intertrial interval of 15-30 s). Trace and pseudoconditioning sessions lasted ~1 h and consisted of the same number of CSs and USs. A significant blink response produced a change in voltage 10 ms in duration and 4 SD of the mean baseline voltage. For trace-conditioned animals, conditioned responses (CRs) were defined as significant eyeblinks that were anticipatory of the US, that is, occurring within the 300-ms period prior to US onset. This ensured that the conditioned responses in the trace-conditioned group were not due to sensitization. However, conditioned responses for the pseudoconditioned group were defined as significant blinks occurring within the 600-ms period following CS onset. Trace-conditioned animals were considered learners if the percentage of CRs exceeded 50% on any day, and animals were considered learning impaired (nonlearners) if the daily CR percentage never exceeded 25%. Animals that failed to show 25% CRs by the 10th day of training were administered an additional 5 days of training. All of the nonlearners failed to exceed 25% CRs during these 15 days of training.

Single-neuron recording

Single neurons were recorded from rabbits during trace eyeblink conditioning using surgically implanted nonmoveable electrodes that were cemented in place. This minimized the amount of drift or movement of the electrode tip during a single training session. Each implanted recording electrode consisted of six channels with a total diameter of ~80 µm. Each channel was a Teflon-coated tungsten microwire (18 µm diam bare; 25-µm diam coated). The channels were bonded tightly together in parallel with epoxylite to form a 25-µm center-to-center spacing. The tip of the electrode was cut at a 45° angle with sharp scissors to maximize the number of single neurons recorded from the electrode. During recording, two-wire stereotrode combinations were selected from the implanted probe that provided the largest and most heterogeneous ensembles of single neurons (2-12 neurons). This is an enhanced version of the stereotrode technique, which has been shown to allow large numbers of single neurons to be recorded and separated with much greater accuracy than single electrodes (McNaughton et al. 1983). Similar ensemble techniques have been used for recording tightly clustered groups of single neurons from a single probe (e.g., Apkarian et al. 2000).

Single-neuron analog signals were amplified (10,000 times), filtered (band-pass, 300 Hz to 10 kHz), and collected with a DT 2821 Data Translation board (Marlboro, MA) attached to a 200-MHz Pentium computer, which sampled each channel at 30 kHz. Single-neuron data were collected 1 s prior and 2.75 s following CS onset using software from DataWave Technologies (Longmont, CO). The software recorded 1.5-ms epochs of data whenever a single neuron discharged a definable action potential. The action potentials of each of the different single neurons recorded on an electrode were separated off-line using a template-matching program developed in our laboratory. This software allowed template windows to be defined for each single neuron's characteristic waveform. All action potential waveforms that fell within the boundaries of a single template-window belonged to an individual single neuron. The template window could account for any unique segment along the single-neuron waveform, and the template window could be minimized anywhere along the waveform to exclude other electrophysiological data that did not fit the exact shape of an individual single neuron. The template could be widened to account for drifting and changes in the single-neuron waveform across a recording session. The software compared the waveform of every single-neuron discharge to all other electrophysiological activity on the probe. All action-potential waveforms on a probe were also compared visually to ensure that the characteristic waveform of each individually defined single neuron was different from the waveforms of all other defined single neurons on the probe. This conservative approach ensured that the ensembles recorded from each probe were made up of unique single neurons that could be accurately followed throughout a single training session. Individual hippocampal pyramidal cells have been reported to exhibit complex spikes within a burst of activity where the action potentials of a single neuron decrease in height (Ranck 1973). Based on parameters described by Quirk and Wilson (1999), the software was able to track patterns of activity that might represent complex spike activity. This prevented the complex spike activity of a single neuron from overlapping with more than one individually defined single neuron.

Single-neuron activity was analyzed from a single day of training only if the ensemble of single neurons on a stereotrode remained consistent throughout the entire 1-h training session. This ensured that the electrode did not drift during the recording session, which might produce an overlap of activity from more than one single neuron. However, it is important to note that the configuration of single neurons on a stereotrode changed from 1 day of recording to the next in almost all cases. A conservative approach was used, and neurons were treated as the same neuron on consecutive days only if the same template window yielded the same configuration of single neurons on one probe. This does not rule out the possibility that small drifts in the electrode between recording sessions may have allowed new configurations of single neurons to form which included one or two of the neurons from the previous recording day.

Following spike separation, an average waveform was computed for each single neuron to determine if the neuron was a pyramidal or theta cell. Action-potential widths were calculated from each average waveform as the peak time minus the valley time. Pyramidal cells were separated from theta cells using measurements of action-potential width and background firing rate. Using criteria similar to those described by Fox and Ranck (1981), cells with a spike duration 0.3 ms and background firing rate <6 Hz were classified as pyramidal cells, and cells with a spike duration <0.3 ms and a background firing rate 6 Hz were classified as interneurons.

Analyses

All statistical analyses were performed with the aid of Microsoft Visual Basic routines developed in our laboratory and Minitab statistical software v10.0 (State College, PA). Analyses of background firing rate were performed by calculating the mean single-neuron discharge rate prior to the delivery of each trial used in training. Changes in single-neuron action potential firing were measured using standard t-test scores. For each neuron, standard test scores were computed for time periods from 200 to 600 ms in duration following either the CS or US to capture discrete short-latency increases or decreases in activity. The standard test scores were computed by subtracting the number of action potentials in the period preceding CS onset from the number of action potentials in the period following CS or US onset. The difference calculated for each period was divided by the sample standard deviation during baseline. The test score measures could then be averaged across trials or across neurons or to compare changes in activity across days of training.

