3.1.1.

Pyramidal neurons

After the first imaging of the CA1 place cell activity,⁷ calcium imaging of CA1 pyramidal neurons during behavior has been widely reported. Imaging demonstrated that pyramidal neurons in superficial sublayer form more stable place maps, whereas those in deep sublayer are preferentially modulated during goal-oriented learning.⁶¹ Imaging of the CA1 neuron activity during reward location alternation task in VR identified a subset of pyramidal cells that are involved in reward coding.⁶² Hippocampal place maps display a phenomenon called over-representation, in which place cells accumulate at locations with behaviorally relevant features such as reward and landmarks. Calcium imaging of CA1 pyramidal neurons in mice navigating a virtual linear track with reward and a landmark placed at distinct locations demonstrated that over-representation of these locations occurs with different time courses and molecular mechanisms involving the synaptic scaffold protein Shank2 and primarily owing to selective stabilization and accumulation of relevant place cells.⁴⁵ Additionally, when mice were retrained under a new condition in which the reward was shifted to match the landmark location [Fig. 4(a)], a subset of place cells that were active at the initial reward site shifted their place fields to the new reward site, along with delayed recruitment of nearby place cells [Fig. 4(b)].⁴⁶ Collectively, these findings reveal a basic comprehensive picture of cellular dynamics associated with hippocampal cognitive map formation and plasticity.

Fig. 4

Plasticity mechanism for the CA1 cognitive maps. (a) CA1 map reorganization induced by reward relocation. Mice were first trained on the standard linear track (Pre). Once training was completed, mice were retrained in a new arrangement in which the location of reward delivery was shifted to match the location of the landmark (Rearr 1–5) (top). Examples of place cell maps imaged in the same animal in the preceding control (immediately before the reward relocation), early (first retraining session), and late (fifth retraining session) phases of the reward-rearrangement task (bottom). (b) A model for CA1 map plasticity. Immediately after reward relocation, the fraction of place cells decreases due to decreased reward cell formation and stabilization, and a subpopulation of reward cells moves their fields to the new reward location (early). After a few sessions, the stability and formation of reward+landmark cells increase, as do recruitment from the outside place cells to reward+landmark cells (late). Adapted from Mizuta et al.⁴⁶

Recently, an olfactory VR system, using rapid flow controllers and an online algorithm to produce precise odorant distributions, demonstrated that odor-guided virtual navigation behavior engages CA1 place cells similar to those found in visual virtual environments.⁷¹ They further revealed that mice navigate the same multisensory environment using either vision, odor, or both, and hippocampal spatial mapping relies on the modality used for navigation.⁷² An odor-cued navigation task also exhibited that mice develop different neuronal spatial representations depending on task rules.⁷³ Thus, multisensory VR environments and navigation tasks that depend on non-visual modalities widen a potential for investigating mechanisms underlying hippocampal sensory-spatial and memory processing.

While hippocampal neurons encode physical variables such as space, they can also encode more abstract cognitive variables such as learned knowledge. CA1 neurons jointly encode accumulated evidence with spatial position as demonstrated by two-photon calcium imaging of CA1 pyramidal neurons in mice performing a decision-making task in a virtual T-maze.⁶³ This suggests that the hippocampus creates task-specific low-dimensional neural manifold that contain a geometric representation of learned knowledge.

Donald O. Hebb, a Canadian psychologist, postulated the concept of “cell assembly” as a group of neurons that are repeatedly coactivated and support various cognitive processes, including memory. Hippocampal neuronal populations are known to show structured activity, such as sequential activity and synchronous activity, even under spatial cue-deprived environments, and two-photon imaging of these preconfigured neuronal ensemble activity has provided important insights into their organization. Villette et al.⁸³ imaged the activity of CA1 pyramidal cell populations while mice ran on a treadmill without external cues and rewards. They found that the dynamics of a hippocampal population code spontaneously displays recurring sequences in the presence of self-motion cues, suggesting that such a functional structure may be an internal cognitive template that shapes behavior. In their follow-up study, Malvache et al.⁸⁴ conducted simultaneous two-photon calcium imaging with a contralateral local field potential (LFP) recording to investigate patterns of synchronous activation of the CA1 ensembles in mice alternating running and rest on a treadmill. They observed that anatomically intermingled and functionally orthogonal assemblies undergo recurrent reactivation. Additionally, these assemblies represented discrete temporal segments of neuronal sequences observed during runs, and multiple assemblies could be combined into longer chains of replay.

