Functional imaging of hippocampal place cells at cellular resolution during virtual navigation

Abstract

Spatial navigation is a widely employed behavior in rodent studies of neuronal circuits underlying cognition, learning and memory. In vivo microscopy combined with genetically-encoded indicators provides important new tools to study neuronal circuits, but has been technically difficult to apply during navigation. We describe methods to image the activity of hippocampal CA1 neurons with sub-cellular resolution in behaving mice. Neurons expressing the genetically encoded calcium indicator GCaMP3 were imaged through a chronic hippocampal window. Head-fixed mice performed spatial behaviors within a setup combining a virtual reality system and a custom built two-photon microscope. Populations of place cells were optically identified, and the correlation between the location of their place fields in the virtual environment and their anatomical location in the local circuit was measured. The combination of virtual reality and high-resolution functional imaging should allow for a new generation of studies to probe neuronal circuit dynamics during behavior.

The location-specific firing of hippocampal place cells¹ during navigation represents a salient neural correlate of spatial information in the mammalian brain. More generally, place cells and other hippocampal neurons are thought to be involved in the encoding of episodic memories², making the behaving animal the most appropriate setting in which to study their activity patterns. The firing properties of hippocampal neurons in behaving rodents have traditionally been studied using extracellular electrode techniques, which has led to their characterization in numerous experimental paradigms (previously reviewed^3,4). However, electrophysiological methods have limitations in characterizing precise spatio-temporal activity patterns, targeting genetically labeled subpopulations and monitoring sub-cellular compartments.

For example, it is unclear whether hippocampal neurons with similar place fields are spatially organized within the hippocampus. While organization on spatial scales ranging from tens of microns⁵ to millimeters^6,7 has been reported, a separate study found no organization on either of these scales⁸. This question has been difficult to address because extracellular electrodes cannot report the precise anatomical location of the recorded cells^6-9. Indirect methods, such as immediate early gene (IEG) expression^5,8, are also problematic because the relationship between gene expression and neuronal spiking is not well established. Tetrode recordings have revealed no correlation between place field location and the cell's anatomical location⁸. In contrast, IEG studies have provided evidence for micro-clustering of cells that were active in restricted places within a larger environment⁵.

High spatial resolution optical imaging of neuronal activity can address these and other questions by providing a full spatio-temporal view of neuronal population activity within a micro-circuit, by recording from genetically-labeled neuronal subtypes, and by monitoring sub-cellular compartments such as dendritic branches and spines. However, no methods currently exist to perform high-resolution functional imaging of neurons in the hippocampus of awake, navigating mice. To achieve this goal, three obstacles must be overcome: cellular resolution imaging in the brain of a mobile mouse, imaging more than a millimeter beneath the cortical surface, and imaging compatible with navigation behavior. Two-photon microscopy¹⁰ in combination with calcium sensitive dyes has been used to provide cellular resolution functional imaging in awake mobile mice on a spherical treadmill^11,12. Chronic functional imaging in this preparation has been made possible using the genetically encoded calcium indicator GCaMP3¹³. In an anesthetized acute mouse preparation, sub-cellular resolution hippocampal imaging of non-activity dependent fluorescent probes has been demonstrated¹⁴. Recently, a virtual reality system has been developed to allow for the study of spatial behaviors in head-fixed mice¹⁵.

Here, we combined and extended the capabilities of these existing methods to make possible high-resolution functional imaging of the hippocampus in mice navigating in a virtual reality environment. A preparation was developed for chronic imaging of CA1 neuronal activity with cellular and sub-cellular resolution in awake mice head-restrained on a spherical treadmill. This was combined with a custom two-photon microscope and light blocking methods designed for imaging in the presence of the visual display surrounding the treadmill that defined a virtual environment. Using these methods, we optically recorded from populations of ∼80–100 GCaMP3-expressing neurons in the CA1 region of the hippocampus while the mouse was running along a linear track. Subpopulations of the recorded neurons were identified as place cells. Because our methods could precisely determine the physical location of the place cells within the hippocampus, we were able to show that for cells separated by more than a few tens of microns, no strong relationship existed in our experimental paradigm between the location of place fields in the virtual reality environment and the position of the corresponding place cells in CA1. Nearby cells (<∼35 micron separation) showed enhanced correlation, although the possibility that this is produced by optical intermixing of signals from adjacent cells could not be excluded. Additionally, the ability to record place related activity in putative hippocampal interneurons and CA1 apical dendrites is demonstrated.

Results

Combining two-photon microscopy and mouse virtual reality

Our apparatus (Fig. 1a, b) was designed around our previously described virtual reality system¹⁵. The limbs of a head-restrained mouse rested on a spherical treadmill¹². A toroidal screen subtending a mouse's visual field surrounded the treadmill and displayed a computer-generated image of a virtual environment^15,16. Ball movements recorded with an optical computer mouse provided information on running speed and direction which were used by a computer program implementing the virtual environment to update position and view angle. As previously reported¹⁵, head-restrained mice were trained using operant conditioning to perform the task of running back and forth along a 180 cm-long virtual linear track (Fig. 1c). The mice received water rewards at the ends of the track after successfully traversing the full track length; consecutive rewards at the same end were not available.

Figure 1.

Experimental setup. a. The experimental apparatus consisting of a spherical treadmill, a virtual reality apparatus (projector, RM: reflecting mirror, AAM: angular amplification mirror, toroidal screen and a optical computer mouse to record ball rotation) and a custom two-photon microscope (Ti:S: titanium:sapphire laser, LP: long pass filter, X-Y: galvanometers, SL: scan lens, M: mirror, TL: tube lens, DM: dichroic mirror, CL: collection lens, L: biconcave lens, BP: bandpass filter, FL: focusing lens, PMT: photomultiplier tube, Sliding stage: used to move microscope for treadmill access, X-Y translation: moves treadmill and mouse, Z-translation: objective focus control, Rubber tube (shown in cross-section): for light shielding). b. Photograph of experimental setup. c. View from one end of the virtual linear track (top). Top view of the linear track (bottom). d. View of materials used to block background light from entering the microscope objective hole. Hippocampal imaging window can also be seen. e. Detailed view of hippocampal imaging window (from boxed region in d.) f. In vivo two-photon images at different depths through the hippocampal window.

A two-photon microscope that could fit within the geometric constraints of our virtual reality apparatus without obstructing the mouse's view of the display was designed and constructed (Fig. 1a). The physical dimensions of the design were most severely constrained by the small distance (∼13cm) between the top of the head-plate on the mouse's head and the bottom of the reflecting mirror in the virtual reality projection path. Additionally, this microscope was designed to be completely shielded from the bright light of the virtual reality projection display so that the smaller number of photons from the fluorescent probe could be detected by the photomultiplier tubes (PMTs) without contamination. Additional light blocking measures were then implemented at the laser input port and the hole for the microscope objective (see Methods, Fig. 1a,d,e) such that the amount of detected background light was less than ∼5% of the baseline fluorescence level from labeled cells.

