How do neurons combine distinct information streams and form long-lasting associations? Dendritic plateau potentials may allow the integration and storage of coincident location and contextual information in hippocampal neurons.
How does the brain integrate the myriad of distinct information streams that animals encounter, such as color, smell, location, context and past events, to achieve a holistic experience? Broadly speaking, this challenge, known as the binding problem, consists of two parts. First, how is information combined? Second, how is combined information stored and at what spatial scale do these processes occur? Possible solutions to the first problem vary widely and range from large populations of neurons communicating via exotic quantum entanglement1 to more well-known mechanisms of integration and regenerative firing in the somata of individual cells or in dendritic branches2,3,4. In comparison, solutions to the second problem are more concrete, nearly always involving synaptic plasticity. When presynaptic glutamate release activating NMDA receptors occurs with postsynaptic regenerative dendritic signals5, synaptic potentiation often results6,7. Thus, solutions to both problems exist on the dendritic scale, and one attractive hypothesis is that the brain might solve both parts of the binding problem simultaneously using regenerative dendritic firing. Bittner et al.8 now provide evidence for just this sort of elegant solution. The authors show that the conjunctive activation of two separate information streams in the hippocampus of behaving mice activates regenerative dendritic signals, called plateau potentials, that can amplify neuronal output (combining the information streams) and lead to synaptic potentiation (storing the association).
Hippocampal CA1 pyramidal neurons receive input from entorhinal cortex layer 3 (EC3) on their distal tuft branches and input from CA3 on their proximal branches. These inputs are thought to represent two distinct information streams, with EC3 and CA3 inputs carrying spatial and contextual information, respectively. Binding of these two information streams is thought to lead to the context-dependent place firing (place fields) observed in CA1 neurons in navigating rodents. For example, in some place cells, place firing is contingent on task context, such as whether the animal will be turning left or right in an upcoming turn of a T maze9. Thus, these cells often only respond when the relevant location and contextual information are present together. But how are these information streams combined and stored together in CA1 neurons of navigating mice? Strong hints come from compelling brain slice experiments carried out previously in the Magee laboratory.
These previous studies showed that coincident stimulation of CA3 and EC inputs results in the initiation of large regenerative events, known as plateau potentials, in CA1 distal dendrites10. These dendritic plateau potentials generate prevalent Ca2+ influx, burst firing output and NMDA-dependent synaptic potentiation, exactly the mechanism that could be used by hippocampal neurons to bind (combine and store) information during behavior. However, nearly nothing was known about the occurrence or relevance of plateau potentials in the hippocampus during behavior.
Here the study from Bittner et al.8 greatly advances our understanding. The authors made challenging whole-cell recordings from CA1 neurons in head-restrained mice navigating along a band treadmill with different tactile cues along its length. Many neurons had place fields along this track, and most periodically displayed high-frequency bursts of action potentials (APs) associated with large somatic membrane potential depolarizations, which the authors convincingly determined to be plateau potentials11,12. Having shown that plateaus do indeed occur naturally during behavior, they next wanted to see whether coincident input from EC3 and CA3 might be generating the plateaus, as in their previous slice experiments.
To complement their already technically challenging in vivo measurements, Bittner et al.8 then added an extracellular recording electrode to measure the CA1 local field potential (LFP) during whole-cell recordings. Hippocampal LFP oscillates at theta frequency, and the two input streams are thought to be preferentially active during different theta phases, EC3 inputs active near the peak and CA3 near the trough13. The authors astutely observed that neurons firing APs closer in time to the theta peak tended to generate more and longer duration plateau potentials. They suggest that, in these neurons, APs are caused by CA3 inputs arriving closer in time to the EC3 inputs, generating the depolarization necessary for plateaus. Although a logical conclusion given their prior slice work, it is still possible that strong input from EC3 alone could drive the firing and plateaus near the theta peak10.
