Francesca Fardo

New Open dataset alert:
🧠 Introducing "Spacetop" – a massive multimodal fMRI dataset that bridges naturalistic and experimental neuroscience!

N = 101 x 6 hours each = 606 functional iso-hours combining movies, pain, faces, theory-of-mind and other cognitive tasks!

🧵below

I'm deeply grateful to everyone who contributed! Especially first-author, Jesper Fischer Ehmsen, and co-last author @micahgallen.com, whose brilliance, dedication, and generosity brought our vision to life in ways far beyond what a single tweet could capture!

Huge thanks to our funders @erc.europa.eu and @lundbeckfonden.bsky.social for supporting this project. Thanks also to our research centre, CFIN, without whom we could not have collected this data. This work was the result of incredible teamwork between my lab and @the-ecg.bsky.social.

Key takeaways: (1) the brain interprets sensations based on expectations, making harmless inputs painful when uncertainty is high. (2) Individual computational fingerprints during thermosensory learning reflect distinct brain microstructural correlates.

Using quantitative MRI, we linked thermal learning parameters to brain microstructure. Notably, decision temperature correlated with iron content in the Subnucleus Reticularis Dorsalis (SRD), a key structure in descending pain modulation.

In contrast, second-order uncertainty, reflecting how unsure you are about the precision of your own predictions, actually intensified the illusion of pain. This means that when thermal expectations are uncertain, the brain interprets these ambiguous stimuli as more painful!

We compared multiple computational models and found a two-level Hierarchical Gaussian Filter best explained our data. First-level uncertainty shaped temperature sensations: stronger expectations of cold amplified cold feelings, while stronger warm predictions enhanced warmth.

We manipulated uncertainty in two ways: (1) continuously changing cue-stimulus associations via probabilistic reversals; (2) introducing ambiguous trials combining harmless cool & warm stimuli, eliciting the "thermal grill illusion" of pain!

To answer this question, over 250 participants learned associations between auditory cues and warm or cold stimuli, and rated their experiences of cold, warmth, and pain.

We know expectations shape how we experience temperature and pain. But what happens when we're unsure what's coming next? Our key question: can uncertainty about expectations lead the brain to mistakenly perceive harmless temperatures as painful?

Thermosensory predictive coding underpins an illusion of pain Computational modeling reveals how uncertainty transforms harmless stimuli into perceptions of pain.

Excited to share our latest publication, out now in @ScienceAdvances: “Thermosensory predictive coding underpins an illusion of pain.” www.science.org/doi/10.1126/.... Read the full thread for details!

The Thermal Grill Elicits Central Sensitization The thermal grill, in which innocuous warm and cool stimuli are interlaced, can produce a paradoxical burning pain sensation—the thermal grill illusion (TGI). While the mechanisms underlying TGI remai...

My brilliant student Matt Cormie became interested in the thermal grill, and designed a study to see if the TG can elicit primary/secondary hyperalgesia and/or brush allodynia. These could help us parse out the mechanisms involved.

www.biorxiv.org/content/10.1...

Cold hands, altered ownership? We link disturbed body ownership in stroke patients to reduced hand temperature and impaired thermoception. The right Insula & Parietal hubs are key players. We are excited to share our new findings
@naturecomms.bsky.social
▶️ rdcu.be/d5SDU

All data, code, and materials will soon be available here: github.com/Body-Pain-Pe... Thanks for reading, and we are excited to hear your thoughts on the study!

This was a monumental effort - the culmination of a decade of collaboration with @micahgallen.com, and our first shared last-author paper. It would not have been possible without the amazing work of @jesperfischer.bsky.social!

In summary: we learn about thermal contingencies using hierarchical Bayesian mechanisms, where the precision of our expectations guides both veridical pain sensation and illusions of pain. Individual differences in this precision weighting are linked to TGI response and brain microstructure!

We were intrigued to find that R2* in bilateral, basolateral amygdala was related to the uncertainty modulation of the thermal grill index. This region is important for affect and pain conditioning, and could help explain how we infer pain sensations from harmless inputs.

We find that individual variance in thermosensory computations (e.g., decision temperature and hierarchical uncertainty) index the microstructural features of the somatosensory cortex, insula, amygdala, and the brainstem Subnucleus Reticularis Dorsalis (SRD).

Next, leveraging our large sample, we conducted a whole brain quantitative MRI analysis using the MPM sequences developed collaborators at UCL, relating individual fingerprints of thermosensory computations to brain maps indexing cortical myelination and iron.

This means that the more you incorporate uncertainty into your illusory pain ratings, the more sensitive to the TGI you are in general! In the future, uncertainty-weighted TGI may provide a useful way to quantify disordered pain.

Clinically, responsiveness to the TGI is often used as an indicator of thermo-nociceptive function. We calculated an uncertainty modulation of thermal grill illusion index (UMTI), and found that it was highly correlated with individual differences in TGI response.

So cool and warm percepts are precision-weighted, but what about the illusion? Remarkably, we find that the intensity of the TGI is specifically increased by estimation uncertainty. When we are unable to make reliable thermal predictions, we infer greater pain.

Next, using our hierarchical ZOIB regression approach, we predicted VAS ratings using trial by trial variance in prediction and estimation uncertainty. We find that the precision (i.e., inverse uncertainty) of thermosensory predictions increases felt cold and warm sensations.

The model does an excellent job of explaining trial by trial learning, where the 1st level encodes thermal prediction uncertainty (e.g., probability of warm vs cold stimuli) and the 2nd encodes estimation uncertainty (e.g., uncertainty about cue-stimulus associations).

To develop a precision-weighted predictive coding model of thermosensation, we adapted the hierarchical gaussian filter (HGF) to our task. Bayesian model comparison, posterior predictive checks, and cross-validation all supported a 2-level learning model.

So how did TGI trials influence thermal learning? We hypothesized that TGI would reinforce associations depending on if they were perceived as predominately warm or cold. We found participants were highly stable in this tendency, which significantly explained post TGI errors.

Positive control 3): Do predictions influence subjective ratings? Hierarchical zero one inflated beta regression (ZOIB) of VAS ratings revealed that cold and warm ratings are increased when stimuli matched predictions. Participants expectations shifted their felt sensations!

Positive control 2): Did catch trials successfully elicit the TGI? Yes! Catch trials were rated as significantly more warm and burning, characteristic of the TGI, than innocuous cold and warm trials. Same temperatures, but participants feel more pain!

Positive control 1): does cue validity (predictive, nonpredictive, antipredictive) influence participants' predictions and reactions times? Yes! Participants successfully learned the cue-stimulus associations. Responses to TGI trials resembled veridical prediction error trials.

By presenting these stimuli together, the thermal grill illusion elicits a feeling of burning, tingling pain. These catch trials acted as ambiguous stimuli, and allowed us to assess how learned thermal expectations modulate the intensity of both innocuous and painful sensations.