Read on Twitter
Now that @danieljamesyon's smashing commentary has come out I think it's a good time to (temporarily) break my Twitter ban to make a thread about our latest paper in @CurrentBiology with @jamesbisby, @camillalnord , @NeilBurgess10 & @profgeraintrees 
https://www.sciencedirect.com/science/article/pii/S0960982220315864
https://www.sciencedirect.com/science/article/pii/S0960982220315864
Seminal theoretical models suggest that Noradrenaline plays an important role in how organisms learn when the world is uncertain (c.f. Yu & Dayan - and the gif below), but direct evidence for this is somewhat lacking in humans.
We assessed the effects of Propranolol (a beta-adrenergic antagonist) on the computational & physiological mechanisms of learning under volatile conditions. Here we are dealing with sensory learning (what am I seeing?) as opposed to reward maximisation (will I win 50p?).
The task manipulates expectations and sensory noise. Both noise and unexpectedness slow down reaction times, but Propranolol augments the interaction between these factors (more reliance on expectations when noise is high).
We fit a hierarchical Bayesian learning model to this behavioural data to examine learning directly. Propranolol slowed down probabilistic learning, meaning ppts are less likely to update in the face of new information (relying more on prior beliefs).
Prior work has shown that both prediction errors and volatility are encoded in pupil size, thought to reflect the activity of the LC-NA system. By blocking NA we extinguished the phasic effects of both, without changing the canonical pupil response - a nice validation.
Finally we measured blood pressure to confirm the autonomic effects of Propranolol. Baseline systolic bp predicted estimated volatility during the task in all ppts. Note, higher estimates of volatility can be though of as adaptive (since this is a volatile environment).
It will be interesting to see how this relationship uncouples when 1. volatility is being misestimated and 2. when bp is not in a healthy range.
This work has implications for prior studies - for example, in this paper ( https://www.nature.com/articles/nn.4615) our measures might be sensitive to anxiety (50% of autistic adults have a diagnosed anxiety disorder). We hope to understand this relationship better in our @cam_raa study.
That's all for now folks. Thanks to @wellcometrust for funding the research, and @Autistica and the Lister Institute for funding me/the lab. Now back to blocking Twitter for the good of my wellbeing :)
