Spatial navigation requires landmark coding from two perspectives, relying on viewpoint-invariant and self-referenced representations. The brain encodes information within each reference frame, but their interactions and functional dependency remains unclear. Here we investigate the relationship between neurons in rat retrosplenial cortex (RSC) and entorhinal cortex (MEC) that increase firing near boundaries of space. Border cells in RSC specifically encode walls, but not objects, and are sensitive to the animal’s direction to nearby borders. These egocentric representations are generated independent of visual or whisker sensation, but depend on inputs from MEC that contains allocentric spatial cells. Pharmaco- and optogenetic inhibition of MEC cells led to a disruption of border coding in RSC, but not vice versa, indicating allocentric-to-egocentric transformation. Finally, RSC border cells fire prospective to the animal’s next motion, unlike those in MEC, revealing the MEC-RSC pathway as an extended border coding circuit that implements coordinate transformation to guide navigation behavior.

For tetrode recordings, rats were unilaterally implanted with a hyperdrive that contained 28 individually adjustable tetrodes made from 17-μm polyimide-coated platinum-iridium (90-10%; California Fine Wire; plated with gold to impedances below 150 kΩ at 1 kHz). The tetrode bundle consisted of 30-gauge stainless steel cannulae, soldered together in a 14x2 rectangular shape for recordings of the entire RSC, 7x4 for anterior RSC. Tetrodes were implanted alongside the anteroposterior axis, starting at (AP) -2.5 mm posterior from bregma until -4 mm to -6.5 mm, (ML) 0.8 mm lateral from the midline, (DV) 1.0 mm below the dura, and at a 25° angle in a coronal plane pointing to the midline in order the get underneath the superior sagittal sinus. For animals with id #50, #97, #167 and #224, the electrode was implanted in the right hemisphere, while for animals #246-#294 it was implanted in the left hemisphere.

All data processing steps were performed in MatLab (MathWorks). Neural signals were acquired and amplified using two 64-channel RHD2164 headstages (Intan technologies), combined with an OpenEphys acquisition system, sampling data at 15 kHz. Neuronal spikes were detected by passing a digitally band-pass filtered LFP (0.6-6 kHz) through the 'Kilosort' algorithm to isolate individual spikes and assign them to separate clusters based on waveform properties (https://github.com/cortex-lab/KiloSort). Clusters were manually checked and adjusted in autocorrelograms and for waveform characteristics in principal component space to obtain well-isolated single units, discarding any multi-unit or noise clusters. Tetrodes were moved a minimum distance of 80 µm between recording days to find a new set of neurons for the next recording session.

We refer the reader to the associated paper for details on behavioural data acquisition, as well as a detailed account on how data vectors were generated. These data files include the following data variables for all individual neurons:

• Animal ID: animal identifier that corresponds with those mentioned in figure legends of the paper.
• Session ID: session identifier that corresponds to the recording day number.
• EMD scores: Earth Mover's Distance (EMD), the main metric for border cell classification.
• Global FR: Overall spiking rate in spikes/sec, calculated as the total number of spikes, divided by the duration of a recording session.
• Spatial rate map: Firing rate map in 2D space.
• Boundary rate maps: Firing rate map in egocentric border space, with bins corresponding to the angle (3rd dimension) and distance (4th dimension) of a boundary relative to the animal's position.
• Spatial correlations: Pearson's correlation between the spatial rate maps of the first and last regular recording session on a given day.
• Allocentric MVL: Mean Vector Length (MVL) values in allocentric head-direction space.
• Allocentric MVL stat: MVL statistic, corresponding to the percentile of the MVL relative to a 1000-fold time-shuffled null-distribution.
• Egocentric MVL: MVL values of angles in egocentric boundary space.
• Egocentric MVL stat: MVL statistic, corresponding to the percentile of the MVL relative to a 1000-fold time-shuffled null-distribution.