How do psychedelics affect the brain?

A primer on the REBUS/CANAL theory

Arthur Juliani
15 min readMay 2, 2023

In recent years there has been a renaissance of research into the use of psychedelics for the treatment of a wide array of psychopathologies. Mental health concerns as various as mood disorders, anxiety disorders, body-image disorders, addiction, and others have all been shown to be amenable to psychedelic therapy [1]. In many cases, results suggest that one or two sessions of psychedelic assisted psychotherapy can lead to lasting reduction in symptoms, even when other mainline therapies have failed. Indeed, a recent high quality double-blind study has shown the classic psychedelic drug psilocybin to be as effective at reducing symptoms of major depressive disorder as SSRIs [2]. These promising results have led to an increase of interest from academic, commercial, and legal entities. With all of this interest, it is increasingly likely that psilocybin will become a legal and available treatment option for individuals with depression and potentially other related issues in the coming years.

Figure 1. Effects of Psilocybin vs Escitalopram on ratings of major depression over time. (Carhart-Harris et al., 2021)

With the impending sea change in the legal status and therapeutic availability of these substances, it is more important than ever to understand their mechanisms of action on the brain and wider nervous system. Despite the groundbreaking work that has been done in the field of psychedelic neuroscience over the past decade, there is still a considerable amount to be learned before a complete account can be given for how and why psychedelic therapy produces the clinical outcomes that it does. Given that the subjective effects of psychedelics are heavily dose and context dependent [3], understanding the underlying action of these drugs is essential to better ensure that therapeutic outcomes can be positive. This line of work is especially important because having a better understanding of how and why psychedelics work can help to inform not only more nuanced treatment protocols, but ultimately the development of novel drugs themselves, drugs which have the potential to contribute to more positive clinical outcomes and may be better suited to treat a wider range of individuals.

This article is written with the goal of providing a high-level summary of the current state of research into psychedelics and their effect on the brain, both from an empirical and a theoretical perspective. First we survey the current empirical understanding of psychedelics’ action, which involves understanding the serotonin receptors that psychedelics act on, as well as in what brain regions these receptors are expressed. It also involves understanding how psychedelics impact brain activity more broadly, both during and after administration. Next we turn to a pair of prominent computational theories which attempt to take the empirical evidence and use it to develop a working model of psychedelic action. Key among these theories is the RElaxed Beliefs Under pSychedelics (REBUS) model, proposed by Carhart-Harris and Friston [4]. The REBUS model is based on the hierarchical predictive processing (HPP) framework for understanding brain activity [5], which will be briefly explained as well, because of its foundational nature to understanding later theories. As the acronym suggests, REBUS attributes the therapeutic action of psychedelic drugs to their ability to relax the high-level belief landscape instantiated in the brain, thus making it more amenable to updating with new evidence from both the body as well as the outside world. A more recent theory from Carhart-Harris and colleagues builds on REBUS to propose that canalized beliefs — overly-rigid beliefs encoded in the cortex — may be the primary underlying factor behind all of psychopathology [6]. Here we summarize both theories, discuss their implications for psychedelic therapy, and briefly address how they might be further refined in the future.

Neuroscience of Psychedelics

The classic psychedelics include mescaline, LSD, psilocybin, and DMT. What all of these molecules have in common is their action on the serotonin system within the brain. In particular, these drugs are all believed to exert their action by primarily acting as an agonist of the serotonin 5-HT2A receptor [7]. This mechanism of action has been elucidated by co-administering a 5-HT2A antagonist drug with a psychedelic. Whereas the psychedelic alone results in the typical subjective effects, the psychedelic plus the antagonist results in no reported effects, making it clear the 5-HT2A receptors play an essential role. The 5-HT2A receptor is expressed throughout the mammalian nervous system and is responsible for increasing the excitability of the neuron, making it more likely to fire.

Figure 2. Expression of 5-HT2A receptors in the brain. Red corresponds to greater density. (Beliveau et al., 2017)

The locations of neurons which express 5-HT2A receptors is not uniform however. Certain brain regions contain neurons which express them much more densely than others, and this differential in expression is responsible for the unique effects of psychedelics. 5-HT2A receptors are densely expressed in the thalamus, prefrontal cortex, and claustrum, among other regions [8]. We mention these three regions in particular, because each plays a key role in the dynamics of perception, cognition, affect, and behavior, and therefore is thought to be heavily involved in the effects of psychedelics [9]. The thalamus is involved in the gating of sensory and other subcortical information into the cortex. The cortex is the region of the brain thought to be generally responsible for cognition in a broad sense. The prefrontal cortex in particular is responsible for high-level cognition, including the formation and maintenance of various conceptual abstractions related to self-awareness. Finally, the claustrum is involved in the orchestration of information processing within the cortex, with different cortical columns connected to one another through the claustrum.