A hierarchical clustering analysis was used to classify the response profile of each single neuron recorded during trace eyeblink conditioning. This is a multivariate statistical technique where each observation, or single neuron in this case, is assumed to be a unique cluster for the first step of the analysis. In the first step, the two most similar observations are joined together to form a cluster. In the second step, either a third observation joins the two previous observations or two of these observations join together into a different cluster. The analysis proceeds by joining together observations that are most similar, each step results in one less cluster than the preceding step. The result of the analysis is a limited number of unique clusters of observations, where the observations within each cluster share the most similarities. Theoretical and mathematical descriptions of this analysis are supplied by Everitt and Dunn (1991), and Lance and William (1967). Briefly, the clustering analysis in this study used a Pearson distance matrix where the distance between observations i and k are as follows with v_j the variance of variable j
The clustering analysis used a Ward linkage method (Ward 1963). At each step of the analysis, clusters are merged so as to produce the smallest increase in the sum-of squares error term. Each trace conditioned single neuron was hierarchically clustered using the discharge activity summed across a single 80-trial session into 300-ms bins. Specifically, the clustering analysis used the following five measures of activity to classify single neurons: the bin preceding the onset of the CS, the two bins following the onset of the CS, and the two bins following the onset of the US.

The hierarchical clustering analysis sought to classify cells based on their activity during trace conditioning. Single-neuron responses recorded during pseudoconditioning were considered a unique class of responses because they were not recorded during any CS-US trials; therefore responses from these cells were not included in the initial clustering analyses. Therefore, a final categorization of all trace and pseudoconditioned cells was conducted by matching the average trial activity of each cell to a template of each category produced by the original hierarchical clustering analysis. The template for each category was created by summing the 80 trials of each trace conditioned single neuron's daily activity into 100-ms bins and averaging each summed bin across all neurons within a category. An average profile of single-neuron activity was then created for each cell recorded during trace conditioning using 100-ms bins. For pseudoconditioning this was done by summing across the pseudoconditioned trials of each single neuron's daily activity into 100-ms bins for the following periods: the baseline period prior to CS onset on CS-alone trials, the 600-ms period following CS onset on CS-alone trials, and the period following the onset of the US on US-alone trials. This created a daily average of pseudoconditioned activity to the CSs and USs that could then be matched to one of the trace conditioned templates from the clustering analysis. A dependent t-test was used to measure the difference between the 100-ms bins of the template and the daily average of a single neuron's pseudoconditioned activity. Both trace and pseudoconditioned cells were then matched into the trace conditioned category with the least difference in activity.

To examine the relationship of single-cell firing patterns within local ensembles, cross-correlation histograms (adapted from Perkel et al. 1967), also called cross-correlograms (CCRs), were constructed from the activity of pairs of cells recorded from each single stereotrode. These CCRs were constructed by measuring the time interval (resolution of 1 ms) between each action potential-discharge of one cell of the pair with each discharge of the other cell. Intervals were measured for all spike discharges between the two cells, and the interval durations were plotted in a CCR using binwidths ranging from 10 to 100 ms. The height of each bin within the CCR was adjusted by dividing it by the total number of action potential discharges recorded from both of the cells within the pair. This correction of CCR height normalized for larger measures of coincident activity that might be due to faster firing rates of one cell or both cells within the pair and thus allowed for the level of coincidence to be compared between cell pairs. The magnitude of correlated activity or coincidence was compared between cell pairs by examining the height of the two tallest adjacent bins of the adjusted CCR. This method for calculating a cross-correlation coefficient was in part based on methods used previously by Eggermont (1992) and Roy and Alloway (1999). The goal of this study was to examine differences in coincident firing between groups. Therefore this normalized cross-correlation measure was not used for judging the significance of a CCR but rather for comparing the magnitude of coincident activity between groups.

Although the number of cells recorded from an individual stereotrode ranged from 2 to 12 cells, analyses were limited to the stereotrode-ensembles that contained 4 cells. The magnitude of coincident activity was calculated for each possible cell pair within a stereotrode ensemble. This approach allowed us to examine how one specific neuron within the stereotrode ensemble coordinated single-neuron discharges with the other cells within the ensemble. An average magnitude of coincident activity was also calculated for each stereotrode ensemble, and because each cross-correlation coefficient was normalized for firing rate, the magnitude of ensemble-correlated activity could be compared between groups.

Statistical tests used ANOVAs with a general linear model. Repeated-measures ANOVAs were used to examine behavioral data across days of training. However, factorial rather than repeated-measures ANOVAs were used to measure changes in single-neuron activity across days of training because only 26 individual neurons could be tracked from 1 day of training to the next. Significant interactions were subjected to follow-up one-way ANOVAs using the MSerror term from the main ANOVA. Significant main effects were followed up with Newman-Keuls post hoc tests. An alpha of 0.05 was required for all significant analyses.

Histology

Marking lesions were placed at the tips of all electrodes by passing DC current (25 µA) for 20 s. Animals were overdosed with pentobarbital sodium and perfused transcardially with saline (0.9% NaCl) followed by 10% formaldehyde. Brains were then frozen, sectioned coronally (50 µm thick), mounted on albumin/gelatin-coated slides, and stained with neutral red. A light microscope (×25 and ×50) was then used to locate electrode tips.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Histology

All of the animals used in this study had electrode tips placed directly in the pyramidal cell body layer of the CA1 area of the dorsal hippocampus. Data from two animals were not included in analyses because the electrode tips were placed 100-200 µm below the pyramidal cell layer.

Trace eyeblink conditioning

Greater than half of the aged animals showed (n = 5) few if any CRs, similar to the young pseudoconditioned group (n = 5). These aged nonlearners failed to show daily CR percentages 25% even though behavioral training was administered for 15 days. The remaining aged animals (n = 4) showed normal acquisition very similar to that of the young animals (n = 6). Figure 1A shows the mean percentage of trace eyeblink CRs across the 10 days of training for these four groups: young, aged learners, aged nonlearners, and young pseudoconditioned. A repeated measures analysis of these data revealed a significant group × days interaction, F(27, 144) = 4.8, P = 0.0001. Follow-up tests showed that the young group exhibited a greater percentage of CRs than the aged nonlearners and the pseudoconditioned group during the last 3 days of training. These tests also revealed that the aged learners showed a greater percentage of CRs than the aged nonlearners and the pseudoconditioned group during the last 4 days of training. The level of pseudoconditioning shown in this study was similar to the level of pseudoconditioned eyeblink responses shown in other rabbit studies (for review, see Gormezano et al. 1983).

