## Place fields analysis

The advent of cutting-edge in vivo electrophysiology and neuroimaging techniques poses strong limitations to both the mobility of the subjects, and the design of behavioral apparatuses and data acquisition systems. One solution to these shortcomings is the implementation of virtual reality (VR) systems, which succeed portions of sensory inputs from real environments all the while reducing the physical size of the behavioral apparatus. In closed-loop VR systems, mechanical inputs are coupled with visual, auditory, tactile and even olfactory feedbacks to deliver a virtual representation of the world. Programmed logics executed by a computer allows precise controls over the rules and designs of the environment. Consequently, VR systems do not act merely as workarounds to the imposed restraints, but decrease variability in the sensory inputs and expand the possibility of manipulations otherwise impossible in the real world. Studies focused on neural integration can benefit from distorting the relationships across sensory and motor signals.

For every deconvoled time series, we excluded epochs below a certain threshold of running speed. The threshold was defined as the valley between the two local maxima (one of which being 0 cm/s) in the absolute running speed distribution or the first local minima in case of multiple peaks (Takahashi et al. 2013).

The 100-cm virtual track was divided into 100 equally spaced positional bins. The mean activity was computed for each bin across all trials, and normalized by the occupancy (i.e. number of samples per bins). Neurons with an average positional firing rate of less than 0.03 dF/F s^{-1} were excluded from subsequent analysis.

Two criteria were used to identify place cells: One based on spatial information content and another based on spatial vector tuning specificity. The spatial information (SI) of each cell was computed as follow:

where for *n* location bins, *p _{i}* is the probability of occupancy in the

*p*-th bin, λ

_{i}*p*is the mean inferred firing rate in the

_{i}*p*-th bin and λ is the overall firing rate across all location bins. A distribution of

_{i}*SI*was subsequently generated by applying circular shifts of random intervals to the time series and by recalculating the

*SI*accordingly for a total of 1000 times. The

*p*-value for each neuron was reported as the percentage of

*SI*from the shifted distribution that exceed the true

*SI*of the neuron.