Nonpharmacological amelioration of age-related learning deficits: The impact of hippocampal θ-triggered training
- Yukiko Asaka*,
- Kristin N. Mauldin,
- Amy L. Griffin†,
- Matthew A. Seager‡,
- Elizabeth Shurell, and
- Stephen D. Berry§
-
Communicated by Richard F. Thompson, University of Southern California, Los Angeles, CA, July 29, 2005 (received for review February 3, 2005)
Abstract
Age-related learning deficits are often attributed to deterioration of hippocampal function. Conversely, a well studied index of hippocampal activity, the θ rhythm, is known to enhance hippocampal plasticity and accelerate learning rate in young subjects, suggesting that manipulations of θ activity might be used as a means to counteract impairments related to the aging process. Here, young and older rabbits were given eyeblink conditioning trials either when exhibiting hippocampal θ (θ+) or regardless of hippocampal activity (yoked control). Although, as expected, older-yoked control animals showed a learning deficit, the older θ+ group learned as fast as young controls, demonstrating that aging deficits, at least in eyeblink classical conditioning, can be overcome by giving trials during episodes of hippocampal θ activity. The use of several learning criteria showed that the benefits of hippocampal θ occur in multiple phases of learning that may depend on different cognitive or motor processes. Whereas there was a benefit of θ-triggered training in both age groups during the early phase of acquisition, the enhancement persisted in older animals, peaking during later performance. These findings have implications for theories of age-related memory deficits and may contribute to the development of beneficial treatments.
Clinical applications derived from basic neuroscience are a crucial source of ideas and methods to address disorders of learning and memory, including amelioration of the growing economic and personal impact of age-related learning deficits. One promising line of inquiry includes recent observations of oscillatory neurobiological potentials that have earned them increased study as correlates, perhaps mechanisms, of cognitive processes including perception and learning (1–6). Much current work is devoted to theoretical and empirical studies of these phenomena, especially neocortical γ (40 Hz; ref. 6) and hippocampal γ and θ (3–8 Hz; refs. 7–9). Behavioral studies of Pavlovian or classical eyeblink (EB) conditioning have yielded a productive convergence of empirical observations and theoretical interpretations between human and animal studies, especially in the involvement of essential cerebellar and modulatory hippocampal systems (see refs. 10 and 11 for review). This growing neurobiological foundation derives, in part, from precise manipulation and/or localized recording of electrophysiological events in awake, undrugged animals during the acquisition and performance of widely used and well-defined cognitive tasks. In brief, EB conditioning in the trace paradigm starts with onset of a conditioned stimulus (CS, e.g., tone), and CS offset occurs before the unconditioned stimulus (US, e.g., airpuff) onset, creating a period when there is no stimulus present between the CS and US (trace period). Conditioned responses (CRs) occur to the CS and are adaptively timed to precede and overlap the US. Especially important for this study, the trace paradigm has been shown to depend on the integrity of hippocampal processes (see below).
With respect to hippocampal function, animal conditioning studies show that hippocampal θ activity (2–8 Hz oscillatory field potentials) accurately predicts behavioral acquisition rate (1, 5, 12) and that hippocampal dysfunction (e.g., via lesions or pharmacological manipulations) significantly impairs learning (13–21). Moreover, age-related learning deficits are highly correlated with age-related biophysical changes of hippocampal pyramidal neurons, including reduced postsynaptic neuronal excitability (22–24) and reduced hippocampal long-term potentiation (25–27). This relationship between hippocampal activity and learning is strongest for training conditions that depend on intact hippocampal function, such as spatial memory tasks (25, 28, 29) or the trace EB paradigm (17, 30–32). These learning paradigms are often referred to as hippocampus-dependent learning tasks. In the aging literature, trace EB conditioning seems to be highly indicative of functional senescence, revealing the impact of aging on behavioral and neurobiological learning systems in both animal and human subjects (refs. 33–39; see also refs. 40–42). The consistent and impressive covariation between hippocampal and behavioral processes in the trace EB paradigm suggests an important modulatory role for the hippocampal/medial temporal system in associative learning.