A window for chronic imaging of CA1 neurons in awake mice

The CA1 region of the hippocampus is more than a millimeter below the cortical surface and cannot be directly imaged using two-photon microscopy¹⁷. Instead, the overlying cortex was removed by careful aspiration (down to the external capsule) and replaced with a metal cannula with a coverslip sealing one of the openings (See Methods, Fig. 1e). This created a chronic hippocampal window that allowed for direct imaging of the hippocampus.

Genetically encoded calcium indicators were used to optically record the activity of CA1 neurons. Hippocampal injections of adeno-associated virus AAV2/1-synapsin-1-GCaMP3 (a moderate, neuron-specific promoter¹⁸ in a serotype with efficient gene delivery to neurons¹⁹) resulted in expression of GCaMP3 in a large population of CA1 neurons. Two-photon imaging of the GCaMP3 expressing area at a shallow depth through the hippocampal window showed axons in the external capsule/alveus which appeared as a dense plexus of fibers running parallel to the hippocampal surface (Fig. 1f, top). Inspection of images acquired in stratum pyramidale (∼130–160 microns below the external capsule) revealed the densely packed neurons of this cell body layer. Voids of fluorescence indicative of unlabeled cells were not seen, suggesting near ubiquitous local labeling (Fig. 1f, middle, note that GCaMP3 is excluded from the nucleus causing a “donut-like” appearance of the cell bodies¹³). All pyramidal neuron cell bodies in stratum pyramidale have a prominent apical dendrite projecting ventral into stratum radiatum (deeper into the hippocampus). Imaging perpendicular to the long axis of these dendrites in stratum radiatum revealed their cross-section as small dots (Fig. 1f, bottom) (see Supplemental Movie 1 for a z-series through the labeled CA1 region).

Optical identification of CA1 place cells

A timeline was developed for the sequence of steps necessary to image hippocampal activity in mice trained to perform the spatial behavior task. Mice were first implanted with a metal headplate to allow for head-restraint on the spherical treadmill and behavioral training. They were placed on water scheduling for several days before behavioral training in the virtual reality apparatus began. Once the mice were proficient at the task (∼2 rewards/minute, ∼7–10 days of training), the GCaMP3 virus was injected, followed the next day by the hippocampal window implantation. The mice were returned to behavioral training for ∼5–7 days until the GCaMP3 expression level produced an acceptable signal-to-noise ratio level for imaging calcium transients. Using the hippocampal imaging window and viral delivery of GCaMP3, it was possible to image activity in CA1 neurons repeatedly over the course of many (∼3–4) weeks.

To optically identify and characterize place cells, we collected time-series movies (∼64ms/frame) of fields of view (∼200×100 microns) in the CA1 region of the hippocampus containing ∼80–100 neurons (Fig. 2a, left) while the mice navigated the virtual linear track. Time-series at a single location in CA1 were acquired for ∼9–13 minutes. In all, 47 time-series from ∼10 different labeled regions in 4 mice were analyzed. During the time-series acquisitions, rewards were acquired at a rate of 3.2±0.7 rewards/min with a mean distance run between rewards of 2.5±0.5 m, similar to reward rates during our previous electrophysiology experiments¹⁵. To extract the fluorescence versus time traces for the individual neurons, the movies were first corrected for brain motion that occurred during the acquisition using a 2-D cross correlation algorithm (see Methods). It was difficult to manually draw regions of interest (ROIs) around individual neurons because CA1 cell bodies are tightly packed together, GCaMP3 resides only in the cytoplasm (near the cell edges), and the axial extent of the imaging focal spot is likely >∼4 microns²⁰. Instead, an automated cell identification method based on independent component analysis (ICA) and principal component analysis (PCA)^21,22 was used. Regions of interest for 10 neurons identified using this procedure are shown in red for the example field of view shown in Fig. 2a (right). ΔF/F traces (Fig. 2b, black traces with red segments) were extracted from the ICA/PCA-defined ROIs. ΔF/F traces revealed a baseline periodically interrupted by calcium transients that varied in amplitude (mean peak ΔF/F=28±32%), consistent with a difference in the number of underlying action potentials^13,23,24, and varied in duration (mean transient duration=1.2±1.1s), consistent with the summation of multiple transients^12,13,25. Significant transients with <5% false positive error rates were identified^11,12 and used in all subsequent analyses (see Methods) (Fig. 2b,c red traces). These traces were taken as a surrogate measure of spiking activity and are referred to as the temporal activity pattern of the neurons.

Figure 2.

Imaging CA1 place cells in the dorsal hippocampus. a. Two-photon image of neuron cell bodies in stratum pyramidale of CA1 labeled with GCaMP3. The indicator is excluded from the nucleus¹³. ROIs for example cells are shown in red (right). b. GCaMP3 baseline subtracted ΔF/F traces are shown in black for a subset of the cells labeled in a (right). Red traces indicate significant calcium transients with <5% false positive error rates (see Methods). The position of the mouse along the virtual linear track and reward times are shown at the bottom. c. Expanded view of boxed region in (b). d. Mean ΔF/F versus linear track position for a subset of the cells labeled in a (right). e. A plot of mean ΔF/F versus linear track position for all of the cells labeled in a (right). f. Place cells are colored according to the location of their place fields along the virtual linear track. Only place cells with significant place fields during running in the positive direction are shown here.

In the 47 time-series datasets, we analyzed epochs in which the mouse exhibited goal directed behavior to receive a reward (high reward rate epochs, see Methods). In nearly all of these datasets many individual neurons were found to have location specific activity. From the example field of view shown in Fig. 2a, the temporal activity patterns of four neurons are shown in Fig. 2b along with mouse position. When the mouse ran in the positive direction (left to right as seen in Fig. 1c, bottom), these four cells were each active at a different specific location. This can be seen in the individual traversal shown in Fig. 2c and in the plot of mean activity versus linear track position averaged over all 21 positive direction traversals (Fig. 2d, see Fig. 2e for plots from all 10 cells shown in Fig. 2a). These plots reveal well-defined and statistically significant fields of neuronal activity (p<0.05, p-value from bootstrapping, see Methods). The spatially restricted activity of these neurons and the well-defined shape of the activity fields were very similar to extracellular recording measurements of place cells in real and virtual reality linear tracks^15,26,27. Therefore, we concluded that these neurons are place cells. In addition to identifying the place field for each of the place cells, imaging also reveals their exact relative anatomical location (Fig. 2f).

Consistent with previous studies^15,26, optically identified place cells had different place fields in the same environment depending on the direction of running (Fig. 2b, Fig. 3). To analyze directionality of activity during place field traversals, the fluorescence time-series were divided into epochs defined by running direction. The mean fluorescence versus track position was calculated separately for each direction. A cell that had a place field during at least one direction of running was called a place cell; each place cell could therefore have more than one (up to two) place field. Many place cells had different place fields depending on the running direction (Fig. 3a, closed arrowheads) or had a place field for only one running direction (Fig. 3a, open arrowheads). This high degree of directionality (Fig. 3b) was consistent across all datasets and was quantified for all place fields using a directionality index (0.83±0.25, Fig. 3c, 1=high directionality, 0=no directionality, see Methods for definition).

Figure 3.

