Title: Closed-loop gain control of visual landmarks reveals that pre/parasubiculum neurons are more closely bound to visual input than medial entorhinal cortex neurons

Animals combine environmental landmarks and self-motion inputs to navigate their surroundings. The process of using self-motion inputs to derive an estimate of position is called path integration (PI) . PI is a noisy computation that accumulates errors, which can be corrected by landmarks. Recent studies in CA1 place cells have shown that landmarks can act as a teaching signal to recalibrate the gain of the PI system when the system is confronted with sustained conflicts with landmarks (Jayakumar, Madhav et al., 2019). The medial entorhinal cortex (MEC) is believed to be the locus of the PI computation in the medial temporal lobe (McNaughton et al., 2006). However, it is unknown whether and where landmark information exists in the medial temporal lobe to correct for PI errors. We performed simultaneous tetrode recordings from MEC and neighboring medial temporal lobe regions, presubiculum (PrS) and parasubiculum (PaS), in Long-Evans rats (n=4) under conditions of persistent conflict between landmarks and self-motion inputs. This conflict was produced by gradually rotating an array of landmarks in a virtual reality environment as a function of the rat’s speed, producing an illusion that the rat was moving slower or faster than its actual speed. In a subset of sessions where a large conflict was introduced, a striking dissociation between MEC and PrS/PaS neurons was observed: the firing fields of all MEC neurons broke away from landmarks while the firing fields of all PrS/PaS neurons remained strongly tied to the landmarks. This strong control of landmarks over PrS/PaS neurons persisted even under extreme landmark manipulations, such as sudden jumps in the closed-loop gain controlling landmark movement. Furthermore, the firing rates of these PrS/PaS neurons were strongly correlated to the brightness of the landmarks, and consequently about ~30% of these cells stopped firing in the absence of landmarks. In open-field recordings, ~60% of these cells had single/multi-lobed head-direction tuning curves. These results show that the MEC neurons reflected a combination of landmarks and self-motion inputs , whereas PrS/PaS neuronal responses were solely driven by landmarks. It is known from prior work that PrS/PaS neurons provide extensive monosynaptic synaptic inputs to MEC (Canto et al., 2012), and lesions of dorsal PrS results in degradation of landmark control over head direction cells in the anterior dorsal thalamic nucleus (Goodridge et al., 1997). Taken together with these prior studies, our findings point to the possibility of PrS/PaS neurons carrying the landmark information to the MEC to enable the correction of PI errors.