Homeostatic shutdown of long-term potentiation in the adult hippocampus

Abstract

Homeostasis is a key concept in biology. It enables ecosystems, organisms, organs, and cells to adjust their operating range to values that ensure optimal performance. Homeostatic regulation of the strength of neuronal connections has been shown to play an important role in the development of the nervous system. Here we investigate whether mature neurons also possess mechanisms to prevent the strengthening of input synapses once the limit of their operating range has been reached. Using electrophysiological recordings in hippocampal slices, we show that such a mechanism exists but comes into play only after a considerable number of synapses have been potentiated. Thus, adult neurons can sustain a substantial amount of synaptic strengthening but, once a certain threshold of potentiation is exceeded, homeostatic regulation ensures that no further strengthening can occur.

Neurons possess a variety of mechanisms for homeostatic regulation of ion concentrations, osmotic conditions, oxygen consumption, and other metabolic processes. However, neurons not only have to monitor and control basic metabolic parameters but also have to regulate neural activity carefully. Homeostatic compensatory mechanisms are well documented during the development of the nervous system (1, 2). Such mechanisms should remain important after functional synaptic connections are established and neuronal circuits have attained a mature state, because the adult nervous system continuously needs to adapt to environmental changes.

The cellular underpinnings of learning and memory include mechanisms that alter the efficacy of synaptic connections (3, 4). Such mechanisms, although necessary for the storage of information, run the risk of strengthening the inputs to an individual neuron to an extent that it, or an entire neuronal network, moves beyond their optimal operating range. To ensure proper functioning, neural networks have to balance two seemingly conflicting requirements: change and stability. A dynamic equilibrium between these opposing but complementary processes is especially important in a highly plastic brain structure, such as the hippocampus, that has been implicated in different forms of memory (5, 6).

Hebb's postulate of activity-dependent synaptic plasticity (7) was experimentally confirmed with the discovery of long-term potentiation (LTP) in the hippocampus (8). Subsequent work has posited that, to ensure stability of the neuronal network, this very ability of neurons to undergo changes in synaptic strength should itself be subject to activity-dependent regulation (9), as now experimentally demonstrated with the phenomenon of “metaplasticity” (10). Metaplasticity is a form of homeostasis in which the history of previous neuronal activation influences the direction and degree of synaptic plasticity elicited by a given stimulus.

Hippocampal pyramidal cells of the CA1 region each receive thousands of excitatory inputs with the potential for activity-dependent increases in synaptic efficacy (11). Without mechanisms for limiting total synaptic strength, their physiological balance might be compromised and, if too many of their inputs were potentiated, selective differences in synaptic weights would diminish, reducing the information-storage capacity of neuronal circuits (12, 13). We therefore hypothesized that CA1 neurons might possess homeostatic mechanisms to regulate total synaptic strength. We set out to investigate whether saturating potentiation in one set of CA1 hippocampal synapses might diminish or even prevent further potentiation of a different set of synapses on the same neurons. Because it is unclear what proportion of synapses must be potentiated before homeostatic regulation comes into play, we considered both electrical tetanization of the Schaffer collateral input to area CA1 and a chemical potentiation procedure that activates a high proportion of excitatory afferents throughout CA1. Although this might be perceived as more “artificial,” chemical potentiation utilizes induction mechanisms similar to that of electrical tetanization (see, e.g., refs. 14–17) and critically should realize the maximal potentiation that individual neurons can sustain. Our series of extra- and intracellular experiments demonstrate that synaptic activation can cause a heterosynaptic shut down of LTP, but the operating range of hippocampal neurons appears to be exceptionally broad.

Results

Probing Homeostasis on a Population of Neurons: Normal LTP Despite Saturation of Synaptic Inputs.

