Introduction

The hippocampus plays a critical role in spatial and contextual memory processes by forming cognitive maps of environments (Lisman et al., 2017; Moser et al., 2008; O’Keefe and Dostrovsky, 1971) that can support effective navigation towards learned salient locations (Robinson et al., 2020). During navigation many pyramidal cells (PCs) in the dorsal hippocampal CA1 area (CA1PCs) exhibit action potential (AP) firing that is restricted to spatial locations, the cell’s place fields (PFs). In different environments the activity of CA1PCs reorganizes to collectively form distinct representations, i.e. creating specific maps of each environment. In a novel environment, some CA1PCs are active already during the first traversal of their PFs, but many cells become place cells after some delay (Dong et al., 2021). Furthermore, spatial maps also gradually reorganize during continuous or repeated exploration in already familiar environments due to changes in the activity of individual PCs, including the recruitment of new place cells into the representation (Dong et al., 2021; Keinath et al., 2022; Mankin et al., 2012; Rubin et al., 2015; Ziv et al., 2013). Despite the fundamental importance of the dynamic hippocampal population code in spatial navigation, the mechanisms initiating and mediating new PF formations (PFFs) are still poorly understood.

How can place cells acquire their PFs? While cells that are already active at the first traversal of their PFs may simply receive sufficient active synaptic inputs ab ovo, delayed PFF is assumed to require dynamic changes in synaptic inputs to or the intrinsic excitability of CA1PCs. The cellular mechanisms underlying de novo formation of PFs have been elusive due to difficulties in measuring the subthreshold activity pattern that drives AP firing. However, recent studies have shed light on possible mechanisms. Most strikingly, in vivo patch-clamp recordings in head-fixed mice running on a familiar cue-rich treadmill revealed the spontaneous appearance of new PFs in previously silent cells, following the sudden occurrence of a robust, prolonged depolarization driving complex spike burst (CSB) firing (Bittner et al., 2015). Such CSBs are thought to be driven by dendritic Ca²⁺ spikes (plateau potentials) mediated by voltage-gated Ca²⁺ channels and NMDA receptors (Grienberger et al., 2014; Takahashi and Magee, 2009). The formation of new PFs by plateau potentials is thought to be mediated by synaptic plasticity obeying a non-Hebbian rule, whereby the excitatory input synapses activated in a ∼2-second-long time window around the CSB are rapidly potentiated. This form of plasticity was named behavioral time scale synaptic plasticity (BTSP; Bittner et al., 2017; Milstein et al., 2021). After initial electrophysiological demonstration, spontaneous PFF with BTSP- like characteristics was observed in CA1PCs by in vivo [Ca²⁺] imaging as well (Adoff et al., 2021; Grienberger and Magee, 2022; Priestley et al., 2022; Rolotti et al., 2022). Confirming the mechanism, PFF with features similar to spontaneous BTSP-like PFF could also be induced artificially by strong sustained electrical or optical depolarization of a cell, a stimulus assumed to evoke Ca²⁺ plateau potentials (Bittner et al., 2015; Bittner et al., 2017; Diamantaki et al., 2018; Fan et al., 2023; Milstein et al., 2021; Rolotti et al., 2022). In summary, BTSP evoked by CSBs has been proposed to serve as the main mechanism driving de novo PFF; however, its triggering mechanisms are enigmatic and its prevalence under different behavioral conditions is not yet known.

Whereas BTSP is a suitable and attractive candidate, other mechanisms may also contribute to PFF. In vivo patch-clamp recordings by Cohen et al. (2017) from CA1PCs in head-fixed mice running in a novel virtual environment showed that most of the delayed PFFs did not begin with CSBs, but with a weak subthreshold spatially tuned depolarization. This result suggests the involvement of synaptic plasticity mechanisms other than BTSP, such as e.g. spike timing dependent plasticity (STDP; Bi and Poo, 1998; Magee and Johnston, 1997; Markram et al., 1997) or cooperative plasticity mechanisms operating locally in dendrites (Golding et al., 2002; Kim et al., 2015; Mago et al., 2020). Furthermore, network mechanisms may also drive the emergence of new PFs. It has been shown that optogenetic stimulation of a small subset of CA1PCs at a particular location induced persistent PFF or remapping in non-stimulated cells, suggesting that changes in synaptic coupling and circuit dynamics of PCs and inhibitory interneurons can also lead to reorganization of spatial maps by unmasking latent PFs (McKenzie et al., 2021). Also pointing to a network effect, optical induction of BTSP-like PFF has been shown to be restricted by lateral inhibition, strongly limiting the number of CA1PCs that can simultaneously form new PFs by BTSP (Rolotti et al., 2022). Altogether, a comprehensive understanding of the contribution of possibly diverse cellular mechanisms underlying PFF during various forms of remapping processes is lacking. BTSP has several unique features when probed by somatic [Ca²⁺] imaging (Grienberger and Magee, 2022; Priestley et al., 2022). First, the prolonged CSB results in large [Ca²⁺] transient during the initial PFF event, typically followed by weaker Ca²⁺ signals on consecutive traversals through the PF. Second, due to the long and asymmetric temporal kernel of the plasticity (favoring potentiation of inputs active 1-2 seconds before the CSB) a substantial backward shift in the spatial position of the PF center can be observed on linear tracks after the formation lap. Third, the width of the new PF is generally proportional to the running speed of the animal during the PFF event. In contrast, PFF by other synaptic or network mechanisms is not expected to exhibit such properties. Using these features, we aimed to characterize PFFs as mediated by BTSP- or non-BTSP-like mechanisms in large populations of CA1PCs. We performed two-photon [Ca²⁺] imaging of CA1PCs in head-fixed mice performing a spatial navigation task in familiar and partially novel virtual corridors. We find that BTSP is prevalent but is not the only mechanism driving PFFs, as many new PFs formed without exhibiting any of the characteristics of BTSP. Furthermore, we show that robust CSB-like somatic activity is often ineffective in inducing new PFFs, suggesting that additional factors are required than just CSBs to successfully initiate new PFs.

