Calbindin-Expressing CA1 Pyramidal Neurons Encode Spatial Information More Efficiently

Abstract Hippocampal pyramidal neurons (PNs) are traditionally conceptualized as homogeneous population. For the past few years, cumulating evidence has revealed the structural and functional heterogeneity of hippocampal pyramidal neurons. But the in vivo neuronal firing pattern of molecularly identified pyramidal neuron subclasses is still absent. In this study, we investigated the firing patterns of hippocampal PNs based on different expression profile of Calbindin (CB) during a spatial shuttle task in free moving male mice. We found that CB+ place cells can represent spatial information more efficiently than CB− place cells, albeit lower firing rates during running epochs. Furthermore, a subset of CB+ PNs shifted their theta firing phase during rapid-eye movement (REM) sleep states compared with running states. Although CB− PNs are more actively engaged in ripple oscillations, CB+ PNs showed stronger ripple modulation during slow-wave sleep (SWS). Our results pointed out the heterogeneity in neuronal representation between hippocampal CB+ and CB− PNs. Particularly, CB+ PNs encode spatial information more efficiently, which might be contributed by stronger afferents from the lateral entorhinal cortex to CB+ PNs.

Based on the relative somatic location of PNs to the border of stratum radiatum (S.R.), hippocampal CA1 PNs can be subdivided into superficial and deep layer neurons, with superficial neurons closer to S.R. Most neurons in the superficial layer of the dorsal CA1 area express Calbindin (CB), a Calcium-binding protein, while deep layer PNs did not. Superficial PNs also have more complex apical dendritic arborizations compared with deep PNs (Bannister and Larkman, 1995;Graves et al., 2012;Li et al., 2017). These two PN subtypes receive different excitatory synaptic inputs from several afferents including lateral, medial entorhinal cortex (MEC), and also CA2 (Kohara et al., 2014;Li et al., 2017;Masurkar et al., 2017). LEC preferentially projects to superficial PNs, while CA2 preferentially connects to deep PNs (Kohara et al., 2014;Li et al., 2017). They also differ in IPSC responses, induced either by CA3 or CA2 stimulation (Kohara et al., 2014;Valero et al., 2015). Deep PNs displayed larger IPSC responses, probably because of more inhibitory innervation by Parvalbumin positive interneurons on deep PNs than their superficial peers (Lee et al., 2014).
In addition, growing evidence revealed heterogeneity in neural activity of hippocampal PNs during hippocampus dependent behavior tasks. Using silicon probes, hippocampal PNs can be divided into superficial and deep layer PNs based on their somatic location relative to the middle of stratum pyramidale, which is defined by the largest amplitude of ripple oscillation (Mizuseki et al., 2011). Deep and superficial PNs showed significant differences in firing rate, phase-locked firing with hippocampal theta and ripple oscillations (Mizuseki et al., 2011;Stark et al., 2014;Danielson et al., 2016). A subset of deep PNs shifted their firing phase during rapid-eye movement (REM) sleep, while superficial PNs are consistently phase-locked with theta oscillation either during running or REM sleep (Mizuseki et al., 2011). During hippocampal ripple oscillations, superficial PNs are more active than deep PNs at least in anesthetized headfixed animals (Valero et al., 2015). Deep PNs are more actively engaged in spatial navigation compared with superficial PNs in a two-photon Ca 21 imaging experiment (Danielson et al., 2016). In our previous study, we have shown that the direct projection from LEC to Calbindin-expressing (CB1) hippocampal pyramidal neurons plays an essential role in olfactory associative learning (Li et al., 2017).
Despite the progress, it remains unclear whether and how the anatomic and biophysical variations of PNs lead to their distinct roles in hippocampal functions. Specifically, the functional involvement of CB1 and CBÀ hippocampal PNs during different behavior tasks/states is still elusive. To answer this, we combined optogenetic tagging with multichannel in vivo recording and investigated the firing profiles of hippocampal CB1 and CBÀ PNs in freely moving mice under different behavior states.

Animals
Two lines of transgenic mice were used in this study, the Calb2-IRES-Cre line (Jax No.010774) and Ai35 mice (Rosa-CAG-LSL-Arch-GFP-WPRE, Jax No.012735). They were generously shared by our collaborator, Prof. Xiaohui Zhang from Beijing Normal University. The two mouse lines were crossed to selectively express Arch in CB1 neurons (Chow et al., 2010). This approach had also been reported in our previous study to identify different types of pyramidal neurons in the CA1 area of the hippocampus (Li et al., 2017). Ten male transgenic mice were used for in vivo electrophysiological recordings in this study. All procedures were approved by the Animal Advisory Committee at East China Normal University (No. AR201404009) and were performed in accordance with the National Institutes of Health Guidelines for the Care and Use of Laboratory Animals. Six mice with considerable good yields of neurons were used in further data analysis.