Place cells differ depending on the running direction in the linear track. a. An example imaging field in which the place cells are colored according to the location of their place fields along the virtual linear track. Significant place fields during running in the positive direction are shown on the left and negative running direction are shown on the right. Example place cells with different place fields or no place fields depending on the running direction are highlighted with closed arrowheads or open arrowheads, respectively. b. A plot of mean ΔF/F versus linear track position for the positive direction place cells labeled in a (left) during running in the positive (left) and negative (right) directions. c. Histogram of directionality index for all place fields.

Although only high reward rate periods were analyzed in the Results, we also analyzed place fields during different behavioral contexts^28-30 including the low reward rate periods (see Supplemental Fig. 1). We found the main results of this research to be highly similar using either method of analysis (Supplemental Fig. 2).

Across our 47 time-series datasets, 808 place cells and 881 place fields in those place cells were identified. The mean total number of place cells per imaging field was 17±9 and the mean total number of place fields per imaging field was 19±10 (number of place cells with 1 place field: 735, 2 place fields: 73; 18±6 complete runs through each place field). The calcium transients that defined these place fields had a mean duration of 1.1±0.9 seconds (Fig. 4a) and a mean peak ΔF/F of 35±29% (Fig. 4b). The distribution of place fields as a function of position along the track was inhomogeneous (Fig. 4c), as seen previously³¹, with a larger number centered between the reward sites and the towers (tall distal cues located ∼60cm from each end of the track, see Fig. 1c) than in the middle of the track. The mean width of the optically identified place fields was 50±19 cm (Fig. 4d), which is ∼20% larger than the mean width of place fields defined by firing rate found in extracellular recordings under similar conditions (41±14cm)¹⁵. To investigate this broadening, we convolved spike train electrophysiology recordings with model calcium transient waveforms (see Methods) and observed a similar increase of the resultant place field width (∼20%, see Supplemental Fig. 3b,e). This result is consistent with the idea that our place fields are defined by underlying spiking activity that is convolved with the time-broadening effects of intracellular calcium dynamics (Supplemental Fig. 3).

Figure 4.

Characterization of place cell calcium transients and place fields. Histograms of place cell transient widths (a) and transient peak ΔF/F (b) are shown for periods of mouse movement along the virtual track. Histograms of place field position along the linear track (c) and place field widths (d) are also shown.

We found that when optically identified place cells are active, the activity most frequently occurs in the place field; however, the place cells are not necessarily active during every traversal of the place field. This is similar to previous tetrode study reports of “excess activity variance”²⁸ in place cells. Fig. 5a shows the activity trace (temporal activity pattern) for each individual traversal of the mouse in the positive running direction for a subset of the place cells shown in Fig. 2. While place cells 2,3,6 and 9 are reliably active at the same location during nearly every traversal (active during >75% of traversals through the place field and active >66% of the time spent in the place field), place cell 10 demonstrates a high degree of activity variance. This cell is reliably active at the same location, but not reliably active during each traversal (active during 38% of traversals through the place field and active 35% of the time spent in the place field). For all of the 881 place fields, the place cells were active during 65±18% of traversals through the place field (Fig. 5c) and active 53±17% of the time spent traversing through the place field (Fig. 5d).

Figure 5.

Place cell activity variability in place fields. a. Temporal activity pattern versus virtual linear track position traces for a subset of the cells shown in Fig. 2 a (right). Each of the 21 positive running direction track traversals is shown for each of the cells. b. Mean and SD of ΔF/F versus linear track position for the traces shown in (a). Histograms of the probability that a place cell is active during traversals through the place field (c) and of the percentage of place field traversal time that the cell had a significant calcium transient (d).

Comparison to electrophysiology

The properties of place cells measured optically in hippocampal window mice (cortical excavation and region infected with AAV2/1-synapsin-1-GCaMP3) were compared to the same properties measured previously with electrophysiology in control mice (no cortical excavation or virus infection)¹⁵: 1. The place field width in the hippocampal window mice (50±19 cm) is very similar to the control mice (41±14 cm,¹⁵), with the effects of intracellular calcium dynamics in the presence of GCaMP3 capable of explaining the ∼20% broadening of the fields (Supplemental Fig. 3) and 2. The percentage of neurons that are place cells (∼15–20%) in hippocampal window mice is similar to, though slightly less than, our whole-cell patch clamp recording (∼20–30%,¹⁵). We also directly compared electrophysiologically-measured place cell properties and local field potentials (LFPs) in hippocampal window mice (cortical excavation and region infected with AAV2/1-synapsin-1-GCaMP3) to the same measurements in control mice (no cortical excavation or virus infection)¹⁵. The LFP recorded theta frequency in hippocampal window mice (7.3±0.3 Hz, n=4 mice) (Supplemental Fig. 4a) is very similar (p=0.32, two-tailed t test) to control mice (7.4±0.2 Hz, n=11 mice). The phase precession of spike times relative to LFP recorded theta observed in control mice (Δphase=−73±48°, p<0.01; correlation between phase and position of spikes: correlation coefficient=−0.2±0.1, p<0.01; n=10 place fields, 3 mice,¹⁵) was highly similar in hippocampal window mice (Δphase=−104±87°, p<0.01; correlation between phase and position of spikes: correlation coefficient=−0.4±0.3, p<0.01; n=3 place fields, 2 mice) (Supplemental Fig. 4g). Finally, theta-modulated high frequency (>50Hz) bursts of action potentials with decreasing spike amplitude that were characteristic of place cells in control mice were also observed in the hippocampal window mice (see inter-spike intervals in Supplemental Fig. 4c,e). These measurements demonstrate the similarity between place cells and CA1 network dynamics in hippocampal window mice compared to control rodents in both real and virtual environments^15,26,27. Because our hippocampal window mice required cortical excavation and AAV2/1-synapsin-1-GCaMP3 infection, these results also demonstrate that neither potential perturbation from viral infection or GCaMP3 expression, nor the cortical excavation, significantly altered the hippocampal CA1 dynamics.

The anatomical organization of CA1 place cells

We next examined the anatomical layout of place cells on the micron scale (see Fig. 6a for example fields of view). To quantify the relationship between anatomical location and place field location, all pairwise distances between place cells (distance between each cell's ROI center of mass) and pairwise distances between the peaks in their place fields were calculated separately for each running direction for each imaging field. The results from all 47 datasets were then combined and the mean distance between place fields was plotted as a function of mean distance between the place cells (Fig. 6b). For cells spaced over a wide range of distances, but excluding near-neighbors <∼35 microns apart, a statistically insignificant relationship was found between the mean distance between place fields and the mean distance between the place cells (Spearman's rank correlation coefficient of 0.57, p=0.2). Therefore, on average, the distance between place cells in the physical space of the hippocampus was not related to the distance between their place fields along the track when the place cells were >∼35 microns apart (though some exceptions to this average rule were seen, see for example Supplemental Fig. 1c left). The data point corresponding to the most closely neighboring place cells (<∼35 microns apart) was significantly lower (p<0.025, two-tailed t test) than the points for all other distances between cells. This could be interpreted as fine scale spatial clustering of place cells with similar place fields, however residual crosstalk, residual brain motion or a common neuropil signal between neighboring ROIs could also possibly explain this result (see Discussion).

