Neuronal signature of spatial decision-making during navigation by freely moving rats by using calcium imaging

Significance This manuscript addresses a longstanding problem about how place cell firing contributes to navigation in spatial environments, and data derived from on-line calcium imaging offer a new solution. These data support the view that prior to entering an arena where an animal will navigate to only one of several possible places on each day, a population of spatially responsive cells, including place cells, fire during the last 5 s. This population is shown to be predictive of the destination and trajectory that the animal is about to take.


Supplementary Discussion
Surgical considerations. While CA3-CA1 connections form the intrahippocampal Schaffer collaterals, extrahippocapal connections reach the hippocampus via three major routes: the fimbria/fornix, the angular bundle/perforant path and the temporoammonic alvear pathway. The fimbria/fornix contains the main bundle of afferent and efferent connections to and from subcortical regions. Most connections from the entorhinal cortex (EC) reach the hippocampus either via the performant path (pp) or the alvear path (ap) (1). The angular bundle is located between the EC and the presubiculum and parasubiculum. It is the main route taken by axons originating in the EC as they travel to the septotemporal levels of the other hippocampal fields, particularly the dentate gyrus, hippocampus and subiculum. These projections enter the hippocampal formation by perforate the subiculum (pp), proceeding intrahippocampally to innervate the dentate gyrus (DG) and CA3 and CA1 neurons (2). In CA1 pyramidal neurons, EC projections are disproportionally located in the stratum lacunosum moleculare adjacent to the DG (3). In our aspiration procedure, the Shaffer collateral, the fimbria and perforant path are unaffected, so that the main afferent paths to the CA1 region remain intact.
However, at more septal levels, the number of entorhinal fibers that take the alvear pathway is higher (4). In particular, in the septal portion of the hippocampal most of the EC fibers reach CA1 via the alveus by travelling parallel to the alveus then perforating the pyramidal cell layer (5). Because these fibers are medial to the location of our GRIN lens implant, the majority of the EC inputs to the hippocampus should be spared (see Figure SD1 below redrawn from Deller et al. 1996, showing EC anterograde tracts).
It has been reported that place cells activity and ripple generations are relatively robust to even large disruptions of EC inputs such as MEC lesions (6,7), which would suggest that the majority of sensory inputs to the hippocampus would be spared by the surgical procedure to image the rat CA1 and the placement of the small 0.9 mm GRIN lens. Nevertheless, in experiments where EC inputs are critical (8) this aspect should be taken into consideration. Indeed, the procedure described by (9) does report that cell identification is possible without aspirating the alvear fibers, albeit maybe at the expense of a lower effective resolution due to light scattering through myelinated fibers. Nonetheless, we and the authors of the other study (9) will have concerns about damage to stratum oriens but our histology shows minimal damage. A side-by-side comparison of the two technical approaches (aspirating vs keeping alvear fibers) would probably be useful.
The aspiration of callosal fibers is, however, necessary for a successful imaging, as also in (9). The corpus callosum contains commissural fibers connecting, among other regions, the hippocampal formation between the two hemispheres. The position of the GRIN lens implant is expected to spare the dorsal hippocampal commissure connecting the dorsal hippocampi, but it cannot be excluded that the implant can affect interhippocampal communication. Although poorly studied with respect to this aspect, the rat hippocampus seem to have little lateralization; moreover, interhemispheric communication does not seem to severely impact the performance in mnemonic tasks (10,11), but see (12).
Behaviour setup considerations. The everyday arena allows to model episodic-like memories in rodents (13,14). A key facet of the task is that the position of the reward changes from day to day in an unpredictable way.
However, the global spatial context remains unchanged. Thus, on any day, there is nothing to learn about the spatial context and we therefore had every reason to expect place cell fields to remain stable. The animal does, however, have to update its representation of the most recent spatial location and not use a consolidated spatial memory.
In our experiment, we used a modified version of the everyday arena task to accommodate the requirements of calcium imaging, mainly to maintain 3 possible sandwells. Although the sandwells are not at equal distance from the animals, this was deliberate to avoid symmetry and we see no reason why this might bias the animal's preference or its patterns of cell firing. It is now common practice in the everyday arena task to use multiple sandwells configurations in the same experiments (14).
In replay experiments using electrophysiology, a plethora of maze configurations have been used. Linear tracks and variants thereof (such as Z-tracks) are by far the most commonly used experimental setups (15,16). Two-choice or multiple-choice W mazes have also been employed, where the distance of the possible end goals is typically equivalent from the animal's starting point (17)(18)(19). These configurations can be easily modelled and linearized in the analysis (19) although at the expense of limiting the choices of the animals in solving the task.
As discussed in the main text, this may also bias how much animals focus on the different locations, which can be reflected in the neural reactivation. Other experiments have privileged a less structured behavioural setup, which allows the animals greater flexibility during the decision making, and allows for a more naturalistic behaviour. In such cases, two-dimensional environments are typically chosen, with either a gridlike or a random (or quasi-random) distribution of the goals (20)(21)(22)(23). These reflect the random positioning of goals with respect to the animals' starting points, as is the case in the most widely used behavioural setup to study navigation and spatial memory such as the Barnes maze, the watermaze and the Cheeseboard maze (20,24). In our experiment, we chose the latter approach, and indeed our sandwell configuration is very close to the one used in (20). Future work will address if, and to what extent, the configuration of the arena or other experimental features affect the representation of space and the prospective coding during decision making.       their average event rate and their signal-to-noise ratio. Maps correspond to S18 and are essential equivalent for Sessions 19-21. S18 S18 S19 S20 S21 S19 S20 S21 S18 S1 S2 S1 S2 S9 S10 S18 S19 S19 S20 S21 S9 S10 S18 S19       T2  T1  Trial   T3 T4 T5 T6  T2  T1  Trial   T3 T4 T5 T6   T2   T1