Behavior training
There were two experimental setups in this study: the home cage recording of spontaneous neuronal firing across different behavior states and a U-shaped track for the recording of place cells. The home cage was a 28 Â 45 cm 2 , 20-cm-high plastic box where the animal was housed with free access to food and water. The recordings typically last for several hours to cover different spontaneous behaviors, including quiet waking, REM sleep and slow-wave sleep (SWS). After the recording, the mouse was removed from the recording setup and kept in the same home cage for later experiments.
The U-shaped track has three arms, the left, the central and the right arm. The total length of the three arms is 114 cm with a uniform width of 5.5 cm. There were different visual cues along the track and a water port at the end of the left and right arm. Before training, mice were kept under a water restriction protocol. The animals were trained to run back and forth on the track to get a 10-ms water reward at the water ports. The amount of water reward was controlled by a solenoid valve, which was run by a program written with Labview (Labview 8.6, National Instruments). Track training and place cell recording were conducted in a dim environment surrounded by black curtains.

Drivable optrode
An optrode was made of a 200-mm diameter optic fiber [numerical aperture (NA) = 0.39; Thorlabs], and surrounded by eight tetrodes. The optic fiber was later connected to a laser stimulator for optogenetic stimulation. The tips of the tetrodes extended ;0.5 mm beyond the tip of the optical fiber. The optrode bundle was attached to a set of screws and nuts which can be driven by rotation of the screw, with each full turn corresponding to 280 mm in depth penetration. The detail of the assembly of the microdrive can be found from our previous study (Lin et al., 2006). A 64-channel microdrive with two optrode bundles was used in one mouse, each optrode targeting one hemisphere of the hippocampus. The other five mice were implanted with a 32-channel optrode (eight tetrodes).

Surgery
The detailed procedure of the optrode implantation surgery can be found from our recently publication (Ma et al., 2020). In short, the scalp of the mouse was removed after anesthetization (pentobarbital sodium, 40 mg/kg bodyweight) and then mounted onto the stereotactic frame (Stoelting). Body temperature of the mouse was closely monitored and kept constant by a thermoregulation device (FHC). Multiple screws were mounted onto the skull (avoiding the hippocampal area), serving as the foundation for dental cement. One of the screws that were mounted above the cerebellum also served as ground. One or two craniotomies were made above the left and right hippocampus at these positions (in mm): À2.3 AP, 12.0 ML. After removal of the dura, the microdrive was slowly lowered into the brain so that the optrodes reached a depth of ;0.9 mm from the cortical surface. After the optrode insertion, the microdrive was secured on the skull with dental cement. We wrap a piece of copper mesh around the entire microdrive, serving as a Faraday cage and also protecting the microdrive from potential scratching damage. After surgery, the mouse was housed in the home cage with free access to food and water, on a 12/12 h light/dark cycle. The animals were allowed to recover for at least 72 h before any experiments.

Ephys recording
On the day of the electrophysiological (ephys) recording, a helium-filled mylar balloon was tied to the cables to alleviate the weight, enabling the mouse to move freely. All the in vivo ephys recordings were made with a Plexon MAP system (Plexon). The signals were filtered through the preamplifier to record neuronal spiking activities and local field potentials (LFPs) separately. The spike signals were filtered from 400 to 7000 Hz and sampled at 40 kHz, while the LFP signals were filtered from 0.7 to 300 Hz and sampled at 1 kHz. The optrode was advanced manually by ;35 mm every 3 d before they reach stratum pyramidale of the dorsal CA1 area of the hippocampus. The position of the tetrode tips was estimated by observing apparent sharp wave-ripple events in the LFP signals during slow-wave sleep in the home cage. We began the recording after most of the tetrodes have reached stratum pyramidale and spontaneous firings of different neurons were observed. Animal behavior was simultaneously recorded along ephys data by a camera on top of the recording arena. After the ephys recording experiments, the animals were killed for histologic staining of brain slices with 1% cresyl violet to confirm the position of the optrode.

Optogenetic stimulation
The optic fiber within the optrode was coupled to an external fiber using standard FC connectors via the ceramic sleeve and then connected to a Diode-Pumped Solid State (DPSS) Laser (589 nm, Inper). Laser power at the output end of the optic fiber was 10 mW/mm 2 . In each 30min standard ephys recording session, two or three times of continues optogenetic stimulation was delivered, each lasting for 2 or 3 min. The optogenetic stimulation protocol was preprogramed with the DPSS Laser and the signal was synchronized with the ephys recording via a digital input cable connected to the Plexon MAP system.

Spike sorting and data selection
All the ephys and external signals were stored in a single .plx file. Spike sorting was performed manually using Plexon Offline Sorter (version 2.7.3) in a 2D and sometimes 3D feature space. The degree to which the selected unit clusters were separated in the 2D or 3D space was determined by a multivariate ANOVA (MANOVA) test. The smaller the p-value, the more confident one can be of the conclusion that the clusters were in fact distinct and represent different units. Also, during the spike sorting process, sorting quality was closely monitored after each sorting operation by the following built-in statistics in Offline Sorter: J3 and Pseudo-F. For example, the majority of the neurons in Figure 1 were recorded from one tetrode. The MANOVA statistics were: p = 1.61e-77, F = 13.85 in a 2D cluster space, and p = 1.77e-117, F = 14.10 in a 3D cluster space. J3 statistics in a 2D and 3D space were 7.38 and 6.36. Pseudo-F statistics in a 2D and 3D space were 5442.2 and 4686.16.
CA1 pyramidal neurons and interneurons were identified based on their action potential waveforms and corresponding firing rates (typically ,6 Hz for pyramidal neurons and .10 Hz for interneurons). A total number of 327 pyramidal neurons and six interneurons in the running track and 373 pyramidal neurons and 8 interneurons in home cage were sorted from six mice. Only pyramidal neurons were used in this study. Pyramidal neurons with an average firing rate below 0.01 Hz in the home cage recording are excluded from the dataset. Place cell selection criteria was described below.