Results

Place modulated activity of CA1PCs in mice navigating in a virtual corridor

In vivo two-photon [Ca²⁺] imaging of CA1PCs was performed from the dorsal hippocampus of head-restrained double transgenic mice (Thy1-GCaMP6s and Cre-dependent td-Tomato), injected with diluted Cre-recombinase expressing AAV (Figure 1A-B). Mice were trained to run in an ∼8- meter-long virtual corridor for a water reward. The wall of the virtual corridor contained 6 well-discernable visual landmarks, the last of which indicated the reward zone, where mice were required to lick for water reward (Figure 1C). Well-trained mice ran with a relatively constant speed of ∼30-40 cm / s through the corridor and slowed down and started licking just before entering the reward zone (Figure 1D), indicating that they learned the position of the reward zone. Imaging was performed in well-trained mice across multiple days (n = 163 imaging sessions of 45 mice). The imaging field of view contained ∼500-1000 CA1PC somata (Figure 1B), which showed highly variable activities during the session (Figure 1E - H); some cells were silent whereas others showed moderate to high activity rates. Plotting the activity as a function of the animal’s location in the virtual corridor showed high variability in the spatial tuning of active cells (Figure 1I, for PF determination see Methods). The PFs of place cells tiled the entire length of the corridor with enrichments around the visual landmarks (Figure 1J). Place cells had variable number of PFs with widely different widths, covering a variable fraction of the virtual corridor (Figure 1K). These results demonstrate that the activity of CA1PCs in our experimental conditions is comparable to that observed in freely moving animals exploring large environments (Eliav et al., 2021; Lee et al., 2020; Rich et al., 2014).

Figure. 1

BTSP and non-BTSP-like mechanisms are both prevalent PFF mechanisms

During navigation in the familiar virtual corridor, new PFs often appeared suddenly at various spatial locations during the session (Figures 2, 3). To investigate the possible mechanisms underlying such a robust change of activity, we examined the dynamics of the newly formed PFs. If BTSP is the mechanism generating new PFs in CA1PCs, the PFs are expected to have the following characteristics: first, since BTSP is induced by a Ca²⁺ plateau potential and consequential prolonged CSB (Bittner et al., 2015), we expect that the amplitude of the [Ca²⁺] transient in the formation lap is larger than that in the following laps. Second, because of the asymmetrical temporal kernel of BTSP (Bittner et al., 2017), the center of mass (COM) of the new PF is expected to shift to the backward direction compared to the peak activity in the formation lap. To quantify these dynamics, we calculated two parameters for each newly formed PF: the formation lap gain and the initial COM shift (see Methods and Priestley et al., 2022). We then categorized all newly formed PFs in our recordings based on these two parameters. Using a conservative approach, we categorized a new PF to be formed by a BTSP-like mechanism if it had both positive gain and negative shift values (Figure 2A; n = 310 new PFs), whereas new PFs exhibiting neither positive gain nor negative shift were considered as non-BTSP-like events (Figure 2B; n = 59). All other newly formed PFs (no-gain with backward shift and gain without backward shift) were excluded from further analysis. This analysis thus demonstrated that, whereas a large fraction of newly formed PFs were generated by a BTSP- like mechanism, a substantial fraction of new PFF was produced by a different process.

Figure 2.

Figure 3.

Comparison of the [Ca²⁺] transients of BTSP-like (n = 310) and non-BTSP-like (n = 59) new PFs in the laps before, during and after their formation showed that, apart from the strongly different [Ca²⁺] signal amplitudes in the formation lap, [Ca²⁺] transients within the PF on the subsequent laps had similar amplitudes (Figure 2C) and ratios of active laps (0.86 ± 0.03 vs. 0.86 ± 0.06), indicating that the basic characteristics of activity in the stabilized new PFs were comparable between the two groups. Furthermore, both types of new PFs exhibited similar distributions along the virtual track, tiling the whole corridor with enrichments around the visual landmarks (Figure 3A).

A further characteristic feature of the BTSP mechanism is that, due to its seconds-wide temporal kernel, the width of the new PF is proportional to the running speed of the animal during the formation lap, i.e. the potentiated inputs cover a wider spatial sector if the animal travels longer distance during the plasticity time window (Bittner et al., 2017). As an independent confirmation that newly formed PFs in our categorization of PF groups genuinely differed in the mechanisms underlying their formation, we found a significant positive correlation between formation lap running speed and PF width in the BTSP-like but not in the non-BTSP-like PF category (Figure 3B). Because of the difference in the number of events in the two groups, we randomly subsampled the BTSP-like events to the sample size of the non-BTSP-like PFF events 10 000 times and performed regression analysis. This bootstrapping revealed that both the r and p values of the fit to the non-BTSP data fall outside the 95% confidence interval of the bootstrapped BTSP values, therefore we conclude that the lack of significant correlation between the running speed and the PF width for the non-BTSP-like events is not the consequence of the smaller sample size.

