This hypothesis paper proposes that psychedelic agents like LSD and psilocybin induce altered states of consciousness by activating the 5-HT2A receptor system, leading to a state of synthetic surprise. This concept is based on recent understandings of serotonin's role in signaling surprise and is framed within the predictive coding framework, where surprise is seen as a mismatch between expectations and sensory input. The paper suggests that psychedelics disrupt maladaptive patterns by dynamically interacting with top-down expectations and sensory data, with implications for their clinical use, particularly emphasizing their ability to induce surprise to promote therapeutic effects.
Psychedelic agents, such as LSD and psilocybin, induce marked alterations in consciousness via activation of the 5-HT2A receptor (5-HT2ARs). We hypothesize that psychedelics enforce a state of synthetic surprise through the biased activation of the 5-HTRs system. This idea is informed by recent insights into the role of 5-HT in signaling surprise. The effects on consciousness, explained by the cognitive penetrability of perception, can be described within the predictive coding framework where surprise corresponds to prediction error, the mismatch between predictions and actual sensory input. Crucially, the precision afforded to the prediction error determines its effect on priors, enabling a dynamic interaction between top-down expectations and incoming sensory data. By integrating recent findings on predictive coding circuitry and 5-HT2ARs transcriptomic data, we propose a biological implementation with emphasis on the role of inhibitory interneurons. Implications arise for the clinical use of psychedelics, which may rely primarily on their inherent capacity to induce surprise in order to disrupt maladaptive patterns.
De Filippo and colleagues introduce a theoretical account that frames the psychedelic state as an imposed affective condition they term "synthetic surprise." Building on longstanding evidence that classic psychedelics act via the serotonin (5-HT) system, the paper integrates recent experimental findings linking 5-HT signalling to surprise, prediction error and uncertainty with the predictive coding framework and the concept of cognitive penetrability of perception. The authors argue that psychedelics do not simply relax high-level priors or globally increase cortical excitability, but instead bias 5-HT receptor activation in a way that artificially increases prediction-error signalling, with downstream perceptual and emotional consequences. The paper sets out to develop this synthetic surprise hypothesis, to contrast it explicitly with alternative predictive-coding based accounts, and to propose a plausible neural implementation that emphasises inhibitory interneurons. In addition to the conceptual synthesis, the investigators report an analysis of mouse whole-brain transcriptomic data (from the Allen Institute) to explore receptor expression patterns and provide an online visualiser of 5-HT receptor RNA expression to support the proposed circuitry-level implementation.
Papers cited by this study that are also in Blossom
Anand, A., Mathew, S. J., Sanacora, G. et al. · New England Journal of Medicine (2023)
Aqil, M., Roseman, L. · Neuropharmacology (2022)
Barrett, F. S., Griffiths, R. R. · Current Topics in Behavioral Neurosciences (2017)
Bienemann, B., Ruschel, N. S., Campos, M. L. et al. · PLOS ONE (2020)
This work is a theoretical review and synthesis rather than an experimental trial. The authors review empirical literature from animal and human studies linking 5-HT function to surprise, prediction error, learning and perceptual change, and they situate those findings within predictive coding and affective realism frameworks. They explicitly contrast their proposal with other predictive-coding accounts and draw on physiological, neuroimaging, electrophysiological, optogenetic and pharmacological results reported across species. Complementing the conceptual synthesis, the investigators conducted an analysis of publicly available mouse transcriptomic data from the Allen Institute cell-type atlas and examined a MERFISH dataset to map 5-HT receptor RNA expression across cortical cell classes and layers. They adopted a conservative expression threshold (a cell was considered to express 5-HT receptor RNA if log(CPM) > 3.5) and defined a "5-HT receptor–enriched supertype" as one in which at least 50% of cells express the receptor RNA. Findings from this transcriptomic interrogation were used to inform the proposed circuit-level implementation, and an online visualisation tool based on the MERFISH data was produced for exploration of receptor expression patterns.
