Thread by @jmacshine

Everywhere you look in systems neuroscience, you see evidence for low-dimensional manifolds -- but how do these distributed patterns of neural activity relate to neurobiology?

In a recent pre-print, @DrBMunn @eli_j_muller @GabWainstein and I tested the hypothesis that different arms of the ascending arousal system should differentially change attractor landscape dynamics -- i.e., the way that the low-dimensional brain state changes over time.

What is an attractor landscape? The idea is that the low-dimensional, distributed activity of the nervous system can be conceptualised as a ball rolling down a rugged, hilly landscape. You can read more about the concept here: macshine.github.io/publications/2… if you're interested.

We were interested in whether the noradrenergic locus coeruleus (LC) and the cholinergic basal nucleus of Meynert (BNM) caused different changes to the way that the attractor landscape changes over time.

In previous work, I've argued that these systems play a crucial role in balancing integration and segregation in the brain macshine.github.io/publications/2…. Mapping these ideas onto the attractor landscape framework, BNM should deepen attractor valleys and LC should flatten the landscape.

This is analogous to an idea from chemistry, in which catalysts can change the Activation Energy (Ea) needed to allow different chemicals to react together. LC should decrease the Ea (i.e., make reactions easier) and BNM should increase it (i.e., make it harder to react)

We had lots of theoretical ideas about how this would work, but we needed a way to test it. Unfortunately, most of recordings of the LC and BNM involve local interrogations of the nervous system, but we needed access to data from the whole brain in order to map out the manifolds

To solve this problem, we turned to a high-resolution (7T) resting state fMRI dataset -- @DrBMunn extracted BOLD time series from the LC and BNM and identified peaks in their relative activity. We then tracked brain activity around those peaks to test our different hypotheses.

Confirming the predictions of previous theoretical work (macshine.github.io/publications/2…), we found that peaks in LC preceded cross-network integration, whereas peaks in BNM preceded an increase in network segregation/modularity.

But we wanted to go a little deeper and track attractor landscape dynamics around the LC/BNM peaks. Again, @DrBMunn used his physics super-powers to quantify the shape of the attractor landscape.

He tracked the similarity of contiguous brain states using Mean Squared Displacement, and related this to the probability that each MSD was seen across the whole scan -- the inverse of the state probability gives the state energy: high state probability = low energy (and v.v.)

As we predicted, we found that, relative to baseline time-points within the session, LC peaks caused the landscape to flatten (i.e., navigated the brain to low-energy states with increased MSD) and BNM peaks did the opposite.

We were still left with a problem -- did these patterns relate to particular conscious states, or were they epiphenomena of the poorly-constrained resting state???

To solve this, we turned to an awesome open dataset from @NeuroWendy, in which they scanned a small cohort of experienced meditators while they practices breath-awareness mediation. They were asked to push a button whenever they noticed that they had lost their focus.

Our prediction was that we should see an increase in LC activity/landscape flattening before the button press (when they noticed that they weren't on task), and an increase in BNM activity/landscape deepening after the button press (when they re-focussed on their breath)

We found evidence for our hypothesis, suggesting that the neuromodulatory-mediated changes in the distributed, low-dimensional activity patterns in the brain is related (at least in part) to our conscious awareness of changes in our brain state.

This was a really fun project to work on, and nicely connected our different backgrounds. The preprint is here if you'd like to read it: biorxiv.org/content/10.110…. We'd love to hear any thoughts/feedback.