Introduction
Storage of memories must involve some long-term physical changes in the brain. But where should we look for these changes? Most neuroscience textbooks will tell you that memory storage involves modification of the strength of synaptic connections between neurons. This view, proposed over 50 years ago by the psychologist Donald Hebb, has been enormously influential and forms the basis of learning in many neural network models. Much experimental work has also supported this idea, particularly in simpler systems such as Aplysia, as well as in mammalian systems where Hebbian long-term potentiation and depression (LTP and LTD) have been demonstrated at many synaptic connections and shown to be correlated with learning. Nevertheless, there remains some controversy about whether these synaptic changes alone are both necessary and sufficient to fully account for memory storage in the brain1.
In this issue, Frick and colleagues2 demonstrate a new locus for activity-dependent changes in neural circuits: long-term changes in dendritic excitability. The past decade has seen a revolution in our understanding of how dendrites help neurons process the thousands of synaptic inputs they receive. Direct dendritic recordings, combined with imaging experiments, have shown that dendrites express just about every known type of voltage-gated channel, often at different densities from the soma. These channels endow dendrites with the power to support active back-propagation of action potentials (APs) and even to generate local spikes, both of which can trigger increases in dendritic calcium by activating dendritic voltage-gated calcium channels3. These forms of excitability can also be important in the induction of synaptic plasticity, with dendritic spikes and back-propagating APs being involved in gating LTP induction at distal and more proximal synapses, respectively4, 5. Dendritic voltage-gated channels are also subject to modulation by several neurotransmitters and second-messenger systems, thus allowing the integrative properties of dendrites to be varied over a wide range6. Given that LTP induction activates many of the same second-messenger systems, can the same induction protocol also trigger long-term changes in dendritic excitability?
Frick and colleagues2 addressed this question by directly monitoring dendritic excitability before and after LTP induction in CA1 pyramidal cells using a powerful combination of techniques: dendritic patch-clamp recording and calcium imaging. This is an impressive feat, as it represents the first time that (usually ephemeral) dendritic patch-clamp recordings have been combined with long-duration plasticity experiments. The authors showed that LTP induction is accompanied by a persistent increase in the amplitude of back-propagating APs near the site of synaptic stimulation. This was associated with an increase in the dendritic calcium signal triggered by the back-propagating AP that was localized to a 
100-
m-wide region surrounding the stimulation site (Fig. 1).
Figure 1: Plastic dendrites.
(a) Schematic illustration of the induction of long-term potentiation (LTP) at distal synapses in a CA1 pyramidal neuron using a theta-burst protocol (TBP). After LTP induction, a zone of enhanced dendritic excitability (red) exists near the potentiated synapses. (b) Mechanisms and consequences of enhanced dendritic excitability2. Top, somatic EPSPs recorded before and after LTP induction. Second row, cell-attached dendritic recording showing that activation of dendritic A-type potassium channels near the potentiated synapses (using a voltage step from -72 mV to + 57 mV) is inhibited after LTP induction. This is associated with an enhancement of AP back-propagation into this dendritic region (third row) and an increase in the local dendritic calcium signal triggered by the back-propagating AP (fourth row).
Full size image (38 KB)To explore the mechanisms underlying this enhancement of dendritic excitability, the authors performed cell-attached dendritic recordings of voltage-gated potassium channels, which showed a decrease in the activity of rapidly inactivating or A-type potassium channels near the resting potential after LTP induction. This was due to changes in the properties of the channels rather than a decrease in channel density, as it was associated with a hyperpolarizing shift in the voltage dependence of inactivation. The mechanism fits neatly with previous results from the same group showing that A-type potassium channels can regulate the size of back-propagating APs and limit dendritic excitability in general7.