View larger version (22K): [in this window] [in a new window]   Fig. 1. A: daily mean percentage of trace eyeblink conditioned responses (CRs) for young rabbits (5-6 mo; n = 6), aged animals (30-33 mo) that acquired trace eyeblink CRs (n = 4), aged animals that did not acquire trace eyeblink CRs (n = 5), and young animals that received pseudoconditioning (n = 5). Bars indicate SE. Greater than half of the aged animals showed few if any CRs, similar to the young pseudoconditioned group, while the remaining aged animals showed normal acquisition very similar to that of the young animals. B: voltage measures for eyeblink responses (closure upward) on conditioned-unconditioned stimulus (CS-US) trials (black) and US-alone trials (gray) averaged across the first 4 days of training and across all of the animals from each group. All groups showed similar unconditioned responses to the US-alone trials, and the 3 trace-conditioned groups showed similar reflex facilitation on CS-US trials. Pseudoconditioned animals received no CS-US trials. Each voltage measure is 1,750 ms in duration. Calibration bars = 1 V and 300 ms.

Each animal received four US-alone test trials during each of the first 4 days of training. Figure 1B shows that the average unconditioned behavioral responses (non-CR responses) on the CS-US trials and the US-alone trials during the first 4 days of training were identical for the three groups that received trace eyeblink conditioning. This figure shows that the average unconditioned responses of the three trace groups were augmented similarly on CS-US trials. This effect is similar to reflex facilitation described by Gormezano et al. (1983) and Weisz and McInerney (1990). Thus the CS and US information was able to enter the CNS of all trace conditioned groups; however, the aged nonlearners were still unable to make the necessary associations between these stimuli for learning to occur. The three trace-conditioned groups also showed behavioral responses to the US-alone trials that were similar to the pseudoconditioned group. One-way ANOVAs compared the unconditioned behavioral responses on CS-US trials during the first 4 days of training for the three trace-conditioned groups on the following measures: maximum amplitude, area under the curve, response onset latency, and peak voltage latency. None of these ANOVAs revealed significant effects. One-way ANOVAs also compared all four groups on the unconditioned behavioral responses on US-alone trials during the first 4 days of training. The only significant ANOVA was on the measure of peak voltage latency, F(3, 17) = 3.57, P = 0.04. The follow-up test showed that the pseudoconditioned animals showed a shorter peak latency compared with the trace-conditioned animals. The result of this analysis was most likely due to the pseudoconditioned animals having received numerous additional US-alone trials prior to the US-alone test trials.

Average single-neuron response profiles

A total of 2,159 pyramidal cells were recorded from the four groups of animals. Each cell was recorded throughout one entire session from at least 1 of the 10 days of training shown in Fig. 1. A total of 16 other cells were excluded from the analyses because they exhibited firing characteristics similar to interneurons (see Fox and Ranck 1981). The young, aged learners, aged nonlearners, and pseudoconditioned groups each contained 693, 563, 361, and 542 pyramidal cells, respectively. Almost all of these single neurons were tracked for only a single day of training. Thus on average, 54 cells were recorded for each of the 10 days of training for each group. The channel configuration of the electrodes was optimized prior to each day of recording to sample the largest ensemble of CA1 neurons. Furthermore, the number and firing pattern of the pyramidal cells on each stereotrode changed significantly from one day of training to the next. As a result in most cases it was difficult to track isolated pyramidal cells from one day of training to the next. However, 26 of the 2,159 neurons were tracked for two successive days of training. It is possible that many additional neurons were not tracked as the same single neuron because of significant changes in the waveform and firing pattern from one day of training to the next. Regardless, this does not detract from the goal of this study, to describe the different CA1 single-neuron response types that occur during trace eyeblink conditioning and to understand how the encoding of learning-related information within these ensembles is affected in aged animals that are unable to learn the trace eyeblink task.

The single-neuron activity of each of the four groups was characterized initially by examining the average single-neuron response pattern during the early and late stages of training. This allowed for an examination of single-neuron activity before and after the acquisition of the trace eyeblink conditioned response. Figure 2 shows perievent histograms of spike discharge activity averaged across all of the single neurons recorded from the first 3 and last 3 days of training. Each of the groups exhibited an average increase in single-neuron activity following the CS and US both early and late in training. However, early in training the single-neuron increase following the CS and US was much smaller in the aged nonlearners. A factorial ANOVA with a group and days factor (early vs. late) was applied to the change score measures of single-neuron activity during the 400-ms period following the onset of the CS. The interaction was significant, F(3, 1269) = 4.23, P = 0.006, and follow-up ANOVA and post hoc tests revealed that the single-neuron increase in activity following the CS was significantly smaller for the aged nonlearners compared with the other groups early in training, F(3, 640) = 4.20, P = 0.007. A factorial ANOVA for the 400-ms period following the US revealed a similar interaction, F(3, 1258) = 3.55, P = 0.014. Follow-up ANOVA and post hoc tests revealed that the single-neuron increase in activity following the US was significantly smaller for the aged nonlearners compared with the other groups early in training, F(3, 640) = 5.50, P = 0.001. These analyses show that the average increase in activity of CA1 pyramidal cells is reduced early in training in aged animals that are not able to acquire trace eyeblink conditioning.

View larger version (41K): [in this window] [in a new window]   Fig. 2. Perievent histograms (10-ms bins) of spike discharge activity averaged across all of the single neurons recorded from the first and last 3 days of training. For each histogram, action potentials (spikes) from each cell were summed across the CS-US trials of a single training session, then averaged across cells. The duration of each histogram is 3,750 ms, and the duration of the prestimulus period prior to the onset of the CS is 1,000 ms. Increases in single-neuron discharge to the CS and US were smaller in the aged nonlearners compared with the other groups early in training.

Heterogeneous stereotrode ensembles

To further understand the altered CA1 pyramidal cell activity in the aged nonlearners, the ensemble profile of single neurons recorded on each of the stereotrodes was examined. One or two stereotrodes were implanted in each animal, and the ensemble profile on each stereotrode almost always changed from one day to the next. Therefore each of the 10 days of recording was treated as a new stereotrode ensemble. This approach allowed 354 different stereotrode ensembles to be analyzed in this study. The number of neurons isolated from each stereotrode ensemble recorded in CA1 ranged from 2 to 12, with an average of 6.5 neurons/stereotrode. The group averages of neurons/stereotrode were nearly identical, as confirmed by a nonsignificant one-way ANOVA. The smallest average was obtained from the aged learners [6.2 ± 2.2 (SD) neurons/stereotrode], and the largest average was from the young group (6.6 ± 2.6 neurons/stereotrode). This suggests that group differences in single-neuron activity were not due to differences in the sampling or the morphological distribution of cells.