Figure 6.

Spatial organization of place cells in dorsal CA1. a. Example images from different fields of view in which the place cells are colored according to the location of their place fields along the virtual linear track. Each image shows place cells with significant place fields during running in either the positive or negative direction. b. Plot of mean place field-place field distance versus mean place cell-place cell distance averaged over all 47 time-series. The error bars represent SE. c. Plot of mean cell-cell temporal activity pattern correlation versus mean cell-cell distance averaged over all 47 time-series for all place cells (grey) and all neurons (black). The error bars represent SE.

We next studied whether neuron-neuron temporal activity pattern correlation varied as a function of interneuronal distance. For all place cells in each imaging field, all pairwise temporal activity pattern correlations and pairwise distances between place cells were calculated and then the results from all 47 datasets were combined. When the mean temporal correlation was plotted as a function of mean distance between the place cells (Fig. 6c, grey trace), a statistically insignificant relationship was found (Spearman's rank correlation coefficient of −0.25, p=0.5) for cells spaced >∼35 microns apart. Again, the data point corresponding to the near-neighbor place cells was found to be statistically different (p<0.035, two-tailed t test) than all but one of the other data points. When this calculation was repeated for all neurons, including both place cells and non-place cells, the overall level of correlation was lower than for the place cells alone and a statistically significant relationship was found between neuron correlation and distance between the neurons (Spearman's rank correlation coefficient of −0.95, p<10⁻⁵) (Fig. 6c, black trace), regardless of whether the data point for the most closely neighboring neurons was included.

Imaging activity in dendrites and putative interneurons

To test whether our imaging method had the resolution and signal-to-noise ratio capable of recording activity patterns from dendrites, time-series were acquired ∼75 microns ventral to stratum pyramidale (Fig. 7a, dashed line). In this plane, apical dendrites from the overlying pyramidal neurons could be identified (individual bright spots in Fig. 7b). When the mean activity versus linear track position was plotted, well-defined place fields were found in many of the apical dendrites (Fig. 7c). Dendrites were imaged in 4 fields of view (∼50 dendrites per field) in 3 mice; spatially modulated dendritic activity patterns were seen in all of the fields. Back-propagating action potentials are known to invade the dendritic arbors of CA1 pyramidal neurons^32,33. Therefore, it is likely that the activity in the apical dendrites is due to back-propagating action potential-induced opening of voltage gated calcium channels rather than the smaller calcium transients associated with synaptic input.

Figure 7.

Imaging place related activity in dendrites and putative interneurons. a,b. A two-photon image (b) of a field of view ∼75 microns ventral to the stratum pyramidale cell body layer (dashed line in a). Bright spots in (b) are a cross-section through the apical dendrites from the overlying CA1 neurons. c. Mean ΔF/F versus linear track position for the dendrites labeled in the image in (b). d,e. A two-photon image (e) of a field of view ∼50 microns dorsal to the stratum pyramidale cell body layer (dashed line in d). Sparsely distributed cell bodies in (e) are assumed interneurons. f. Mean ΔF/F versus linear track position for the interneurons labeled in the image in (e).

To demonstrate the ability of our methods to record from interneurons, the imaging plane was adjusted to ∼50 microns dorsal to stratum pyramidale (Fig. 7d, dashed line) where sparsely distributed cell bodies were observed (Fig. 7e). The neurons in this region of stratum oriens have been previously identified as interneurons, suggesting that these cells are likely Basket, oriens lacunosum moleculare or Axo-axonic interneurons³⁴. As in stratum pyramidale, neurons with spatially modulated activity patterns were found in stratum oriens as well (Fig. 7f), some with “off” fields (Fig. 7f, cell 1) as has been previously observed with extracellular recording^35,36. Interneurons were imaged in 4 fields of view (∼2–3 interneurons per field) in 2 mice; spatially modulated interneuron activity patterns were seen in all of the fields.

Discussion

Our method relies on several combined technologies. Surgical methods were developed to implant a hippocampal window that allowed for chronic sub-cellular resolution imaging within the CA1 region of the hippocampus in behaving mice. The combination of this window with a custom two-photon microscope and background light suppression methods allowed for imaging in mice interacting with a recently described visual virtual reality system. Using the recently developed genetically encoded calcium indicator GCaMP3, our methods allowed for the study of spatially modulated activity patterns of pyramidal neurons in stratum pyramidale, putative interneurons in stratum oriens, and apical dendrites in stratum radiatum. By imaging the activity of populations of ∼80–100 neurons in stratum pyramidale in mice trained to navigate along a virtual linear track, place cells were identified that had characteristics very similar to spiking rate-defined place cells in both real and virtual environments^15,26,27,37.

Tetrodes are currently the most common method used to record CA1 place cell activity in behaving mice. Tetrode array recordings combined with spike sorting procedures can be used to identify many (∼5) single units per tetrode^38,39. While sub-millisecond temporal resolution is a major advantage of this method compared to imaging, the current limitations of tetrode methods are the low spatial resolution (>100s of microns) and sparse sampling within the micro-circuitry^38,40.

Our imaging method has several advantages over tetrode methods. For example, it is possible to report the precise anatomical position of functionally identified neurons within the micro-circuitry. Imaging also allows for functional recording from sub-cellular compartments, identifying all neurons (even silent cells), and can take advantage of the growing number of available genetic tools⁴¹. It should be possible to image subcellular dynamics such as signal transduction⁴² and structural plasticity⁴³ in the context of learning and memory during behavioral paradigms. Additionally, optical methods may allow for identifying the functional properties of specific neurons in a large population, followed by either the subsequent reconstruction of the underlying connections⁴⁴ or modification of their activity⁴⁵. Finally, imaging methods can allow for the unambiguous identification and recording of the same neurons over many weeks. Although this was not the focus of this research, to address the technical feasibility of such studies we have imaged the same region of CA1 over multiple days (see Supplemental Fig. 5).

Here it was assumed that calcium transients could be used as a proxy for spiking activity. Sodium action potentials have been shown to generate calcium transients in GCaMP-3 expressing hippocampal pyramidal neurons in brain slices. Calcium transient amplitude was found to increase linearly with the number of evoked action potentials, until saturation at >∼10 action potentials¹³. A quantitatively similar relationship between spiking activity and calcium transient amplitude was found for neocortical neurons in brain slice and in vivo in both anesthetized and awake mice¹³. It is therefore reasonable to assume that the spiking activity of the CA1 neurons in vivo generated calcium transients similar to what was observed previously in slices¹³. Using these previous calcium transient measurements¹³ and our extracellular place cell recordings, we calculated the spike induced calcium transients that would be expected in our imaging experiments (Supplemental Fig. 3). The characteristics of the optically defined place fields are consistent with and can be fully explained by calcium transients induced by sodium action potentials. Therefore, the contribution of other sources to our calcium transients, such as calcium influx due to synaptic input or dendritic calcium spikes, is likely small.

It is unlikely that our methods allow for detecting single action potentials¹³, determining firing rates or reliably counting the number of spikes. While these limitations likely do not pose a problem for identifying place fields in place cells due to the dramatic increase in spiking rate in the place field and the ability to average over many traversals, it is still possible that place cells with low activity levels may not be detected.