The central conceptual finding is the synthetic surprise hypothesis: psychedelics bias 5-HT receptor activation to impose a state akin to surprise (an unsigned prediction error) on the brain's inferential machinery. Under this account, the subjective feelings of awe, novelty and heightened connectedness reported under psychedelics are primary affective consequences that drive downstream perceptual changes via cognitive penetrability and predictive processing. The model emphasises an increase in prediction-error signalling itself, rather than a primary reduction in the precision of high-level priors; the downstream effect on priors depends on precision weighting, which is modulated by other neuromodulators. Empirical regularities consistent with the model are reviewed. Psychedelics consistently reduce cortical alpha power, a neurophysiological change also observed during exposure to novel stimuli. Cardiovascular and autonomic responses associated with surprise (increased heart rate, blood pressure, pupil dilation, skin conductance) are commonly elicited by LSD and psilocybin. Some electrophysiological indices of surprisal, such as mismatch negativity (MMN), have been reported to diminish under psychedelics in some studies, although results are mixed. Behaviourally, low-dose psychedelics and 5-HT manipulations affect reversal learning and cognitive flexibility in ways that align with the role of 5-HT in signalling unexpected outcomes. At the level of neuromodulation and cells, the review collects converging evidence that 5-HT neurons in the raphe respond to unsigned violations of expectation and to measures of uncertainty; fast-scan voltammetry and optogenetic studies show phasic 5-HT responses linked to surprise and sensory uncertainty. From the transcriptomic analysis, the authors report enrichment of 5-HT receptor RNA in cortical inhibitory subclasses, most notably SST (somatostatin-expressing) interneurons: a stable fraction of SST interneurons across cortical layers expressed 5-HT receptor RNA (reported as 63.76 ± 0.98%). Sst44, a marker associated with an SST subset implicated in error signalling during navigation, was present in a substantial fraction (≈40%) of Calb2+ and Hpse+ neurons, and many SST 5-HT receptor–positive supertypes were distributed across cortex and layers. Lamp5 and PV interneuron classes also showed notable 5-HT receptor RNA expression, and some excitatory subclasses exhibited high prevalence (>50%) in specific cortical areas. The authors provide these transcriptomic observations as anatomical support for a circuit in which 5-HTR activation recruits inhibitory interneurons. Using these inputs, a circuit-level proposal is advanced: exaggerated activation of SST interneurons by biased 5-HT receptor agonism leads to over-inhibition of neurons that normally compute negative prediction errors, producing an unbalanced positive prediction-error signal. This imbalance could account for common psychedelic phenomena (e.g. perceived movement of static objects, increased access to remote priors, enhanced imaginative suggestibility) and for the mix of heightened belief updating and, in some conditions, strengthened priors depending on precision weighting. The review notes that other interneuron classes and excitatory subtypes also express 5-HT receptor RNA, so multiple cell classes may contribute to net effects.
De Filippo and colleagues interpret their synthesis to mean that psychedelics should be understood primarily as agents that artificially impose a surprise-like state on the brain's predictive machinery. They argue this perspective helps reconcile features of the psychedelic experience that are difficult to accommodate under alternative formulations that emphasise a uniform relaxation of high-level priors. In particular, the synthetic surprise model can explain enhanced imaginative suggestibility and culturally consistent hallucinations by allowing top-down priors to shape perception when precision relationships favour such influence; whether priors are weakened or strengthened depends on the precision afforded to prediction errors, which in turn is shaped by other neuromodulators. The authors position their proposal relative to prior predictive-coding models by stressing two main differences: the locus of change (an increase in prediction-error signalling versus a loss of prior precision) and the proposed cellular implementation (a shift in emphasis from excitatory pyramidal gain to engagement of inhibitory interneurons, notably SST cells). They marshal electrophysiological, optogenetic and imaging findings and their transcriptomic analysis as convergent support, but acknowledge that current data are incomplete. Key limitations and uncertainties are explicitly acknowledged. The extracted text notes incomplete mapping of 5-HT receptor effects across cortical cell classes and regions, unresolved functional selectivity of 5-HT receptors, and mixed empirical results for some signatures (for example MMN and oscillatory-band changes beyond alpha). The authors caution that the exact downstream impact on precision weighting remains complex because acetylcholine, dopamine and other neuromodulatory systems also contribute. They further recognise that transcriptomic enrichment does not prove functional activation across all cortical areas. For future research and clinical implications, the paper proposes several empirical tests: behavioural paradigms that probe the influence of priors on perception under psychedelics; pharmacological antagonism experiments to test the behavioural role of specific 5-HT receptors; systematic circuit-mapping of receptor effects across inhibitory and excitatory subclasses; and exploration of non-pharmacological ways to elicit unexpected uncertainty. Clinically, the authors tentatively suggest that the therapeutic effects of psychedelics in disorders like depression may derive from their ability to disrupt "locked-in" maladaptive priors by inducing surprise, but they also note novelty effects may wane with repeated exposure and that further mechanistic and translational work is required to validate the hypothesis. The synthetic surprise model is presented as a falsifiable framework intended to guide such empirical investigations.