The experiments were performed in rat hippocampal slices by using classical NMDA-receptor-dependent CA3–CA1 LTP. Field excitatory postsynaptic potentials (fEPSPs) were measured in the stratum radiatum of the CA1 region, whereas stimuli were elicited in two independent pathways (see Methods and Supporting Text, which is published as supporting information on the PNAS web site). Multiple tetanizations of one pathway were applied to induce saturated LTP, with saturation operationally defined as the inability to induce further LTP on the same pathway with the same tetanus (saturated pathway, 156 ± 6% 1 hr and 144 ± 9% 6 hr after the last tetanus, n = 12, Fig. 1a). The second, or test, pathway was monitored continuously to ensure that it experienced no LTP when the saturated pathway was tetanized. One hour later, the test pathway was probed for its ability to undergo LTP. The classical principle of input specificity indicates that induction of LTP on one pathway is independent of that on another. But if a homeostatic mechanism were to limit the total amount of potentiation, prior saturation might block or at least reduce LTP in the other subset of synapses. We found no evidence for this assumption. Robust LTP was observed in the test pathway of a magnitude no smaller than that seen in control slices without previous saturation (LTP in test pathway, 138 ± 5%; LTP in control slices, 136 ± 6%; both 1 hr after LTP induction, P > 0.10; Fig. 1 a and b). Long-term recordings showed that late LTP was also unaffected by the earlier saturating stimulus (6 hr after tetanus: 122 ± 6% in the test pathway, 121 ± 7% in the control slices, P > 0.10; Fig. 1b).

Fig. 1.

Saturation of one pathway did not block or reduce LTP in another independent pathway. (a) Multiple tetanization (open arrows) of one Schaffer collateral pathway induced saturated and long-lasting LTP (saturated pathway, 156 ± 6% 1 hr and 144 ± 9% 6 hr after the last tetanus, n = 12). One hour later, a second independent pathway in the same slice (test pathway) received a single tetanus (filled arrow) to probe its capability to undergo LTP. The test pathway showed robust long-lasting LTP as well (138 ± 5% 1 hr and 122 ± 6% 6 hr after LTP induction, n = 12). (b) The amount of LTP in the test pathway did not differ from LTP observed in control slices without previous saturation (136 ± 6% 1 hr and 121 ± 7% 6 hr after LTP induction, n = 10, P > 0.10). Arrow, tetanus to the test pathway or to control slices. (c) Example of an intracellular saturation experiment. Open arrows, tetani to the saturated pathway; filled arrow, tetanus to the test pathway. For the plot, amplitudes of two consecutive EPSPs were averaged; action potentials are indicated by vertical bars. (a–c) Representative fEPSPs were taken before and after LTP induction at the time points indicated in the graphs (averaged over five consecutive sweeps). Error bars indicate ± SEM. (d) Correlation analysis for eight intracellular saturation experiments. The amount of LTP in the saturated pathway is plotted against the amount of potentiation in the test pathway 1 hr after the (last) tetanus. The homeostatic hypothesis predicts a negative correlation, but this was not observed (correlation coefficient r² = 0.08, P = 0.49).

At first glance, these observations argue against homeostatic regulation of synaptic efficacy, but they are not fully conclusive. One reason is that the populations of CA1 neurons stimulated by the two pathways might not be sufficiently congruent. Many neurons will receive input from both pathways (these are the neurons of interest), but there might also be neurons that receive input from the test pathway only. These would, of course, exhibit normal LTP in response to the test tetanus. Their contribution to the fEPSPs might therefore mask a potential homeostatic effect. We therefore performed a second series of experiments, recording from individual CA1 neurons with sharp intracellular electrodes.

Probing Homeostasis on a Single Neuron: Normal LTP Despite Saturation of Synaptic Inputs.

The experimental stimulating protocol was very similar to that used for the extracellular recordings (see Supporting Text). We found that again, despite robust LTP on the saturated pathway (151 ± 8%, 1 hr after the tetanus, n = 8), single CA1 neurons could still exhibit additional potentiation in response to a tetanus in the test pathway (124 ± 7%, 1 hr after the tetanus, n = 8; see example in Fig. 1c), arguing against a homeostatic effect. Moreover, correlation analysis did not reveal a negative correlation between the magnitude of LTP in the saturated and test pathways (Fig. 1d; correlation coefficient r² = 0.08, P = 0.49); if anything, there was a trend for a positive correlation (r² = 0.87, P < 0.01, n = 6, with two outliers removed), again arguing against a homeostatic effect.