Identification of CB1 and CB2 PNs
During each recording session, a brief yellow laser stimulation was delivered to inhibit neuronal firings of CB1 PNs. We calculated the laser inhibition index to identify CB1 PNs from the data pool. Laser inhibition index was calculated as follow: Average firing rate during laser stimulation Average firing rate non laser stimulation p 100%: A neuron was marked as CB1 whenever the neuronal laser inhibition index exceeds 80%. In our dataset, the total number of CBÀ PNs was considerably higher than CB1 PNs (number of recorded neurons reported in Fig.  1D). This was largely because of technical limitations of tetrode recordings. CBÀ PNs reside at the deep layer of stratum pyramidale (next to stratum Oriens), which was more accessible for our tetrode bundles. In order to maximize the total number of neurons recorded, we chose to start the recording session once the tetrode bundle had reached the stratum pyramidale, hence more deep layer pyramidal neurons, presumably CBÀ PNs, were recorded.

Behavior states selection
We selected three behavior states for further firing pattern analysis of CB1 and CBÀ PNs: active running (RUN), REM sleep (REM) and slow wave sleep (SWS). An overhead camera recorded animal behavior simultaneously with ephys recording. RUN state was selected during active running in the track with a running speed over 5 cm/s. Behavior video recordings were used to identify sleep states, which were further divided into REM and SWS states based on distinct local field potential patterns of theta and ripple oscillations.

Data analysis Burst analysis
We used the MaxInterval method to detect burst events (NeuroExplorer, Nex Technologies). A burst should at least contain two consecutive spikes. For bursts with more than two spikes, the interval between the first two spikes should be ,10 ms, while the interval between any two consecutive spikes should be ,20 ms. The duration of a burst should be at least .3 ms (two spikes minimum). The interburst interval should be .20 ms. Burst frequency is defined as the number of bursts over a time range divided by that duration. Burst index was defined as the total number of bursting spikes divided by the number of overall spikes committed by that neuron.

Detection of theta and ripple oscillations
We used the same method reported in our previous report to detect different oscillation patterns (Ma et al., 2020). In brief, the original LFP was bandpass filtered (4-12 Hz) for the detection of theta oscillations. Theta epochs were detected by calculating the power ratio of the theta (5-10 Hz) and d (2-4 Hz) band by sliding a 2-s window. Epochs with more than three consecutive time windows in which the ratio was .4 were identified as theta episodes (Csicsvari et al., 1999).
To detect ripple events, the original LFP was first bandpass filtered (100-250 Hz). The power of the filtered signal was calculated by sliding a 10-ms window every 1 ms. The threshold for ripple detection was set to 5 SD above the background mean power. The beginning and end of each ripple epoch were identified by sliding the time window forward and backward, by the threshold of 2 SD above the background mean power (Csicsvari et al., 1999).

Phase analysis
The theta or ripple bandpass filtered LFP was first decomposed into instantaneous amplitude rðtÞ and phase f t ð Þ components by using a Hilbert transform: yðtÞ ¼ Reðr ðtÞe jf ðtÞ Þ: Given the neuronal spike train t i j i ¼ 1; 2; :::; n f g , spike phase was calculated by We define oscillation peaks at 0°and 360°and troughs at 180°throughout the paper. The mean direction and mean resultant length of the phases of a given neuron's spikes were taken as the preferred firing phase and modulation Figure 1. In vivo identification and basic firing pattern of hippocampal pyramidal neuron subtypes. A, Two recording setups used in our experiment. Left, Home cage. Right, U-shape running track. B, Eight putative pyramidal neurons sorted from a single tetrode. Each color denotes one single unit. Corresponding tetrode waveforms were illustrated at the bottom. The orange cluster in the upper right corner represents one CB1 neuron. Scale bar: 0.2 mV. C, Neuronal firing sequence of 30 simultaneously recorded pyramidal neurons for a 1000-s recording period in home cage. Vertical bar on the right denotes eight neurons recorded from one single tetrode, as illustrated in B. Each dot represents one action potential. The firing of CB1 PNs were largely inhibited on optogenetic stimulation (yellow shaded areas). Bottom, Average firing rate of these neurons showing population responses of PNs to laser stimulation. Blue, CBÀ PNs; orange, CB1 PNs. D, Top, Ratio of CBÀ PNs recorded from each tetrode in running track as a function of that in home cage. Each dot represents result from one tetrode, while each color denotes individual animal (n = 6 mice). Bottom, Ratio of active CB1 and CBÀ PNs recorded from two recording setups. Note the significant decrease of the ratio of CB1 neurons recorded during running (**p = 6.6e-3, x 2 test). depth of that neuron, respectively. To evaluate the presence of phase locking, we performed Rayleigh's test for circular uniformity to compute the significance of phase locked firing. Only neurons that showed an average firing rate over 0.1 Hz during theta or ripple epochs, and significantly modulated by the oscillations (Rayleigh's test, p , 0.05) were included in the analyses. To avoid theta/ripple phase variability as a function of recording depth, the electrode with the largest ripple power (that is, the middle of the CA1 pyramidal layer) was used as reference for detecting theta/ ripple phase.