The psychedelic experience, characterized by profound alterations in perception, cognition, and self-awareness, has captivated human curiosity for centuries. From ancient rituals to modern scienti ic investigations, the exploration of altered states of consciousness through the use of psychedelics has been an intriguing and enduring topic. Psychedelics, such as lysergic acid diethylamide (LSD), psilocybin, mescaline and N,N-dimethyltryptamine (DMT), have the ability to induce remarkable and often transformative subjective experiences. The psychedelic experience can be described as a departure from ordinary waking consciousness. It leads to an altered state of awareness that transcends the boundaries of conventional perception and cognition. Participants often report vivid visual and auditory alterations, a noetic quality, synesthesia, profound shifts in emotions, a sense of awe, altered sense of time, and a heightened sense of connection with the self, others, and the surrounding environment. Psychedelics seem to have the potential to unravel the fabric of our conventional reality, unveiling its deeply contingent and subjective nature. The study of psychedelics and their effects on consciousness has experienced a resurgence in recent years, driven by the exploration of their potential clinical applications, after decades of limited research because of legal and regulatory restrictions. In this work we propose a new model to explain the effect of psychedelic compounds on cognitive functions informed by recent experimental observations linking 5-HT to surprise, prediction error and uncertainty. Three concepts inherently related to each other. Activation of 5-HT Rs is universally recognized as the mechanism underlying the subjective experience provoked by psychedelics. According to our hypothesis the psychedelic state should be considered as an alteration of the native emotional state of surprise, which we refer to as synthetic surprise. Here the quali ier synthetic is used deliberately to underscore the signi icant distinctions in comparison to the natural surprise emotion. Synthetic surprise is posited to emerge from the partial activation of the 5-HTR system, in contrast to the response observed under physiological conditions following the release of endogenous 5-HT. We can assume that native 5-HT activates all 5-HTRs according to speci ic binding af inities. Psychedelics deviate from this pattern by selectively activating only a subset of recep-We review the literature supporting the notion that 5-HT signals surprise/prediction error.
We posit that activation of 5-HT receptors can elicit a synthetic surprise response. • 2A Synthetic surprise, via cognitive penetrability of perception and predictive coding, can explain the psychedelic state.
We propose a possible biological implementation via activation of interneurons.
We provide an online 5-HT RNA visualizer, based on the mouse transcriptomic atlas from the Allen Institute.
"No account of the universe in its totality can be inal which leaves these other forms ofconsciousness quite disregarded."William2A tors. Moreover, in some cases, binding kinetics are remarkably different. The activation of different patterns of 5-HTRs can certainly elicit profoundly different downstream effects. For example, 5-HT R and 5-HT R have opposing effect on the membrane potential, causing respectively inhibition and excitationand can co-localize on the same neuron (Amargós-. The proposed model is built upon two theoretical frameworks: the cognitive permeability of perception and the predictive coding hypothesis. These two concepts are interconnected, as cognitive permeability can be effectively understood and described within the framework of predictive coding. Notably, surprise, in the context of predictive coding, refers to the magnitude of the prediction error or the level of mismatch between the predicted and actual sensory input. Surprise plays a crucial role in the predictive coding framework as it drives learning and adaptation processes. The brain constantly re ines its internal models to align with sensory inputs, thereby minimizing prediction errors. This adaptation improves understanding of the environment and the accuracy of future predictions. Various models have been proposed to explain the effect of psychedelics. Some theories, like the cortico-striato-thalamocorticaland the cortico-claustrocortical model, focus on implementation, highlighting altered circuits without placing signi icant emphasis on explaining the psychological effects. Other theories, like the strong prior hypothesis to explain hallucinations, on the contrary focus on the psychological effects. The REBUS model, built also on the foundation of predictive coding, attempts to provide explanations for both the psychological effects and the speci ic alterations in neural circuits induced by psychedelics. We can outline the synthetic surprise model against the backdrop of the REBUS model, given the similar foundation, as an alternative perspective with a notably different implementation. At the heart of the REBUS model is the concept that psychedelics exert their effects by diminishing the precision assigned to high-level beliefs or priors, thereby increasing the precision of prediction errors and facilitating the low of bottom-up sensory information. In contrast, the synthetic surprise model posits an increase in the prediction error signal itself, rather than in its precision. This distinction has signi icant implications. A steady prediction error signal is anticipated to increase the expected uncertainty, therefore lowering the ability of the errors to in luence priors, corresponding to a strengthening of priors. Accordingly, learning rates are