In recent years, some studies have combined optogenetics with multiphoton imaging. However, they are typically manipulations that stimulate or inhibit brain regions outside the FOV^94,95 or full field stimulation or inhibition of a particular neuronal subtype within the FOV.⁷⁶ Thus far, limited studies have conducted imaging while optogenetically manipulating neuronal activity at the single cell level within the FOV. Rickgauer et al.³⁵ conducted such an experiment first. Technically, they performed dual-wavelength two-photon excitation to observe the activity of GCaMP3-labeled CA1 pyramidal neuron populations at 920 nm and photostimulate neurons expressing a red-shifted optogenetic actuator C1V1(E122T/E162T) in the same FOV at 1064 nm in mice exploring a virtual environment. They manipulated task-modulated activity of the CA1 place cells by artificially biasing natural place-field activity and revealed subthreshold dynamics and local interactions between multiple neurons underlying place field formation using this system. Recently, light modulator-mediated two-photon holographic optogenetics that allows simultaneous activation of several place cells has also been developed.⁸⁹ Targeted stimulation of specific place cell ensembles biased the spatial behavior of mice in a virtual environment, thereby providing a direct causal evidence linking place cell activity and spatial navigation.

Finally, two-photon calcium imaging of CA1 pyramidal neurons during non-spatial tasks^64–68 and that of developing CA1 hippocampus in living mice^69,70 have provided valuable knowledge into their implication in learning and memory, as well as in postnatal development of functional hippocampal circuits.

3.1.2.

Inhibitory interneurons

In the hippocampus, network excitation computation and plasticity are controlled through inhibition. ETL-mediated multiplane calcium imaging in a virtual track running task found a larger fraction of locomotion-activated somatostatin (SOM)-expressing and parvalbumin (PV)-expressing interneurons and a small fraction of the immobility-activated similar interneuron subtypes.⁷⁴ Furthermore, in their follow-up study, they elucidated that the activity of both CA1 SOM- and PV-expressing interneurons during a goal-directed spatial navigation task was at first strongly suppressed upon exposure to a new environment but recovered as mice grew accustomed to it.⁷⁵ The modulation of activity of SOM-expressing interneurons was context-independent, and their recovery from the suppression was blocked when behavioral recovery was prevented.

Pyramidal neuron disinhibition is provided by the interneuron-specific interneurons that express the vasoactive intestinal polypeptide (VIP) and selectively innervate other inhibitory interneurons. Calcium imaging of CA1 VIP-expressing interneurons suggests that their activity was modulated by velocity during locomotion and by reward and location during goal-oriented learning.⁷⁶ Inhibition of VIP-expressing interneurons showed that disinhibition mediated by this interneuron subtype is necessary for reward learning and goal-directed enrichment of place cells, demonstrating that their behavioral state- and task demand-dependent disinhibition supports goal-oriented learning.

The relationship between various interneuron activities and electrical network activity in the hippocampus was shown via a two-photon calcium imaging study combined with LFP recordings. Two-photon calcium imaging of VIP-expressing interneurons and LFP recording in awake locomoting mice revealed their significant speed modulation and preferential recruitment during theta-run epochs but not during ripples.⁷⁷ Moreover, fast 3D acousto-optical imaging of the basal dendrites of the PV-expressing interneurons combined with ipsilateral LFP field recording in awake mice visualized regenerative calcium spikes initiated at variable hot spots and propagated in both directions.⁷⁸ During SWR doublets within a short temporal window, a supralinear dendritic summation emerged, suggesting their role as a dendritic spatiotemporal coincidence detector.

The dynamics of large-scale interneuron populations from the stratum oriens/alveus border to the stratum lacunosum moleculare in CA1 during navigation and learning was characterized via an acousto-optic deflector-based 3D two-photon functional imaging and post hoc immunohistochemistry.³⁷ Identification and characterization of six molecularly defined interneuron subtypes indicated that a subset of CA1 interneurons exhibit spatially-tuned activity. Moreover, prominent reward modulation of SOM-expressing interneuron activity during goal-directed spatial learning was demonstrated. These and other findings collectively yield important insights into the functional organization of local inhibitory circuits.

3.1.3.

Astrocytes

Calcium dynamics in astrocytes influence nearby neuron and synapse activity, and also affect behavior. Two-photon calcium imaging revealed that hippocampal astrocytes responded to an anxiogenic environment through intracellular calcium activity, reflecting a mouse’s affective state during a virtual elevated plus maze task.⁸⁰ Optogenetic activation that increases astrocytic intracellular calcium elicited anxiolytic behaviors, which was mediated by a mechanism where ATP released from activated astrocytes increased excitatory synaptic transmission in dentate granule cells.

