Dendritic mechanisms of hippocampal place field formation

Abstract

Place cells in the hippocampus are thought to form a cognitive map of space and a memory of places. How this map forms when animals are exposed to novel environments has been the subject of a great deal of research. Numerous technical advances over the past decade greatly increased our understanding of the precise mechanisms underlying place field formation. In particular, it is now possible to connect cellular and circuit mechanisms of integration, firing, and plasticity discovered in brain slices, to processes taking place in vivo as animals learn and encode novel environments. Here, we focus on recent results and describe the dendritic mechanisms most likely responsible for the formation of place fields. We also discuss key open questions that are likely to be answered in the coming years.

Introduction

Place cells are a subset of the excitatory pyramidal neurons in the hippocampus that fire at specific locations within an animal’s local environment (place field)[1]. Ensembles of place cells tile the environment with place fields, with orthogonal ensembles encoding different environments. During sleep and wakeful rest, place cells can reactivate in the same (or reverse) sequence as during the preceding experience [2–6], leading to the idea that orthogonal place cell ensembles represent distinct maps of different environments, and a cellular substrate of spatial memory.

Here we focus on the cellular and circuit mechanisms that contribute to the formation of new place fields in the CA1 region of the hippocampus when animals are exposed to novel environments (Figure 1). The mechanisms underlying place field and spatial memory formation may generalize to other types of memories encoded in the hippocampus (contextual fear, trace, etc) [7,8], thus revealing general principles of hippocampal memory formation.

Figure 1.

Hippocampal place field formation at the ensemble level and active dendritic signaling in CA1 pyramidal neurons. (a) Switching from a familiar to a novel environment causes global remapping of CA1 place fields (colored neurons indicate place cells). CA1 neurons in the novel environment can either be silent (no place field), form a place field after some time or experience (delayed place field), or form a place field immediately (instant place field). (b) CA1 pyramidal neurons receive input from CA2 and CA3 on their basal and proximal apical dendrites, and entorhinal layer 3 cortical inputs on their tuft dendrites (Left). These inputs can cause a number of postsynaptic responses (from left to right): local NMDA spikes in basal, oblique apical, and tuft dendrites in the absence (or presence, not shown) of somatic AP firing; Somatic firing without any branch spiking in the dendrites; Somatic firing with bAPs that cause global branch spiking in the dendrites; Somatic burst firing with co-occurring plateau potential generated in the distal apical dendrite that causes global branch spiking (calcium spike). These distinct responses are shown in isolation, but combinations of these can also co-occur. EC: Entorhinal cortex. EC3: Entorhinal cortex layer 3. DG: Dentate gyrus. bAPs: backpropogating action potentials.

When animals navigate in a familiar environment, ~25% of CA1 neurons display place field firing [9,10]. When animals are switched to a novel environment, “global remapping” takes place, where a random ~25% of neurons across the population encode the novel environment [11–17]. These cells cannot be predicted based on the cells encoding the familiar environment. Further, if two place cells each have a place field in both environments, the location of these place fields with respect to one another in the environments is also unpredictable. Nevertheless, 3 general functional cell types appear in the novel environment that together form the new map (Figure 1B) [17–20]: 1. Instant place cells that begin somatic action potential (AP) firing in the place field the first time the animal traverses the field location, 2. Delayed-onset place cells that begin somatic AP firing in the place field only after the animal traverses the field location multiple times, and 3. Silent cells that rarely fire somatic APs. Here we treat each functional cell type (instant, delayed and silent, Figure 1B) separately, and describe the mechanisms most likely responsible for their formation. We will also discuss dendritic mechanisms that may determine the precision and stability of place fields after formation.