Concentrated PFF associated with representational switch

We next examined the distribution of new PFF events across imaging sessions and laps. While in most of the sessions the number of newly formed PFs was moderate, few sessions contained an unusually large number of such events (Figure 3C). When we examined the distribution of new PFFs in these sessions across individual laps, we found that the majority of PFF events occurred within a short period over a limited number (3-10) of consecutive laps (Figure 3D). These concentrated PFF periods were termed as PFF ‘surges’, which included PFs formed both by BTSP- and non-BTSP-like mechanisms (non-BTSP / BTSP ratio: 0.57). The unusually high rate of new PFF events during PFF surges led us to investigate whether they were associated with a more general network-wide reorganization of the CA1PC representation of the virtual corridor. Indeed, we found that the lap-by-lap correlation of activity in the whole population of imaged CA1PCs underwent a drastic and rapid switch between two stable states before and after the PFF surge. A closer look indicated that the initial representation began to disorganize as the surge started to ramp up, and a new coherent representation started at the lap with the peak PFF rate. In contrast, no such representational switches were observed in sessions without PFF surges (Figure 3D bottom). Finally, when we compared the [Ca²⁺] dynamics of PFFs during surges with those of PFF outside the surges, we found similar [Ca²⁺] amplitudes in both the BTSP-like and non-BTSP-like groups, except for higher Ca²⁺ signals in the first lap after the formation lap in the non-BTSP group during surges compared to outside surges (Figure 3E). The period including the PFF surge and representational switch was not accompanied by any noticeable change in the behavior of the animals as assessed by running speed and lick rate. Altogether these results reveal unique, short epochs during task-engaged behavior when rapid reorganization of the population activity of CA1PCs occurs, in part due to new PFFs by both BTSP- and non-BTSP-like mechanisms.

New PFF in a novel environment is mediated by both BTSP- and non-BTSP-like mechanisms

CA1PCs are known to rapidly remap in novel environments to create a specific representation of newly explored locations. We next sought to determine the contribution of BTSP- and non-BTSP- like mechanisms to new PFFs in a novel environment by extending the familiar virtual corridor with a 6-m-long segment decorated with never-seen visual landmarks during the imaging session. Animals traversed this novel segment with an average speed comparable to that in the familiar portion of the maze. The maze extension did not induce a detectable change in lick frequency. At the end of the familiar segment of the 14-meter-long corridor, the animals reduced their running speed and displayed anticipatory licking just before entering the reward zone similar to their performance in the standard 8-meter-long corridor. Population-level analysis revealed a reorganization of the neuronal representation, as demonstrated by a decreased population vector correlation between the extended and original maze, despite high intra-segment correlations (mean correlation coefficients: original maze odd vs. even: 0.50±0.21, original maze odd vs. extended maze even: 0.08±0.06, extended maze odd vs even: 0.36±0.12; n = 6 mice; Repeated Measures ANOVA followed by Tukey post hoc test: original maze odd-even vs. original maze odd - extended maze even: p < 0.001, extended maze odd - even vs. original maze odd - extended maze even: p = 0.002, original maze odd - even vs. extended maze odd - even: p = 0.114). Consistent with previous studies (Dong et al., 2021; Priestley et al., 2022; Sheffield et al., 2017), we observed many new PFs already during the first traversal in the novel virtual environment, and additional new PFs formed during subsequent laps, although with lower probability (Figure 4A, B). The frequency of new PFFs was the highest in the first few laps for both BTSP- and non-BTSP-like events. Analyzing the somatic [Ca²⁺] transients revealed that their amplitudes were very similar to those observed in the familiar environment for both the BTSP- and non-BTSP-like PFF events during the 1^st traversal (Figure 4C). Interestingly, the relative prevalence of non-BTSP-like PFF was higher in the first laps of the novel environment than that observed in the familiar environments (novel corridor lap 1-2: n = 24 non-BTSP, n = 44 BTSP in 6 mice, non-BTSP/BTSP ratio = 0.55; familiar corridor: n = 59 non-BTSP, n = 310 BTSP in 30 mice, non-BTSP/BTSP ratio = 0.19; Chi-square test: p < 0.001), whereas in the subsequent laps BTSP- like PFF dominated again (novel corridor region, laps 3-31: non-BTSP / BTSP ratio = 0.16).

Figure 4.

Large somatic [Ca²⁺] transients per se are not sufficient to induce new PFs

The above analysis demonstrated an important role for BTSP in new PF formations; however, it also indicated that BTSP is not an exclusive mechanism of PFF, as new PFs could also develop rather gradually without strong initial somatic activity during the formation lap. Finally, we asked the question in reverse: is the occurrence of strong somatic Ca²⁺ activity sufficient to trigger the formation of new PFs? To address this, we selected CA1PCs exhibiting at least one BTSP-like newly formed PF in a given imaging session in the familiar corridor and identified all somatic Ca²⁺ transients in that session that had amplitudes larger than the Ca²⁺ transient initiating the BTSP-like PFF event (if multiple PFs were formed then the largest formation lap [Ca²⁺] transient was used; Figure 5). We found that, in approximately one quarter of the sessions, BTSP-inducing [Ca²⁺] transients could be matched with at least one larger [Ca²⁺] transient that failed to initiate new PFF (‘solitary events’, Figure 5A - B). Our results together suggest that a large somatic [Ca²⁺] transient is neither necessary nor sufficient to induce the formation of new PFs.