known to decrease in the presence expected uncertainty. When priors are more precise than sensory data, they can potentially shape perception, aligning with the strong prior hypothesis for hallucinations. The computing of precision, however, seems to be distributed among several neuromodulators including dopamine and acetylcholine. Notably, dopamine and acetylcholine have been historically associated with the neural regulation of selective attention, a process that has been hypothesized to optimize precision. In summary, while the expected effect of a persistent arti icial prediction error signal is to increase the precision afforded to priors, we do not exclude that the in luence of other neuro-modulators can, in some instances, increase the precision of sensory data, resulting in a relaxation of priors as described in the REBUS model. This interplay enables a dynamic interaction between top-down expectations and sensory information (Fig.). The synthetic surprise model strives to offer a solution to certain discrepancies of the REBUS model. Speci ically, the emphasis of the REBUS model on a reduction in the precision of toplevel priors does not easily account for the enhanced "imaginative suggestibility" observed under psychedelics. "Imaginative suggestibility" is the capacity of an individual to immerse themselves in imaginative scenarios that can in luence their behavior and subjective experience. In a recent placebo-controlled study, it was found that the perceived realism and vividness of imagined scenarios, following detailed auditory instructions, were signi icantly heightened under the in luence of LSD. Complementing this inding, ethnographic studies often note a cultural consistency in hallucinatory experiences, supporting a culturalist perspective on psychedelic hallucinations. Factors beyond the drug itself, including mindset (expectations, preparation, intention) and environment (physical and social settings), signi icantly in luence these hallucinogenic experiences. To account for the in luence of imaginative suggestions on perception, it seems logical to incorporate top-down control of perception. This mechanism allows individuals to align their perceptual experiences with the expectations that arise from external suggestions. It is not trivial to reconcile these observations with the main tenet of the REBUS model, a relaxation of the precision associated with top-level priors. Second, it is widely recognized that hallucinations can occur in healthy individuals under conditions of sensory deprivation, particularly in complete darkness. In such instances, where sensory input is limited or absent, the brain may generate its own perceptual experiences, resulting in hallucinations. This phenomenon highlights that vivid stimuli alone do not typically alter perceptions, but rather hallucinations can arise in the absence of external sensory stimulation, such as in complete darkness. Closed-eyes hallucinations are commonly reported under psychedelicsand it is unclear how a relaxation of priors can explain these phenomena. Both discrepancies can theoretically be explained by abnormal precision assigned to priors. At a neurophysiological level, the REBUS model places signi icant emphasis on the increased spiking activity of deep-layer pyramidal neurons induced by 5-HT R activation. If this were the primary effect of psychedelics, it would suggest a general enhancement of excitatory drive throughout the cortical network. However, the network effects of 5-HT R activation seems to be more intricate and nuanced, and they do not align with a simple increase in gain of excitatory cells. For instance, studies in rodents employing selective activation of 5-HT R through optogeneticor pharmacologicalstimulation consistently show a decrease in stimulus-evoked activity in the visual cortex. Furthermore, 5-HT Rs have been implicated in mediating divisive scaling of evoked responses, leading to a bidirectional modulation by suppressing neurons with strong responses and facilitating neurons with weak responses. Further research, utilizing optogenetic stimulation and fMRI techniques in mice, has established a correlation between the inhibition of activity across the cortex, triggered by 5-HT release, and the expression of 5-HT R. Additionally, activation of 5-HT R in rats with psychedelics inhibited spontaneous population activity in a va- riety of cortical regions. Given that 5-HT R activation typically leads to depolarization and increased spiking, it is reasonable to assume that the reduction in activity of pyramidal neurons is mediated by the activation of inhibitory interneurons induced by 5-HT R. Inhibition or direct interneuron activation by 5-HT R has been reported across various brain structures in addition to visual cortex, including prefrontal cortex (PFC), piriform cortex, cingulate cortex, cochlear nucleus, amygdala, olfactory bulb, entorhinal cortexand hippocampus. In summary, the synthetic surprise model, informed by recent experimental insights regarding 5-HT, differs from the REBUS model in two key aspects. First, it proposes that psychedelics induce primarily an arti icial prediction error, with an impact contingent on its assigned precision. In this context hallucinations can be explained by the strong prior hypothesis. Second, the biological implementation of the synthetic surprise model shifts its focus from excitatory to inhibitory cells. We will start by brie ly introducing the concepts of cognitive permeability of perception and the predictive coding hypothesis. Subsequently we will describe the psychedelic experience with its biological underpinnings, novel indings connecting 5-HT with the prediction error and our model with its implications.
