What Language Is the Brain Speaking?

The prevailing paradigm of brain function assumes that the neural correlates of consciousness are (somehow) to be found in the time-evolution of coordinated brain activity. If so, we would expect that a state of general anesthesia should be marked by severe disruptions to these processes. The question arises what measure of brain activity best captures these changes. Traditionally various indices of the frequency content in a single electroencephalogram (EEG) channel have been used to describe how the brain activity alters with anesthetic-induced loss of responsiveness. However, a big unresolved issue is that different classes of anesthetic drugs can produce the same behavioral effects (unresponsiveness) but with completely different effects on EEG frequencies. Notably, propofol tends to cause the EEG to be dominated by slow waves, whereas ketamine increases the power in high frequencies. This has led to a search for different ways of capturing brain activity involving higher-order multichannel descriptions of brain functional connectivity, source localization, spatial mapping, etc. One approach that provides a spatiotemporal window on brain activity suggests that the brain is progressing through a series of microstates.¹  Microstates are defined as a stereotypical set of four to six topographical patterns of EEG power, each of which is stable for 60 to 120 ms before switching to the next pattern (fig. 1 in Michel and Koenig¹ ). They have been termed the “atoms of thought.”  Brain microstates are distinct from the concept of brain metastability, which describes the capacity of the brain to balance disorder and stability, providing a functional framework for switching from one state to another.

Liang et al. have reported on how propofol and esketamine alter the microstates when patients become unresponsive.  For propofol, they found that the change to unconsciousness was associated with a shift in dominance from posterior microstates to an anterior pattern. For esketamine, the dominance shifted to a prefrontal pattern. These observations are interesting but perhaps unsurprising. The authors then went a step further to examine how propofol and esketamine altered the complexity of the evolution of these microstates using three measures of complexity. Two quantify the degree of randomness (Lempel–Ziv complexity, mean information gain) and are termed type 1 complexity measures. The third one (fluctuation complexity) is probably a better estimator of “complexity,” because it achieves maximal values at a point where there is some structure, but also some irregularity (type 2 complexity) They found that the state of unconsciousness was associated with an increase in mean information gain (randomness) and a decrease in fluctuation complexity (structure)—irrespective of whether propofol or esketamine was used. To help illustrate this, figure 1 shows examples of the probabilities of transitioning between five microstates for six sequential 5-s segments when the subject was awake (top) and when unconscious under propofol (middle). It is apparent that the transition probabilities from when the patient was awake have more extremes of variation (dark blue and yellow) and “structure” than those from when the patient was unconscious under propofol anesthesia (predominantly green-blue), which are more “random” or homogenized (i.e., closer to a uniform probability distribution, where entropy is maximal). Importantly, these findings meant that Liang et al. had found EEG microstate complexity measures that were driven by the change in the state of consciousness of the patient, irrespective of the class of hypnotic drug—acknowledging the proviso that the unconsciousness of ketamine may be different from that of propofol.

Plausibly, the main function of the brain is to generate meaning from information. This process involves the brain representing information in some structured organized way (syntactics) so that it can generate biologic meaning (semantics). If we assume that the prime action of anesthesia is to wreck this process, then we should be able to monitor the deleterious effects of anesthesia on the semantics by detecting disruption at the appropriate syntactic level. Following this language metaphor, we could therefore see anesthesia as disturbing the “words” that make up the language of the brain. There are at least three hierarchical levels that make up the brain’s language. First, the letters of its language might be represented by the fluctuations in electrical fields measured in some way. This is commonly (but imperfectly) achieved using the scalp EEG. Second, the letters form accepted words—analogous to the spatial microstates. Third, the full semantic content is then derived from the correct sequence of words combined into a meaningful sentence. This could be thought of as the complexity of the transition probabilities of the sequence of microstate, driven by the brain’s inherent metastability—as reported by Liang et al., and in similar work by Artoni et al. 

As a simple example, we might take the simple sentence, “Jill turned left.” If we assume that propofol slows the single-channel EEG, we might represent this sort of effect as a repetition of the letters. This might result in a sentence like, “JJll turndd left.” The entropy of the single letters has decreased. In contrast, if we assume the ketamine acts to cause random insertion of the letters, we might get a sentence with increased letter entropy like, “Jxll Zurned Qeft.” At this level of disruption, the syntax has been disturbed, but the meaning may be somewhat discoverable. This would explain why existing monitors of consciousness that rely on single-channel EEG may be imperfect. However, if both propofol and ketamine also act at the level of how the words are joined to form a sentence (manifest as the increase in randomness [mean information gain] and decrease in fluctuation complexity), we might eliminate the spaces between the words, resulting in further degradation of meaning: “JJllturnddleft.” or “JxllZurnedQeft.”

This is somewhat fanciful but illustrates the point that there are many different ways to disrupt the brain’s language syntax, some of which may be evident in the temporal structure of a single EEG channel and others in the complexity of the microstates. Importantly, the change in consciousness state in both propofol and ketamine is driven by the loss of higher-level semantics. All this theorizing seems a long way from the operating room, and our understanding is still at quite a primitive level. However, this paper highlights the fact that the optimal monitor of consciousness for anesthesia will be the one that is quantifying disturbances at whichever syntactic level is most closely linked to the actual generation of the brain semantics (i.e., the “mind”).