Figure 5.

Discussion

In the present study we performed in vivo two-photon [Ca²⁺] imaging from hundreds of dorsal hippocampal CA1PCs while mice ran for a water reward in a familiar and a partially novel virtual environment. During the exploration of familiar and novel corridors, PFF events occurred in CA1PCs spontaneously along the entire corridor. Many of these events had characteristics reminiscent to those reported earlier for BTSP-like PFF events, but a significant fraction of PPFs had small [Ca²⁺] transient amplitudes in the formation lap and did not show the characteristic backward shift in the COM of the newly formed PF. Such non-BTSP-like PFs not only occurred spontaneously in the familiar environment but were also formed during representational switches within familiar and in the first laps of a novel environment. We also observed that in many behavioral sessions when a CA1PC displayed a BTSP-like PFF event, there were [Ca²⁺] transients (‘solitary events’) with amplitudes higher than that of the formation lap of the BTSP-like PFF event, which did not induce novel PFs. These results demonstrate that BTSP-like PFFs are not the only mechanisms of new PFFs and that large somatic [Ca²⁺] transients are not sufficient per se to induce new PFs.

The lack of large initial somatic [Ca²⁺] transients – that are likely associated with strong bursts involving Ca²⁺ plateaus - during non-BTSP-like PFFs suggests two possible scenarios: they either do not involve excitatory synaptic plasticity in CA1PCs but rather result from a change of incoming excitatory or inhibitory synaptic inputs, or they are driven by other synaptic plasticity mechanisms that do not require global Ca²⁺ plateaus and CSBs to increase excitatory synaptic weights. Local synaptic plasticity in individual dendritic branches is a particularly intriguing possibility for the latter scenario, since it has been shown that PFF can be preceded by local dendritic activity in the future PF locations (Sheffield et al., 2017). This activity, presumably produced by clustered activation of Schaffer collateral synapses (Adoff et al., 2021) may induce local synaptic potentiation, whereby the inputs eventually become suprathreshold at the soma manifesting as non-BTSP-like PFF. Importantly, long-term potentiation (LTP) of excitatory synaptic inputs by local dendritic Na⁺ and/or NMDA spikes (d-spikes) can be induced by a single or a few repetitions of d-spikes (Mago et al., 2020; Remy and Spruston, 2007), and thus can be consistent with rapid generation of new PFs similar to BTSP. It remains to be explored whether the temporal coincidence kernel of local d-spike- induced LTP is on millisecond or second time scales, i.e. different or similar to that of BTSP. In contrast to synaptic plasticity induced by d-spikes, LTP induced by cooperative activity of clustered excitatory inputs without evoking APs or d-spikes (Mago et al., 2020) or LTP induced by STDP (Bi and Poo, 1998; Magee and Johnston, 1997; Markram et al., 1997) require larger number of repeated co-activations of pre- and postsynaptic cells and thus may be incompatible with fast PFF and global remapping. However, it cannot be ruled out that non-BTSP-like PFFs occurring after many traversals of the same location in familiar environments may involve STDP. Furthermore, since STDP in CA1PCs has been characterized under in vitro conditions, it is yet unknown whether its induction rules are similar in vivo, especially in the presence of active neuromodulatory systems, which may act as a reinforcement signal to influence STDP rules even at synapses with temporally distant activity via presumed eligibility traces (e.g. Fisher et al., 2017).

By showing that both BTSP- and non-BTSP-like PFFs can underlie changes in spatial tuning of CA1PCs and by dissecting their relative contribution under different remapping conditions, our study reconciles diverse and sometimes contradictory earlier observations. Although BTSP-like PFF events have been suggested recently as a dominant mechanism of new PFFs in the hippocampus (Bittner et al., 2015; Bittner et al., 2017; Grienberger and Magee, 2022; Milstein et al., 2021; Rolotti et al., 2022), previous results also indicated that PFs in a new environment frequently form without CSBs (Cohen et al., 2017). Recording from a large number of CA1PCs, we could estimate the prevalence of BTSP-like and non-BTSP-like PFFs during various forms of remapping (although we note that our estimates likely underrepresent non-BTSP-like PFF due to the limited sensitivity of the genetically encoded Ca²⁺ indicator to detect few APs). We found that BTSP dominates PFFs under conditions when new PFs are born in individual CA1PCs that are part of a relatively stable representation in a familiar environment, i.e. during gradual representational drift. On the contrary, the relative ratio of non-BTSP-like PFF was much higher in the first laps of running in a novel corridor segment. This may be consistent with a global remapping process in response to reconfiguration of the excitatory input patterns arriving to individual CA1PCs from upstream areas. As a result, in some CA1PCs no plasticity would be required to immediately generate a new PF, while in other CA1PCs plasticity of the newly activated inputs (e.g. by activating local d-spikes) may also contribute to non-BTSP-like PFF, with this process possibly facilitated by a change in brain state due to the novelty of the new corridor segment. These processes altogether could increase the contribution of non-BTSP-like PFF to the initial development of the novel representation of the newly explored corridor.