The effects of psychedelics on the brain have been studied using various brain imaging techniques, including fMRI, EEG, and MEG. Between the wide array of studies which have been conducted, there is a set of emerging results which seem to be held in common between modality and drug. The first is that psychedelics seem to increase the computational complexity of brain activity, sometimes referred to as the entropy of the brain activity [10]. This can be measured in various ways, with Lempel–Ziv (LZ) complexity being one popular metric. Measures of LZ complexity of brain activity are higher for individuals undergoing the acute effects of a psychedelic drug than they are for someone who is in a normal everyday waking state. Furthermore, individuals in a normal waking state have a higher LZ complexity score than individuals who are either in non-REM sleep or in an anesthetized state [11].

Figure 3. Measure of entropy in brain activity across various conscious states. (Scott & Carhart-Harris, 2019)

Relatedly, increases in entropy under psychedelics are accompanied by changes in functional connectivity, a measure of how likely different regions of the brain are to communicate with one another. Studies have found that within-network connectivity is decreased while between-network connectivity is increased while under the acute effects of LSD [12]. A brain network which has been of particular interest for psychedelics researchers is the Default Mode Network (DMN), which is responsible for self-referential processing and so-called mind wandering [13]. The DMN shows a reduction in within-network connectivity as well when an individual is on psychedelics, and this has been thought to be particularly relevant to the experience of ego-dissolution which is commonly reported. As self-reported ego-dissolution is correlated with positive therapeutic outcomes of psychedelic therapy [14, 15], this may be a particularly important network to study in the context of these drugs’ effects.

Figure 4. Image of cortical neuron treated with various substances. VEH and TRY are baselines. NMT and DMT are active psychedelics, and result in marked neurogenesis. (Vargas et al., 2023)

After the acute effects of a psychedelic have worn off, there are a series of additional after effects which can last from hours to days to weeks. Most notable among these are changes in systems related to the brain’s neuroplasticity. Brain-derived neurotrophic factor (BDNF) is a protein responsible for neurogenesis, and has been found in increased levels in brains after exposure to a psychedelic drug such as psilocybin [16]. More importantly, there is direct evidence for neurogenesis in cells after exposure to psychedelics which can last up to weeks. Imaging of pyramidal neurons in the cortex shows an increased number and density of dendritic spines in these cells after exposure to DMT, LSD, or psilocybin [17]. Collectively, this evidence suggests that exposure to psychedelics puts the brain into a state of greater plasticity both during and after the experience, though the nature of the plasticity induced in the acute and post acute states differs.

Hierarchical Predictive Processing

Among the various computational models of the brain, hierarchical predictive processing (HPP) has seen significant success in being used to understand and predict brain activity, both at a low and a high-level of abstraction [18]. Within the context of HPP, the brain can be understood as a sequence of generative models, each with the objective of predicting the activity of models lower in the hierarchy, which serve as their input. The extent to which the predictions fail to account for the incoming information from lower levels produces error signals. These errors in prediction are then passed to higher levels as the input to those levels. This simple objective of predicting the activity of other generative models within the brain enables, hypothetically at least, all of the complex representational capacity which we humans seem to be endowed with. Moreover, this principle is general enough to apply not only to human brains, but to the nervous systems of all living animals. Within this framework nervous systems are acting to minimize the free-energy [19], which can be thought of as equivalent to the entropy, surprise, or prediction errors of the system at each level of representation. Ultimately this surprise or potential entropy comes from the outside, either from external sensory information in the world, or from interoceptive information gathered from the body of the organism.

Figure 5. A simple schematic of hierarchical predictive coding in the brain. (Kanai et al., 2019)

In attempting to anticipate this ever changing stream of exogenous information, it makes sense that increasingly complex representations about the belief of the world and body would emerge. What is less immediately apparent, but just as important, is the fact that the ability to meaningfully control the external world is also an emergent property of any system which is capable of acting in the world that is engaged in trying to minimize free energy. It is in this need to control aspects of the world that the process of prediction goes from being simply a passive affair to an active one as well [20]. This becomes clear when we realize that in many cases the best way to predict the outcome of some event is to have a causal influence over that event oneself.

