I often come across papers from other fields (and even within my own field) and think to myself “yeah, but why’s this important?” It’s not that I don’t believe that it’s important, it’s just that I don’t know enough to get why. With that in mind, I thought I’d contextualize my new paper for the lay reader in a few paragraphs. For the tl;dr version: scroll down to the parts IN BOLD AND ALL CAPITALS WHERE IT LOOKS LIKE I’M SHOUTING BUT I’M REALLY NOT where I try to highlight the main findings.
The title is “Context-dependent incremental timing cells in the primate hippocampus.” In normal-speak: I find monkey neurons that seem to help us (yes us—presumably we’re just fancy monkeys) keep track of time in certain contexts by changing their firing rates incrementally. To be fair: I don’t show anything proving how these neurons track time. My former colleague (and paper coauthor) Yuji Naya did that with some amazing previous work. In short, Yuji recorded from multiple brain regions in a task where monkeys had to keep a couple of pictures in order. He found that neurons would incrementally time (by slowly rising or falling in firing rate) during a delay in between when these pictures were shown where the monkey had to remember one picture in preparation for learning the second picture. The key was: this incremental signal was strongest in the hippocampus, and the farther you looked from the hippocampus into other nearby (more “vision-related”) brain regions, the worse the signal got. This plus a bunch of other work points to the hippocampus as important for how our brain keeps track of time.
So, my first question was: do neurons do this in all tasks or just when the monkey has to keep pictures in order? Therefore I went digging through some data taken by Sylvia Wirth where monkeys were tasked to memorize the associations between objects in certain places for reward (think animal pictures in different locations on a computer screen). When I started looking at this data, I saw cells rising or falling in firing rate during a similar delay period as Yuji’s (we look during delay periods because this is when the monkey is keeping the information in memory for the animal to respond later). There isn’t much for standard ways to statistically analyze firing rates that change over time like this, so I used resampling methods (aka bootstrapping aka Monte Carlo) to find groups of these cells (See Fig. S4 for details; here’s another paper dealing with much the same problems I did that might have been useful for my analysis if it’d come out earlier). I also spent a lot of time trying to make sure these changing signals weren’t just signaling anticipation like in other brain regions (e.g. amygdala, parietal cortex). This stuff is all in the paper and kinda boring so look there for details.
Anyway, the key point is: I found a bunch of these cells rising and falling in firing rate while the monkey remembered objects in certain places. AND MANY OF THESE NEURONS DID THIS TIME TRACKING ONLY WHEN CERTAIN OBJECTS WERE IN CERTAIN PLACES. This is where the “contextual” part comes in, as these single neurons seem to be uniting temporal information with contextual information.
Now, at this point you might be like: “what’s the big deal, man/mate/eh? You just found a bunch of moving firing rates.” What I think is really cool about this is when we looked at the animal’s behavior. This task was pretty darn hard for a monkey, so the animal only learned a new object in a new place in 71% of daily sessions. But when I found these time-context cells, the monkey had a 93% change of learning during that session. Therefore, THESE SPECIAL TIME-CONTEXT NEURONS ALMOST EXCLUSIVELY SHOWED UP WHEN THE MONKEY LEARNED NEW CONTEXTS. And since contextual encoding is something known to require the hippocampus as part of our episodic memories, it makes some sense that this part of the brain is keeping track of context. What is novel about this is SINGLE NEURONS SEEM TO BE KEEPING TRACK OF TIME AND CONTEXT SIMULTANEOUSLY.
I also looked at what happened when the monkey got the answer right or wrong. What I found was the timing signal dissipated when the monkey was about to answer incorrectly. And since this incremental timing signal occurs during a delay period where the monkey is keeping the object-place combination in its mind, this could show that THESE TIME-CONTEXT NEURONS MIGHT STRENGTHEN CORRECT MEMORIES BY INCREASING THEIR INCREMENTAL SIGNAL.
For someone outside of neuroscience (or even someone insi…in neuroscience), this might not be so surprising (“Yeah so neurons keep track of things: duh.”). But what these results imply is how complicated the neural code might be. Let me give you an example.
There are multiple studies that use fMRI with humans while they do things like learn new objects in new places (here’s a new one). They’ll look at say 1000 voxels (3-dimensional pixels—each usually about a cubic millimeter with thousands of neurons inside) in the person’s brain and find that maybe 400 go up and 400 go down in activity during the period the object is shown (usually summing activity for as many seconds as the image is shown). Then when the person recalls the same object, maybe 425 neurons will go up and 375 of them will go down. And using some advanced stats methods they can “predict” which object the person is remembering by showing that this similar group of neurons go up or down (this is called training a classifier). People do this kind of thing all the time for individual neurons instead of voxels: measure dozens or hundreds of them and ask if they go up or down and run some linear regression to correlate how the neuronal network changes during a given task.
The problem is: they’re assuming that all neurons do is go “up” or “down” during a single timeframe. While the neurons I show go up or down, if these studies used the neurons I found they’d be throwing away information, since my neurons actually change over time (instead of just peaking and dissipating quickly like we often show in other neurons):
Each color represents a different combination of objects+places. The neuron is tuned to the red object-place combination in this case as the firing rate (FR) increases during the delay period.
Therefore, NEURONS CAN CODE FOR SIGNALS IN TWO DIMENSIONS. Studies that only look at firing rates in a given timeframe are only working in one dimension: firing activity. But if you look at how this neuronal activity changes over time that represents a second dimension of signaling. This doesn’t mean that the studies using only one dimension don’t work: it’s very possible that with enough signal from many voxels/neurons you can accomplish perfectly good decoding for whatever your goal is (like interfacing with a machine). But it’s also possible that until we understand the real code we’ll always be playing a game of “you’re getting warmer” with each and every neuron without ever getting the right answer because we’re not directly measuring how the brain actually works. This is the goal of “basic” (I realize it’s a bad term) research: to understand the fundamental principles of the brain so that in the future we know enough to improve human lives as technology develops.
One great example of this was a recent work where a tetraplegic woman fed herself for the first time using her thoughts. I can’t find the link at the moment but the authors noted that there were about a dozen essential animal studies (starting in the 60s) that led up to this achievement, each one building off the last until we understood enough about the system to tap into how the brain controls movement. Of course, we didn’t know these dozen were the seminal ones until hundreds were done. So, in conclusion, the hope is that work like mine will help inform future neuroscientists/neuroengineers on key principles about how our brain works. But at the moment it’s just an intriguing way that the brain seems to use to encode information.