In a demonstration of the role of endogenous hippocampal oscillations in conditioning, we have previously shown the beneficial effects of training in which the initiation of trials was contingent on the presence of particular patterns of hippocampal slow wave activity (2, 5). In brief, when online spectral analysis ensured that training trials were given only when the hippocampus was generating field potentials in the θ range, young rabbits learned two to four times faster than when trials were given when this activity was absent (non-θ trials). Control groups that were trained irrespective of brain wave activity learned at an intermediate and variable rate. In the current experiment, we investigated whether such hippocampal θ-triggered training could prevent the well-documented impairment of trace EB classical conditioning in older rabbits.
Theoretical models of classical conditioning cite evidence for the participation of different cognitive/behavioral processes during several phases of learning. For example, a prominent behavioral model (43, 44) posits the existence of two phases of the learning process: (i) CS–US contingency detection and (ii) adaptive response timing (related to Pavlov's “inhibition of delay”; typically completed as the subject approaches asymptotic responding). Our progress in the neurobiology of classical conditioning can be enhanced by more precise delineation of behavioral learning processes after, e.g., selective manipulation of relevant brain systems. Therefore, in addition to four traditional learning criteria, this project also included, for the first time in an EB conditioning study, two criteria based on a recently developed state space model (45) that can be used to specify mathematically the earliest phase of learning and the initial appearance of asymptotic responding. These new criteria provide the advantage of being more robust with respect to within-animal variability and to differences in learning trajectories among groups, yet are likely to measure the same cognitive processes as the more traditional criteria. Thus, combining neurobiological manipulations (i.e., triggering training trials contingent on the presence of θ) with a more comprehensive analysis of behavioral learning phases, we report here that θ-triggered training accelerated learning such that older animals trained during hippocampal θ were indistinguishable from young control animals. In addition, the benefit of θ persisted through later phases of learning only in older rabbits, demonstrating a continuing role for the hippocampus in adaptive performance by aging animals.
Materials and Methods
Subjects. Twenty-four New Zealand white rabbits (Oryctolagus cuniculus) were provided by Myrtle's Rabbitry (Thompson Station, TN). The young group consisted of 12 4- to 6-mo-old animals (weight, 2.94–4.00 kg). The older group consisted of 12 22- to 36-mo-old animals (weight: 4.36–5.29 kg). All animals were kept in individual cages with a 12:12-h light/dark cycle and ad lib access to food and water.
Surgery. All animals underwent sterile surgical procedures under anesthesia with i.m. administration of ketamine (50 mg/kg) and xylazine (10 mg/kg). A 6-0 nylon thread was sutured into the surface of the left nictitating membrane and tied to make a small loop. Electrodes (Epoxylite insulated, except 30- to 50-μm exposed tip, 00 insect pins) were placed in CA1 (stratum pyramidale) of the dorsal hippocampus according to stereotaxic coordinates (4.5 mm posterior and 5.5 mm lateral to bregma). The final dorsal-ventral coordinate was determined electrophysiologically (≈3–3.5 mm ventral to dura). Dental cement was used to secure the electrodes in place, and the incision was sutured.
Procedure. After 5 d of postsurgical recovery, each animal was placed in an acrylic (Plexiglas) restrainer and the conditioning chamber for a 45-min adaptation session. The day after the adaptation day, animals in each age group (young and older) were randomly assigned to one of two groups. One group was given EB conditioning trials only when exhibiting hippocampal θ activity (θ+), and the other group received the same number of trials per day with the same inter-trial intervals, regardless of brainwave frequencies (yoked control). This design made a total of four groups: young θ+, young-yoked control, older θ+, and older-yoked control. Because each θ+ animal received trials depending on the state of its hippocampus, the number of trials per session and intertrial intervals were different for each animal. In the EB conditioning literature, manipulation of inter-trial intervals and the number of trials per session have been shown to affect acquisition rate (46). Thus, it was crucial to control for these effects with the use of yoked control groups.