Even high-level goals which aren’t directly tied to minimizing exogenous free energy can still be incorporated into this framework of active inference. Consider for example someone setting the goal for themself to eat a sandwich for lunch. This goal can be understood as a kind of prediction about the future state of the world. Because the goal is not actualized, it produces errors within the generative models of the brain, since it is not in reality the state of the world, and thus is contradicted by sensory evidence. There are two ways for the individual to resolve this source of error, either to change their internal belief about eating the sandwich, or to change the world itself such that the world conforms with the belief. If one assigns a high enough confidence to the belief represented by the goal, then the path the system follows to reduce free energy will necessarily involve changing the external world.

The result of this process of free energy minimization is that a hierarchical set of belief landscapes are eventually learned and refined over time. These landscapes encode both the nature of the currently held beliefs as well as the strength of those beliefs. This sense of confidence in a belief is referred to as the precision of the belief, which in more technical language corresponds to the inverse variance in the variational distribution induced by the generative model representing that belief. Our certainty in beliefs can change over time either through the process of inference or learning. In inference we rapidly update a given belief based on new context-specific information. An example of this would include walking home after work. If I see that a given road is closed, I then have to update my beliefs about the best way home. Given that I have walked through the neighborhood many times, I simply select a different route, and the discrepancy is resolved. As I update the beliefs, I ‘move through’ the belief landscape, from one point to another. More radical updates to beliefs can also happen through the process of learning. Imagine for example that I am in a new city and need to find my way to my hotel. Here the updating of my beliefs happens both regarding where I am, but also more deeply regarding how I would even get around this new location at all. Here the updating of my beliefs involves not only moving through the belief landscape, but also changing the underlying topology of that landscape. Of course, both of these processes are taking place simultaneously throughout the life of an organism.

Relaxed Beliefs Under Psychedelics

Given the HPP framework, we can then ask what happens to these finely-tuned belief landscapes under the effect of psychedelics. We can start to understand this action by considering the brain regions mentioned above which have densely expressed 5-HT2A receptors in their neuron populations. The first of these is the thalamus, which has been hypothesized to decrease in its ability to gate information’s entry into the cortex under the effects of psychedelics [21]. What this means in an HPP framework is that higher precision sensory evidence is entering into the system, producing larger prediction errors at different levels of the generative model. This effect is further catalyzed by the effect of 5-HT2A agonism on the prefrontal cortex, which is thought to represent high-level beliefs about the world, including various self-referential beliefs [22]. Here the activity of the psychedelic excites the neurons in the PFC, leading to a desynchronization of their activity, and effectively disrupting their ability to represent coherent beliefs about the world. This disruption thus prevents the high-level generative model within the HPP framework from making coherent predictions about the beliefs at other representational levels, and likewise reduces the precision of the beliefs represented at the high-level themselves.

The inability of the brain’s high-level generative models to predict the beliefs at lower levels corresponds to an inability to suppress those beliefs, with incoming evidence from lower-level beliefs serving to more greatly impact the updating of beliefs at higher levels. This lack of coherence of high-level representations of beliefs is further exacerbated by a disruption of the normal activity of the claustrum, leading to typically functionally unconnected regions of the predictive hierarchy entering into communication with one another. While the REBUS model has minimized the importance of the claustrum’s role in these effects, other recent models have given it more of a central position [23]. These complex effects within the cortex are all measurable via an increase in complexity of neural activity, as well as in changes in functional connectivity which have been seen in fMRI studies of individuals under the effect of LSD, psilocybin, and DMT.

According to the REBUS model, the ultimate result of this change in cortical dynamics is to relax the belief landscape which the cortex normally represents. In particular, this relaxation is hypothesized to take place at the highest levels of the representational hierarchy in the regions of the brain responsible for self-referential beliefs. Within the HPP framework, this relaxation corresponds to a reduction in the precision of the high-level beliefs being encoded by the PFC. Lower precision in a belief means that the generative model is less confident in it, making it more amenable to updating from conflicting beliefs at other levels of the hierarchy. In this way, it is thought that psychedelics open up a window of time within which high-level beliefs can be more easily updated by evidence from the exogenous world, both of the external senses as well as the interoceptive world of the body. As the effects of the drug wear off, the belief landscape gradually returns to a state of higher precision beliefs once again, thus recalcifying the newly updated beliefs.