Regenerative dendritic events and plasticity

Hippocampal pyramidal neurons can engage activity-dependent synaptic plasticity[21], a mechanism by which spatial information could be encoded and stored[22–24]. NMDA-dependent potentiation requires presynaptic glutamate release paired with a post-synaptic regenerative dendritic event (strong depolarization in the dendrites, here referred to as “branch spikes” [25]) to relieve NMDA channel magnesium block and allow for a large influx of calcium to activate enzymatic plasticity pathways. Brain slice research has revealed numerous types of branch spikes, such as all-pervasive back-propagating APs (bAPs) originating in the soma or axon initial segment[23,26–28], partially invasive bAPs [23,27], plateau potentials [29–31] and local NMDA spikes (local branch spikes) [32–36] (Figure 1B). These distinct branch spikes are caused by varying engagement of distinct types of voltage- and ligand-gated ion channels, such as Na⁺-, Ca²⁺-, K⁺-, AMPA, and NMDA-channels. The efficacy of synaptic inputs varies greatly with distance from the soma and location in the arbor [37], thus different input patterns lead to different types of branch spikes. For example, strong synaptic activation coincident in both proximal (CA2 and CA3 input) and distal (Entorhinal perforant path input) portions of the arbor can lead to plateau potentials (somatic AP burst firing coincident with widespread dendritic calcium spikes). Plateau potentials have recently been shown to drive behavioral timescale plasticity—potentiation of inputs active within seconds before or after the plateaus [38]. On the other hand, coincident activation of a small number (~5–20) of synapses on a particular branch can lead to local dendritic spikes, such as NMDA spikes, in the absence of somatic spiking, which can potentiate the coactive synapses [39]. The dendritic membrane potential and the types and patterns of activated synaptic input therefore determine the specific type of branch spike generated. All of these branch spiking mechanisms are likely at work at different times in place cells, operating in concert, independently or sometimes not at all, thus defining “windows” for synaptic plasticity across the arbor.

Impact of novel environment exposure on CA1 state

When an animal enters a novel environment, the following observations of changes to CA1 inputs and state have been observed. First, two-photon (2P) functional imaging of dendritic targeting interneuron axons in CA1 Stratum Oriens has shown that dendritic inhibition of pyramidal neuron basal dendrites is reduced, leading to increased dendritic excitability. At the same time, somatic targeting interneuron axons increase the level of somatic inhibition[17]. After several minutes of exploration in the novel environment, familiar environment inhibition levels return [17]. Intracellular recordings from place cells established that the subthreshold hill of depolarization associated with place field firing [10] is larger in novel environments compared to familiar [12], consistent with the idea of a net reduction in inhibition. The CA1 region contains a variety of interneuron subtypes [40], some involved in direct pyramidal neuron inhibition and others in disinhibition [41]—novel environment exposure likely leads to changes in both. Second, the sensory experience associated with a familiar vs novel environment is different, leading to activation of different sensory afferents from the periphery (Figure 1B). Given that CA1 is many synapses from the periphery, and the likelihood that only a fraction of the afferents are activated in any given environment, it is likely that different environments lead to the activation of different subsets of CA1 afferents; however, for afferents from CA3, pattern completion may maintain some input pattern similarity. Third, the firing rate of CA1 pyramidal neurons increases for many minutes across the population [12,17,42,43], which could be due to increased excitatory synaptic input, a net reduction in inhibition, changes in neuromodulatory input or a combination of these.

Instant place cells

In some CA1 cells, somatic AP firing occurs in the place field on the first traversal of the novel environment (Figure 2). In these cells, the new sensory configuration likely leads to the activation of many afferents that impinge on already strong (pre-strengthened) synapses. The sum of all synaptic inputs is large enough upon initial exposure to drive AP firing at a particular location (new place field). This somatic AP firing can co-occur with increased regenerative dendritic events; however, any potentiation produced by these events cannot be required for the coincident firing in the new place field. Correspondingly, additional evidence has shown that a subset of place cells still form instant place fields in mice in which NMDA-dependent synaptic plasticity has been perturbed or blocked [17,44–47]. Further, some cells can be predicted to express place fields prior to exposure to a novel environment based on their likelihood to display somatic burst firing [48] or their participation in pre-play [49], in which firing before novel environment exposure predicts upcoming place field firing sequences. This again suggests that synaptic plasticity plays little to no role in the appearance of these place fields since their inputs were already able to drive firing prior to the experience. In some instant cells, strong synaptic activation likely occurs in both proximal (CA2 and CA3 input) and distal (EC3 input) portions of the arbor, generating plateau potentials [29,30] that may potentiate weaker synapses activated in or near the new field location. Through behavioral timescale plasticity rules [38] (Figure 2; left neuron), such events may be critically important for binding together place or context related inputs occurring close to, but initially outside of, the new field. These scenarios are supported by intracellular place cell recordings that observed plateau potentials [12] and 2P dendritic imaging that observed all-pervasive dendritic calcium transients [17] (indicative of calcium spikes and burst firing [30]) during the first occurrence of AP firing in many instant place fields.