During virtual navigation, concurrent two-photon calcium imaging of CA1 astrocytes and neurons demonstrated that astrocytes display calcium events carrying spatial information in topographically organized cellular subregions.⁸¹ Astrocyte-encoded spatial information was complementary and synergistic to that carried by neurons, which improved decoding when astrocytic signals were considered. A recent study further demonstrated that hippocampal astrocytes not only convey spatial information but also encode reward location.⁸² Chronic imaging of calcium activity of CA1 astrocytes during virtual navigation demonstrated that they display experience-dependent, persistent ramping activity toward the reward location in a familiar environment, with which the mouse trajectory was sufficiently decoded.

3.1.4.

Alzheimer’s disease and other disease models

Alzheimer’s disease (AD), a progressive neurological and neurodegenerative disorder, primarily affects the temporal lobe including the hippocampus, and is characterized by cognitive deficits and memory impairment. Hence, hippocampal cellular resolution imaging of AD mouse models can inevitably contribute to deepening our understanding of its pathophysiology. A study that imaged spontaneous CA1 pyramidal neuron activity in anesthetized amyloid precursor protein (APP)/presenilin-1 (PS1) transgenic AD model mice observed increased fractions of silent and hyperactive neurons near amyloid-$\beta$ ($\mathrm{A}\beta$) plaques stained with thioflavin-S in 6–7-month-old mice.⁸⁵ Interestingly, the increase in hyperactive cells was already present before plaque formation in 1–2-month-old mice, and direct application of soluble $\mathrm{A}\beta$ elicited neuronal hyperactivation in wild-type mice, suggesting the crucial role of soluble $\mathrm{A}\beta$ in the early functional impairment in AD.

The oriens-lacunosum-moleculare (O-LM) interneurons is a type of SOM-expressing interneurons that controls CA1 pyramidal neuron activity. In vivo two-photon imaging of GFP-labeled O-LM neurons in APP/PS1 transgenic AD model mice revealed progressive axon loss and structural plasticity impairment of dendritic spines.⁸⁶ Fear learning-induced, cholinergic input-dependent spine gain was also affected. Moreover, in these mice, calcium imaging demonstrated decreased O-LM interneuron responses to aversive stimulation. Thus, structural and functional imaging of this interneuron subtype identified impaired local circuit function and remodeling associated with memory impairment in AD.

Poll et al. used mice acquired from crossing of c-fos promoter GFP transgenic mice with APP/PS1 transgenic AD model mice and imaged the dynamics of their GFP-expressing CA1 pyramidal neurons over several days after contextual fear conditioning to assess the neuronal correlates of memory trace, referred to as an “engram,” in AD model mice.⁸⁷ Learning-induced GFP expression and partial reactivation of engram cells during recall is preserved in AD mice. However, additional ensembles that encoded novel contextual information interfered with the intact engram and memory recall, indicating that memory trace interference is a key process that underlies memory deficits in AD.

To better comprehend memory deficits and associated hippocampal dysfunction in AD, it is vital to explore hippocampal representations of spatiotemporal information during behavior and learning. Chronic calcium imaging of the CA1 pyramidal neurons in App knock-in AD model mice navigating a virtual environment showed that place cells were preferentially and progressively reduced near $\mathrm{A}\beta$ aggregates, whereas time cells remained unaltered.⁴⁷ These results reveal the differential impact of $\mathrm{A}\beta$ aggregates on two major modalities of episodic memory.

To date, only a few studies have examined abnormal hippocampal neural circuit activity in mouse models of brain disorders other than AD using two-photon imaging. Calcium imaging of an awake, chronic mouse model of temporal lobe epilepsy demonstrated that GABAergic neurons are preferentially recruited during spontaneous interictal activity in the CA1 region.⁷⁹ A study investigated a neural substrate of cognitive deficits in $\mathit{Df}(\mathit{16}{)}^{+/-}$ mice, a mouse model of 22q11.2 deletion that is a common genetic risk factor for cognitive dysfunction and schizophrenia, via two-photon calcium imaging of CA1 pyramidal neurons.⁸⁸ This revealed a significant deficit in goal-oriented learning and reduced stability and a lack of reorganization of place cell maps toward the goal location. Another study imaged the formation and plasticity of a CA1 place cell map during spatial learning using VR in a mouse model of autism spectrum disorder (ASD) that lacks the synaptic scaffold protein Shank2.⁴⁵ The Shank2-deficient ASD model mice ran more, received more rewards, and exhibited more enhanced goal-directed behavior than the wild-type mice. Interestingly, in the CA1 of these mice, the learning-induced over-representation of landmarks was absent, while the over-representation of rewards was substantially increased. This indicates that abnormally distorted hippocampal cognitive mapping may underlie cognitive and behavioral abnormalities in a subset of ASDs. Altogether, these findings collectively indicate that while a mouse model of 22q11.2 deletion syndrome shows deficits in hippocampal reward over-representation and goal-oriented learning,⁸⁸ a specific deletion of the Shank2 gene leads to substantially augmented reward over-representation and enhanced goal-directed behavior.⁴⁵ The hippocampal reward representation and related behavior can thus be altered bidirectionally in mutually opposite directions in different mouse models of neurodevelopmental disorders.