The result led us to ask whether the saturation of one pathway is quite the same as the induction of LTP at a high proportion of a neuron's afferents. As Moser et al. (12) have noted in vivo, it may be that the total number of potentiated synapses remain within the working range after tetanization of a single pathway, and thus below the threshold required to trigger homeostasis.

Homeostatic Control of LTP After Widespread Synaptic Strengthening.

Chemical LTP is an alternative induction protocol that should result in a much larger proportion of synapses on individual neurons being potentiated. We used a brief bath application of a potentiation medium containing increased potassium, reduced magnesium, and 25 mM of the potassium-channel blocker tetraethylammonium (refs. 14 and 18; see Supporting Text). The induced LTP was similar to electrical LTP with respect to duration, pharmacology, and mutual occlusion (see Fig. 4, which is published as supporting information on the PNAS web site; see also refs. 15–17).

To induce chemical LTP in a high proportion, but not in all, of a neuron's afferents, a local superfusion technique (19) was used to protect a small group of synapses from the potentiation medium (Fig. 2; Fig. 5, which is published as supporting information on the PNAS web site). The superfusion medium was similar to the normal recording solution but contained a higher calcium concentration (10 mM). To isolate the synapses within the superfusion spot pharmacologically, synaptic transmission outside the superfusion spot could be blocked by replacing the normal extracellular medium with a solution containing reduced calcium (1.2 mM) and low cadmium (5 μM).

Fig. 2.

Homeostatic control of LTP after widespread synaptic strengthening. A large proportion of synapses were potentiated by inducing chemical LTP throughout the slice except within a superfusion spot. The superfused and therefore unpotentiated synapses were then probed for their capability to express LTP. (a and b) Experimental design and representative example of a test (a) and control experiment (b). Inside- and outside-spot synapses are synapses in or outside the superfusion spot. BLOCK, bath application of blocking medium; POT, bath application of potentiation medium. Blocking and potentiation medium have access to the outside- but not the inside-spot synapses. A tetanus was applied at time point zero (arrow). (c) Population data show no LTP could be induced after widespread synaptic strengthening (test experiments: 98 ± 3% 1 hr after tetanus, n = 6), but normal LTP was expressed (control experiments, 136 ± 11% 1 hr after tetanus, n = 5) without prior chemical LTP. The difference between control and test experiments is significant (P < 0.05). Not all of the eight test and eight control experiments lasted until 1 hr after tetanus. Shaded area in c corresponds to shaded areas in a and b. Error bars indicate ± SEM. (d–f) Same as a–c, but AP5 (100 μM) and verapamil (30 μM) were present in the superfusion solution 20 min before, during, and 40 min after the bath application of potentiation medium to prevent undesired potentiation of the superfused synapses. Open arrowheads, transient depressions in fEPSP amplitude due to the exchange of the superfusions solutions (see Supporting Text). This second series of superfusion experiments with pharmacological protection of the superfusion spot confirmed the observation of homeostatic shutdown of LTP; no LTP after widespread synaptic strengthening (test experiments, 100 ± 7% 1 hr after tetanus, n = 6) and normal LTP in control experiments (124 ± 8% 1 hr after tetanus, n = 6). The difference between test and control experiments is significant (P < 0.05). Representative fEPSPs averaged from five consecutive stimuli were taken at the time points specified in the graph.