Figure 2.

Dendritic mechanisms of instant place field formation, precision, and stability. The top box shows two possible modes of somatic firing and co-occurring dendritic branch spiking during the first traversal of this cell’s place field in a novel environment (instant place field). The mechanism driving instant place field firing is through activation of pre-strengthened pathways (red and black spines) that are sufficient to cause somatic firing. If the cell fires in burst mode, it can lead to behavioral timescale synaptic plasticity, although this is not necessary for place field firing on subsequent laps. The bottom box depicts the average branch spike prevalence that could occur across the dendritic arbor over many traversals of a cell’s place field once the environment has become familiar; 3 different example cells are shown. A low average branch spike prevalence is associated with place fields that are imprecise and lack stability, i.e. they tend to disappear over time, and vice versa.

In a subset of the instant place cells, strong synaptic activation likely occurs in either proximal or distal portions of the arbor, or weaker activation occurs in both, leading to regular somatic firing and branch spikes that are less widespread in the new field compared to plateau potentials (Figure 2; right neuron). This scenario is supported by observations from intracellular place cell recordings showing many place fields form during regular firing without plateau potentials [12], and also from 2P functional imaging of place cell dendrites showing that some place fields form with only partial branch spiking across the arbor [17].

Instant place cells that form with more widespread branch spiking will be more likely to potentiate weaker synapses activated in the place field, and further increase the prevalence of branch spikes (i.e. greater chance of clustered synapse activation or boosting of bAPs further into the arbor [50,51]), thus increasing the prevalence of plasticity “windows” and possibly increasing amplification of synaptic input [33,52,53]. An increased prevalence of branch spiking during subsequent place field firing could therefore lead to more precise place fields due to dendritic spike amplification of inputs[54,55] or through a potentiation and renormalization process [25,56,57]. Further, increased prevalence of branch spikes may provide positive reinforcement of synaptic strength, thus increasing place field stability over time [58]. This idea is supported by a study measuring place cell dendrites, which established that the greater the average prevalence of branch spiking across the arbor over many traversals of the cell’s place field, the more stable and precise the place field [25]. Interestingly, some place cells have more than one place field in the same environment [59,60] and the average prevalence of branch spiking was found to vary significantly between different place fields of the same place cell [25]. If such branch spiking is indicative of the mechanisms involved in forming the fields, it supports the interesting possibility that different place fields of the same cell can form through different mechanisms (for example, one instant field and one delayed field).