Ripple participation and inhibition index
For each neuron, the fraction of ripples with neuronal spike(s) was calculated as ripple participation to represent the extent of neuronal involvement in ripple oscillations. The ripple inhibition index was defined as follow to describe the degree of inhibition of neuronal firing after ripple peak: Where postripple firing rate (FR) was the average firing rate for the 100 ms after ripple peak, baseline FR was the baseline neuronal firing rate excluding the 100-ms period before and after ripple peak.

Place cell analysis
A total number of 327 pyramidal neurons fitted the criteria for data selection from 6 mice in the track running experiments (66 CB1 PNs and 261 CBÀ PNs). To characterize place fields of CA1 pyramidal neurons, we first linearized and then binned the U-shape track with 1 cm width bin. Average firing rate of each running direction (clockwise and counterclockwise) was calculated by dividing the total number of spikes in each bin by the occupancy time in that bin. The place field of a neuron was the bins where the neuron displayed the highest firing rate and all contiguous bins exceeded 20% of peak firing rate. Neurons that showed a peak firing rate over 2 Hz with a clear place field preference were defined as place cells. Furthermore, cells with a place field covering more than half of the track were excluded from the analysis because of low place specificity. Following these criteria, we identified 27 CB1 and 163 CBÀ place cells from the data pool.

Spatial information
Spatial information content was calculated in bits per second (I second ) using the following formula: Where N was the number of bins of the track, l i was the mean firing rate of a neuron in the i-th bin, l was the overall mean firing rate of the neuron, p i was the probability that animal being in the i-th bin (occupancy in the i-th bin divided by the total recording time).
Spatial information content can also be calculated in bits per spike (I spike ) when divided by the overall firing rate of the neuron: Sparsity and selectivity Place cell firing sparsity measures the fraction of the environment in which a neuron was active. The sparsity index was calculated following Skaggs (Skaggs et al., 1996) as: Spatial firing selectivity measures how concentrated the neuron's activity was. Neurons with no spatial tuning will have a selectivity of 1. Higher selectivity index represents more tightly concentrated place cell firing. Selectivity index was defined as:

Selectivity ¼
Maximum firing rate Average firing rate :

Phase precession
We used a circular-linear regression method (Maier et al., 2011) to quantify properties of theta phase precession of place cells. Phase precession was calculated within each place field of place cells. Linearized track was binned into 1-cm width bin and neuronal firing rate was calculated for each bin. The place field of a neuron was the bins where the neuron displayed the highest firing rate and all contiguous bins exceeded 20% of peak firing rate. Place field was normalized so that 0 represented the beginning and 1 represented the end of each place field. Phase precession was defined as significant negative linear-circular correlation (p , 0.05, linear-circular correlation test) between the animal position in the normalized place field and the theta firing phase (Berens, 2009).

Quantification and statistical analysis
All statistical analyses were performed in MATLAB (MathWorks). Data were presented as mean 6 SEM unless stated otherwise. Shapiro-Wilk test and Kolmogorov-Smirnov test were used to determine whether sample distribution was standard normal distribution. If normality was uncertain, nonparametric tests were used. Details of the statistical tests and the resultant p-values were listed in the main text and figure legends. In each box-plot figure, the central mark indicates the median, and the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. The whiskers extend to the most extreme data points not considered outliers. Outlier values were always included in the statistical analysis, although they were not represented in the plots. Circular statistics toolbox for MATLAB was used for comparison in polar coordinates (Berens, 2009). Rayleigh's test was applied to

Basic firing patterns of hippocampal CB1 and CB2 PN subtypes under different behavior states
We performed multichannel in vivo ephys recordings in 10 transgenic mice that had Arch-GFP selectively expressed in CB1 PNs. Ephys data were collected in two behavior setups: home cage and a U-shaped track (Fig.  1A). Well-isolated pyramidal neurons with an average firing rate over 0.01 Hz were selected for further analysis (Fig. 1B). During each recording session, a brief yellow laser stimulation of 3 min was delivered to inhibit the neuronal firing of CB1 PNs. Based on the firing inhibition profile during light stimulation, we categorized all the sorted pyramidal neurons as CB1 and CBÀ PNs (Fig. 1C). We collected a total of 116 CB1 and 257 CBÀ PNs in home cage recording, 66 CB1 and 261 CBÀ PNs in the running track. The number of CBÀ PNs recorded from each tetrode was higher than CB1 PNs ( Fig. 1D; also, for detailed explanation, see Materials and Methods). We selected three distinct behavioral states for further analysis, RUN, SWS, and REM sleep, based on local field potential (LFP) patterns ( Fig. 2A) and simultaneously recorded animal behavior videos.
We found that the firing rate of CBÀ neurons were significantly higher than that of CB1 PNs during active running, but not during sleep states (Table 2; Fig. 2B). Although the two PN subtypes showed similar kinetics of burst firings (Fig. 2D, including burst frequency, duration, number of spikes and ISI within burst), CB1 PNs showed a higher burst index than CBÀ PNs under all three behavior states (Table 2; Fig. 2C). Note that CB1 PNs showed a longer burst duration than that of CBÀ PNs during SWS (Fig. 2D).