Cortical excavation was used to expose the hippocampus for acute, anesthetized cat electrophysiology experiments nearly 50 years ago⁴⁶. More recently, cortical excavation was combined with a polystyrene tube filled with agarose and sealed with a coverslip to facilitate two-photon imaging of dendritic spines in an acute, anesthetized mouse preparation¹⁴. Here, these methods were used as a starting point to develop a chronic hippocampal window designed for imaging in behaving mice. The stainless steel cannula and coverslip directly bonded to the external capsule surface with Kwik-sil formed a rigid support structure that minimized brain motion during animal movements and allowed for repeated imaging of the hippocampus for weeks. We found that removing the cortex overlying the hippocampus did not detectably alter the mouse's performance of the task, or the hippocampal dynamics and place cell properties that were measured in our preparation compared to control mice.

Miniaturized head-mounted microscopes^47,48 may allow for hippocampal imaging in freely moving animals. Such experiments would benefit from the natural array of inputs, as opposed to the lack of vestibular input and potentially altered gait of our mice. Our methods, however, have the advantages of not requiring a miniaturized microscope, easy combination with electrophysiology, and the potential to manipulate specific environmental cues using virtual reality in ways that would be difficult or impossible in real environments.

We used an ICA/PCA algorithm to identify individual neurons based on both their spatial and temporal characteristics²¹. This method, however, could not identify silent neurons and occasionally missed active cells, making the estimation of the number of potentially active neurons difficult. Therefore, it was difficult to quantify the exact fraction of neurons that were place cells. Based simply on the morphology of neurons in the fields of view, we estimate that our fields contained ∼80–100 neurons, meaning that ∼15–20% of the neurons were place cells, slightly less than previous estimates¹⁵. In addition to the uncertain number of potentially active neurons, this slight difference could also be due to differences in the recording methods and differences in the definitions of place fields.

The ICA/PCA algorithm was successful in limiting the crosstalk between neighboring ROIs. However, when studying the spatio-temporal organization of neurons in CA1, the possibility of residual crosstalk, residual brain motion or a common neuropil signal between the most closely neighboring cells (<∼35 microns) could not be ruled out. This is one possible explanation for the statistically outlying data points in Fig. 6b,c corresponding to the most closely neighboring cells. An alternative explanation for these outlying data points is that they represent small clusters of functionally similar neurons, as recent studies have suggested^5,9.

For place cells that were further separated (>∼35 microns), the distance between their place fields along the track was statistically unrelated to the distance between their positions in the hippocampus. A few previous electrode recording studies found anatomically organized clusters of functionally similar hippocampal neurons on the 0.6–1mm spatial scale^6,7, while a separate study measuring CA1 place cell location concluded that there was a lack of anatomical spatial organization⁸. Because of the low spatial resolution and sparse sampling within micro-circuitry³⁸, these previous methods provided only an indirect measure of the spatial organization of place cells in CA1. In contrast, using our methods, we were able to directly measure the spatial organization. While possible crosstalk concerns did not allow us to unambiguously measure the organization down to the finest scale, it was possible to conclude that place cells are anatomically distributed down to a scale of at least ∼35 microns.

For place cells separated by at least this scale, there was also no relationship between the correlation between their temporal activity patterns and the physical distance between them. Interestingly, for all neurons, a statistically significant decrease in temporal correlation was found between the neurons the further they were from each other in the physical space of the hippocampus. It is interesting to note that while this decrease is significant, both the overall level of correlation and the rate of correlation decrease as a function of distance are nearly an order of magnitude smaller than has been observed in the motor cortex of behaving mice¹¹.

Finally, we note that spatio-temporal organization can occur in many forms, most of which were not examined here. The methods outlined in this research should allow for future studies to search more thoroughly for micro-organization within the hippocampus.

Materials and Methods

Two-photon microscope and data acquisition

Our microscope design is shown in Fig. 1a. Focusing control was achieved by mounting the microscope objective on a single-axis piezo-motorized translation stage (Z-translation; Agilis AG-LS25) within the microscope. X-Y control was achieved by mounting the spherical treadmill and mouse head-restraint on a manual actuator-driven translation stage.

The microscope was built out of custom machined and black anodized aluminum parts and was designed to entirely enclose all of the internal optics except for one hole for the microscope objective and one hole for the entry of the excitation laser beam. The first step in mitigating stray light entry into the microscope was to create precise fitting lips at every joint between any two pieces of the outer shell of the microscope. This had the effect of eliminating nearly all light entry through the outer body of the enclosed microscope. The first major point of entry for stray light into the microscope was then the excitation laser input port. A long pass colored glass filter (Thorlabs, 780 long pass) covered this hole to allow only the excitation laser light to pass, but not the shorter wavelength light from the projection system. The second point of entry for stray light was the hole for the microscope objective (Olympus 40×, 0.8 NA). To combat light entry at this point, one end of a loose tube of thin black rubber was attached to the microscope around the objective (Fig. 1d). The other end of the tube was attached to a metal ring that was machined to tightly fit around a separate ring that was cemented to the top of the headplate and centered around the craniotomy. The flexibility of the thin rubber allowed for movement of the microscope objective with respect to the craniotomy while maintaining the light-tight connections at both ends of the rubber tube. It was also possible for stray light (especially longer wavelength red light) to enter the microscope by traveling through the animal and subsequently through the exposed skull between the craniotomy and the inner edge of the metal ring that was attached to the animal's skull. This area was therefore covered with opaque dental cement (Meta-bond, Parkell, made opaque by adding India Ink at ∼5% vol/vol) so that the only remaining source of stray light through the microscope objective hole was through the hippocampal window itself (Fig. 1e). The light through this window proved to be inconsequential in the green imaging channel (<∼5% of the baseline fluorescence level from labeled cells). Though red fluorophores were not used in this research, we found that a red light blocking filter (e.g. 500/40, Chroma) in front of the projector was needed to reduce the amount of stray light in the red imaging channel to <∼5% of the baseline fluorescence level from labeled cells.

The Ti:sapphire excitation laser (Chameleon Ultra II, Coherent) was operated at 920 nm (∼30–50 mW average power at the sample). Green GCaMP3 fluorescence was isolated using a bandpass filter (Semrock, 542/50) and detected using a GaAsP PMT (1077P–40, Hamamatsu). ScanImage (v3.6)⁴⁹ was used for microscope control and acquisition. Images (256×64 pixels, ∼200×100 micron field of view) were acquired at 15.6 Hz. During the ∼9–13 minutes of time-series acquisition at a single location, photo-bleaching was observed at a mean rate of ∼1–2%/minute.

Our spherical treadmill and virtual reality system has been previously described^{12, 15}. We used a Digidata (Axon Instruments, 1440A) data acquisition system to record (Clampex 10.2) and synchronize position in the linear track, reward timing, and two-photon image frame timing (using the command signal to the slow galvanometer). Instead of the MX-1000 computer mouse used previously, here we used an MX-518 (Logitech) mouse to record ball rotation. The ball movements measured using the LED based MX-518 mouse were less sensitive to the exact distance between the sensor and ball surface compared to the MX-1000. Additionally, the MX-518 mouse can be adjusted between three gain settings, the lowest of which allowed for measurements of larger ball velocity without saturation compared to the MX-1000.

