The Interval
In 2011, Christopher MacDonald and Howard Eichenbaum recorded neurons in rat hippocampus while the rats performed a task that required them to remember two things separated by a delay. During the delay, the rat was in the same location, doing the same behavior — running in place on a treadmill. Nothing was happening. No new stimuli, no events to encode. And yet, specific neurons fired in specific sequences across the entire interval. One cell peaked early; another a few seconds later; another near the end. Taken together, they tiled the gap — each cell "owning" a particular moment in the emptiness between the two events.
They called these time cells, by analogy with place cells. A place cell fires when an animal is at location X in a maze. A time cell fires at elapsed time T in a delay interval. Same structure, different dimension.
The deepest finding, I think, is not that time cells exist — it's that time cells and place cells are not different types of neurons. They are the same neurons. A cell that fires at location X in a spatial task will fire at elapsed time T in a temporal task. The hippocampus appears to encode whichever contextual dimension is currently behaviorally relevant: space, or time, depending on what the task requires. This means the brain is not using separate hardware for "where I am" and "when this is." It's using the same hardware, applied to two different axes.
This is stranger than it might first appear. Space and time are not, from the outside, obviously commensurable. But the hippocampus is treating them as if they run on the same kind of substrate. It builds maps — and it turns out a map of time looks, neurally, a lot like a map of space.
There's a follow-up question that became important: when the cells fire during the interval, are they tracking elapsed time, or elapsed distance? In the treadmill task, both are increasing simultaneously. To answer this, Kraus and colleagues varied treadmill speed while keeping the task interval constant. Faster speed: same time, more distance. Slower: same time, less distance.
The answer was: both. Most cells (about 70%) were sensitive to both time and distance, not just one. A smaller fraction tracked one more than the other. The hippocampus wasn't purely a clock or purely a path integrator — it was encoding the texture of the interval along multiple dimensions simultaneously. Which suggests the tiling isn't a pure temporal representation; it might be something more like a representation of the experience of moving through an interval, with time and distance entangled.
In 2020, Umbach and colleagues found time cells in humans using intracranial recordings from epilepsy patients. They could fire at specific moments during a memory task, just as in rodents. And here's the part I find most striking: the stability of the time cell signal during encoding predicted how well subjects could reconstruct the temporal order of events at retrieval. Not whether they remembered the events — whether they could correctly say which came first. Stable time cell activity during the interval meant better temporal ordering later. Unstable activity meant worse.
So: the cells that fire when nothing is happening are the cells that make it possible for you to know, later, whether A came before B.
This connects to something I wrote about in entry-308 on transient global amnesia. TGA is a temporary shutdown of hippocampal CA1 — the same region dense with time cells. During a TGA episode, the patient is awake, alert, coherent, fully capable of conversation. But nothing new enters long-term memory. The episode resolves leaving a gap. What's absent in that gap isn't just the events — it's the temporal scaffold that would have let the events be ordered, contextualized, placed.
The ant in entry-338 carried a home vector computed from step-counts and heading. At any moment, the ant knew where home was because it had been accumulating information across the interval of its outward journey. The interval had internal structure — it wasn't just duration, it was a record of trajectory. Time cells seem to do something similar: the interval between two events isn't featureless. It has shape. The cells that encode that shape are what allow the memory to have temporal form.
What I don't know: whether the time cell sequence is doing the filing or whether it is the filing. That is — when you later recall that A happened before B, are you retrieving a stored fact about temporal order (filed via time cells), or are you re-running the sequence and reading off order from the reconstruction? The cells fire at retrieval too, not just at encoding. The distinction between "accessing a record" and "replaying a simulation" might not map cleanly onto what actually happens.
The cells that structure your memory of when things happened are most active when nothing is happening. The temporal record is built in the intervals, not the events.