The authors then cleverly shifted the timing of CA1 firing either closer to or further from the peak of theta. To do this, they took their LFP signal and played it back in real time into the intracellularly recorded cell in the form of a depolarizing current. With a variable delay, they could essentially control when CA1 cells fired in relation to LFP theta. When the added depolarization occurred closer in time to presumed EC3 input (near the theta peak), increased plateau probability and duration resulted. This finding demonstrates that plateau generation is sensitive to the timing of additional depolarization in relation to theta, which, as the authors suggest, likely comes from CA3 input during natural place firing.
The authors then turned their attention to the necessity of the EC3 pathway for plateau potential generation. They switched off these inputs to CA1 cells in a precise manner by expressing a light-activated ion pump (Archaerhodopsin) in EC3 pyramidal cells and observed reduced plateau potential probability and duration. Thus, EC3 inputs are likely necessary for the generation of plateau potentials. Furthermore, slice experiments showed that pairing EC3 inputs with backpropagating APs caused plateau potentials that originated in the region of the CA1 dendrites that receive EC3 input.
Insightfully putting all of their above results together, the authors suggest that CA3 input primarily drives somatic firing in CA1 neurons, which then backpropagates into tuft dendrites, allowing an interaction with EC3 inputs that generates plateau potentials (Fig. 1). The plateaus then forward-propagate to the soma and cause burst firing. This is a likely possibility given the data presented. However, it would also be of interest to investigate whether EC3 inputs alone are sufficient to generate plateaus. This would answer the question of whether integration of the two inputs streams is required, or only sufficient, for plateau generation.
With a potential solution to the binding problem in hand, the next question is whether plateau potentials can alter cellular firing output and lead to stored changes in feature encoding, as would be expected for such a solution. In this context, Bittner et al.8 noticed that plateau occurrence increased the firing rate of CA1 neurons in existing place fields. Plateaus may therefore be acting to amplify CA1 neuron firing when CA3 and EC3 inputs are coincident, effectively modulating the feature encoding of CA1 cells when two information streams are present. However, it is also possible that such 'gain modulation' is not the main function of plateau potentials and it is actually a byproduct of their primary role in inducing and maintaining synaptic strength.
Indeed, in what is perhaps their most provocative result, Bittner et al.8 were able to artificially induce new place fields by evoking plateau potentials in silent cells (a process that they also observed spontaneously in six non-place cells). After generating plateaus at the same track location on approximately five sequential traversals, a new place field formed and no further stimulation was required. Indirect evidence from measurements of somatic membrane potential fluctuations suggests that synaptic potentiation led to the new place fields. This is the first time individual hippocampal neurons have been manipulated experimentally to create new, long-lasting place cells. Combining these manipulations of Bittner et al.8 with recent advances in imaging methods5,6 should make it possible in future experiments to directly determine whether plateaus induce new place fields by increasing the strength of specific synapses.
A fascinating observation made by the authors was that place fields can be formed by plateau induction at any location on the track. This suggests that each CA1 cell receives significant input at every location along the track. This is a remarkable result that goes against the common idea that CA1 neurons form place fields by receiving appreciable input only at specific track locations. Instead, certain cells likely express place fields through potentiation of specific subsets of the inputs, possibly by plateau potentials. Thus, the relative timing of inputs from EC3 and CA3 arriving on CA1 dendrites may determine the context-dependent spatial map.
Bittner et al.8 suggest that CA1 receives information regarding the current location of the animal in its environment and combines this with contextual information (Fig. 1). When appropriately timed, these different information streams generate plateau potentials and subsequent burst firing in CA1 neurons, which may represent the binding of simultaneous information in the animal's experience. A further (and perhaps the main) function of the plateau is to induce synaptic potentiation that serves to store this bound information. Subsequent activation of either of the strengthened pathways would then be expected to cause firing in a cell that now represents the bound information of both pathways. One can imagine that this integrated signal could further be combined with other information downstream, allowing the brain to bind multimodal representations to create a rich holistic experience. The authors suggest this integration mechanism may be a general core function of pyramidal cells across many brain regions. Other examples already exist14, suggesting that dendritic plateau potentials are the brain's general solution to the binding problem.