Hippocampal activity from the electrode tip was filtered (0.5–22 Hz) and digitized at 100 Hz by labview software (National Instruments, Austin, TX), which sampled data for 640-ms scrolling time intervals, updated every 160 ms, performed a fast Fourier transform, and computed a power spectrum. A θ ratio was calculated from the output of the power spectrum (3.5- to 8.5-Hz activity in the numerator and 0.5- to 3.5-Hz and 8.5- to 22-Hz activity in the denominator). These values were validated in previous experiments through visual inspection of raw data (see refs. 2 and 5). The software was programmed to trigger trials in the θ+ group when the θ ratio exceeded 1, for three consecutive calculations (960 ms total pretrial duration). Regardless of treatment group, all animals received trace EB conditioning (“500-ms trace”; ref. 32) for 1.5 h/d. An 85 dB, 1-kHz tone CS was presented for 100 ms, followed by a 500-ms stimulus-free trace period, and a 100-ms air puff to the cornea as an US. All of the responses that occurred after CS onset and before US onset were considered CRs. Animals were trained until they reached a behavioral criterion of 80% CRs or for at least 15 d for young and 30 d for older animals.
Data Analysis. Neural analysis. In addition to the online spectral analysis of hippocampal field potentials, 1 min of free-running slow wave activity was sampled (100 Hz) before each training session and was recorded for off-line analysis. As in previous studies relating pretraining θ to learning rate (1, 12), these samples were filtered (1–25 Hz; Krohn-Hite model 3700 filter, Brockton, MA) and converted to digital values (Keithley System 570) before being analyzed with a zero-crossing program (asystant plus, Asyst Software Technologies, Rochester, NY), a method that determines the predominance of selected frequency bands independently of the amplitude of the local field potential. To examine whether pretraining θ contributes to any possible effects of θ triggering during trials, a percentage of θ was computed as in a previous study for each rabbit's pretraining sample on the first day of training [(2–8 Hz)/(0–2 Hz and 8–22 Hz) × 100, see ref. 1]. The percentage of θ was compared among the treatment groups with a single-factor ANOVA.
Behavioral analysis. Trials to reach several behavioral criteria were used to subdivide the learning process according to Prokasy's (43, 44) two-phase model. In brief, phase 1 represents detection of the contingency between the CS and the US, and phase 2 represents refining and adapting the response to the CS so that the CR is timed to anticipate arrival of the US. Traditionally, the end of phase 1 is signaled by the occurrence of early CRs (e.g., the fifth or, here because of variability in older rabbits, the 10th CR). Phase 2 has most often been estimated by the performance of eight CRs in any nine consecutive trials, evidence that asymptotic responding has begun. Additionally, we looked at a criterion used in the aging literature that represents a clear plateau of stable performance (80% CRs in a single session; see ref. 32). A recently proposed state space model (45) outlines a more mathematically grounded means of assessing these same points on the learning curve. A 5% baseline criterion (estimating the end of phase 1) was defined as the trial on which the ideal observer was 95% certain that the animal would give CRs at a rate above chance (5%) for the rest of the experiment. The 0.05 plateau criterion was the point, after passing the 5% criterion, at which the change in the probability of a correct response was <5% for 10 consecutive trials, thus estimating the onset of asymptotic responding. The relationships among these criteria can be seen on a sample learning curve (see Fig. 1 A). For analysis, the fifth CR, 5% baseline, 10th CR, eight/nine CRs, 0.05 plateau, and 80% criteria were numbered criteria 1–6, respectively. The total number of trials the animals took to reach each of the criteria was compared by using a mixed design ANOVA (two treatment conditions × six criteria × two ages). When there was a significant two-way interaction, post hoc independent (between subjects factors) or paired-samples (within subjects factors) t tests were performed. This analysis allowed for the assessment of early and late phases of learning as a function of age and the benefit of θ-triggered training for each factor. To illustrate the trajectory of learning, each group's mean correct response probability over trials was plotted according to a normalized baseline in which the x axis was adjusted according to the group's average number of trials to the 80% CRs criterion (see Fig. 1B).