Efficient representation of spatial information by CB1 place cells
Since the discovery of place cells in rat hippocampus, neuronal representation of spatial information is one of the most fundamental and widely studied function of the hippocampus (O' Keefe and Dostrovsky, 1971;Moser et al., 2017). The higher firing rate of CBÀ PNs during running led us to speculate potential functional difference in spatial representation. To test this, we recorded place cell activities while the mice was trained to shuttle back and forth on a U-shape track. Yellow light stimulation was delivered during running on the track to inhibit neuronal firing of CB1, but not CBÀ PNs (Fig. 3A). We collected a total of 66 CB1 and 261 CBÀ PNs from 6 mice during track running. Among the data pool, we identified 27 CB1 and 163 CBÀ place cells, with 54 and 367 place fields for each place cell subtype (for data selection criteria, see Materials and Methods). Place cell activities of both CB1 and CBÀ PNs can cover the whole track (Fig. 3B), while place fields formed on each running direction were slightly different for both PN subtypes (Fig. 3C, two CB1 and four CBÀ PNs on each running direction are illustrated). CB1 PNs were less likely to be place cells than their negative peers, with only 40.9% of the  (Fig. 3D). The average number and size of place fields formed by each place cell subtypes are comparable (Table 2; Fig. 3E,G). Moreover, CB1 place cells were more likely to form double place fields (typically one on each running direction) than CBÀ place cells (Fig. 3F). We next compared the spatial firing patterns of both place cell subtypes. Although they showed no difference in average and peak firing rate within place fields (Table 2; Fig.  3H), CB1 place cells exhibited higher firing rate inside than outside of their place fields compared with that of CBÀ place cells (Table 2; Fig. 3I). Consistently, the activity of CB1 place cells were more refined on the track, with higher spatial information per spike than CBÀ place cells (Table 2; Fig. 3J,K). Furthermore, the two place cell subtypes showed no difference in all aspects of phase precession characteristics ( Fig. 3L-N).
Overall, we found that CB1 place cells encode spatial information more efficiently, as they fire in a more condensed manner inside and outside of place fields, with higher spatial information carried by each spike than that of CBÀ place cells.
A subset of CB1 PNs shifted theta firing phase during REM sleep but not during RUN Previous study has reported that during REM sleep theta oscillation, a subset of hippocampal pyramidal neurons shifted their preferred theta firing phase, compared with that of running thetas (Mizuseki et al., 2011). We wondered whether it holds true for pyramidal neurons with distinct Calbindin expression profile. To do this, we calculated theta firing phase for each neuron during RUN and REM thetas (Fig. 4A,B). We found that firing activities of both PN subtypes are phase-locked to the ascending phase (180-360°) of theta oscillation during RUN (Fig. 4A, C), but they differ significantly in preferred theta phase (Fig. 4C).
During REM sleep theta oscillations, while the firing of most CBÀ PNs remained phase-locked to the ascending phase of theta oscillations as during RUN states, CB1 PNs shifted their theta firing phase dramatically (Fig. 4B,  D). To determine the degree of theta phase shift, we compared the theta phase distribution of each PN subtypes during RUN and REM sleep. We found that the theta phase of CB1 PNs shifted significantly between RUN and REM sleep (Fig. 4E), while no such theta phase shift was observed in CBÀ PNs (Fig. 4F).
We also calculated the strength of theta phase-locking of both PN subtypes during the two theta states. CB1 PNs were more strongly phase-locked to theta oscillations than CBÀ PNs during both RUN and REM sleep (Fig. 4C,D). Furthermore, both PN subtypes were more deeply phase-locked to REM sleep theta than RUN theta oscillations (Fig. 4E,F). CB1 PNs are more strongly modulated by ripple, despite lower engagement during SWS Activity of hippocampal pyramidal neurons during slow wave sleep, especially ripple oscillations is critical for hippocampal function of learning and memory (Girardeau et al., 2009;Buzsáki, 2015). So we looked into neuronal firing dynamics of both PN subtypes during ripple oscillations. We found that both PN subtypes increased their firing when ripple oscillations were prominent (Fig. 5A), but they were differentially engaged during ripples. First of all, CBÀ PNs were more actively engaged in ripples (Fig. 5B,C), as indicated by stronger participation in ripples, higher firing rate during ripples, and higher firing rate ratio inside and outside ripples of CBÀ PNs (Table 2; Fig.  5D-F). Such elevated activity of CBÀ PNs during ripples was not because of difference in baseline firing rate since the average firing rate of both PN subtypes are similar during SWS (Fig. 2B). It is also not caused by burst spikes since the deletion of burst spike did not change such elevated activity of CBÀ PNs (data not shown).
Second, CB1 PNs participated earlier in ripples than CBÀ PNs (Table 2; Fig. 5H). They were also more strongly suppressed after ripple peak as indicated by significantly lower postripple firing rate and higher ripple inhibition index of CB1 PNs than that of CBÀ PNs (Table 2; Fig. 5B,I).
Finally, we compared the modulation of ripple phaselocked firing between CB1 and CBÀ PNs during SWS. Neuronal firing activities of both PN subtypes were phase-locked to the trough of ripple oscillations with no significant phase-locking difference (Fig. 6A-C). Although CBÀ PNs are more actively engaged in ripples, CB1 PNs showed stronger phase-locked firing compared with CBÀ PNs (Fig. 6D).