Our results are consistent with previous reports of increased PFF during initial exposure to novel environments (Priestley et al., 2022) and around newly introduced salient cues in a familiar environment (Grienberger and Magee, 2022), although certain differences may be apparent. In particular, Priestley et al (2022), who classified BTSPs similarly to our method but pooled all PFF events not clearly BTSP as an ‘other’ group, observed increased fraction of BTSP-like PFF on the first trial in a novel environment compared to a familiar environment (Priestley et al., 2022). Here, however, we have found an increase in the prevalence of non-BTSP-like PFF events during the initial exposure to the novel corridor. Yet, these results are not directly comparable for the following reasons: first, we defined the non-BTSP-like group as a subclass of ‘other’ PFFs based on multiple feature criteria to clearly separate them from BTSP-like PFFs; second, in the familiar environment we only analyzed PFs that were formed newly after an silent period in at least 16 laps, whereas (Priestley et al., 2022) analyzed all PFs including those with established activity from the first lap in an environment.

Interestingly, we also observed spontaneous surges of coordinated new PFF events during short imaging epochs involving a few laps in the familiar corridor. These PFF surges included a high fraction of both non-BTSP- and BTSP-like PFF events, and were associated with global remapping of the CA1 representation. The precise trigger and relevance of these representational ‘switches’ in a well-learned, invariant environment is currently unknown, but we speculate that – in the absence of alterations of sensory cues and task demands – they may indicate a change in the internal state of the animal, or a transition between relying on different signals for navigation (e.g. sensory cues vs self-motion), assigning distinct internal contexts to the same environment (Sanders et al., 2020). Although different CA1 population codes across subsequent (interrupted) explorations of the same maze have been demonstrated (Sheintuch et al., 2020), to our knowledge similar representational switches during a continuous behavioral epoch in an unchanged environment have only been described in the entorhinal cortex (EC; Low et al., 2021). It is possible that the representation switches we observed originate in the EC, which thus provides globally remapped inputs to the hippocampus similarly to the situation in a novel corridor. Overall, the higher ratio of non-BTSP- like PFFs in the initial laps of the novel environment and during representational switches (both ∼0.55) than that in the familiar environment (0.19) suggests that distinct cellular/synaptic mechanisms might contribute to gradual representational drifts and abrupt global remapping.

While sudden large Ca²⁺ activity at a location where a cell was previously inactive could often induce BTSP-like PFF, our results also demonstrate that a large fraction of even stronger Ca²⁺ activities failed to initiate PFF in the same cell. If we assume that the large (and often prolonged) somatic Ca²⁺ signals represent CSBs, this indicates either that the synaptic plasticity induced by the underlying plateau was not sufficiently strong to evoke somatic firing, or that the induction of plasticity may be gated by additional conditional factor(s), for example modulatory signals or inhibition. This latter possibility highlights the importance of elucidating the molecular components contributing to the initiation, expression and regulatory mechanisms governing the long-term increase in synaptic strength by BTSP. While research efforts of the past three-four decades provided an enormous amount of information regarding the molecular mechanisms underlying Hebbian synaptic plasticity, the molecular events taking place during BTSP are almost completely unknown. Further research will be crucial to reveal the similarities and differences in the molecular mechanisms underlying these fundamentally different synaptic plasticities, which will allow us to design molecular modifications that specifically alter only one of them to uncover its consequences on new PFFs and animal behaviors.

Materials and methods

Animals

A total of 45 (23 male and 22 female) Ai9xGP4.3 double-transgenic mice were used in this study. These animals were also involved in separate experiments unrelated to the current investigation. Mice were housed in the vivarium of the HUN-REN Institute of Experimental Medicine under a standard 12-hour light/dark cycle with ad libitum access to water and food. Water intake was restricted during the experiments (details provided below). All procedures adhered to the regulations outlined in the Hungarian Act of Animal Care and Experimentation (40/2013 [II.14]) and received approval from the Animal Committee of the Institute of Experimental Medicine, Budapest.

Virus injection

Mice were first anesthetized with an intraperitoneal injection of a mixture of ketamine hydrochloride (CP-ketamine, 62.5 mg/kg), xylazine (CP-Xylazine, 12.5 mg/kg), and promethazine hydrochloride (Pipolphen, 6.25 mg/kg). Local anesthesia was achieved by subcutaneous injection of ropivacaine hydrochloride (Naropine, 50-100 μl, 2 mg/ml). To protect the eyes, Opticorn A ointment was applied. The animal’s head was then fixed in a stereotaxic frame, and a craniotomy of ∼0.5 mm in diameter was made above the injection site. Four times 25 nl AAV vector (pENN.AAV.hSyn.Cre.WPRE.hGH, 3.3 x 10¹⁰ GC/ml, Addgene #105553-AAV9) was injected into the left hemisphere at coordinates AP: −2, DV: −1.1, ML: 1.8. with 2.3 nl/sec bouts separated by 1- minute intervals. Following surgery, buprenorphine hydrochloride (Bupaq, 0.1 mg/kg) and carprofen (Rimadyl, 25 mg/kg) were administered subcutaneously for analgesia and preventing inflammation. Carprofen (35 ng/ml) was also supplemented in the drinking-water for the following two days of postoperative care.