The sequence of steps was as follows. Throughout the experiment, baseline stimulation was performed by a stimulating electrode in the Schaffer collaterals; by using an electrode whose tip was positioned in the superfusion spot, fEPSPs were recorded in the stratum radiatum of the CA1 region. During an initial period in which synaptic transmission was enabled within the superfusion spot while it was blocked in the rest of the slice, the positions of the stimulating electrode, the superfusion spot, and the recording electrode were adjusted to achieve an optimal postsynaptic response. After changing the blocking solution back to normal medium, the signal recovered (first 50–100 min of the experiments; data not displayed in Fig. 2). The chemical potentiation medium was then bath-applied for 10 min to induce LTP throughout the slice except at those synapses in the superfusion spot (Fig. 2a; approximately −100 to −90 min). Chemical LTP was monitored for 1 hr, and the blocking solution then washed in again for the outside-spot synapses, effectively isolating the synapses within the spot. This decreased the fEPSP amplitude, because the outside-spot fraction of synapses was silenced, resulting in a signal that reflected only the strength of the inside-spot synapses (Fig. 2a, from around −30 min on). In the decisive last phase of the experiment (shaded part of Fig. 2 a and c), these test synapses were then probed with an electrical LTP stimulus for their ability to undergo potentiation. After recording a stable baseline, a tetanus was applied to the inside-spot synapses. Fig. 2c (light-blue symbols) that displays only this last phase shows that LTP at these synapses did not occur (test experiments, 98 ± 3%, 1 hr after tetanus, n = 6). To check that this lack of LTP was not a trivial consequence of failing to reach the cooperativity threshold for LTP induction (20) within the small population of inside-spot synapses, control experiments were performed according to the same protocol but without chemical potentiation. These showed normal LTP (Fig. 2 b and c, dark-blue symbols; control experiments, 136 ± 11%, 1 hr after tetanus, n = 5, P < 0.05).

However, comparison of the signal amplitudes during the first and second blocking periods revealed an unexpected increase of the inside-spot fEPSP after chemical LTP induction outside (145 ± 8%, n = 8; data not shown), indicating that potentiating medium might have leaked into the superfusion spot, thereby potentiating some synapses at the border. Apparent LTP shutdown might then be no more than homosynaptic occlusion (as in Fig. 4c) rather than heterosynaptic homeostasis. We therefore repeated the entire experiment under circumstances in which the induction of LTP in the spot was pharmacologically prevented (Fig. 2 d–f). Chemical LTP can be completely blocked by the NMDA-receptor antagonist AP5 together with the l-type calcium channel blocker verapamil (100 and 30 μM, respectively; Fig. 3; see also Supporting Text). AP5 and verapamil were therefore added to the superfusion medium 20 min before and during and until 40 min after the potentiation medium was applied to the outside-spot synapses. No increase of fEPSP amplitude occurred from the first to the second blocking period (87 ± 10%, n = 6; data not shown). We still observed that the test synapses failed to undergo LTP after the outside-spot synapses had been potentiated chemically (100 ± 7%, n = 6), whereas, under control conditions (without potentiation of the outside-spot synapses), normal LTP occurred (124 ± 8%, n = 6). This difference was statistically significant (P < 0.05, 1 hr after tetanus; Fig. 2f). This second series of superfusion experiments, performed under strict pharmacological control, confirmed the observation of homeostatic shutdown after widespread synaptic strengthening.

Fig. 3.

Shutdown of LTP is not due to unspecific effects of the potentiation medium. Chemical LTP is blocked by AP5 (100 μM) and verapamil (30 μM, 107 ± 5% 1 hr after LTP induction). After the washout of these drugs, LTP could be induced normally by a tetanus (131 ± 5% 1 hr after tetanus, n = 7). AP5 and verapamil were present in the extracellular medium 20 min before and during and 40 min after the bath application of potentiation medium. The timing of the experiment (delay between chemical LTP and electrical LTP) and the concentrations of the drugs were precisely the same as in the original superfusion experiments. Representative fEPSPs averaged from five consecutive stimuli were taken at the time points specified in the graph. Arrow, tetanus. Error bars indicate ± SEM.

Numerous studies performed by others (e.g., refs. 14–17), as well as our own experiments (see Fig. 4), indicate that electrical and chemical LTP are similar. To rule out that an additional unspecific effect of the potentiation medium may have caused the shutdown of LTP, we performed a control in which we inverted the experimental logic: AP5 and verapamil were now both added to the potentiation medium. This should block the induction of chemical LTP and, therefore later electrical LTP should still occur in normal extracellular recordings without superfusion. Adding AP5 (100 μM) and verapamil (30 μM) to the potentiation medium blocked chemical LTP (107 ± 5% 1 hr after LTP induction, n = 7; Fig. 3). As predicted, after the washout of AP5 and verapamil, electrically induced LTP could still occur (131 ± 5% 1 hr after tetanus, n = 7). This demonstrates that the lack of LTP in the inside-spot synapses cannot be explained by the potentiation solution causing harm to the cells or interfering with fundamental physiological parameters.