Discussion
In this study, we systematically investigated the in vivo firing patterns of two hippocampal PN subtypes based on Calbindin expression profile in free-moving mice. The subclassification of hippocampal pyramidal neurons by Calbindin expression profile may provide us with more precise classification and deeper understanding of the functional heterogeneity of hippocampal pyramidal neurons. We found that CBÀ PNs showed higher firing rates during running, while CB1 PNs displayed a higher tendency of burst firing regardless of behavior states. We also found that CB1 PNs can represent spatial information more efficiently: they are less likely to form place cells compared with their CBÀ counterparts, but with more spikes inside place fields. The firing of CB1 place cells also carry more information during spatial navigation. During REM sleep, a subset of CB1 PNs shifted their theta firing phase, while CBÀ PNs remained phase-locked to theta through. CBÀ PNs showed significantly higher participation in ripple oscillations during SWS.

Using Calbindin as a molecular marker to differentiate hippocampal PNs
Calbindin has long been used as a molecular marker to identify neurons in the brain (Baimbridge and Miller, 1982;Sloviter, 1989). Recent studies have used the Calbindin expression profile to identify CB1 and CBÀ PNs with different afferents and neural circuits underlying different behaviors (Li et al., 2017;Pi et al., 2020). On the other hand, functional investigation of Calbindin, a Ca 21 binding protein, is scarce. One study has shown that Calbindin equips hippocampal neurons with mobile, high-affinity Ca 21 -binding sites that slow and reduce global Ca 21 signals (Müller et al., 2005), indicating that Calbindin might reduce the activity of hippocampal pyramidal neurons. Another study recorded from dentate granule cells of Calbindin knock-out mice with patch clamp recording. They reported hyperexcitability of dentate gyrus granule cells in Calbindin knockout mice (Kim et al., 2021). In our study, we used Calbindin as a molecular marker to distinguish superficial layer CB1 and deep layer CBÀ PNs. We found a significantly lower firing rate of CB1 PNs than that of CBÀ PNs during running (Fig. 2B) and hippocampal ripple oscillations (Fig.  5E). These results are consistent with the two aforementioned studies. The functional heterogeneity of CB1 and CBÀ PNs in our study could rise from their relative somatic location within the stratum pyramidale and hence their distinct afferents with other brain   regions. Meanwhile, the physiological function of Calbindin may also be a contributing factor.
The Calbindin expression dichotomy versus the superficial/deep dichotomy in neural coding heterogeneity Neural coding heterogeneity of hippocampal deep and superficial layer pyramidal neurons have been reported recently (Mizuseki et al., 2011;Danielson et al., 2016;Fernández-Ruiz et al., 2021). Burst firings are different from single spikes in a way that they might play a different role in transmitting sensory information (Krahe and Gabbiani, 2004). We have shown that CB1 PNs showed a higher burst index regardless of behavior states, while Kenji Mizuseki found that more deep layer pyramidal neurons (presumably CBÀ PNs) showed higher burst index than superficial PNs (Mizuseki et al., 2011). We reasoned that on one hand, CB1 PNs receive more excitatory inputs from LEC and less inhibitory innervation from local parvalbumin positive basket cells (Lee et al., 2014;Li et al., 2017), which could lead to more burst spikes. On the other hand, CBÀ PNs showed a significantly higher firing rate during running, which as a denominator for calculating burst index, could result in significantly lower burst index during RUN.
Studies had shown similar spatial representation characteristics of deep and superficial layer pyramidal neurons including higher in/out place field firing rate ratio and spatial information per spike (Mizuseki et al., 2011;Fernández-Ruiz et al., 2021), which we confirmed in our comparison between CB1 and CBÀ PNs. The medial entorhinal cortex (MEC) is essential for spatial navigation of animals. A previous study has shown that strong MEC and weak LEC inputs favor deep PNs with more SLM spines in CA1c (close to CA2). In CA1a (near subiculum), strong LEC and weak MEC inputs favor superficial PNs with more SLM spines (Masurkar et al., 2017). Our previous study has shown that the MEC projects equally to CB1 and CBÀ PNs, while the lateral entorhinal cortex projects almost exclusively to CB1 PNs in dorsal CA1 (Li et al., 2017). In our experiments, we placed our recording electrodes in CA1b (mid-CA1) and found that CB1 PNs can represent spatial information more efficiently. The underlying mechanisms of such efficient spatial representation of CB1 PNs, e.g., the potential contribution of LEC and/or other brain areas that project differentially to CB1 and CBÀ PNs, is worth further investigation.
Interestingly, we found no significant difference in phase precession parameters between CB1 and CBÀ PNs, inconsistent with a recent study (Fernández-Ruiz et al., 2021). In our study, mice were trained to run freely back and forth in a U-shaped track, while others recorded from head-fixed mice running on a belt consisting of a cue-rich and cue-poor zones. The difference in self-motion and external visual cues between the two experiment setups may contribute to such controversy.
Theta oscillation dominates animal active running states, representing an online learning state (Buzsáki, 2002). During such exploration, neurons fire at specific phase of each theta cycle, representing distinct phase of information encoding process. Encoding occurs at the trough and rising slope of theta, when current sinks are strong in SLM, where entorhinal input terminates, and currents in layers receiving CA3 input are weak (Hasselmo, 2006). Indeed, we found that both CB1 and CBÀ PNs are phaselocked to trough of theta oscillations during running on the track, also consistent with previous report (Mizuseki et al., 2011). Theta oscillations were also prominent during REM sleeps. We found that a subset of CB1 PNs shifted their theta firing phase during REM sleep, CBÀ PNs remained the same theta firing phase as during running. Kenji Mizuseki had reported similar results with one subtype of PNs remained their theta firing phase while the other shifted their theta phase during REM sleep, except that they found the REM-shifting neurons were located in the deep layer of Stratum Pyramidale, presumably CBÀ PN subtypes. The somatic distribution of CB1 and CBÀ PNs may not follow the laminar distribution of deep and superficial layers (Baimbridge et al., 1991;Klausberger and Somogyi, 2008). In our study, optogenetic inhibition enabled us to identify CB1 and CBÀ PNs more accurately, thus allowing us to investigate the differences in their firing patterns more precisely.

Distinct involvement of CB1 and CB2 PNs during hippocampal ripple oscillations
Hippocampal ripple oscillations are high frequency transient network events recorded in the hippocampus during periods of immobility and slow wave sleep (Ylinen et al., 1995). Selective disruption of ripples during postlearning sleep results in impairment of behavior performance (Girardeau et al., 2009;Jadhav et al., 2012), indicating that hippocampal ripple oscillations are critical for memory consolidation. Pyramidal neurons in the hippocampus fire at specific phase of each ripple cycle (English et al., 2014;Hulse et al., 2016;Gan et al., 2017). The involvement of neuronal firing of different PN subtypes with ripple oscillations has not been fully addressed. In our continued H, Average firing rate (aFR) and peak firing rate (pFR) within place fields (n.s., not significant, p = 0.615 for aFR, 0.606 for pFR, Mann-Whitney test). I, Firing rate ratio inside and outside place fields. CB1 place cells fire significantly more spikes inside than outside the place fields than that of CBÀ place cells (***p = 1e-3, Mann-Whitney test). J, Neuronal firing sparsity and selectivity of place cells (**p = 9e-3; n.s., not significant, p = 0.379, Mann-Whitney test). K, Spatial information calculated by every spike or second of both place cell subtypes. CB1 place cells carry more information per spike than CBÀ place cells (**p = 5e-3; n.s., not significant, p = 0.203, Mann-Whitney test). L, Example phase precession of both place cell subtypes during one place field traverse. Two normalized theta cycles are shown for clarity. M, Phase precession of CB1 and CBÀ place cells. Two theta cycles are shown. N, No significant difference found in phase precession parameters, including slope, onset, and range (p = 0.934, p = 0.823, p = 0.778, Mann-Whitney test). study, we found that CB1 PNs are activated earlier during ripples than CBÀ PNs, while CBÀ PNs showed a significant higher participation in ripples than CB1 PNs. Previous report had pointed out hippocampal CA2 region as an initiation zone for sharp-wave ripples (SPW-R) in which the activity of CA2 neurons preceded SPW-Rrelated population activities in CA3 and CA1 (Oliva et al., 2016). Also, optogenetic activation of CA2 terminals results in significantly larger EPSC in deep layer CA1 pyramidal neurons than superficial layer neurons (Kohara et al., 2014). These observations might be the mechanism underlying the much stronger participation in ripple oscillations of CBÀ PNs than that of CB1 PNs. Another study had reported that hippocampal superficial layer pyramidal neurons are depolarized and increased their firing rate during ripple oscillations, while deep layer pyramidal neurons are hyperpolarized and decreased their firing rate during ripples (Valero et al., Figure 4. A subset of CB1 PNs shifted their theta firing phase during REM sleep. A, Example theta firing phase distribution of both PN subtypes during RUN state (theta peak = 0°, 360°, theta trough = 180°; bin size: 30°). B, Same as in A, but for preferred theta firing phase during REM sleep theta oscillations. C, Left, Distribution of preferred theta phase of both PN subtypes during RUN state (CB1 PNs, n = 43; CBÀ PNs, n = 199; p = 2.2e-4, Watson's U 2 test; bin size: 30°). Top trace indicates idealized reference theta cycle. Right, Theta modulation depth of CB1 and CBÀ PNs during RUN (**p = 8.6e-3 Mann-Whitney test). D, Same as in C, but for population distribution of preferred theta firing phase during REM sleep. Note the bimodal distribution of theta phase preference of CB1 PNs (p = 0.099, Watson's U 2 test; bin size: 30°). Right, Theta modulation depth of both PN subtypes during REM sleep (**p = 1.4e-3, Mann-Whitney test). E, Left, Comparison of preferred theta firing phase of CB1 PNs during RUN and REM states. The preferred theta firing phase shifted significantly between the two states (p = 9.1e-3, Watson's U 2 test, bin size: 30°). Right, Modulation depth of CB1 PNs during REM sleep theta is significantly deeper than RUN theta states (****p = 2.3e-8, Mann-Whitney test). F, Same as in E, but for comparison of preferred theta firing phase of CBÀ PNs during RUN and REM states. No significant theta phase shift was observed between the two theta states (p = 0.446, Watson's U 2 test, bin size: 30°). Right, Theta modulation depth of CBÀ PNs during RUN and REM states (****p = 1.7e-26, Mann-Whitney test). Figure 5. CBÀ PNs are more actively engaged in hippocampal ripple oscillations during SWS. A, Top, Neuronal firing sequence of ten CB1 (blue dots) and ten CBÀ (orange dots) PNs along with hippocampal LFP (black trace) and bandpass filtered ripple oscillation (gray trace with ripple events highlighted in dark blue) during SWS. Scale bar: 0.2 mV, 0.5 s. B, Normalized neuronal firing of both PN subtypes during ripple oscillations (n = 3888 ripple events), sorted by peak firing time (n = 116 CB1 PNs, and 257 CBÀ PNs). Time zero represents ripple peak (white broken line). Note that CBÀ PNs exhibited higher firing rate before ripple peak, while CB1 PNs showed lower firing rate after LFP ripple peak. C, Normalized firing rate of each PN subtype population during hippocampal ripple oscillations. Time zero denotes ripple peak. Note that population activity of CB1 PNs are briefly suppressed after ripple peak. Bin size: 10 ms. D, Ripple participation of both PN subtypes. CBÀ PNs are more actively engaged in ripple oscillations (****p = 4.2e-9, Mann-Whitney test). E, Average firing rate of both neuron subtypes during ripple oscillations (****p = 1.4e-9, Mann-2015). Distinct recording setups could be vital for such controversy of results, in which they recorded from head fixed rats under urethane anesthetization, while we recorded from free moving mice, which is closer to natural physiological states.

