Ultra-slow mechanical stimulation of olfactory epithelium modulates consciousness by slowing cerebral rhythms in humans

The coupling between respiration and neural activity within olfactory areas and hippocampus has recently been unambiguously demonstrated, its neurophysiological basis sustained by the well-assessed mechanical sensitivity of the olfactory epithelium. We herein hypothesize that this coupling reverberates to the whole brain, possibly modulating the subject’s behavior and state of consciousness. The olfactory epithelium of 12 healthy subjects was stimulated with periodical odorless air-delivery (frequency 0.05 Hz, 8 s on, 12 off). Cortical electrical activity (High Density-EEG) and perceived state of consciousness have been studied. The stimulation induced i) an enhancement of delta-theta EEG activity over the whole cortex mainly involving the Limbic System and Default Mode Network structures, ii) a reversal of the overall information flow directionality from wake-like postero-anterior to NREM sleep-like antero-posterior, iii) the perception of having experienced an Altered State of Consciousness. These findings could shed further light via a neurophenomenological approach on the links between respiration, cerebral activity and subjective experience, suggesting a plausible neurophysiological basis for interpreting altered states of consciousness induced by respiration-based meditative practices.

SECTION A: air delivery device setup and mode of operation and polygraphic system overview. . Before conducting each comparison, the feature value was normalized to its corresponding pre-period value. Normalization, except were otherwise stated, was always obtained as the ratio between post-and pre-values. The between-session comparisons were performed after a proper normalization to avoid possible confounding effects stemming from differences in electrodes impedances values between the two sessions. Such differences usually lead to non-physiological differences in EEG voltages, strongly biasing the results of between-sessions comparisons.

SECTION C: analysis of breathing rhythms
The subjects' breathing signals were acquired using a piezo-resistive respiratory belt placed on their abdomen. Signals were acquired for both conditions (NS-SC) and periods (pre-post).
Respiratory signals were acquired for each subject, session and period, (sampling rate 500Hz, online band-pass filtering 0.01-20Hz). Three subjects were discarded from the analysis as their respiratory signals were largely affected by noise making it impossible to extract their breathing rhythm. For each retained subject and period, the signal was divided in 60s epochs with a 50% overlap between contiguous ones. Each epoch was submitted to a Hamming-windowed Fast Fourier Transform. Mean spectrum density as a function of frequency were obtained for each period averaging between epochs pertaining to the period itself, its peak denoting the breathing frequency as shown in the exemplary figure presented below. Figure C1. An exemplary case of breathing signal (first plot) and the related mean spectral density (second plot) is presented. The peak frequency of the mean spectrum density is of 0.1167Hz (red dot) corresponding to a breathing rhythm of 7 breaths/minute.  For the sake of completeness, even if the omnibus tests were not significant, planned post-hoc were performed: i) NS post-versus pre-ii) SC post-versus pre-and iii) post-NS versus post-SC.
For each comparison the t-value was extracted. 500 random permutations of the original dataset were performed under the null-hypothesis of no between-condition effect. Under the null, for each subject the value for condition1 could be attributed to condition2 and vice-versa. For each permutation the corresponding t-value was extracted (in absolute value for two-tails significance assessment). We obtained thus the t-values distribution under the null-hypothesis. The original tvalue significance was obtained as the ratio between the number of randomly generated t-values exceeding the absolute value of the real one and the number of randomizations. This number was set to 500 as with 9 subjects and two conditions the maximum number of possible different randomizations would have been 2^9=512.

SECTION E: Randomization tests
-No assumption about distributions normality or homoscedasticity.
-Null-hypothesis: no effect or treatments/conditions…. (i.e. under the null the value associated with a subject/feature is independent from the condition).
-Any statistic can be used depending on the dataset organization and on the formulated null-hypothesis.
-The output of the chosen statistic is not compared to its tabled distributions but rather to outputs obtained calculating the statistics of choice on datasets generated by the repeated randomization of data across groups (treatments/conditions).
-Significance is obtained as the ratio between the number of outputs (in absolute value) generated from the randomizations higher than the real output (taken in absolute value) and the total number of randomizations.
As an example, let us consider the one-way ANOVA with hubs as an eight-levels between-factor.
-The F-value statistics is extracted.
-We formulate the null-hypothesis of no significant hubs-effect.
-Under the null-hypothesis values related to hubs can be relabeled randomly (i.e. a value from hub1 attributed to hub3, a value from hub4 to hub8 and so on).
-One thousand random relabeling are performed collecting their F-values.
-The significance of the original F-value is obtained as the ratio between the randomly generated F-values exceeding the real one, and the total number of relabeling.  Of the 128 electrodes of the EGI sensor net, 21 located on the cheeks and forehead were removed as they mostly contributed to movement-related noise (red dots in Figure H). The 107 retained channels were then re-referenced to the mastoids' (black dots in Figure H) average reducing the analyzed electrodes to 105.

SECTION I: Statistical non-Parametric Mapping, SnPM.
Let us briefly introduce the basis and rationale of SnPM: assume without loss of generality to have collected an EEG feature for each electrode in two different conditions. For each electrode, a paired t-test between the conditions is conducted and its t-value is retained. As the test is applied to multiple electrodes, a single-threshold SnPM procedure is used to assess the significance of each t-test, considering the multiple comparison issue. Let us consider the null-hypothesis of no significant condition-effect: under this null-hypothesis the labeling of the collected feature can be changed randomly (i.e. a feature estimated for condition 1 can be assigned to condition 2 and vice-versa). Based on this assumption, 1000 random relabeling are made, and for each of them, the t-value related to each single comparison is extracted. For each relabeling, only the maximum t-value (in absolute value, for two-tailed significance assessment) among simultaneous comparisons is retained. In this manner, the maximum t-value distribution under the nullhypothesis of no significant condition-effect is extracted. The significance of each original t-value is then estimated as the ratio between the number of t-values of the null-distribution exceeding the original t-value (in absolute value) and the number of relabeling.

SECTION J: Granger Causality in the frequency domain
Herein we briefly rehearse the mathematical basis and hypotheses underlying the spectral domain Granger Causality: let x and y be two wide-sense stationary time-series, the Granger Causality (1).
In (1) a k are the 2x2 matrices of the model coefficients, n is the model order, e x (t) and e y (t) are the residuals. The restricted model is mathematically described by: In equation (2), b k are the model coefficients, n is the model order and ê x (t) is the residual. Please note that for both models t 0 represents the time-lag between consecutive samples. Coefficients The unrestricted model regression coefficients can be then expressed as: