Place cell activity in a 4-room environment

--PART II--

How do place cells represent complex, multicompartmented spaces?

Preprint (same as for place cells and connectivity): https://www.biorxiv.org/content/10.1101/2020.10.20.346130v1.article-metrics

Thread:
1/

In simple spaces, place cells are (reproducibly) active in 1 or a few specific locations (='place fields'). They are thought to support spatial memory.
1 cell with sound:
https://figshare.com/articles/media/Grieves_et_al_2016_Place_cell_recording_with_sound/9911555
10 cells:


How do they represent complex spaces?

2/

In 1998, place cells were recorded in rats foraging in 2 boxes connected in parallel: same orientation, same visual cues. Contrary to two disconnected environments, the 2 maps were more similar than chance (with some variability):
https://www.jneurosci.org/content/18/20/8455
3/

Interestingly, if two identical compartments are connected vertically instead of horizontally, place fields do not repeat any more:

https://journals.physiology.org/doi/full/10.1152/jn.00132.2005
4/

Extending this to 4 parallel rooms, @hugospiers, @drkjjeffery and others found in 2015 that the maps were *very* similar (still, with some variability), making it impossible to decode the current compartment from place cell firing:

https://academic.oup.com/cercor/article/25/1/10/363850
5/

Are rats lost if their place cells (PCs) repeat? YES shows @rmgrieves in a parallel maze (2016).

This was not the case in a 'radial' maze, in which there was also almost no place field repetition!

(Note that behaviour & PCs were tested separately)

https://onlinelibrary.wiley.com/doi/full/10.1002/hipo.22496
6/

The results are explained by the Boundary Vector Cell (BVC) model ( @tom_hartley @caswellcaswell &others) stating that PCs receive inputs from BVCs (in subiculum) reacting to distance to walls of a specific orientation.

@rmgrieves modelled it (2018):
https://doi.org/10.1080/13875868.2018.1437621

7/

So far: in complex (4 rooms) geometrically-overlapping environments, place fields tend to repeat and rats get lost. This happens even if a visual, distal cue is present (Grieves et al 2015) and does not change with experience.

Is geometry really so important for place cells?

8/

IN OUR EXPERIMENT - finally, I know - we used geometrically-identical boxes, but with different entry points into each box.

Perhaps place fields would mainly repeat (local coding), or mainly not repeat (global coding):

(note, this is our definition, definitions can vary)

9/

We found that they mainly express a global code!

We saw all numbers and configurations of place fields and correlations between boxes were low: see how our correlation curves compare to previous experiments above, Tweets 5&6.

(also, this did not change with connectivity)

10/

However, data were still more correlated than chance! We generated artificial correlations from 1 (no), 2, 3 or 4 repeating fields & found that data was closest to two; looking closer, we find a small population of 2,3,4,+ repeating fields: thus, few PCs fully repeating.

11/

Could we decode box identity from place cell firing, unlike in Spiers et al 2015?

YES! says Adam Harris from @summerfieldlab showing we can even successfully decode *quadrants* of boxes from PCs (sessions with >15PCs):

(also, this did not change with connectivity either)

12/

Such encoding of global location matches rats' behaviour in the task: they knew which doors were open or locked, could locate a goal and plan a path to it better than chance (significantly better in some conditions):

13/

Conclusion:
- cells with repeating & non repeating fields can coexist, as seen before
- despite the repeating geometry of the 4 rooms, we find a mainly global encoding of space, which is (to our knowledge) new.
- coexistence of the 2 codes probably useful for brain?

14/

Why do we see this? Many possibilities:
- prominent distal cues (more than in past exps)
- presence of spatial task (Grieves 2016 had a task, not in recorded rats though)
- local cues (rats differentiating doors from dummy doors, using bells, etc.)
-...

15/

*My* preferred explanation is that the configuration of entry points helps differentiating compartments; if boxes were connected in parallel I would predict repetition in our conditions.

Relates to recent work of @atkeinath & @markpaulbrandon
16/

This was not the main goal of our study, but we hope it will contribute to the now vast literature on place cells in complex environments and place field repetition!

If you're interested, check our review on place field repetition:
https://journals.physiology.org/doi/full/10.1152/jn.00933.2016

17/

I did not mention here:
- CA3
- grid cells (see Carpenter et al with @caswellcaswell)
- HD cells (important to avoid place field repetition)
- the rest of the brain (sorry)
- many other interesting studies (also sorry)
Feel free to add relevant references to this tweet!

18/

Finally, this study was funded by @HumanBrainProj: many thanks to them!

(It would have been even better if they had continued to fund us during the pandemic for the analysis and write-up but that's the research system for you: often too short, non-extendable funding...)

19/

Thanks for reading and do your thing: Questions, comments, criticisms, etc.

Final poll: do you find this more or less interesting than our connectivity results?

End/

and for Part 1 of the results, see this other thread (same preprint): https://twitter.com/ElDuvelle/status/1318856844887154688?s=20