Abstract
Representational states of our world are woven together into a tapestry of connected statescalled the cognitive map (CM) — a model of our external world (Tolman, 1948). Historically,
investigations have focused on CM states representing the current, immediate features in the
animal’s environment4. But in the past few decades, work has endeavored to extend this
understanding of the map in two directions—(1) how does the CM represent previous and
upcoming states needed to both contextualize present states in the past and to flexibly
plan upcoming states. This is known to involve both sequences of CM cells stretching
ahead/behind the animal (Skaggs & McNaughton, 1996) and recently, cells that incorporate
the past or future in their receptive field (Ferbinteanu & Shapiro, 2003; Frank, Brown, &
Wilson, 2000; Howard et al., 2014; Sarel et al., 2017; Wood et al., 2000). (2) how does the
cognitive map emerge from multi-brain area communication and dialogue?
In this work, we attempt to study both questions above using two brain areas, the hip-pocampus and prefrontal cortex, key players in memory-guided navigation (Floresco, Sea-
mans, & Phillips, 1997; Maharjan et al., 2018). We looked for cells that represent upcoming
states (goals and spaces; Chapters 2 & 3) or the past (Chapter 3). And we attempted to
study how these areas might cooperate through communication subspaces to empower the
emergent cognitive map (Chapter 4).
The first study (Chapter 2) focuses on representations of future, namely goal states(Ormond & O’Keefe, 2022; Sarel et al., 2017). It examines hippocampal and prefrontal
4i.e. percepts as transformed information streaming from the sensorium—the spinal cord and cranial
nerves; additionally, transformed features may be inclusive of the animal’s action/motorium states.
neuronal populations during a spatial memory task, decoding goal information and locationswith a convolutional neural network. This analysis highlights robust goal decoding facilitated
by CA1 and PFC activity, with a nuanced observation that place encoding predominates over
goal coding in these regions.
The second study (Chapter 3) delves into spatiotemporal coding in CA1-PFC place cells(Dotson & Yartsev, 2021), revealing some place cells preferably represent past and future
space in rats. Critically we find modulation of non-local temporal fields between cue-guided
and memory-guided conditions, underscoring a possible influence of task demands on tem-
poral coding.
We lastly examine communication. Chapter 4 investigates how hippocampal-prefrontalcommunication subspaces (CS) might support memory-guided behavior and to what extent
prior communication-driven network patterns influence CS (Semedo et al., 2019). It demon-
strates that aligned subspace activity is a strong predictor of task outcomes, contrasting with
the ineffectiveness of orthogonal activity. Further, the study discerns how different rhythms,
notably theta and ripples, uniquely influence subspace dimensionality and performance. The
alignment of communication with theta rhythm amplitude exhibits a progressive adaptation
over learning epochs, illuminating complex dynamics in cross-regional interactions during
spatial memory tasks.
Collectively, these studies help form a picture of the dynamics of retro-prospective infor-mation, how they differ with task demands, and how rhythms shape communication-related
spiking subserving this CM system and the downstream behavior.