Implant surgery

After virus injection (median: 7, range: 2 to 32 days) animals were anesthetized with Isoflurane (1.5% in carbogen, flow rate 0.8 l/min). A 3 mm diameter craniotomy was created over the virus injection site and the cortex above the hippocampus was gently aspirated. A 3 mm diameter 1.7 mm long stainless-steel cannula with a glass coverslip attached to the bottom was inserted and glued to the surrounding skull. A custom-made 3D printed titanium head bar was cemented to the skull together with the cannula. Buprenorphine hydrochloride (Bupaq, 0.1 mg/kg) and carprofen (Rimadyl, 25 mg/kg) were administered subcutaneously for analgesia and preventing inflammation. During the 2 days of postoperative care, the animal’s drinking water was supplemented with 35 ng/ml of carprofen.

Behavioral training

Water-restriction started 24 hours before behavioral training (median: 4, range: 2 - 45 days after surgery). Throughout training days, animals received 5% sucrose solution as a reward during behavioral sessions. If an animal’s weight fell below 85% of its starting weight by the end of the training/imaging session, the animal was provided free access to water for 10 minutes in its home cage. On the first day of training animals were habituated to head-fixation in the behavioral setup and were encouraged to run on a virtual reality attached treadmill by receiving rewards via a lick-port at regular intervals. A virtual corridor was displayed on three flat screens placed in front and on the two sides of the animal’s head and synchronized with the movement of the treadmill (PyOgre virtual reality and Luigs & Neumann treadmill logged and synchronized via a LabVIEW software). The animal was also rewarded for licking the lick-port in the designated reward zone. Upon reaching the end of the corridor, animals were instantly teleported to the start. On the following days, the distance between running-related rewards was gradually increased until animals relied solely on operant rewards in the reward zone at the end of the corridor. Subsequently, the length of the virtual environment was gradually extended from 2 to ∼8 meters by adding new segments to the beginning of the corridor. Animals were trained until they completed at least 50 laps within 30 minutes (median = 22, range 11 - 99 days).

In vivo two-photon [Ca²⁺] imaging

Two-photon [Ca²⁺] imaging of CA1PCs was performed in vivo in the CA1 region of the left dorsal hippocampus in well-trained animals running typically 50 laps per session. A Femto-2D dual two-photon microscope equipped with a Nikon LWD 16X objective lens (NA = 0.80) was used and the laser power was set to ∼70 mW at the specimen, with a wavelength of 920 nm. Images of 512 by 505 pixels were acquired at a frame rate of ∼31 Hz with a spatial resolution of 0.89 x 0.89 µm / pixel.

Analysis of in vivo [Ca²⁺] imaging data

Motion correction, ROI detection, fluorescent trace extraction, and neuronal activity deconvolution were performed using the Suite2p software, followed by analysis with custom-written Python scripts. Suite2p-identified ROIs exhibiting unrealistic shapes were excluded from the study based on the consensus of three human experts (JM, ZN, MS) blinded to the fluorescent activity of these ROIs.

To account for background fluorescence originating from neuropil (tissue consisting of neuronal and glial processes surrounding the cell bodies), we performed neuropil correction. The corrected fluorescence (F) for each ROI was calculated by subtracting a scaled version (0.7x) of the Suite2p-estimated neuropil fluorescence (F_neu) from the raw fluorescence values (F_raw): F = F_raw – 0.7 X F_neu. For a conservative estimation of inferred neuronal activities, we used the Suite2p spks output containing neuronal activity events deconvolved from the underlying fluorescence traces. We filtered the spks output by the following process. Local noise was assessed around each non-zero spks value. Local maxima of the neuropil-subtracted fluorescence (F) were identified between each non-zero spks value and the following nineth frame. These maxima were replaced with the average value of three neighboring fluorescence data points around the peak (F_max). Next, a linear baseline was fit to the fluorescence data points from the -30th to the -20th frame preceding each non-zero spks value. The local noise of the fluorescence (SD_local) was calculated as the root-mean-square error (RMSE) of the linear fit. The baseline fluorescence (F_base) for each activity was defined as the linear prediction at the -20th frame position. Global noise was characterized by the standard deviation (SD_global) of F values within the range 1.5 times below and 0.5 times above the median F value of the entire fluorescence trace. A non-zero spks value was considered valid only if the corresponding fluorescence transient amplitude (F_max - F_base) exceeded both the local and global noise. The resulting, noise-reduced inferred activity values were used for all subsequent analyses.

For comparing the amplitude of [Ca²⁺] transients in laps before, during and after PFFs, we have also calculated the change in fluorescence relative to baseline (ΔF/F). Each neuropil-corrected fluorescence value was divided by the median of the preceding 1000 fluorescence values. For values earlier than the 1000^th time point we used the following 1000 timepoints for median calculation. The neuropil-subtracted ΔF/F traces from multiple ROIs during PF transitions were aligned, if at least one inferred activity occurred during the transition. The ΔF/F traces were smoothed using a 5-point wide symmetrical moving average. The difference between each smoothed point and the 4th point before it was then calculated. The position of the maximum in this difference curve, corresponding to the steepest rising phase of the fluorescence transient, was used for trace alignment. For place field transitions without inferred events, the traces were aligned to their midpoint. Following trace alignment, an asymmetrical window encompassing −50 to 150 data points around the alignment point was isolated. The mean ΔF/F trace within this window across all transitions was calculated, and the position of its maximum determined. The peak amplitude for each trace was then calculated by subtracting the median ΔF/F value of the isolated segment from the ΔF/F value at the maximum of the mean trace. Finally, average and standard deviation of these peak amplitudes were calculated across all traces.

