Heart Rate Scales with Prediction Error

2486 

Abstract Submission 

Maria Azanova^1,2, Lina Skora^3,4, Esra Al^5,1, Vadim Nikulin¹, Arno Villringer^1,6,7

¹Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany, ²Max Planck School of Cognition, Leipzig, Germany, ³Heinrich-Heine-Universität Düsseldorf, Düsseldorf, Germany, ⁴University of Sussex, Brighton, United Kingdom, ⁵Columbia University, New York City, NY, ⁶University Hospital Leipzig, Leipzig, Germany, ⁷Humboldt University Berlin, Berlin, Germany

Maria Azanova  
Max Planck Institute for Human Cognitive and Brain Sciences|Max Planck School of Cognition
Leipzig, Germany|Leipzig, Germany

Lina Skora  
Heinrich-Heine-Universität Düsseldorf|University of Sussex
Düsseldorf, Germany|Brighton, United Kingdom

Esra Al  
Columbia University|Max Planck Institute for Human Cognitive and Brain Sciences
New York City, NY|Leipzig, Germany

Vadim Nikulin  
Max Planck Institute for Human Cognitive and Brain Sciences
Leipzig, Germany

Arno Villringer  
Max Planck Institute for Human Cognitive and Brain Sciences|University Hospital Leipzig|Humboldt University Berlin
Leipzig, Germany|Leipzig, Germany|Berlin, Germany

Recent conceptualizations of cardiac deceleration in adaptive environments suggest that heart slowing could improve the precision of perception by reducing the frequency of noisy events associated with heartbeats (Skora et al., 2022). For instance, threat-related cardiac deceleration as part of freezing may not merely reflect a fear state but rather enable the integration of autonomic and central nervous reactions to coordinate an appropriate response (Roelofs & Dayan, 2023). Indeed, parts of the neural circuitry underlying freezing are also prominently involved in decision-making, learning, and interoception (e.g., ACC: Seamans & Floresco, 2022; insula: Zhao et al., 2023). This shared neural architecture suggests a more general role of heart rate in the integration of information. To further the knowledge of cardio-behavioral states involved in learning, we studied single-trial cardiac and neural reactions to feedback during a probabilistic learning task with non-threatening stimuli. We asked: (1) how exactly, if at all, does heart rate scale with prediction error? (2) how does it relate to the strength of the neural reaction to feedback?

To this end, we recorded combined EEG and ECG during the task in 37 participants. We used computational modeling (Q-learning) to extract prediction errors and Bayesian mixed linear models to obtain the results. We identified the Region of Practical Equivalence by assuming that effects of maximal change in prediction error can not be smaller than 1 ms change in IBI interval; in standardized coefficients, this resulted in 0.005 bounds. To study cardiac responses, we examined the change in the inter-beat interval (IBI) of the first heartbeat after feedback relative to the previous IBI. To study neural responses, we computed average amplitudes of feedback-related negativity (FRN) and P300, brain-evoked potentials that occur at around 270 and 380 ms post-feedback, respectively. We used the pseudo-trial approach (Wainio-Theberge et al., 2021) to confirm that observed ERP differences are not due to cardiac artifacts by adding random feedback triggers to the data and repeating the analysis.

We found that heart rate slowed down more with larger prediction errors, even after controlling for the heart cycle phase at the moment of feedback presentation, its valence, as well as for the change and magnitude of the previous IBI: mode of standardized coefficient = .018, 95% highest density interval (HDI) = [.007,.032], Bayes Factor for coefficient >0 (BF>0) = 999, BF for null coefficient = 0.1. Furthermore, only when feedback appeared earlier in heart cycle phase (i.e., systole), change in heart rate was associated with the frontal P3a component of P300 (mode = -.058, HDI = [-.1,.-.16], BF<0 = 269.27, BF null =.12), and less so with FRN (mode = -.015, HDI = [-.06,.024], BF<0 = 4.41, BF null = 3.12) or posterior P3b component (mode = .04, HDI = [-.001,.08], BF>0 = 30.62, BF null = .84). Finally, we observed that heart rate change weakly predicted change of decision in the next trial with the same outcome-predictive stimulus (mode = .25, HDI = [.04,.488], BF>0 = 97.77, BF null = .34), indicating a direct relation to choices. All reported results were insensitive to a range of normal priors with 0 mean and uncertainty scaling from .05 to 1 sd.

Therefore, we show it is possible to detect instant IBI scaling with prediction error in the order of tens of ms. Our findings provide insight into the mechanistic basis of the integration of cardio-behavioral states in learning and prediction. In particular, it seems that instantaneous heart rate change is mainly associated with frontal dopaminergic attentional involvement (Polich, 2007). This has further implications for understanding how cognitive processes are affected by various disturbances in heart-brain interactions, such as ones observed in anxiety (Tumati et al., 2021) and arrhythmia (Kumral et al., 2022).

Decision Making ²

EEG/MEG Modeling and Analysis

Perception and Attention Other ¹

Electroencephaolography (EEG)

Learning

Perception

Somatosensory

Other - Heart-Brain Interactions; Prediction Error