Mean change score measures of single-neuron firing were used to examine the profile of single-neuron responses on the stereotrodes implanted in CA1. These measures provided a descriptive analysis of the increases (excitatory) and decreases (inhibitory) in single-neuron activity that make up the profile of each stereotrode ensemble. Examination of the daily mean change in single-neuron activity following the onset of the CS revealed that 79% of the 354 stereotrode recordings used in this study had at least one excitatory and one inhibitory cell within their ensemble. A similar examination of the activity following the CS revealed that 48% of the stereotrodes had at least two excitatory and two inhibitory cells within the ensemble, and 25% had at least three excitatory and three inhibitory cells within the ensemble. Examination of the daily mean change in single-neuron activity following the onset of the US revealed similar heterogeneous stereotrode profiles. The measures of activity following the US revealed that 75% of the stereotrodes had at least one excitatory and one inhibitory cell, 42% had at least two excitatory cells and two inhibitory cells, and 22% had at least three excitatory and three inhibitory cells within an ensemble. Thus the stereotrode ensembles consisted of heterogeneous profiles of single-neuron activity. There was no group difference in the amount of heterogeneity on the stereotrodes. These analyses of inhibition and excitation suggest that the overall group analyses from Fig. 2 may be due in large part to a specific class or classes of cells. Therefore the aim of our subsequent analyses was to characterize the different profiles of single-neuron activity that occurred during trace eyeblink conditioning.

Figure 3 shows an example of a heterogeneous ensemble of 10 individual single neurons recorded simultaneously from a single stereotrode from a young animal on a single day of training (day 5 in this example). A total of 12 neurons were recorded from this stereotrode ensemble; however, only 10 of the neurons are displayed because 2 of the neurons never fired during the 3.75-s window of analysis shown in the figure. These neurons were recorded simultaneously from the tip of an implanted stereotrode that never moved during training. The perievent time histograms show that at least three of these neurons (neurons 4, 6, and 9) show on average excitatory increases in activity to the US; however, three other neurons in this ensemble show inhibitory decreases in activity (neurons 2, 5, and 7) immediately after US onset. Figure 4 demonstrates how multiple-neuron average analyses overlook a significant part of the information about the ensemble. The summed activity of the individual single neurons from Fig. 3 (shown in Fig. 4A), similar to a multiple-unit record, is largely dictated by the activity of the number 4 single neuron in Fig. 3. This ensemble effect could account for the early findings of Berger and Thompson (1978), who used multiple-unit recordings and found that CA1 cells show large bimodal increases in activity to the stimuli used in eyeblink conditioning. However, our previous work (McEchron and Disterhoft 1997) and especially the present investigation showed that very few single neurons exhibit this pattern of activity. Figure 4, A and B, also suggests that a small number of neurons with faster firing rates can have a significant influence on multineuron or ensemble analyses. Figure 4C shows the number of action potentials summed across all of the neurons shown in Fig. 3 excluding neurons 4 and 6. Figure 4C demonstrates that neurons with very low firing rates (e.g., neuron 6) can also influence the overall shape of multineuron or ensemble analyses. Thus single-neuron response patterns of activity in CA1 are very heterogeneous and ensemble activity can be influenced by the excitatory or inhibitory pattern of activity of an individual single-neuron response pattern (e.g., Fig. 4C) or the overall firing rate of a single neuron (e.g., Fig. 4, A and B).

View larger version (50K): [in this window] [in a new window]   Fig. 3. Histograms from 10 individual pyramidal cells simultaneously recorded during a single day of training (day 5) from a single nonmovable stereotrode implanted in a young rabbit. A total of 12 neurons were recorded from this stereotrode ensemble; however, only 10 of the neurons fired during the 3.75-s window of analysis shown here. Each histogram shows the number of action potentials in 10-ms bins summed across 80 trials. The duration of each histogram is 3,750 ms, and the duration of the baseline period prior to CS onset equals 1,000 ms. This figure demonstrates how heterogeneous response profiles were obtained for individual single neurons recorded from a local ensemble of a single stereotrode. At least 3 of these neurons (neurons 4, 6, and 9) show excitatory increases in activity to the US; however, 3 other neurons in this ensemble show inhibitory decreases in activity (neurons 2, 5, and 7) immediately after US onset. Heterogeneous ensembles similar to this example were common. For example, the daily mean change in single-neuron activity following the onset of the CS revealed that 79% of the 354 stereotrode recordings used in this study had 1 excitatory and 1 inhibitory cell within their ensemble, and 48% had at least 2 excitatory and 2 inhibitory cells within the ensemble.

View larger version (28K): [in this window] [in a new window]   Fig. 4. A: histogram shows the number of action potentials in 10-ms bins summed across all ten of the neurons shown in Fig. 3. B: histogram shows the number of action potentials summed across all of the neurons shown in Fig. 3 excluding neuron 4. C: histogram shows the number of action potentials summed across all of the neurons shown in Fig. 3 excluding neurons 4 and 6. This figure demonstrates how multiple neuron average analyses overlook a significant part of the information about the ensemble. The summed activity of the individual single neurons from Fig. 3, similar to a multiple-unit record, is largely dictated by the activity of the faster firing single neuron 4. This demonstrates that a small number of neurons with faster firing rates can have a significant influence on multineuron or ensemble analyses. However, B and C demonstrate how neuron 6 with a very low firing rate can also influence the overall shape of multi-neuron or ensemble analyses. Thus single-neuron response patterns of activity in CA1 are very heterogeneous and ensemble activity can be influenced by the excitatory or inhibitory pattern of activity of an individual single neuron or the overall firing rate of a single neuron. The duration of each histogram is 3,750 ms, and the duration of the baseline period prior to CS onset equals 1,000 ms. D: average eyeblink response (closure upward) on day 5 from the young rabbit from which these 10 cells were recorded. Although this animal showed only 5% CRs on this day of training, during the last 5 days of training this animal acquired >50% CRs. Calibration bar equals 0.5 V.