Delayed place cells

In some CA1 cells, place fields form only after the animal traverses the place field location multiple times (Figure 3). These fields could form via different mechanisms: 1. synaptic potentiation or 2. a change in input pattern. Initially in both cases, the new sensory configuration in the novel environment likely leads to the activation of many afferents synapsing across the dendrites of CA1 neurons. Given that there are many more weak synapses than strong ones, the pattern of synaptic activation in the majority of neurons is likely biased in favor of weak synapses [61]. The sum of all synaptic inputs is therefore initially too small to drive somatic firing in these neurons. In some delayed place cells (Figure 3; left neuron), some afferents may be active coincidently at the same environment location, and converge onto synapses in the same branch of the arbor. This synchronous and clustered synaptic input can drive NMDA spikes that begin a powerful potentiation process in the small group of activated synapses that is independent of somatic AP spiking (Figure 3; left neuron). Thus, clustered coincident input is an efficient mechanism for reaching local spike threshold with a small number of inputs [62], facilitating place specificity without somatic AP spiking. This process may occur at many different sets of clustered synapses across the arbor, either all simultaneously occurring on a particular traversal of the future place field, or spread out in time over different traversals. After a few minutes and traversals of the future place field, the sum of the now potentiated synaptic inputs becomes large enough to drive somatic AP firing, forming the new place field. This is supported by a recent study that found localized dendritic calcium transients (likely NMDA spikes) occurring at the environment location where the future somatic place field formed [17]. Further, it was observed that mice with NMDA receptors functionally inactivated in CA1 [46,63] appeared to have fewer place cells across the population compared to wild-type mice, suggesting NMDA-dependent plasticity is required for the formation of a subset of place cells [17,47]. ~2 putative NMDA spikes occurred in the basal dendrites on each traversal of the future somatic place field prior to its appearance. This sparsity of dendritic branch spikes fits with a previous model of hippocampal information encoding and storage[64]. In this model, the storage capacity of a CA1 neuron is increased as the number of synapses potentiated to store the memory decreases. Interestingly, the model used local dendritic branch spikes to drive potentiation of small subsets of inputs, consistent with the experimental 2P data described above.

Figure 3.

Dendritic mechanisms of delayed place field formation, precision, and stability. The top 2 boxes depict CA1 pyramidal cells that initially lack place fields during the first few traversals of a novel environment. In one case (top left box), clustered synaptic input at 5 locations across the arbor generates 5 local NMDA spikes. This occurs at a particular spatial location in the animal’s environment. The clustered and co-activated synapses that caused the NMDA spikes become potentiated. On subsequent traversals of that location in the environment, reactivation of the same inputs is now sufficient to drive somatic firing (either regular or burst firing), causing a delayed place field to appear. The next steps are the same as those described in Figure 2. In the other case (top right box), the initially silent neuron forms a delayed place field through a change in the activation of inputs onto the neuron. This could occur through changes in attention associated with specific behaviors like head-scanning. The subsequent steps leading to place field formation depend on the pattern of inputs, and could either require no synaptic plasticity (left middle box) or could generate local NMDA spikes that cause synaptic plasticity (top left box).

For a subset of delayed place cells (Figure 3; right neuron), plasticity in presynaptic populations, or changes to which sensory features of the environment the animal is attending, may result in the activation of previously inactive afferents. If these coactive afferents arrive onto weak, but spatially clustered synapses, then the above-described delayed place field formation process could occur. If the inputs arrive onto numerous strong synapses, then the sum of all synaptic inputs could be large enough to drive action potential firing without the need for CA1 synaptic plasticity, which would appear as a newly formed delayed-onset place field. Indeed, delayed-onset place cells have been observed in mice in which NMDA dependent synaptic plasticity has been perturbed or blocked [17,44–47], supporting the idea that synaptic plasticity is not required for the formation of all delayed place cells. Along these lines, attentional changes associated with head-scanning at a particular location during novel environment exposure can cause some pyramidal neurons to fire, which will continue to fire at that location on subsequent laps (new place field)[20] [65].

Once delayed place cells begin somatic AP firing in the new field, further synaptic potentiation facilitated by NMDA spikes, bAPs or plateau potentials is expected to refine the precision and stability of the fields using analogous processes as proposed above for instant place cells (Figure 2). Intracellular and juxtacellular stimulation of plateau potentials in silent cells has been shown to lead to place field formation at the environment location of stimulation, demonstrating that these events are powerful synaptic potentiation signals, and are sufficient to drive new place field formation. However, when plateau potentials are observed during the first occurrence of firing in naturally formed (not stimulated) place fields, any potentiation produced by these somatic AP burst firing events could not have been required to drive initial place field firing. Accordingly, a recent study using juxtacellular stimulation of silent cells suggests that burst firing with plateau potentials is sufficient but possibly not necessary in the formation of artificial place fields. This also supports the idea that other plasticity mechanisms such as bAPs and local dendritic spikes are involved in driving place field formation[66]. Together, these findings suggest that delayed onset place fields appear in novel environments following either NMDA-spike-dependent potentiation of weak synapses or changes in input patterns that activate pre-strengthened synapses. These processes lead to either regular or burst firing in the soma, both of which can be associated with branch spikes that strengthen synapses to maintain or modify the place field.