Mouse training, hippocampal window and AAV injections

All experiments were performed in compliance with the Guide for the Care and Use of Laboratory Animals (http://www.nap.edu/readingroom/books/labrats/). Specific protocols were approved by the Princeton University Institutional Animal Care and Use Committee. Imaging experiments were performed on 5 male C57BL-6 mice (postnatal day ∼50). First, mice were anesthetized (Isoflurane, ∼1%), a headplate was attached to the skull using opaque metabond and the stereotactic coordinates overlying the hippocampus were marked (1.8mm lateral, 2.0mm caudal of Bregma). Any exposed skull was covered with a silicone elastomer (Kwik-Sil, World Precision Instruments). Anesthesia was then removed and the mouse recovered. Beginning the next day, the mice were put on a water restriction regiment such that they received 1 mL of water per day. After ∼5–7 days, final body weight was ∼80% of the before restriction weight. Training then began in the virtual linear track (one ∼45 minute session per day) and continued until the mice routinely ran back and forth along the linear track to achieve a high reward rate (>∼2 rewards/minute); this required ∼10 training sessions. The mice were trained the same and the learning was the same as shown previously¹⁵.

The mice were then anesthetized and a small (∼0.5–1.0 mm) craniotomy was made at the previously marked spot on the skull overlying the dorsal hippocampus. Three injections of a solution containing AAV2/1-synapsin-1-GCaMP3 were made (∼30 nL/injection, ∼5 minutes/injection) ∼200 microns apart at a depth of ∼1300 microns below the dural surface using a beveled glass micropipette (∼2 MΩ after beveling). The craniotomy and any exposed skull were again covered with Kwik-Sil. The mice then recovered from anesthesia.

24–48 hours later, the mice were re-anesthetized and a trephine drill (Fine Science Tools) was used to cut a ∼2.7–2.8mm diameter craniotomy centered over the previously made small craniotomy. The dura was then removed with forceps and aspiration was used to slowly remove the cortex within the craniotomy. The removal was accomplished very slowly with aspiration of ∼50–100 microns of tissue at a time followed by repeated irrigation with saline until any bleeding stopped. These steps continued until the external capsule was exposed. The cortex and top most layers of the external capsule were “peeled away” such that the remaining external capsule was never physically touched. To reduce animal movement-induced brain motion that could interfere with time-series image acquisition, the surface of the external capsule was allowed to dry until tacky, a small drop of uncured Kwik-Sil was applied to the surface, and then the cannula was inserted (with a few hundred microns left above the skull surface) and cemented to the skull using opaque Meta-bond. When cured, the small drop of Kwik-sil had the effect of bonding the cannula to the external capsule and reducing in-focal plane (X-Y) brain motion. In some cases, the cannula did not fit through the craniotomy and a hand drill was used to slightly widen the hole. After this hippocampal window implantation surgery, the mice typically woke up after ∼10 minutes and were walking around the cage within ∼15 minutes. The next day, training in the virtual linear track resumed as before the surgery. Approximately 7 days after the window surgery, imaging experiments began. The experiments took place every 2–3 days for up to ∼3–4 weeks in the same mouse (up to ∼4–5 weeks after the window surgery). GCaMP3 expression reached a somewhat steady state level ∼14 days post injection. A small fraction of nuclear filled GCaMP3 neurons with altered physiology were previously described in the cortex¹³. A similar small fraction of nuclear filled neurons were observed here in CA1 (starting ∼2–3 weeks post injection). These neurons often generated large, long duration calcium transients that rarely defined a place field.

The cannula was composed of a 1.5mm segment of a 2.77mm outer diameter and 2.31mm inner diameter thin walled stainless steel tube (Small Parts Inc.). A 2.5mm diameter round coverglass (Erie Scientific, Thermo Scientific) was cemented to one end of the tubing using UV curable adhesive (Norland Products).

Does cortical excavation alter hippocampal dynamics and place cell properties?

In order to image the CA1 region of the hippocampus with cellular resolution, it was necessary to unilaterally remove part of the overlying cortex (including parietal cortex and parts of visual and hindlimb sensory cortex). This raises the concern that cellular and network properties within the hippocampus may be altered either by direct mechanical trauma to the hippocampus or by altering upstream inputs to the hippocampus due to the cortical lesion (although the excavated cortical regions do not provide strong direct projections to the hippocampal region). Regarding mechanical trauma, our surgical procedures allow for the removal of the overlying cortex without the need to ever physically touch the surface of the hippocampus. The cortex and the top most axons in the external capsule were “peeled away” from the remaining external capsule without applying direct pressure or inadvertent mechanical trauma to the hippocampus. Furthermore, the external capsule itself provides a protective barrier to the underlying CA1 region of the hippocampus. Therefore, overt signs of damage, such as vesiculated dendrites or reduced or aberrant neuronal activity observed via calcium transients, were not observed. While overt signs of damage were easily avoided, more subtle effects due to alterations of the upstream circuits projecting to the hippocampus were still possible. However, the properties of CA1 place cells measured electrophysiologically and optically in our hippocampal window mice (cortical excavation and region infected with AAV2/1-synapsin-1-GCaMP3) were found to be highly similar to the same properties measured previously with electrophysiology in control mice (no cortical excavation or virus infection) (see Results). Additionally, no significant difference was observed in the task performance of the same mice in the 4 days before the hippocampal window surgery compared to the 7 days after the surgery (1.7±1.2 vs. 1.8±0.6 rewards/min, respectively, p=0.51, two-tailed t test, n=5 mice, one ∼45 minute session per mouse per day).

Bolus loading of calcium indicators in CA1

We first attempted to use multi-cell bolus loading⁵⁰ of the exogenous indicator Oregon Green Bapta-1-AM. While we were able to label large populations of CA1 neurons and record their activity, this method had the drawback of only labeling cells for a few hours. Additionally, we were only able to label cells immediately before the cannula was implanted, meaning that the imaging experiments had to be performed on the same day as the cortical excavation procedure and during the short time window after the mice awoke from surgery but before the cell labeling disappeared. During these periods, the mice rarely performed well at the task. A few place cells per field could be identified during periods of running, but the percentage of cells that were place cells (<∼5%) and the width of the place fields (∼2–3 times greater than expected from our electrophysiology measurements¹⁵) indicated altered hippocampal place cell and/or network dynamics. Because of these limitations, and the advantages of genetically encoded indicators, we used GCaMP3¹³.

Data analysis

Analysis was performed using ImageJ (1.40 g) and custom scripts written in MatLab (version 7). All data in the text and figures are presented as mean±SD, except in the plots shown in Fig. 6b,c and Supplemental Fig. 2g,h where the error bars represent SE.