Heterogeneous single-neuron response profiles

To further understand these heterogeneous ensembles, it was necessary to go beyond the simple excitatory/inhibitory dimension of activity and describe single neurons based on the change in activity across the entire trace eyeblink conditioning trial. Thus a multidimensional approach was used to understand the learning-related activity of specific classes of cells within the ensemble. An hierarchical clustering analysis categorized each single neuron based on the average firing pattern of activity during the trace eyeblink conditioning trial. Specifically, this analysis compared the daily single-neuron firing-pattern of each cell (summed across the 80 CS-US trials of a single day of training) to the daily activity of all of the other cells recorded during trace eyeblink conditioning. The analysis then determined the similarity of each of the single-neuron firing patterns to classify these single-neuron response profiles into a limited number of unique categories.

These clustering analyses sought to classify cells based on their activity during trace conditioning, therefore cells from pseudoconditioned animals were not included in the initial analysis. The hierarchical clustering analysis was performed on the daily average of activity of 1,367 single neurons recorded from the young and aged animals that received trace eyeblink conditioning. The original hierarchical analysis produced nine clusters; however, analysis of the daily average of activity of each of these clusters revealed that several clusters had identical average patterns of activity for the trace eyeblink trial. Therefore several identical clusters were combined into the next highest cluster within the hierarchical tree. This process revealed six unique patterns of activity. Each trace conditioned cell was then matched to a template that corresponded to one of the six patterns of activity. The average daily responses of these cell categories are shown in Fig. 5, A-F. An additional 250 cells, shown in Fig. 5G, were not included in the clustering analysis because they fired three spikes or less during all 80 trace eyeblink conditioning trials of a single day of training. These cells were considered a unique category of cells because they tended not to fire after the onset of the CS and US. Furthermore, when these cells were included in the original clustering analysis, they were categorized along with cell types in C and D, which had distinctively higher background firing rates compared with the cell type in G. The firing pattern of the cell type in G, shown in Fig. 5 appears to be inhibitory; however, the possibility cannot be ruled out that this effect could be a random placement of a very small number of spike discharges.

View larger version (128K): [in this window] [in a new window]   Fig. 5. Average perievent histograms (10-ms bins) for seven different pyramidal cell response profiles A-G recorded from young and aged rabbits during trace conditioning. A hierarchical clustering analysis applied to 1367 pyramidal cells recorded during trace conditioning revealed 6 different types of single-neurons responses (A-F) during the trace eyeblink conditioning trial. An additional 250 cells (histogram G) were not included in the clustering analysis because they fired 3 spikes during all 80 trace eyeblink trials of a single day of training. For each histogram, action potentials (spikes) from each cell were summed across a single training session, then averaged across cells. The duration of each histogram is 3,750 ms, and the duration of the baseline period prior to CS onset equals 1,000 ms. The dendogram (bottom right) shows the distance, sometimes called similarity, between the single-cell response categories. For example, F cell responses are dissimilar to A-E. Distance units are from the Pearson distance matrix used in the hierarchical analysis. The percentage of cells belonging to each single-cell response type is listed below the dendogram for each of the trace-conditioned groups.

The dendogram in Fig. 5 shows the resulting distance of the single-cell response categories A-F that resulted from the clustering analysis. These distances can also be viewed as the amount of dissimilarity between the single-cell categories (see Everitt and Dunn 1991). Categories A and B are similar to each other and C and D are similar to each other, but A-D are more similar to each other compared with E. Finally, category F appears to have the least similarity with the other categories in the analysis. Below the dendogram is the percentage of cells from each of the trace conditioned groups belonging to each single-cell response category A-F. For each single-cell response category, a ² test compared the percentage of cells falling into each of the trace conditioned groups. There was no group difference in the percentage of cells falling into cell categories A-F; however, the aged nonlearners had a significantly higher percentage (22%) of cells in the single-cell response category G compared with the other trace-conditioned groups, the young (9.2%) and aged learners (17.2%), ² = 6.125, P < 0.05. Analyses also revealed that there was no significant difference in the expression of any one of the cell types between the early and late stages of training (first and last 5 days of training).

Several interesting observations can be made from the cellular response categories shown in Fig. 5. Type F single neurons made up a very small percentage of the neurons encountered in this study. An example histogram of an individual type F neuron is shown as neuron 4 in Fig. 3. Although there were a small percentage of these neurons encountered, 28% of the stereotrode ensembles contained at least one type F single neuron. This suggests that a small percentage of these faster firing neurons could have a significant impact on the pattern of the multiunit ensemble activity in CA1, as shown by neuron 4 in Fig. 3. Although type F neurons fired much faster than the other single-neuron response types, their average firing rate (i.e., 0.75 ± 0.65 Hz) was still well below the firing rate of theta cells (i.e., 6-8 Hz) as described by Ranck (1973). Almost half of the single neurons showed an excitatory increase in activity to the US (48.3%; types A, C, E, and F), while only a quarter of the single neurons show some type of excitation to the CS (25.4%; types A, B, E, and F). This suggests that many of the cells in CA1 encode information about the US, whereas far fewer cells play a role in encoding CS information. In contrast to these increases, more than two-thirds of the cells show some type of inhibitory decrease in activity during trace eyeblink conditioning (67.2%; types C, D, and G). This shows that the majority of the cells in CA1 do not transmit bimodal increases in unit activity, rather the cells in CA1 appear to be composed of heterogeneous ensembles or networks of individual cells with unique roles in transmitting information. It is also interesting to note that 26 neurons were tracked for two consecutive days of training. The response categories (A-G) of 21 of these 26 neurons remained the same for two successive days, suggesting that pyramidal cells have relatively specific and inflexible response patterns with respect to trace conditioning. Although the response type remained consistent for these 21 cells, the magnitude of the excitatory or inhibitory response did change from one day to the next.