These findings suggest that the induction of long-term synaptic plasticity is accompanied by a parallel memory trace, stored in the functional state of dendritic voltage-gated channels8. The idea that synaptic plasticity can go hand-in-hand with plasticity of intrinsic excitability is not unprecedented. Indeed, the first report describing LTP showed that it was accompanied by enhanced postsynaptic excitability9. This phenomenon is now known as EPSP-spike (or E-S) potentiation and has been studied intensively in recent years, primarily using somatic recordings10. A recent study11 showed that induction of LTP and LTD in CA1 pyramidal cells is associated with bidirectional changes in the summation of EPSPs, with the input specificity of the effect implicating changes in dendritic channels. Another study12 showed that induction of LTP in cultured pyramidal neurons is associated with increases in presynaptic excitability, indicating that potentiated synapses can trigger retrograde intrinsic plasticity. In this context, the importance of the paper by Frick and colleagues is that it is the first to show directly that LTP causes local modulation of dendritic excitability that can be traced to modulation of a specific channel subtype.
Where to next? First, the mechanisms underlying the enhanced dendritic excitability require deeper exploration. Like the synaptic plasticity, both the enhancement of dendritic excitability and the potassium-channel modulation were blocked by NMDA receptor antagonists, consistent with the idea that the modulation of excitability depends on the induction of synaptic plasticity. However, the causality of this relationship needs to be demonstrated more directly. Furthermore, dendritic channels can be regulated by neurotransmitters independently of synaptic plasticity6—does this modulation occur via convergent or parallel second-messenger pathways? Identification of the second-messenger pathways involved should also help to clarify the factors that determine the spatial range of the plasticity. Is the observed spatial spread of the calcium signal enhancement due to messenger diffusion, or does it simply reflect the distribution of active synapses or the nonlinearity of AP back-propagation? Can the enhanced excitability ultimately be restricted to single spines? Finally, Frick and colleagues did not rule out changes in the properties or densities of other dendritic voltage-gated channels. Given recent evidence for activity-dependent modulation of dendritic calcium channels13 and Ih channels (hyperpolarization-activated nonselective cation channels)11, it seems likely that synaptic plasticity may target a range of channel types, perhaps differentially with different plasticity-induction protocols.
The rich functional implications of this study are now also open for exploration. Why is such parallel memory storage mechanism necessary, and what are its benefits (and costs)? On one hand, an additional storage mechanism, particularly one that is spatially restricted, provides the neuron with many additional degrees of freedom for plasticity and enhances the storage capacity of the brain14. But the reciprocal link between the back-propagating AP and the induction of synaptic plasticity2, 5 also provides a sophisticated, tuneable positive feedback mechanism that can promote the induction of subsequent synaptic plasticity. Thus, dendritic excitability can be a substrate for metaplasticity, encoding the recent history of synaptic plasticity and neurotransmitter modulation. The spatial extent of the enhanced excitability also provides a means for spreading this form of metaplasticity to neighboring inactive synapses. Ultimately, however, such a form of amplification is inherently unstable and requires compensatory homeostatic mechanisms15. In that vein, it will be interesting to confirm whether LTD induction is accompanied by a corresponding downregulation of local dendritic excitability11. Ultimately, the significance of these findings for memory storage must be evaluated in the context of behavior, where there already exists considerable evidence for changes in neuronal excitability associated with learning and conditioning in many species10.
The plasticity of dendritic excitability naturally also has important consequences for subsequent dendritic processing of synaptic inputs. The suppression of A-type potassium channels can promote EPSP summation (compare ref. 11) and regenerative synaptic interactions in dendrites, and thus enhance the impact of EPSPs on somatic output, as expected in models of E-S potentiation. Furthermore, the enhancement of the back-propagating AP can also lower the threshold for triggering dendritic calcium spikes, which can lead to axonal burst generation3. In this way, the enhancement of dendritic excitability associated with LTP provides a means of further amplifying the impact of synaptic plasticity on neuronal output.
Finally, in a wider sense, the study by Frick and colleagues is significant because it links three flourishing fields at the heart of cellular and systems neuroscience: LTP, intrinsic plasticity and dendritic excitability. Examining the mechanisms and functional consequences of this dendritic marriage between synaptic and intrinsic plasticity promises to lead to a deeper understanding of memory storage and dendritic function in the brain.