Learning curve and trajectories. (A) Representative trace of a behavioral learning curve for one of the animals in the older θ+ group used in the study. This graph was generated by a program based on the state space model of learning by Smith et al. (45). The black line is the learning curve estimate, and the gray lines are the associated 90% confidence intervals. The black circles at the top represent trials where a CR was given, and the black circles at the bottom represent trials where no CR was given. To allow comparison of the traditional criteria and the criteria implemented in this study, the number of trials it took this animal to reach each of the criteria was calculated and plotted on this curve. The within-subject variance and the number of trials taken to reach criterion 2 (5% baseline) is shown above the figure. (B) Learning trajectories by group. Each curve represents the average probability of a correct response across trials, normalized according to each group's mean trials to the end of training (criterion 6).
Histology. To verify the electrode locations, all animals were killed at the end of the experiment by an overdose of sodium pentobarbital (Euthasol, 0.22 mg/kg i.v.). Before the lethal injection, animals were lightly tranquilized with 1–1.5 cc of anesthesia (ketamine, 50 mg/kg and xylazine, 10 mg/kg i.m.), and the two electrode locations were marked by direct current lesion (200 μA for 10 s). Animals were perfused intracardially with 0.9% saline followed by a 10% formalin solution. Coronal sections were sliced from the frozen brains (70 μm each) and embedded on gelatin-coated slides. Prussian blue staining was done for iron deposits from the electrodes and safranin was a counterstain for cell bodies.
Results
Control Groups Show Aging Deficits. As shown in Figs. 1B and 2A, comparison of the young- and older-yoked control groups revealed substantial differences in acquisition rate as a function of age. A mixed-design ANOVA revealed a significant age by criterion interaction [F (5, 35) = 3.46, P = 0.012], and post hoc Student t tests showed that rabbits in the older-yoked control group took significantly more trials to reach later criteria (eight/nine CRs, 0.05 plateau, and 80% CRs criteria: criteria 4–6) than the young-yoked control group (eight/nine CRs criterion, 658.6 ± 109.2 vs. 222.3 ± 75.31 trials, P = 0.008; 0.05 plateau criterion, 607.6 ± 95.34 vs. 229.5 ± 82.55 trials, P = 0.011; and 80% CRs criterion, 808.8 + 129.23 vs. 2,671.75 ± 92.85 trials, P = 0.007). However, mean number of trials to early criteria between older- and young-yoked control groups did not differ significantly for the fifth CR (P = 0.27) or 10th CR criteria (P = 0.13), although they differed marginally for the 5% baseline (P = 0.057; criteria 1, 3, and 2, respectively).
Trials to criterion. (A) Average number of trials to all of the criteria for the young θ+, young-yoked control, older-yoked control, and older θ+ animals. Error bars indicate + SEM. (B) Average differences between the number of trials to each criteria for the θ+ and yoked control groups for the older and young animals. Error bars indicate + SEM.
θ-Contingent Trial Presentation Ameliorates Impairment. The θ-triggered training accelerated acquisition in both young and older animals, with the benefit greatly enhanced in the older animals (see Figs. 1B and 2). Specifically, the older θ+ animals reached all of the criteria except the fifth CR significantly faster than their yoked controls (for 5% baseline: 141.5 ± 35.38 vs. 299.83 ± 53.28 trials, P = 0.006; for 10th CR: 176.5 ± 37.72. vs. 273.17 ± 56.19 trials, P = 0.04; for eight/nine CRs: 251 ± 61.33 vs. 658.6 ± 109.15 trials, P = 0.0045; for 0.05 plateau: 314 ± 96.16 vs. 607.6 ± 95.34 trials, P = 0.011; and for 80% CRs: 409.8 ± 131.07 vs. 808.8 ± 129.23 trials, P = 0.011; student t tests with one tail). In addition, when comparing the older θ+ group to the young-yoked group, the expected age-related difference in acquisition rate was virtually eliminated by making trial presentation for older rabbits contingent on the presence of hippocampal θ (for fifth CR: 146.33 ± 35.68 vs. 142 ± 63.17 trials, P = 0.94; for 5% baseline: 141.5 ± 35.38 vs. 140.25 ± 76.74 trials, P = 0.98; for 10th CR: 176.5 ± 37.72 vs. 164.75 ± 70.78 trials, P = 0.87; for 8/9 CRs: 251 ± 61.33 vs. 222.25 ± 75.31 trials, P = 0.76; for 0.05 plateau criterion: 314 ± 96.16 vs. 229.5 ± 82.55 trials, P = 0.54; and for 80% CRs: 409.8 ± 131.07 vs. 261.75 ± 92.85 trials, P = 0.42; student t tests with one tail, see Fig. 2 A). The learning trajectory plot (Fig. 1B) supports these conclusions in displaying substantially different curves as a function of age and θ treatment. Intertrial intervals for older θ+ and young θ+ were not significantly different [F (1, 8) = 2.884, P = 0.128], indicating that EB conditioning did not alter the frequency of θ episodes preferentially in one group.
