Increased breathlessness in post-COVID syndrome despite normal breathing patterns in a rebreathing challenge

Figures

• 

  Fig. 1. (a) Internal model and inference (based on Petzschner et al. (2017)). The brain holds a probabilistic model of how the received respiratory input has been generated. This includes the probabilities of different respiratory body states (prior) and the probabilities to observe the received respiratory input conditioned on the possible respiratory body states (likelihood). By inverting this model, the brain can infer the underlying respiratory state. This inference corresponds to applying Bayes theorem. (b) Bayes theorem describes how to optimally combine different sources of noisy data. It is commonly applied to interoception and symptom perception and suggests how the brain should combine noisy sensory (respiratory) data (likelihood) with model-based predictions (a priori knowledge, prior) to yield an optimal estimate of the underlying respiratory body state (posterior). Prior and likelihood are conceived as probability distributions to assign (un)certainty to each of these. The width (i.e., variance) of the likelihood function corresponds to the brain’s uncertainty (i.e., the signal’s reliability) only based on the sensory signal. The variance of the prior corresponds to the…

• 

  Fig. 2. Participants that were excluded from data analysis.

• 

  Fig. 3. Results rebreathing experiment. (a) Exemplary time course of inhaled CO2 concentration from one individual. First, participants breathe room air, i.e., inhaled CO2 concentration is low. After a valve switch, they rebreathe from a rebreathing bag, initially filled with 5% CO2 and 95% O2. This constitutes a strong respiratory stimulus. After 150 s the breathing circuit is opened again, and participants breathe room air. The exact rebreathing and recovery start was defined based on the inhaled CO2 concentration of each individual (see Methods). (b) Differences in breathlessness reports for all patients and healthy participants (left), only those that were not hyperventilating (middle) and patients with versus without breathlessness that were not hyperventilating (right). (c) Breathing patterns and FetCO2 for all three group comparisons. Solid lines – mean, dashed lines – standard error of the mean.

• 

  Fig. 4. Possible computational mechanisms that could explain increased breathlessness despite normal breathing patterns. (A) Different internal models for breathlessness perception and breathing patterns. The same sensory input and its reliability is assumed for breathing patterns and breathlessness perception, represented as the same likelihood function. However, for breathlessness a higher and more reliable prior (i.e., higher mean value and less variance/higher precision) is assumed. According to Bayes theorem, the posterior (blue curve) is a precision-weighted combination of likelihood and prior. Thus, due to the higher and more reliable prior for breathlessness perception than breathing patterns, the resulting posterior is shifted towards higher breathlessness levels, even though sensory input (likelihood) is the same as for breathing patterns. The inferred bodily state (posterior) is thus closer to the actual sensory input in breathing patterns and closer to the prior in symptom perception. While the respiratory state is correctly inferred for breathing patterns, an incorrect internal model, e.g., with a strong weighting of erroneous priors, could lead to strong breathlessness…