Hippocampus Neurons Cycle Possible Futures

This paper Constant Sub-second Cycling between Representations of Possible Futures in the Hippocampus looks to be important in understanding the role of Theta cycles.


One factoid to consider when evaluating this paper - how does this fit with patient HM?

From what I have rad, other than not being able to form new memories and some retrograde amnesia he tested fairly normal on many cognitive tests. He was able to select alternatives based on memory without a hippocampus.

I would be very cautious with suggestion that the hippocampus has a role beyond memory formation.

1 Like

I found a bioRxiv preprint. The information was not easy to piece together but I understand a little.

The mentioned “Possible Futures” part most sounds like two or more alternated choices being morphed/summed to a direction vector field map of where to go. When hungry the direction of food will be signaling at each alternating cycle, causing propagation of traveling waves that direct turns towards. When given two equal choices there can be a quick change in mind or sometimes amusing indecision typical of animals.

In at least an allocentric representation I use: its own position, speed and direction is always the current but that also has to compensate for signal propagation time and is a future prediction based on sensory input state. Everything else in the representation (apparently at theta rate?) alternates between predicted current, and a greater amount in the future for adding awareness of when at a bad place to be at a given time then make it get out of the way. What does the pushing away from danger is the wave direction increasingly pointing the opposite way when nearby hazardous places stop propagating signal. There is no outwardly sent signal to stay away, doing nothing makes an impassible barrier that traveling waves cannot travel through, so of course local field directs it around or away from, instead of straight through that place.

For better speed performance traveling waves can be propagated at a twice or so faster rate than the possible future states are being alternated, but not so fast the frames don’t blend into to a steady average field direction.

For a more resolved and error free map several maps at different scales (like we have) need to be summed together. This at the same time generates a unique representation for anywhere as Matt explained in HTM School, though in this case the same idea would be filling voids where there would otherwise be ambiguous direction readings in a vector map.

The paper suggests to me that how often “Possible Futures” are alternated in proportion to (beta?) propagation speed of traveling waves should be made a constant named something like “ThetaRate” or whatever else works for everyone here.

You made me have to think about that. This was helpful:

What patient HM most seems to have lost is memory of physically moving around in the environment, which would seem to be expected to make it difficult to recall what they earlier did. After the operation they could not recall or map out streets around them, only recalled earlier left/right directions to get around the town where they grew up.

What I call “motor memory” was still intact. They could with it navigate to what they can see, or procedurally remember how to find. But without the primary “navigation network” part to map out events as motion through the environment all ability to recall past experience is gone, their amnesia.

Where given a choice between two directions and they were say hungry they can remember that the kitchen is to the left of an opening in one of the walls, but probably not sense upcoming motion through a narrow hall behind it into room with something to navigate around to reach what they want from the bottom left corner behind old jar of pickles in the refrigerator.

It’s as though an addressing mechanism to form long term memories was removed from an otherwise still functional memory system. All the movement rich episodic sensory information like gets sent to address 000000000000000000000000 or 111111111111111111111111 in RAM instead of saved as a time series to later compress for archiving elsewhere in the brain, while sleeping.

I am not sure that I accept your key/access proposal. In this video this patient does not seem to have any trouble with maps and routes in his older memories.
I have read much of the HM studies and he also had ready access to older memories.

I see the EC/HC as a buffer with very good oneshot learning for both what comes from the cortex and the amygdala. This memory is transferred back to the cortex in sleep.

Yes, I agree. He could recall previously stored memories, problem was forming new ones. Be a big help to know exactly why. I think it’s like you said, nothing could be transferred back to the cortex in sleep. Even when awake their new memories only lasted for 15 minutes.

HM and related patient also had Hebbian learning intact.
HM could learn skills like mirror writing even though he could not remember learning the task.
This point to the one-shot nature of the EC/HC being an important part of why it is special.
I see the EC/HC picking up episodic memories and then hammering them back into the regular cortex until it sticks during spindle wave activation.

Mirror writing would be a eye-hand coordination task I expect his intact “motor memory” would have little or no problem with.

I see the EC/HC picking up episodic memories and then hammering them back into the regular cortex until it sticks during spindle wave activation.

That makes sense. I was speculating his episodic memories were no longer being pieced together then sent to the regular cortex. Even when fully awake there was no memory of what they did minutes ago. Do you have a summary?

This is the essence of episodic memory - the sequence of perceptions in a cluster. I don’t know what are the start and ends of an episode but I suspect it has to do with attention. This one bit (episodic memory) seems to be the main purpose of the EC/HC complex.

All the bits about motion planning seem to be a red herring. Looking at what people do without the EC/HC should show that these functions are handled outside this area. If you watch videos of what functions are preserved in these people the only real deficit is in the episodic memory area. spacial perception and manipulation is fully preserved.

The Dynamic Grouping of Hippocampal Neural Activity During Cognitive Control of Two Spatial Frames paper I modeled the 2D “navigation network” from mentions this that might be helpful for another clue:


Cognitive control is the ability to coordinate multiple streams of information to prevent confusion and select appropriate behavioral responses, especially when presented with competing alternatives. Despite its theoretical and clinical significance, the neural mechanisms of cognitive control are poorly understood. Using a two-frame place avoidance task and partial hippocampal inactivation, we confirmed that intact hippocampal function is necessary for coordinating two streams of spatial information. Rats were placed on a continuously rotating arena and trained to organize their behavior according to two concurrently relevant spatial frames: one stationary, the other rotating. We then studied how information about locations in these two spatial frames is organized in the action potential discharge of ensembles of hippocampal cells. Both streams of information were represented in neuronal discharge—place cell activity was organized according to both spatial frames, but almost all cells preferentially represented locations in one of the two spatial frames. At any given time, most coactive cells tended to represent locations in the same spatial frame, reducing the risk of interference between the two information streams. An ensemble’s preference to represent locations in one or the other spatial frame alternated within a session, but at each moment, location in the more behaviorally relevant spatial frame was more likely to be represented. This discharge organized into transient groups of coactive neurons that fired together within 25 ms to represent locations in the same spatial frame. These findings show that dynamic grouping, the transient coactivation of neural subpopulations that represent the same stream of information, can coordinate representations of concurrent information streams and avoid confusion, demonstrating neural-ensemble correlates of cognitive control in hippocampus.

This seems to be indicating that the various information streams cannot be assembled into a representation of himself interacting with the world. That alone may be enough to explain the amnesia, but I’m only taking a best guess on that one.