Single neurons from pseudoconditioned animals were not included in the clustering analysis to ensure that the single-neuron response profile of each category was defined by only trace-conditioned activity. However, it was necessary to compare trace-conditionted and pseudoconditioned activity for specific categories of single neurons. Therefore, both trace and pseudoconditioned cells were matched to templates corresponding to each of the trace-conditioning categories produced by the original clustering analysis. This allowed cell types from all groups to be analyzed in subsequent analyses.

Baseline levels of single-neuron activity

Analyses of background firing rate were performed by calculating the mean single-neuron discharge rate during the 1-s period prior to the delivery of each trial used in training. Overall, the mean background firing rate of the pyramidal cells was very low, 0.27 ± 0.4 Hz, with a range from 0 to 2.9 Hz. A factorial ANOVA was used to compare baseline firing of the seven single-cell categories shown in Fig. 5 and the four groups of animals in this study. The analysis revealed a significant interaction of group × cell type, F(18, 2086) = 15.91, P = 0.0001. Follow-up one-way ANOVAs and post hoc tests revealed that the difference in baseline activity between the groups was similar for the cell categories A-D. Figure 6 shows that for these specific cell types, the baseline single-neuron firing of the aged nonlearners was significantly less than the other groups. There was no difference in baseline activity between the young and aged learners. However, the pseudoconditioned baseline activity was greater than all of the other groups. The enhanced baseline single-neuron firing in the pseudoconditioned group was most likely the result of the large number of discrete trials and the shortened intertrial interval used during daily training. Although the trace and pseudoconditioned animals received the same number of CSs and USs during a daily session, the pseudoconditioned animals received twice as many discrete trials (CS- and US-alone trials) and had half the amount of time between these trials. Similar group differences in baseline firing can also be seen prior to the CS onset in the histograms in Fig. 2.

View larger version (38K): [in this window] [in a new window]   Fig. 6. Single-cell response categories A-D, shared a similar pattern of group differences in background firing rate. For these specific cell types, the baseline single-neuron firing rate of the aged nonlearners was significantly less than the other groups. There was no difference in baseline activity between the young and aged learners; however, the pseudoconditioned baseline activity was greater than all of the other groups. Baseline activity was measured as the number of action potentials in the 1,000 ms prior to CS onset. Asterisks and brackets indicate groups with significantly different baseline levels of firing at P < 0.05.

Changes in single-neuron activity

Standardized change scores of single-neuron firing were used to examine daily increases and decreases in single-neuron activity for each of the single-cell response categories depicted in Fig. 5. The change scores revealed that the aged nonlearners showed unique increases and decreases in single-neuron activity for specific cell types. A factorial ANOVA was conducted using the change scores computed for the 600-ms period following the onset of the CS. This analysis revealed a significant interaction of group × cell type, F(18, 2131) = 12.14, P = 0.0001. Follow-up analyses for the activity following the CS revealed that group effects were specific to the single-cell response categories B-D, F, and G. A similar factorial ANOVA was conducted using the change scores computed for the 600-ms period following the onset of the US. This analysis revealed a significant interaction of group × cell type, F(18, 2131) = 18.45, P = 0.0001. Follow-up analyses for the activity following the US revealed that group effects were specific to the single-cell response categories B, C, F, and G. Figures 7 and 8 show the changes in activity following the CS and US across days of training for the cell types that showed significant group differences in activity.

View larger version (23K): [in this window] [in a new window]   Fig. 7. Change in single-neuron activity for cell types B and F. These 2 cell types were the excitatory cell types that showed significant group differences in single-neuron firing. This figure shows for cell types B and F, the daily mean change in activity for the 600-ms periods following the CS and US combined. Changes in activity were analyzed for both of these periods together because the pattern of activity across the groups and days was nearly identical. Standard scores measured the change in single-neuron firing during these periods on each trial, and were then averaged across all trials and neurons from a day of training. Top: early in training type B neurons from the young and aged learners showed greater increases in activity following the CS and US compared with the same type of neurons from the aged nonlearners and pseudoconditioned animals. Bottom: type F neurons from all of the groups that received trace eyeblink conditioning showed greater increases in activity following the CS and US compared with the same type of neurons from the pseudoconditioned animals.

View larger version (27K): [in this window] [in a new window]   Fig. 8. Daily mean change in single-neuron activity following the CS and US for the inhibitory cell types C, D, and G. Top: cell types C, D, and G from the aged nonlearners showed smaller decreases in single-neuron firing following the onset of the CS compared with the other groups. Bottom: cell types D and G from the aged nonlearners showed smaller decreases in single-neuron firing following the onset of the US compared with the other groups. Standard scores measured the change in single-neuron firing for the 600-ms periods following the onset of the CS (top) or US (bottom) on each trial. Standard scores were then averaged across all trials and neurons from a day of training.

Figure 7 shows the change in single-neuron activity for cell types B and F, the excitatory cell types that showed significant group differences in single-neuron firing. This figure shows the change in activity for the 600-ms periods following both the CS and US combined. Changes in activity were analyzed for both of these periods together because the pattern of activity across the groups and days was nearly identical for the periods following the CS and US. The top panel of this figure shows that early in training type B neurons from the young and aged learners showed greater increases in activity following the CS and US compared with the same type of neurons from the aged nonlearners and pseudoconditioned animals. A [2 (early, late) × 4 (group)]-factorial ANOVA conducted on the change in type B single-neuron responses in the top panel of Fig. 7 revealed a significant interaction, F(3, 132) = 4.10, P = 0.008. Follow-up tests revealed that during the first half of training the type B cells from the young and aged learners showed a greater increase in activity following the CS and US compared with the cells from the aged nonlearners and pseudoconditioned animals. The bottom panel of Fig. 7 shows that type F neurons from all of the groups that received trace eyeblink conditioning exhibited greater increases in activity following both the CS and US compared with the same type of neurons from the pseudoconditioned animals. A two [early, late ×4 (Group)]-factorial ANOVA conducted on the type F single-neuron responses in the lower panel of Fig. 7 revealed a significant group main effect, F(3, 93) = 13.18, P = 0.0001. Follow-up tests revealed that the type F neurons from all of the groups that received trace eyeblink conditioning showed greater increases in activity following the CS and US compared with the same type of neurons from the pseudoconditioned animals.