Silent Cells

In some CA1 cells, place fields do not form and the cells rarely fire somatic APs (Figure 4). In these cells, the new sensory configuration likely leads to the activation of many afferents impinging on weak synapses and very few on strong synapses. In contrast to delayed place cells, activated afferents are either not active at the same spatial location of the environment, are not appropriately clustered on the dendritic arbor to drive local dendritic spiking, or both. Therefore, either no NMDA spikes occur, they occur in too small of a number, or they occur at different environment locations. They are therefore unable to potentiate a sufficient number of inputs at a particular environment location to drive somatic firing. The idea that NMDA spikes occur in silent cells upon novel environment exposure is supported by data (Authors, unpublished) showing the existence of localized calcium transients, presumed to be NMDA spikes, in silent cells. If head-scanning and changes in attention occur, any newly activated afferents still do not drive a sufficient number of strong synapses on silent cells at a particular location to drive firing or a sufficient number of NMDA spikes. Interestingly, the depolarization of some silent cells through intracellular current injection leads to the emergence of place firing [67]. This suggests the existence of silent place cells with activated pre-strengthened synapses that could have become instant place cells with just a bit more synaptic activation to cause sufficient depolarization for the place field to emerge.

Figure 4.

Possible scenarios that prevent place field formation. The top 2 boxes show somatically silent CA1 pyramidal neurons during the first traversal of a novel environment. In both cases, there is no development of a delayed place field on subsequent traversals of the environment. The failure to form a place field could be due to weak inputs that fail to drive the cell to fire and also fail to generate local NMDA spikes, so no synaptic strengthening takes place (top right box). Alternatively, input is sufficient to cause some NMDA spikes, but there is an insufficient number of these events across the dendritic arbor to cause enough synaptic potentiation to drive subsequent place field firing at the soma. This scenario would also occur if the neuron produced the same frequency of NMDA spikes as delayed onset place fields, but the spikes occurred at various environment locations, with insufficient numbers of overlapping spikes at one particular environment location.

Open questions and conclusions:

1. What are the activity patterns of synaptic input onto a particular place cell, how are these inputs distributed across the dendritic arbor and how do they change during remapping?

2. How many synapses are potentiated during the formation of new fields?

3. What dendritic firing and plasticity mechanisms lead to rate remapping [68]?

4. What is the contribution of genetics and molecular expression in establishing hippocampal connectivity and representations [69,70]?

5. Two distinct types of CA1 pyramidal neurons exist [71], how do these cell types map onto the different functional types of place cells described here?

6. What specific differences in synaptic input and dendritic branch spiking are observed between silent cells and place cells?

7. Other cells across the population fire during navigation, but do not form well defined place fields [59,72]. What role do these cells play in representing a novel environment and what mechanisms lead to their formation?

8. Are the mechanisms that underlie place field formation in CA1 also responsible for CA2 and CA3 place field formation?

9. What role do replay and preplay have on the formation of the cognitive map? Are certain dendrites and synapses biased for activation by preplay? Does replay drive regenerative dendritic events to further strengthen place fields and lead to more stable memories?

10. Are the dendritic mechanisms underlying place field formation engaged during the formation of other memories that are dependent on the hippocampus, such as contextual fear memories?

Highlights.

Recent technical advances make possible in vivo studies of dendritic function.

Place field formation serves as a model for understanding hippocampal memory formation.

Regenerative dendritic events play an active role in place field formation.

Some place fields appear to require synaptic plasticity for their formation.

Some place fields do not appear to require synaptic plasticity for their formation.