Time-series datasets were motion corrected using whole-frame cross-correlation. Although our previously published HMM line-by-line motion correction algorithm¹² likely would provide superior correction, our imaging frame rate (∼15 Hz) was fast enough that in frame distortions were minimal using whole frame correction. Additionally, it was more straightforward to apply the ICA/PCA cell ROI selection algorithm to the whole-frame corrected time-series rather than the HMM corrected time-series in which blank lines made automatic ROI detection difficult. The mean frame to frame in plane (X-Y) Euclidean distance brain motion during all time periods (the mice ran almost continuously during our time-series acquisitions) was 1.8±1.3 microns; the motion during periods of mouse movement along the virtual track used to define place fields (see below) was 1.5±1.2 microns. Out of plane (Z) motion (measured as described previously¹²) was measured in untrained mice running on a spherical treadmill; during periods of running the Z-motion was 0.7±0.2 microns.

ROIs were defined as described previously²¹ (mu=0.5, 150 principal components, 100 independent components, standard deviation threshold =1.25, 60 pixels < area of ROI < 400 pixels). Manual inspection revealed that the ROIs nearly always defined single cell regions. ΔF/F versus time traces were generated for each ROI. Slow time scale changes in the fluorescence traces were removed by examining the distribution of fluorescence in a ∼15 second interval around each sample time point and subtracting the 8% percentile value. The baseline subtracted neuron fluorescence traces were then subjected to the analysis of the ratio of positive to negative going transients of various amplitudes and durations that we described previously¹². We used this analysis to identify significant transients with 5% false positive error rates and generated the significant transient only traces (see Fig. 2b,c red traces and Fig. 5a) that were used for all subsequent analysis as in¹¹. Note that the significant transient only traces are not discontinuous traces; the significant transients are left untouched, but the time points between the significant transients are all set to 0.

Place fields were identified and defined as follows. Long running periods of mouse movement along the virtual track in which the virtual velocity was >8.3 cm/s and the run length was >53 cm (straight run without changing direction or hitting the end of the track) were identified. These periods were first categorized based on the direction of running (positive or negative direction). Positive and negative long running direction periods were then further subdivided into two categories based on the animal's current performance of the task: Segments of time between two rewards in which long running periods of only one direction occurred were defined as high reward rate periods, all other long running periods were defined as low reward rate periods. Timeseries datasets were only included if a mean of at least 10 long running segments during high reward rate periods in the positive and negative directions were completed during timeseries acquisition (this occurred in 47 datasets total). Only high reward rate periods were analyzed in this research, except in Supplemental Fig. 1,2 and 5 where the low reward rate periods were also included as a comparison (note that for the analysis in Supplemental Fig. 1,2 and 5, the requirement of a mean of at least 10 long running/high reward periods was dropped, allowing for the inclusion of additional datasets: 103 datasets total). For each running direction (during high reward rate periods) for each cell, the mean ΔF/F was calculated as a function of virtual position for 80 position bins and this mean fluorescence versus position plot was then smoothed (averaged) over 3 adjacent points. Potential place fields were first identified as contiguous regions of this plot in which all of the points were greater than 25% of the difference between the peak ΔF/F value (for all 80 bins) and the baseline value (mean of the lowest 20 out of 80 ΔF/F values). These potential place field regions then had to satisfy the following criteria: 1. The field must be >18cm in width, 2. The field must have one value of at least 10% mean ΔF/F, 3. The mean in field ΔF/F value must be >3 times the mean out of field ΔF/F value and 4. Significant calcium transients must be present >30% of the time the mouse spent in the place field. Potential place field regions that met these four criteria were then defined as place fields if their p-value from boot strapping was <0.05. Bootstrapping was performed by breaking the ΔF/F trace for each cell into at least 9 segments (determined by the significant transients) that were randomly shuffled. The mean shuffled ΔF/F versus position plots were then subjected to the same criteria outlined above. This process was repeated 1000 times and the p-value was defined as the ratio of the number of times out of 1000 that the random shuffled trace generated a place field that met the above criteria.

A directionality index for mean ΔF/F in a place field in positive and negative running directions (F₊, F_-) was defined as |F₊−F_-|/(F₊+F_-). A directionality index of 0 indicates identical activity in both directions, whereas an index of 1 indicates activity in one direction only.

Note that while very little movement along the width of the virtual linear track was possible, this position as well as the view angle were not controlled for in our analysis. These factors could contribute to increased variability of the place fields. Also note that during time-series acquisition, the mice rarely stopped in the middle (between reward zones) of the track (Supplemental Fig. 1b): only 5±3% of the time per time-series was the virtual velocity <0.05 cm/s in the middle of the track.

Correlation values were defined as the Pearson's correlation coefficient. Correlation versus distance between neuron and place field distance versus distance between neuron plots were generated from imaging fields that had at least 2 place cells and any two cells that had overlapping ROIs were excluded from the analysis. The position of a place field along the virtual track was defined as the position within the field with the maximal ΔF/F value. For the mean place field width calculation, only fields in which neither field edge was at the end of the track were included. This avoided the inclusion of fields that were artificially narrow due to clipping at one end of the track.

Electrophysiology

Extracellular place cell recordings were performed in 4 hippocampal window mice using methods similar to those previously described¹⁵. All procedures for these mice were identical to those used in imaging experiments (water restriction, training, virus injection, surgeries, etc.) except that a small hole (∼0.5 mm) was drilled in the center of the window cover-slip prior to implantation. After implantation, this small hole allowed for electrode access to the hippocampus through the cannula. A tungsten metal electrode (3MΩ impedance at 1KHz, FHC) or glass microelectrode (resistance ∼2.5MΩ, filled with 0.5M NaCl) recorded CA1 extracellular currents that were amplified (DAM80, WPI or AxoClamp 2B) and filtered (3 Hz–10 Khz, Brown-Lee Model 440) before being digitized at 20 KHz with a Digidata (Axon Instruments, 1440A). Note that this preparation was intended to mimic the conditions of the imaging experiments, but did not allow for long duration extracellular recordings; the hole in the coverslip to allow for electrode access likely caused instability due to reduced stabilizing pressure. While these recordings (∼3–5 minute duration per cell) allowed for clear identification of place cells and phase precession, the reduced number of passes through the place field compared to longer recordings led to noisier place field maps and phase precession plots (Supplemental Fig. 4). To reduce low frequency (<∼50Hz) noise induced by charge build-up on the spherical treadmill, for electrical recordings the ball, ball mold, and air input tubes were coated with unscented fabric softener (Ultra Downy®, Procter and Gamble) and allowed to dry overnight.

Spikes were sorted offline using a threshold analysis. At most one unit was isolated from a single recording. Spike waveforms were overlayed and inspected visually to make sure that they matched (Supplemental Fig. 4c,e insets), and ISI distributions were plotted to make sure no spikes fell within the refractory period (1–2 ms; Supplemental Fig. 4c,e). To create firing rate maps (Supplemental Fig. 3b 4b,d), we divided the virtual linear track into 80 bins and calculated the firing rate as the total number of spikes in a bin divided by the total amount of time spent in a bin. To identify place fields, we found groups of adjacent bins with firing rates greater than 0.25 times the rate in the peak bin. We selected only those fields that were larger than 8 bins (18 cm) in length, had mean in-field firing rates of greater than 1.5 Hz, and had mean in-field firing rates more than 3 times larger than the mean out-of-field firing rate. Theta frequency was defined using the lag to the first peak in the autocorrelation of the LFP trace after band-pass filtering between 6–10Hz using an FIR filter.