3.2.1.

Granule cells and mossy cells

The dentate gyrus (DG) is located deeper than CA1 and is thus technically more difficult to image. One of the early studies imaged dentate granule cell activity in awake mice via an imaging cannula that left the hippocampus intact.⁹⁸ Granule cells were labeled with the red indicator R-CaMP1.07 and imaged 600 to $800\mu \mathrm{m}$ below the hippocampal CA1 surface using 1040-nm excitation, although they could also be imaged using the green indicator GCaMP6s with 920-nm excitation. Individual granule cell activity patterns were generally sparse, were heterogeneously associated with locomotion, and varied flexibly across days.

Another early study replaced a portion of the cortex and CA1 with an imaging cannula to image the activity of GCaMP6f-labeled adult-born granule cells (abGCs) and mature granule cells (mGCs).⁹⁹ abGCs exhibited higher activity and less spatial tuning than mGCs. Optogenetic silencing of abGCs showed their active participation in contextual encoding and discrimination. Moreover, abGCs and mGCs were differentially recruited to interictal epileptiform discharges in a mouse model of chronic epilepsy.¹⁰⁰ The same group used this technique further to contrast the activity of mossy cells, another major excitatory principal cell type, in the hilus of the DG with granule cells.¹⁰¹ Their characterization showed that mossy cells are more active than granule cells, exhibited significant but more diffuse spatial tuning than granule cells, and discriminated contexts through robust remapping.

In the DG, granule cells transform information from the entorhinal cortex into a sparse output. Chronic imaging of neuronal activity in the DG, CA3, and CA1 revealed that granule cells in the DG encode stable spatial information over multiple days,¹⁰² whereas pyramidal neurons in the CA1 and CA3 exhibit highly context-specific and constantly changing spatial representations. A follow-up study by Cholvin and Bartos examined whether left-right lateralization in spatial and contextual coding properties existed in granule cells.¹⁰³ They did this by imaging the activity of granule cells in the left or right DG. Despite similar mean activity levels in both hemispheres, the granule cells in the left DG displayed higher spatial tuning, higher across-day stability, less run-to-run variability, and greater remapping, implying that contextual discrimination and contextual generalization is higher in the left and right DG, respectively.

The activity of GCaMP6s-labeled granule cells and jRGECO1a-labeled inputs from the medial entorhinal cortex was imaged via the dual-color in vivo two-photon calcium imaging.¹⁰⁴ This study revealed sparse and synchronized activity of the dentate granule cell populations associated with entorhinal cortex activation in immobile mice. These events incorporated place- and speed-coding granule cells, and these cells were responsive to environmental changes. Furthermore, they were significantly similar to population activity patterns during locomotion.

Imaging of the dentate granule cell activity in mice running on a cue-manipulated treadmill identified robust sensory cue responses in these cells.¹⁰⁵ Such “cue cell” responses are stimulus-specific, more stable than place cell responses, and are associated with the suppression of neighboring cells. This suggests that the DG performs parallel processing of spatial and sensory cue information during the first stage of hippocampal information processing.

3.2.2.

Neurogenesis and astrogenesis

The mechanisms of neurogenesis and astrogenesis in the DG were also explored using two-photon microscopy. Chronic in vivo two-photon imaging was used to follow genetically labeled neural stem cells (NSCs) and their progeny in the mouse hippocampus.^106,107 The studies revealed that achaete-scute homolog 1 (Ascl1)-targeted NSCs underwent limited proliferative activity before they terminally differentiate, whereas a fraction of GLI family zinc finger 1 (Gli1)-targeted NSCs displayed long-term self-renewal in the adult hippocampus.

Dentate astrocytes were deemed to derive from perinatal and adult radial glia-like NSCs (rNSCs). However, a study that assessed the generation and proliferation of astrocytes in the DG across the lifespan identified that the predominant source of newborn astrocytes in the adult DG are not rNSCs, but are locally dividing astrocytes in neurogenic niches.¹⁰⁸ In this study, intravital two-photon imaging of RFP-labeled astrocytes in the adult hippocampus encompassing the dorsal CA1 to the DG showed that astrocytes nearby the hippocampal fissure generated a few astrocytes through local proliferation.