Figure 6. An example of “relaxing beliefs” by flattening a belief landscape.

It is possible, using the HPP framework, to hypothesize about how various psychopathologies could arise from an individual possessing maladaptive high-level priors which have been learned with an overly high precision weighting, thus making them difficult to undo. Likewise, it is possible to understand how psychedelic therapy may act to undo these maladaptive priors. A straightforward account of this can be seen in the example of body dysmorphic disorder. Here an individual interprets the appearance of their body as being overweight, for example, despite the fact that they may be objectively within the normal weight range, or more likely, even on the low end of that range. This individual’s belief in being overweight can be thought of as being encoded in a high-level prior concerning their body image. This prior may be overly precise, and thus will serve to overwrite the sensory evidence to the contrary which an individual might be presented with. Various forms of reinforcement over time may have resulted in the belief acquiring this maladaptive level of precision.

The therapeutic action of psychedelics is then to relax the high-level beliefs being represented in the generative machinery of the brain. This makes it both possible for the individual to interpret themselves more in-line with the sensory evidence, but also importantly to be able to update their high-level prior more towards the evidence they accumulate during and after the psychedelic experience. In the case of a high-level prior corresponding to a distorted sense of the body, its relaxation and updating would ideally make the individual less likely to experience dysmorphia in the future. It is relatively straightforward to see how such an account might be applied to pathologies of addiction, depression, anxiety, and others, all of which can be understood through a lens of overly-rigid high-level priors.

The Canalization Model of Psychopathology

Given the apparent capacity for the REBUS model to account for such a wide range of psychopathologies, it is natural to ask whether there might not be a more fundamental principle at work here. This is exactly what Carhart-Harris and colleagues did in a recent paper published late last year. In that work, they set forth the proposal that overly precise high-level priors can be understood as being the primary factor underlying all of psychopathology, the so-called p-factor [24]. If this proposal were proved to be true, it would be easy to see how and why psychedelics would have potential therapeutic value for a wide range of individuals suffering from apparently diverse psychopathologies.

Appealing to a topographic metaphor, the authors propose the concept of canalization as a way of understanding the development of these inflexible high-level beliefs. Because of the centrality of canalization to their thesis, the overall model of psychopathology is referred to as CANAL. Just as water constantly running down the same slope will make it more likely that water runs down that slope in the same way in the future, repeatedly activating a given belief increases the precision of that belief, making it more likely that they will be activated in the future, and thus ultimately less likely to be amenable to change. In some cases this habit formation can be seen as adaptive, but the authors argue that in many cases it is not. The CANAL model is especially useful for thinking about substance use disorders, which are often the result of given habitual behaviors being consistently reinforced to the point of near automaticity, often with a great detriment to the quality of life of the individual.

Figure 7. Canalization of a landscape as the guiding metaphor for psychopathology. (Carhart-Harris et al., 2022)

It is important to make clear that the hypothesis that canalization corresponds to the p-factor of psychopathology is just that: a hypothesis. As such, it needs empirical verification before it should be used to guide clinical thinking in a broader applied context. Indeed, as the authors allude to, there is good reason to suspect that things may not be quite so simple. This is clear from the fact that there are psychopathologies which may be understood to arise from too-little, rather than too-much stability of beliefs. These include various forms of schizophrenia, bipolar disorder, and depersonalization disorder. It is possible that these disorders may be made to fit within the CANAL model, but doing so would not be straightforward. In fact, psychedelics are often explicitly recommended against for individuals with personal or family histories of schizophrenia, as their use has been linked to the onset of psychosis in such individuals [25]. Because of this, the reality of a psychedelic drug’s capacity for therapeutic action, while real, is more complex than the simple train of causality described by: canalized beliefs are pathological, psychedelics reduce canalization of beliefs, therefore psychedelics produce positive mental health outcomes. The CANAL model provides a very useful starting point upon which to build a more nuanced account of psychopathology and psychedelic drug action, something my colleauges and I hope to do in the near future.

Thank you for reading this far! This article is Part 1 of a multi-part series on psychedelic neuroscience and therapy. Be on the lookout for more articles in the near future.



Arthur Juliani

Interested in artificial intelligence, neuroscience, philosophy, psychedelics, and meditation.