Phase precession was calculated as previously described¹⁵. Briefly, to assign a phase to a spike occurring at time t, we identified, in the filtered trace, the times of the theta peaks immediately preceding and following the spike (t1 and t2, respectively) and calculated the phase as 360(t–t1)/(t2–t1). We circularly shifted the phase of the spikes in 1° steps from 0° to 360°, continuing across the 360° border, and fit a linear regression line to the phase versus position plot at each rotation. We found the rotation with the best fit, such that the sum of squared errors between the fit line and data was minimized, and used the correlation coefficient between phase and position at this rotation as a measure of phase precession.

Computed fluorescence traces (Supplemental Fig. 3c) were created by convolving a spike train with model fluorescence transients (rise time: t1/2 = 52 ms, decay time: t1/2 = 384 ms, peak amplitude taken from Fig. 6C of¹³). All values in these traces below 10% ΔF/F were then set to 0 (Supplemental Fig. 3d) to mimic the significant-transient-only traces generated for the real fluorescence time-series (Fig. 2b,c, red traces).

Supplementary Material

1

Supplemental Figure 1: Place cells for four different experimental context categories. Consistent with previous studies^28-30, the place cells we optically identified had different place fields in the same environment depending on the behavioral context. Place cells had different place fields depending on the direction of running along the track (Fig. 2b, Fig. 3). Furthermore, even when running in the same direction, place cells often had different place fields depending on the animal's current performance of the task (i.e. whether or not the animal was currently exhibiting goal directed running behavior to receive a reward). To analyze place fields during different behavioral contexts, the time-series was divided into epochs defined by four behaviors: 1. Positive running direction high reward rate, 2. Positive running direction low reward rate, 3. Negative running direction high reward rate, and 4. Negative running direction low reward rate. Note that this analysis using four behavioral contexts was only used in Supplemental Fig. 1, 2 and 5; only the two contexts based on running direction (during high reward rate epochs) were used in the rest of our analysis.

An example dataset in which place fields were defined in each context is shown in this supplemental figure. a. Two-photon image of neuron cell bodies in stratum pyramidale of CA1 labeled with GCaMP3. b. The position of the mouse along the virtual linear track versus time. Segments of the trace are colored according to the behavioral context shown in the legend. The reward times are shown at the bottom. c. Images from the field of view in (a) in which the place cells are colored according to the location of their place fields along the virtual linear track. Each image shows place cells with significant place fields during one of the behavioral context periods shown in (b). Although some place cells had the same place field independent of the task performance state (arrows in c for example, compare high reward rate to low reward rate in the same direction), many others had different (closed arrowheads in c for example) or no place fields (open arrowheads in c for example, compare high reward rate to low reward rate in the same direction) depending on the task state.

Supplemental Figure 2: Recalculation of main results using four behavior contexts. The results from Fig. 4, 5, and 6 using the two contexts based on running direction (during high reward rate epochs) were recalculated in this supplemental figure using the four behavioral contexts described in Supplemental Fig. 1. The results using both methods of analysis are highly similar.

For this supplemental figure, 103 time-series from ∼12 different labeled regions in 5 mice were analyzed. During the time-series acquisitions, rewards were acquired at a rate of 2.3±0.9 rewards/min with a mean distance run between rewards of 3.8±1.8 m, Place cells were defined as any neuron with a place field in one or more of the four context categories; each place cell could therefore have more than one (up to four) place field. Across the 103 time-series datasets, 1664 place cells and 1992 place fields in those place cells were identified. The mean total number of place cells per imaging field was 16±10 and the mean total number of place fields per imaging field was 19±13 (number of place cells with 1 place field: 1354, 2 place fields: 292, 3 place fields: 18, 4 place fields: 0). Results in a–d can be compared to Fig. 4a–d, e–f can be compared to Fig. 5c–d, and g–h can be compare to Fig. 6b–c.

Supplemental Figure 3: Comparison of spiking rate versus fluorescence calcium transient defined place fields. a. Spike train from an extracellular recording of a place cell from¹⁵. b. Mean firing rate versus linear track position from the recording in a. c. Fluorescence trace made by convolving the spike train in (a) with calcium transients with the following characteristics: rise time: t1/2 = 52 ms, decay time: t1/2 = 384 ms, peak amplitude taken from previous work¹³ d. Same as c, but all values <10% ΔF/F are set to 0. This creates a trace very similar (in terms of ΔF/F, transient durations and transient rate) to the significant transient-only traces used throughout this research. e. Mean ΔF/F versus linear track position for the computed fluorescence trace shown in (d). f. The position of the mouse along the virtual linear track versus time.

Supplemental Figure 4: Extracellular recordings from AAV2/1-synapsin-1-GCaMP3 infected CA1 regions of hippocampal window mice. a. The LFP recorded theta rhythm. b,d. Mean firing rate versus linear track position for two different place cell examples. The spike overlay (inset) and ISI for these single units are shown in c and e. f. Mean firing rate versus linear track position for a place cell example. g. An example plot of phase (two cycles) with respect to LFP recorded theta versus position on the linear track for all spikes during complete runs through the place field for the cell shown in f.

Supplemental Figure 5: Imaging the same CA1 location over multiple days. a, b. Example fields of view that were imaged over multiple days in the same mouse. The place cells in the fields of view are colored according to the location of their place fields along the virtual linear track. The same behavioral context is shown for the same field over days (positive running direction low reward rate for a, and negative running direction low reward rate for b, see Supplemental Fig. 1 for definitions of behavioral contexts used here). When place cells had place fields over multiple days, the field locations along the linear track were similar (see arrows). However, numerous cells were also observed with significant place fields on only one day; in a, this lead to a dramatic change across the population from day to day. More work is needed to understand the changes seen in a. It is possible that many of the place cells in this field remapped over the course of the many days shown, but it is also possible that the differences are due to technical or biological variability, such as slight differences in the z-plane, changes in the tissue structure due to neuro-genesis, differences in animal behavior leading to changes in view angle or position along the width of the track, or minor changes in the uncontrolled local environment cues (i.e. odor, auditory, etc.).

Supplemental Movie 1: Z-series movie of the hippocampal region labeled with GCaMP3 virus infection. Movie begins in the external capsule/alveus (dense plexus of fibers) and then steps ventral in 5 micron increments through stratum oriens, stratum pyramidale and then ends in stratum radiatum (∼350 microns below the external capsule surface). The field is ∼200×100 microns.

Supplemental Movie 2: Functional two-photon movie of a field of CA1 neurons in a mouse running back and forth along a virtual linear track. The two-photon time-series was acquired at ∼15fps and the movie is displayed at 30fps. The virtual linear track is shown at the bottom of the movie and a “ˆ” indicates the position of the mouse along the linear track. This movie corresponds to the data shown in Fig. 2 (the ∼2 minute time period shown in Fig. 2b). The two-photon movie was made by coloring all of the pixels within a neuron's ROI a red intensity proportional to the value of the significant transient only trace corresponding to each frame (the red color was saturated at 35% changes). This red intensity was added to the still grey-scale image.