Newly formed PF demarcation and categorization

Spatial demarcation

The 832 cm long virtual environment was discretized into 4.24 cm wide space bins and data from bins with average running speed lower than 4 cm/s were excluded. Odd and even laps were separated and used for calculating mean inferred activity in each spatial bin across laps (odd and even tuning curves). ROIs with smaller than 0.25 Pearson’s product-moment correlation coefficient between odd and even tuning curves were excluded from further analysis. We assessed spatial information in each ROI using Skaggs’s metric (Skaggs et al., 1992). Statistical significance was assessed using a bootstrapping procedure performed in Python. The inferred activity data for each lap was circularly shifted with a random amount in time and Skaggs’s information content was recalculated 1500 times to create a null distribution. ROIs with Skaggs’s score exceeding the 99th percentile of the shuffled distribution were selected for further processing (Eliav et al., 2021). Tuning curves for each ROI were calculated as the mean inferred activity in each spatial bin across laps. Peaks and their footprints, the area they cover on the space axis, in the tuning curves were identified using functions from the SciPy library: find_peaks (peak_height = 6), peak_prominences, and peak_widths (rel_height = 0.93). Adjacent peaks were merged if their footprints overlapped or the intervening minimum value between peaks was greater than half the amplitude of the larger peak. Spatial boundaries were calculated as spacebin at the left and right edges of spatial footprints. To identify spatial regions with statistically significant spatial tuning, we employed a localized information metric. First, for each peak identified in the tuning curves, we extended its footprint by half its width in both directions. Within these extended regions, a local Skaggs’ information metric was calculated. To assess the significance of this metric, a null distribution was generated using bootstrapping as above (1500 iterations). The inferred activity for the selected region was circularly permuted with a random time shift in each lap, followed by recalculating the Skaggs’ information content. The 95^th percentile of this null distribution served as the significance criterion to identify regions exhibiting statistically high spatial location selectivity. Regions with statistically high spatial selectivity were considered as PFs if they had at least one inferred activity in at least 30% of the laps.

Temporal demarcation

To ensure a conservative estimate of PF formation time and avoid misidentification of small neuronal activities or noise as formation events, the inferred activity data below 64 a.u. was set to zero. This step excluded potentially ambiguous neuronal activities that might be difficult to confirm visually in the motion-corrected videos (see in Quality control of signal isolation for BTSP and non-BTSP related place field formations).

For analyzing the formation and end laps of newly formed PFs, we used this thresholded inferred activity data, within the previously identified PF spatial boundaries but without filtering based on running speed of the animal. To account for the potential for place fields to exhibit a backward shift in activity relative to the formation lap, we extended the previously identified spatial boundaries in the forward direction (to the right) by 5 space bins or until the end of the corridor. This approach ensured that formation events occurring at locations forward compared to the place field’s spatial center (as expected for BTSP) would be detected. We identified potential PFF events by analyzing the inferred activity data across laps. First, we constructed a PF activity matrix by calculating the average activity for each spatial bin within the PF across all laps. To ensure minimal pre-formation event activity, only laps after the 17^th were considered for potential formation events. A lap was classified as a formation lap only if no more than 10% of the previous laps had a mean activity exceeding 1% of the mean activity of the potential formation lap. In maze extension experiments (Figure 4), the minimum 17 pre-formation laps requirement was ignored. A PF was considered to be terminated if at least five consecutive laps with no activity occurred or the final lap of the session was reached.

Categorization of newly formed PFs

We employed a three-step process to categorize newly formed PFs. First, we excluded PFs that were formed in the first 17 laps. Second, we excluded PFs that exhibited a lifetime < 9 laps. Finally, we addressed activity consistency within PFs. Lap-wise maximum activity rates of the PF activity matrix were normalized to the peak activity rate across all laps. Place fields were excluded if the number of silent laps (laps with normalized activity below 10%) exceeded 8 laps. Place fields were categorized as showing formation lap gain if the peak activity value in the formation lap was greater than the average peak activity value observed across the subsequent three active laps. For further categorization of the newly formed PFs, the center of mass (COM) was calculated for each active lap and smoothed across laps using a running average with a kernel size of 3. The formation point of the PF was defined as the activity COM in the formation lap. Finally, the mean COM value for the first three post formation active laps was calculated. If this mean value was smaller than the formation point, the PF was categorized as exhibiting an initial backward shift. To exclude misinterpretation of occasionally observed continuous drift of the PF COM as BTSP-like initial shift, we also quantified continuous backward drift (termed ‘drift’) in the PF. For this, we first isolated the smoothed COM values for each lap, starting from the fourth active lap after the formation. We then performed a linear regression analysis on these COM value series. If the calculated COM value in the first lap after formation lap extrapolated from the regression line was not larger than the formation points or the Spearman’s rank correlation coefficient between the COM values and their sequential labels was not statistically significant (p < 0.05) with a negative value, we categorized the PF as lacking significant drift.

We categorized PF formation events based on their activity patterns during and after formation. PFFs were considered BTSP-like if they fulfilled all of the following criteria: 1) showed formation lap gain, 2) followed by a backward shift in the place representation, and 3) showed no significant further drift. Conversely, PFFs that lacked both 1) formation lap gain and 2) an initial backward shift during formation, regardless of the PFs’ drift characteristics, were classified as non-BTSP-like. In our initial categorization we found 808 candidate BTSP-like and 165 non-BTSP-like PFFs, which were further analyzed as detailed below. An additional 560 PFs could not be included in the above two groups (238 PFs with formation lap gain but without backward shift; 312 PFs with no formation lap gain but with backward shift); these PFs were excluded from further analysis in this study.