Affect has been known to in luence behavior and judgment since antiquity. Competition between reason and emotion has been a staple philosophical concept throughout history from Plato to Freud. The in luence of emotions can be categorized as either integral or incidental. Integral emotions are derived from the choice at hand, its potential outcomes and repercussions. While they were once viewed as detrimental to decision-making, current consensus sees integral emotions as a potential bene icial guidein line with the trailblazing view proposed by David Hume. Compelling scienti ic validation of this perspective comes from individuals af licted by emotional impairments resulting from injuries to the ventromedial prefrontal cortex (vmPFC). This brain region plays a pivotal role in integrating cognition and emotion. Research shows that such neurological impairments signi icantly diminish both the capacity of patients to experience emotions and the effectiveness of their decision-making processes. Studies show that participants with vmPFC injuries consistently exhibit a tendency to choose riskier inancial options over safer ones, even to the extent of facing bankruptcy in real-money games. Remarkably, despite their cognitive understanding of the suboptimal nature of their choices, these individuals repeatedly make decisions that lead to detrimental outcomes. Physiological measurements, such as galvanic skin response, provide insights into the underlying reasons for this behavior. They suggest that the absence of emotional somatic markers prevents these individuals from experiencing reasonable fear in the face of high-risk scenarios, which is typically present in normal decision-makers. Incidental emotions in luence decisions by carrying over from unrelated situations, usually without awareness. These affective states bias in a systematic normative judgment and decision-making. While a valence-based approach categorizes emotions into positive and negative, suggesting that good moods induce optimism and bad moods pessimism, this approach has its limitations. The Appraisal-Tendency Framework offers a multi-dimensional perspective, suggesting that emotions not only predispose individuals to speci ic behaviors but also to appraise their environment in particular ways. However, biases in judgment can arise when these appraisals stem from incidental emotions. In line with this idea, incidental emotions have an in luence on risk-taking, perceived life-satisfaction (Schwarz and, prosocial behavior, food-choice and consumption. One explanation for the effect of emotions or mood on behavior posits that affect, even when incidental, is an intrinsic facet of perception. In the same way as color is considered to be a property of an object, affect is also incorporated in the perception of reality. This hypothesis is called affective realismand aligns with experimental data showing the impact of affect on perception. Individuals experiencing negative affect exhibit perceptual biases such as heightened perception of sound intensity, increased sensitivity to visual contrast gradients, and a propensity for local rather than global perceptual processing of images. Furthermore, stimuli with negative valence are often perceived as larger in comparison to neutral or positive stimuli, and individual differences in motivation and past experiences in luence perception. For instance, individuals experiencing unpleasant thirst perceive a glass of water as taller, and individuals with spider phobia perceive spiders as larger. Affective realism can be considered a variant of the affect-as-information theorythat states that individuals use their current affective state as a way to gain information from the environment, therefore, blurring the line between incidental and integral emotions. Affective realism differentiates itself by explicitly positing affect as part of the conscious perception of an object or situation and not merely in luencing the judgment process downstream from perception. Some experimental evidence supports this hypothesis, providing empirical validation for the integration of affect within the conscious perceptual experience. Affective realism inherently assumes the cognitive penetrability of perception. Traditionally, cognition, perception, and action were viewed as entirely independent processes. The validity of this perspective has been radically challenged by claims that perception should be considered as a way of acting, and not a purely passive event. Evidence for the top-down in luence on visual perception stems from the organization of the brain inter-regional connectivity, the speed of temporal processing and functional experiments. All visual areas are modulated by upstream inputs, retina included. Top-down in luences originate from various sources, encompassing multiple areas that convey information in accordance with the functional properties speci ic to each area, these in luences have the capacity to modify the information transmitted by neurons in two ways: by modifying the responsiveness of neurons to stimulus properties, or by altering the patterns of correlations within neuronal ensembles. Various studies show that top-down in luences on neuronal activity can occur in less than 50 ms after stimulus onset. Collectively, these