Discussion

Our data show that widespread synaptic strengthening by chemical means results in a shutdown of LTP. Homeostatic regulation of LTP was not observed after conventional electrical potentiation from a single stimulating electrode. Given that both electrical and chemical forms of LTP target the same subcellular machinery (Fig. 4 and refs. 14–17), it is most likely that the important difference between these methods of potentiation lies in the proportion of synapses that are strengthened, rather than in the induction mechanism per se. CA1 cells are contacted by only few terminals from each afferent CA3 cell (21, 22). Accordingly, it may not be feasible, within a hippocampal slice, to activate a sufficiently high proportion of afferents onto an individual cell with electrical stimulation from a single stimulation electrode. In contrast, chemical potentiation ensures widespread activation of the population of available synapses in a slice. We estimate that ≈90% of recorded CA1 synapses were in contact with the chemical potentiation medium, but it is unlikely that all of these activated synapses were successfully potentiated (see Supporting Text; see also ref. 11). The threshold for homeostatic control of LTP may therefore be considerably lower than the 90% of synapses activated, but our experiments were not geared to determine the minimal number of potentiated synapses necessary.

One might argue that chemical LTP is a highly unphysiological way to induce synaptic potentiation. However, both the electrical and chemical forms of inducing LTP are in vitro stimulation protocols that are unphysiological in comparison to the in vivo situation. Because an individual CA1 cell has up to 30,000 excitatory synapses (23) and receives inputs from a very large number of CA3 cells (21), it is not unreasonable to suppose that circumstances may arise quite frequently in individual neurons in which the operating range (i.e., total sustainable synaptic efficacy) of a CA1 cell is approached or even reached. Thus, the dynamic effects of chemical potentiation may, at least in single cells, be more realistic than they may seem initially. In vivo recordings in the hippocampus of behaving rats and monkeys show that a high proportion of the recorded neurons change their activity in response to the respective learning task: in a spatial navigation task, 75% of neurons in the hippocampal CA1 region were found to have a place field (24); in an odor-guided nonmatching-to-sample task, the activity of 72% of CA1 and CA3 neurons was associated with one or more of the variables tested (25); in a location-scene association task, 61% of the recorded hippocampal cells responded in a scene-selective fashion (26). These examples support the idea that hippocampal CA1 neurons in vivo may experience widespread synaptic input as the hippocampus processes information, and widespread synaptic strengthening might be not such an uncommon event. As measured by using the molecular tagging of AMPA receptors, as many as 30% of amygdala neurons actually undergo synaptic plasticity in an associative learning task (27). That a simple learning paradigm such as tone-shock pairing induces widespread plasticity makes it plausible to assume that the complex daily demands upon an animal could result in a similar (or even larger) proportion of neurons in the hippocampus being subject to extensive potentiation.

Still, the capacity of the brain is limited. The critical number of potentiated synapses may be reached under some circumstances in individual neurons, such that further synaptic strengthening has to be shut down to prevent “catastrophic interference” in neural networks (28) and to preserve information already stored (13, 29, 30). Our findings demonstrating homeostatic control of LTP complement and extend earlier in vivo studies that tested the proposed link between LTP and learning (for review, see refs. 4 and 31). Rioult-Pedotti et al. (32) observed that prior skill learning uses an LTP-like mechanism in the motor cortex that strongly diminishes the capacity for further synaptic enhancement. Saturation of LTP in the hippocampus can similarly prevent both further LTP and hippocampus-dependent spatial learning (12, 33). These in vivo studies did not attempt to distinguish between homosynaptic occlusion and heterosynaptic homeostasis, but they strengthen the notion that situations may indeed be experienced where learning-induced plasticity drives a neuronal circuit to its limits, resulting in a shutdown of plasticity.