Figure 8 shows the change in single-neuron activity for cell types C, D, and G, the inhibitory cell types that showed significant group differences in single-neuron firing. This figure shows that the inhibitory cell types from the aged nonlearners exhibited smaller decreases in activity following the CS (top) and US (bottom) compared with the other three groups. A factorial ANOVA conducted on the single-neuron responses of type C, D, and G cells in the top panel of this figure revealed a significant group main effect, F(3, 1461) = 4.04, P = 0.007. Post hoc tests revealed that type C, D, and G cells from the aged nonlearners showed a smaller decrease in single-neuron activity following the CS compared with the other groups. A factorial ANOVA conducted on the type D and G single-neuron responses in the bottom panel of Fig. 8 revealed a significant group main effect, F(3, 941) = 9.66, P = 0.0001. Post hoc tests revealed that type D and G cells from aged nonlearners showed a smaller decrease in single-neuron activity following the US compared with the other groups.

Ensemble single-neuron activity

Cross-correlation analyses were used to examine the coordinated firing of the different cell types (A-G) and to compare the coordinated firing of the single-neuron ensembles among the four groups in this study. Coordinated activity of cell pairs within a stereotrode ensemble was measured using CCRs. Only pairs recorded from the same stereotrode were analyzed with this method. The bins of each CCR were normalized according to the number of action potentials recorded from both cells in the pair. Figure 9, A and B, shows normalized CCRs for two example cell pairs from a single five-neuron steretrode ensemble. This ensemble was obtained from a young animal on the 6th day of training. The CCRs from this stereotrode ensemble are similar to many others recorded in this study, where the activity of one neuron was correlated with the activity of specific neurons within the stereotrode ensemble. For example, Fig. 9A shows a cell type F single neuron, which exhibited a large amount of coincident activity at a very short time delay (near 0) with a cell type D single neuron simultaneously recorded within the ensemble. Figure 9B shows that the same cell type F single neuron exhibited a much smaller amount of coincident activity with a cell type A single neuron simultaneously recorded from the same ensemble.

View larger version (39K): [in this window] [in a new window]   Fig. 9. A and B: example cross-correlograms for 2 cell pairs from a single 5-neuron stereotrode ensemble recorded from a young animal on the 6th day of training. The time delay of coincident events is shown in 10-ms bins, and the height of each bin was normalized by dividing it by the total number of spike discharges recorded from both neurons within each pair. A: a cell type F single neuron (from the cell types shown in Fig. 5), that exhibited a large amount of coincident activity at a very short time delay (near 0) with a cell type D single neuron simultaneously recorded within the ensemble. B: that the same cell type F single neuron exhibited a smaller amount of coincident activity with a cell type A single neuron simultaneously recorded from the same ensemble. C shows the average magnitude of coincident single-neuron activity for each of the four groups. Coincident activity was measured for each of the neuron-pair combinations on a single stereotrode ensemble using CCRs. The CCRs were constructed using binwidths ranging from 10 to 100 ms (x axis). Ensemble mean correlation coefficients were calculated using the tallest 2 adjacent bins from each of the CCRs in the ensembles. Analyses revealed that the coincident activity of the stereotrode ensembles of the aged nonlearners was less than the other groups. D shows ensemble measures of coincidence similar to those in C, but for only the pairs of neurons with a background firing rate of <0.67 Hz. The same group effects were obtained with this analysis, suggesting that the lower coincident activity of the aged nonlearners was not due to faster firing neurons in the other groups. E: ensemble measures of coincidence similar to those in C, but for only the pairs of neurons with similar overall firing rates (<25% difference). The same group effects were seen with this analysis suggesting that these measures of correlated activity were not due to the faster firing rate of one neuron within the neuron pairs. F: ensemble measures of coincidence for each of the cell types described in Fig. 5. Analyses revealed no interaction between group and cell type suggesting that the group effect was similar for each of the cell types. These cell type means were obtained from the measures of coincident activity using the data from the 50-ms bin in A. Bars indicate SE.

The magnitude of correlated activity, or coincidence, was compared between cell pairs on each stereotrode by examining the number of events in the two tallest adjacent bins of the adjusted CCR. This allowed a mean cross-correlation coefficient to be calculated for each cell type (A-G) within a stereotrode ensemble. For example, a stereotrode ensemble that contained the cell types A-C would allow one mean for cell type A to be computed for this stereotrode ensemble, which included the cell pairs A-B and A-C. Means would also be computed for cell types B and C on this stereotrode. Cross-correlation coefficients were computed using 10 different CCR binwidths ranging from 10 to 100 ms. The means were examined using a factorial ANOVA (group × cell type), which was applied individually to the means computed for each binwidth (i.e., 10, 20 ··· 100 ms). These ANOVAs all revealed significant effects of group and cell type but no interaction. The significant group effects for the 10- and 100-ms bin ANOVAs were F(3, 982) = 5.56, P = 0.001 and F(3, 982) = 6.85, P = 0.0001, respectively. The significant cell type effects for the 10- and 100-ms bin ANOVAs were F(6, 982) = 6.70, P = 0.0001 and F(6, 982) = 15.77, P = 0.0001, respectively. The ANOVAs for the other binwidths revealed similar levels of significance but were not reported for purposes of simplicity. Figure 9C shows the mean adjusted correlation coefficient for each of the groups at each of the binwidths. Follow-up tests revealed that the young, pseudoconditioned, and aged learners all showed significantly greater levels of coordinated activity within the stereotrode ensembles compared with the aged nonlearners. The effect was similar for all binwidths. The differences between the young group and the aged learners and the young group and the pseudoconditioned group were not significant but did approach significance at all binwidths.