We emphasize that we employed a conservative approach for categorizing PFFs: we only included PFF events in our analysis if they fulfilled all criteria of either BTSP-like or non-BTSP-like groups and excluded PFFs that failed on at least one criterion. Yet, it should be noted that BTSP might occur without positive gain (if the cell fires CSBs in the PF on multiple laps after the formation lap) as well as without negative shift (since the plasticity kernel extends over zero to positive timings as well). In addition, the nonlinear relationship between electrical activity and [Ca²⁺] signals, as well as the lack of sensitivity of GCaMP6s to detect single APs, may add uncertainty to our classification.

Quality control of signal isolation for BTSP and non-BTSP related PF formations

Automated ROI segmentation by Suite2p can encompass processes from multiple cells which can lead to erroneously reported [Ca²⁺] dynamics during PF activity. To address this and facilitate human confirmation of automated ROI segmentation, we isolated short motion corrected video segments from place cells and their surroundings exhibiting BTSP- (n = 808) and non-BTSP-like (n = 165) PFFs. To quantify the spatial footprints of active neurons during the animal’s PF transitions, we employed a method to assess local coherence of fluorescence signals within the field of view. For each pixel in an isolated video frame, we assigned an intensity by averaging the Pearson’s product-moment correlation coefficients between its time series and the time series of its eight nearest neighbors. In the resultant grayscale image, pixels corresponding to active cells were given higher intensity values, reflecting the synchronized activation of these pixels during neuronal spiking. The contrast of the resulting grayscale image was further enhanced with Scikit-image’s exposure.adjust_sigmoid function (cutoff = 0.8, gain = 20). Finally, to incorporate the temporal aspect of pixel activation patterns, we applied a pseudo-color map. This map assigned a color to each pixel based on the center of mass of its intensity time series. In essence, these spatial footprints showed the shape of active cells with their subregions colored based on their average timing of activity. The spatial footprints of each PF transition were visually inspected and ROIs with multicolored subregions were excluded. All motion corrected video segments for the remaining ROIs were visually inspected.

Detection of solitary [Ca²⁺] transients

Solitary [Ca²⁺] transients were identified by the following process. First, for each BTSP-like PFF event, we calculated the maximum ΔF/F value by analyzing timepoints recorded during the animal’s traversal across the width of the PF in the formation lap extended by 5 space bins to the right but not further than the end of the corridor. This maximum amplitude served as the parameter for peak detection (Scipy, ‘find_peaks’, prominence = 1, distance = 20) to identify peaks equal to or larger than this formation lap peak amplitude. Second, [Ca²⁺] transients within the spatial boundaries of any existing or newly formed PFs were excluded. To further ensure that the detected solitary peak wasn’t part of a short-lived place modulated activity, we calculated the maximum ΔF/F value from the previous and following 5 laps within 11 spatial bins centered around the peak containing space bin. If the maximum ΔF/F in this area exceeded the median plus two SD of the ROIs’ ΔF/F values from the entire session, the solitary event was excluded. For each ROIs only the largest solitary peak was used for calculating the mean waveform, but we note that in many cells multiple [Ca²⁺] transients in a given session fulfilled the criteria for solitary events.

Lap-by-lap population vector correlations

For each Suite2p ROI, the non-thresholded inferred activity data collected during each space bin transition was averaged. Subsequently, Pearson correlation coefficients were calculated between the mean activity of the ROIs at each spatial location for every possible lap pair. Finally, the lap-by-lap PV correlation for a given lap pair was determined by averaging these location-specific correlation coefficients.

Odd-even lap population vector correlations

To evaluate population-wide activity changes upon the extension of the familiar corridor by a new maze segment, we computed population vector correlations between odd and even laps for the first 6 meters of the familiar pre-extension maze (excluding the reward zone) and the 6-meter-long new extension part of the maze. For each ROI, we first calculated the mean non-thresholded inferred activity from each spatial bin for both odd and even laps from the session immediately preceding novel maze segment insertion and during the extended maze experiment. These population vectors were then used to calculate the average pairwise Pearson correlations between intra-corridor and cross-corridor odd-even lap data for each animal. Finally, we compared the means of these correlations across animals for both intra-corridor and cross-corridor population vector correlations.

Statistics

Statistical analyses were performed using Statistica (General Linear Model with repeated measures or mixed ANOVA; software version 14.0.1.25, Tibco) or Python (correlations, bootstrapping). Tukey’s test was applied for post hoc comparisons after a significant effect was demonstrated by the main ANOVA.

Acknowledgements

ZN is the recipient of a European Research Council Advanced Grant (ERC-AG 787157). JKM is supported by a European Research Council Consolidator Grant (ERC-CoG 771849) and the International Research Scholar Program of the Howard Hughes Medical Institute (55008740). The work of JKM and ZN is supported by the National Brain Research Program (NAP 3.0) of the Hungarian Academy of Sciences (NAP2022-I-1/2022). The financial support from these funding bodies is gratefully acknowledged.

Additional information

Author Contributions

MS, GO, IPL, ZN designed the experiments; GO, IPL, MS performed the in vivo imaging experiments, MB developed analysis of PF categories; MS developed analysis tools and analyzed the data guided by ZN and JKM; MS, JKM, ZN wrote the manuscript.

Declaration of Interest

The authors declare no competing interests.