Young θ+ animals learned slightly faster when compared to their yoked controls for the first four criteria (for fifth CR: 47.5 ± 11.66 vs. 142 ± 63.17 trials, P = 0.093; for 5% baseline: 70.5 ± 40.37 vs. 140.25 ± 76.74 trials, P = 0.076; for 10th CR: 79.25 ± 23.56 vs. 164.75 ± 70.78 trials, P = 0.095; and for eight/nine CRs: 144.5 ± 58.29 vs. 222.25 ± 75.31 trials, P = 0.02; student t tests with one tail). In fact, when the criteria were clustered according to early and later phases (criteria 1, 2, and 3 vs. 4, 5, and 6), there was a highly significant interaction of age, θ treatment, and phase, F (1, 26) = 8.344, P = 0.008. Post hoc comparisons substantiate a significant benefit of θ triggering in the early phase for both young, t (11) = 3.287, P = 0.007, and older groups, t (16) = 3.77, P = 0.002. In contrast, the later phase showed no θ-triggering benefit in young, t (11) = 0.899, P = 0.388, but a continuing and clear benefit in the older group, t (15) = 6.619, P = 0.001.
Hippocampal Pretraining θ Did Not Differ Between Groups. To compare these learning effects with previous findings on pretraining θ (2–8 Hz) and acquisition rate (1), the amount of θ was computed for a 1-min pretraining sample on the first day of training. The average percentage of time in θ for each treatment group was not statistically different from each other, F (3, 14) = 1.29, P = 0.32. Thus, pretraining levels of θ, which have been used to predict acquisition rate in the past (1, 12), could not account for the observed differences in learning rates between θ-triggered and yoked groups.
Discussion
The major finding of this study is the virtually complete elimination of an age-related learning deficit in a hippocampus-dependent learning task (i.e., 500-ms trace EB conditioning) by training older rabbits during episodes of hippocampal θ activity. Such results demonstrate an important modulatory role for hippocampal activity, especially oscillations at the θ frequency, in the acquisition of trace EB conditioning and suggest a central role for the hippocampal system in more complex forms of classical conditioning that may be related to declarative memory (47, 48). Furthermore, the results from later phases of the learning process showed that older animals display learning trajectories that differ from young animals in taking much longer to respond adaptively and consistently to the CS and its temporal relationship to the US. The effects of θ triggering in young animals replicate earlier findings in trace EB conditioning (2), and the finding that older-yoked control rabbits took significantly more trials than the young-yoked control rabbits to reach the later phase criteria is consistent with previous aging studies (32, 49). Such results suggest that an age-related EB conditioning deficit may not be a simple, overall disruption of associative learning, but may include specific impairment of later phases of response adaptation and stable performance. Specifically, measures of late-training stable performance (80% CRs criterion) were greatly enhanced by hippocampal θ in older rabbits and were virtually unaffected in young (see Fig. 2B). This phase is long past the acquisition of the CS–US contingency and may be related to loss of cholinergic basal forebrain neuron-mediated cognitive processes such as attention (50, 51) or to hippocampal influences on motor performance itself (e.g., the sensorimotor integration model of hippocampal function, ref. 52). This latter possibility has an interesting connection to models of age-related cerebellar dysfunction (53) in which later phases of EB conditioning (i.e., response adaptation and timing) are susceptible to cerebellar disruption, perhaps Purkinje cell loss. Our data suggest that the θ state of the hippocampus may provide important supporting modulation for cerebellar plasticity and, especially in older animals with late phase learning deficits, can alleviate the impairment.