Figure 9D shows the same data as in Fig. 9C excluding all cell pairs with a background firing rate 0.67 Hz. This criterion represented the mean pyramidal cell background firing rate (0.27 Hz) plus 1 SD (0.4 Hz). Excluding these faster firing cells eliminated 27% of the total 21,200 cell pairs. Nearly identical group effects are shown in Fig. 9, C and D. This provides evidence that the group effect in Fig. 9C was not due to a subset of cells with faster firing rates. The factorial ANOVAs for the data in Fig. 9D were similar to those in Fig. 9C with significant effects of group and cell type but no interaction. The significant group effects for the 10- and 100-ms bin ANOVAs were F(3, 900) = 5.90, P = 0.001 and F(3, 900) = 8.11, P = 0.0001, respectively. The significant cell type effects for the 10- and 100-ms bin ANOVAs were F(6, 900) = 3.83, P = 0.0001 and F(6, 900) = 12.41, P = 0.0001, respectively. Similar to the data in Fig. 9C, follow-up tests for the data in Fig. 9D revealed that the young, pseudoconditioned, and aged learners all show significantly greater levels of coordinated activity within the stereotrode ensembles compared with the aged nonlearners. The effect was the same for all binwidths. The differences between the young group and the aged learners, and the young group and the pseudoconditioned group were not significant but did approach significance at all binwidths.

Figure 9E shows the same data as in Fig. 9C excluding all cell pairs where one cell of the pair fired >25% more action potentials than the other cell. Excluding the pairs with a discordant number of spikes eliminated 81% of the total number of cell pairs. Nearly identical group effects are shown in Fig. 9, C and E. Not shown are analyses that also excluded all cell pairs with one cell of the pair firing >50% more action potentials than the other cell. This excluded only 62% of the total cell pairs and produced group effects identical to those in Fig. 9E. The factorial ANOVAs for the data in Fig. 9E were similar to those in Fig. 9C with significant effects of group and cell type but no interaction. The significant group effects for the 10- and 100-ms bin ANOVAs from Fig. 9C were F(3, 554) = 4.43, P = 0.004 and F(3, 554) = 5.28, P = 0.001, respectively. The significant cell type effects for the 10- and 100-ms bin ANOVAs were F(6, 554) = 3.83, P = 0.037 and F(6, 554) = 10.73, P = 0.0001, respectively. Similar to the data in Fig. 9C, follow-up tests for the data in Fig. 9E revealed that the young, pseudoconditioned, and aged learners all showed significantly greater levels of coordinated activity within the stereotrode ensembles compared with the aged nonlearners. The effect was the same for all binwidths. Because the same effects were revealed with the exclusion of the pairs with discordant firing rates, this provides evidence that the group effect in Fig. 9C was not due to a small number of cells within the ensemble with a much faster firing rate. In summary, Fig. 9, C-E, shows that aged animals that are unable to learn trace eyeblink conditioning show a significantly lower level of coincident pyramidal cell firing within local ensembles in CA1. This effect was not due to differences in background firing between the groups or heterogeneous firing rates within the ensembles.

Analyses of the coincident activity from Fig. 9, C-E, revealed a significant effect of cell type (types A-G in Fig. 5) but no interaction of group and cell type. Means for each of the cell types are shown in Fig. 9F. The pattern of correlation coefficients between each of the cell types shown in Fig. 9F was similar for all of the analyses in Fig. 9, C-E. The means in Fig. 9F were taken from the CCR measures computed using 50-ms bins from the data in Fig. 9C. The ANOVA for the 50-ms bin in Fig. 9C revealed significant effects of group, F(3, 982) = 6.60, P = 0.0001, and cell type, F(6, 982) = 13.99, P = 0.0001. Follow-up tests on the cell type effect revealed that coordinated activity of cell type F with the other cells in the ensemble (pairs that contained 1 cell type F) was greater than the coordinated activity of all other cell type combinations. The follow-up tests also showed that pairs that contained cell type G showed less coordinated activity than all other cell type combinations. Furthermore, pairs that contained cell type C or D showed more coordinated activity than pairs that contained cell type E. These analyses suggest that specific cell types will provide more coincident activity to the ensemble processing of information. For example, cell type F will contribute more coincident events to an ensemble compared with cell type E. This may be due, in part, to the density pattern of action potentials. However, it is very important to note that the deficient coincident activity in the aged nonlearners was not due to a lack of one or several cell types, as analyses from Fig. 5 demonstrated that the cell types (except for G) are expressed evenly among the groups. The analyses of the correlation coefficients in Fig. 9, C-E, all revealed a similar group pattern for each of the cell types (A-G). Finally, the group effects for coincident activity in Fig. 9, C-E, were not due to group differences in firing rate because measures of coincidence were corrected by the number of action potentials discharged by a cell pair, and the Pearson correlation between total number of action potential discharges and the corrected correlation coefficient measure was negligible, r = 0.119.

It is important to note that outlier mean correlation coefficients from two stereotrode ensembles from the young group were not included in the analyses in Fig. 9, C-E, because these means were 4 SD of the mean. Analyses for the data in Fig. 9, C-E, were performed at 10 different binwidths ranging from 10 to 100 ms, and identical effects were revealed at all binwidths. These multiple analyses were performed to ensure that group effects were not due to binwidths that might select for cell pairs with a specific firing rate or pattern. Many ANOVAs were required for this approach, which could inflate the probability of a type I error; however, all significant group effects were below the alpha level of 0.005. The measures of correlated single-neuron firing used in this study were not aimed at examining the overall magnitude of the correlations, nor were they used to examine the significance of the correlated firing of an individual pair of single neurons. On the other hand, the measures of coordinated firing used in this study were designed specifically to correct for faster firing rates of single neurons and to be able to compare the coordinated firing of the different groups.

These cross-correlation analyses suggest that specific cell types provide more coincident activity to the ensemble processing of information. However, regardless of the cell types within the ensemble, aged animals that were unable to learn trace eyeblink conditioning showed a significantly lower level of coincident pyramidal cell firing within local ensembles in CA1.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The results of the present study showed that more than half of the aged rabbits (30-34 mo of age) were unable to acquire trace eyeblink CRs while the remaining aged animals learned this task as well as young animals. Similar to all the other groups of animals, aged nonlearners showed larger unconditioned eyeblink responses on paired trials compared with US-alone trials, suggesting that the animals were able to sense the auditory-CS information. Increases in CA1 pyramidal cell firing to the CS and US were diminished in the aged animals that were learning impaired. The local ensembles of pyramidal cells from all groups exhibited heterogeneous types ^<