The shutdown of LTP belongs to a family of homeostatic processes that act in the nervous system to maintain the stability of neuronal function under ever-changing conditions. There are differences in the timing, target, and trigger of the respective regulatory strategy. In the amygdala, for instance, activity-dependent potentiation of some inputs leads to immediate depression at different synaptic sites on the same neuron and vice versa, thus keeping the net change of synaptic weights roughly balanced (34). In the hippocampal area CA1, heterosynaptic depression is sometimes, but not always, observed (35, 36). Here we did not observe heterosynaptic depression in the superfusion spot after widespread potentiation.

The homeostatic regulation of LTP we describe here is distinct from synaptic scaling (37) and from competitive maintenance of LTP (38). These phenomena both consist of a direct regulation of synaptic strength after global modulation of synaptic activity (synaptic scaling) or local potentiation of synaptic inputs (competitive maintenance). Compared to the former case, the homeostatic mechanism we report operates by preventing additional potentiation: synaptic strength per se is not affected. Compared to the latter, induction of additional potentiation results in the decrease of prior LTP at different synapses. Competitive maintenance is observed under conditions where the levels of plasticity proteins are low (during protein synthesis inhibition or after weak LTP induction). Shutdown of LTP, in contrast, occurs after widespread potentiation, a plasticity regime where a competitive maintenance effect is not to be expected.

Homeostatic shutdown of LTP does relate to the Bienenstock–Cooper–Munro (BCM) theory (9) and the concept of metaplasticity (10, 39). The BCM model proposes that experience-dependent plasticity is regulated by the prior history of neuronal activation, as shown in visual cortex and hippocampus (40, 41). Metaplasticity in the most general sense includes shifts of the modification threshold for LTP (42).

Interpreting our data from the angle of metaplasticity, one could argue that the shutdown of LTP may be explained by a radical right-shift of the Bienenstock–Cooper–Munro (BCM) curve above the LTP threshold. Although we cannot completely exclude this interpretation, we find it unlikely, because it would imply that, after widespread strengthening, the curve would shift such that the 100-Hz tetanus used to probe for LTP would exactly correspond to the zero crossing of the curve (i.e., the LTD/LTP modification threshold). None of the earlier studies that parametrically tested the possible shifts of the BCM curve have given any evidence that the modification threshold can be as high as 100 Hz, finding only values between 3 and 30 Hz (43).

What are the potential mechanisms underlying the shutdown of LTP? This is a matter for future work, but the possibilities fall into two broad categories: resource depletion or active homeostatic regulation. A passive resource depletion model would involve chemical LTP using up limited resources necessary for LTP induction [e.g., AMPA receptors (27) or neurotrophins]. In contrast, direct homeostatic regulation could operate on a network level (e.g., by increasing inhibition) or at the level of an individual neuron. For example, a reduction in excitability affecting the whole neuron might make subsequent potentiation harder (44–47). In contrast to resource depletion, the attractive feature of regulation is that the system might be activated in other circumstances, such as the induction of stress or other conditions in which potentiation should be stopped (48).

Our findings indicate that widespread synaptic strengthening on a population of neurons can prevent further potentiation of their inputs, but this homeostatic shutdown is not ordinarily observed by using conventional stimulation techniques. Both results are telling, because they reflect two faces of the same coin. On the one hand, neurons can be quite robust, allow their synaptic population to regulate their efficacy relatively independently, and thus sustain a substantial increase in total synaptic weight. On the other hand, neurons seem to have evolved strategies of coping with excess synaptic drive. These include, as we have shown, the homeostatic control of LTP.

Methods

Acute hippocampal slices were prepared from male Wistar rats (4–6 weeks old), following standard procedures. Stimulation electrodes were positioned in the Schaffer collateral axons of area CA3; the responses elicited in CA1 pyramidal neurons were recorded either extracellularly or intracellularly. LTP was induced by high-frequency stimulation (tetanus: 100 Hz, 100 pulses) or by chemical potentiation (14). The local superfusion technique had been described in detail (19). Data are presented as mean ± SEM. Two-tailed Student's t tests were used to analyze differences in physiological parameters. The critical value was set at P < 0.05.

Detailed methods are published in Supporting Text.

Abbreviations

LTP

long-term potentiation

fEPSP

field excitatory postsynaptic potential.