In terms of age-related changes in hippocampal cellular properties, extensive anatomical and electrophysiological studies report that although basic membrane and cellular characteristics of dentate gyrus granule cells and CA1 and CA3 pyramidal cells do not change with age (e.g., the resting membrane potential and input resistance; refs. 54 and 55), old animals show an increase in intracellular calcium concentration, resulting from an increase in the number of voltage-gated Ca2+ channels (ref. 56; see ref. 57 for extensive review), and a decrease in postsynaptic neuronal excitability, caused by an enhanced postburst after-hyperpolarization (see ref. 24 for review). Presenting training trials during θ episodes in aging rabbits may compensate for these age-related cellular changes by creating optimal conditions for synaptic plasticity. Green and Woodruff-Pak (58) have emphasized the potential age-related loss of cholinergic inputs rather than deterioration of the hippocampus per se in EB conditioning deficits. It is interesting to note that these age-related changes in hippocampal cellular properties do not seem to affect the amount of pretraining θ, which did not differ between groups, a finding consistent with a previous report (59). It is possible that neurons in older animals may have a different relationship between pretraining baseline θ and stimulus-induced neural and behavioral plasticity than young animals display.
We cannot precisely specify from the current study whether there are other important, learning-related cellular processes that may have accompanied hippocampal θ-triggered trials. However, there is a well documented positive relationship between θ band oscillations and hippocampal synaptic plasticity, including the induction of long-term potentiation (26, 60, 61) and learning-related increases in unit firing rates (2, 12, 62; see also ref. 63). Moreover, there is evidence showing that deficits in hippocampus-dependent tasks are correlated with age-related impairments of hippocampal synaptic transmission and long-term potentiation (25, 64–66). Thus the question of precisely how hippocampal θ-triggered training benefited the older animals in the current study requires further study of how θ is related to such learning-induced neural changes.
Several influential computational and neurobiological models of hippocampal plasticity during learning posit a critical role for θ rhythms (3, 4, 8, 9, 67, 68). The present study demonstrates the efficacy of θ-triggered training as an analytic tool in assessing the modulatory effect of the hippocampus on learning in the young and aging brain. Combined with the use of several criteria to assess behavioral learning, our methods have revealed differences in the timing of hippocampal involvement as a function of age. When considered in light of age- and θ-related differences in learning, these results suggest that distinct cognitive processes, perhaps related to different brain systems, may be involved in at least two separable phases of the associative learning process. The suggestion that degeneration of the cerebellar cortex contributes to aging impairment (53) and that Prokasy's phase 2 represents cognitive processes that can be mapped onto cerebellar circuitry (69) correspond nicely to our results. Other models posit regional differences in cerebellar involvement in early vs. late conditioning (70). This potential relation of learning phases to age-impaired cerebellar substrates is consistent with our finding that hippocampal modulation later in training is especially important in older rabbits. In this way, timing the conditioning stimuli to coincide with hippocampal θ oscillations may participate in the elimination of age-related learning deficits.
Acknowledgments
The current research was partially supported by a Sigma Xi grant-in-aid (to Y.A.). This research was submitted to the Department of Psychology, Miami University, in partial fulfillment of the Doctor of Philosophy degree of Y.A.
Footnotes
-
↵ § To whom correspondence should be addressed. E-mail: berrysd{at}muohio.edu.
-
↵ * Present address: Department of Psychology and Neuroscience, Bowdoin College, 6900 College Station, Brunswick, ME 04011.
-
↵ † Present address: Boston University Center for Memory and Brain, Boston, MA 02215.
-
↵ ‡ Present address: Department of Alzheimer's Research, Merck Research Laboratories, West Point, PA 19486.
-
Author contributions: Y.A. and S.D.B. designed research; Y.A. performed research; Y.A., K.N.M., and E.S. analyzed data; and Y.A., K.N.M., A.L.G., M.A.S., E.S., and S.D.B. wrote the paper.
-
Abbreviations: EB, eyeblink; CS, conditioned stimulus; US, unconditioned stimulus; CR, conditioned response.
- Copyright © 2005, The National Academy of Sciences