Limitations of the study
There are a number of limitations of our study. First, the possibility that Arch expression in pyramidal neurons may affect the firing properties even in the absence of light stimulation. Based on the results of the first paper that introduced archaerhodopsin, the expression of Arch did not alter the basic cellular properties of the neuron, including membrane resistance, resting membrane potential, etc (Chow et al., 2010). We also examined potential change in neuronal firing properties because of prolonged photostimulation. We found no significant difference in population firing rate and action potential waveforms of CB1 and CBÀ PNs before and after photo-stimulation (data not shown).
Second, we used multichannel in vivo ephys recordings (tetrode) in our study, with the combination of optogenetics inhibition. There is certain limitation when categorizing pyramidal neurons. Some CB1 PNs might be misclassified as CBÀ PNs because of unsuccessful or insufficient expression of Arch, leading to a low light inhibition effect. On the other hand, CBÀ PNs could also be misclassified as CB1 PNs if they happened to not fire during the opto-stimulation period. Although the possibility is quite low, we still cannot quite exclude such possibility. Figure 6. Both CB1 and CBÀ PNs are strongly Phase-locked to hippocampal ripple oscillations. A, Phase-locked firing of both PN subtypes with hippocampal ripple oscillations, referenced at LFP ripple peak (orange, CB1 PNs, n = 116; blue, CBÀ PNs, n = 257). Averaged LFP ripple trace is illustrated at the top. Note that neuronal firings of both CB1 and CBÀ PNs are locked to LFP ripple troughs. Bin size: 1 ms. Scale bar: 0.25 mV. B, Polar plots of neuronal firing phase distribution of example CB1 and CBÀ PNs. Bin size: 10°. C, Population ripple phase distribution of both CB1 and CBÀ PNs (bin size: 30°, p = 0.124, Watson's U 2 test). Both PN subtypes are phase locked to ripple trough. D, CB1 PNs are more strongly phase locked to ripple oscillations than CBÀ PNs (*p = 0.027, Mann-Whitney test).
Third, all of our experiments were conducted in male mice. The findings of our study may not extend to females.
Overall, we investigated the basic firing pattern, representation of spatial information, and neuronal firing dynamics of pyramidal neuron subpopulation under hippocampal theta and ripple oscillations. We found prominent heterogeneity of neural activity between CB1 and CBÀ pyramidal neurons at single unit level, implying different roles in hippocampal neural code. Our results pointed out the necessity to pay attention to such functional heterogeneity between hippocampal pyramidal neurons in future investigations.