results demonstrate that the brain possesses the necessary infrastructure to enable the in luence of cognition on low-level perception. The effect of expectation on perception has been investigated in some classic experiments showing a change in the perceived color of a stimulus object shaped by the expectations formed through prior interactions and familiarity. More recent studies have con irmed these results, further pointing to a signi icant modulation of high-level visual memory on color perception. Even when objects are presented in an achromatic format, a discernible neural representation of the associated color can be decoded in the primary visual cortex V1. Observed changes in perception were also linked to activation of areas not part of the early visual system supporting the interaction of perception with memory and attentional processes. Importantly the effect of expectations and attention can be differentiated. Top-down in luences seem to potentially modify expectations, which play a central role in perception. Ful illed expectations are associated to improved object recognition and reduced neural responses. Interestingly, the predictive coding theory offers a convincing theoretical framework to explain these observed effects.
The predictive coding hypothesis has gained signi icant traction in neuroscience as a compelling framework for understanding the mechanisms underlying perception and cognition. This hypothesis posits that the brain actively generates predictions about sensory inputs and uses these predictions to update and re ine its internal models of the world (Von. At its core, the predictive coding hypothesis is rooted in the principle of Bayesian inference. According to this perspective, the brain aims to minimize prediction errors by continuously updating its internal models based on the discrepancy between expected and actual sensory inputs. Furthermore, the concept of perceptual inference and the signi icance of prior expectations in shaping perception have deep historical roots, tracing back to the 11th century with the work of Ibn al-Haytham. He already recognized that the perception of many visible properties relies not merely on the external objects but also on the processes of judgment and inference. Although predictive coding encompasses a variety of interpretations, our focus is on the predictive processing that juxtaposes sensory input with a generative environmental model. This comparison is understood to occur within a hierarchical structure, composed of multiple, successive systems that are involved in prediction and novelty detection. Prediction errors are deeply tied to their precision or uncertainty. This relationship allows us to differentiate the magnitude of prediction errors from their reliability. In the context of a noisy or volatile environment, the continuous signaling of large prediction errors may not necessarily lead to substantial up-dates of expectations, mainly due to the inherent imprecision of the prediction errors. Conversely, even slight deviations between sensory inputs and descending predictions can trigger signi icant updates of conditional expectations when the prediction errors exhibit high precision. The signi icance of precision lies in its ability to minimize surprise about the amplitude of prediction errors, a second order prediction. Precision can be manipulated externally, such as through alterations in the contrast or statistics of the stimulus, or internally, by directing attention to speci ic sensory streams or altering the contextual expectancy of sequential stimuli. Remarkably, attention plays a pivotal role in anticipating precise sensory information or prediction errors. The concept of precision gains particular importance in the interpretation of certain empirical indings, where repetition suppression can be in luenced by contextual factors like attention. The predictive coding hypothesis has profound implications for our understanding of perception and cognition. It posits that perception is an active process, where top-down predictions shape and guide the processing of bottom-up sensory inputs. This framework can account for various perceptual phenomena, including perceptual illusions and the in luence of prior knowledge on perception. Experimental evidence in human using functional neuroimaging has provided support for the predictive coding hypothesis across various sensory modalities. Reports have shown that the brain's response to sensory stimuli is in luenced by the magnitude of prediction errors. Larger prediction errors elicit stronger neural responses, indicating that the brain is sensitive to deviations from its predictions. While empirical evidence has strongly supported the concept of hierarchical inference in the cortex, the precise implementation by cortical neurons remains less understood. One theory posits the existence of error units that directly compare top-down and bottom-up information generating a prediction error signal. When there is a mismatch or discrepancy between the top-down predictions and the incoming sensory inputs, a prediction error signal is generated. This prediction error signal represents the magnitude of the mismatch and serves as a feedback signal to update the existing predictions. The error units play a crucial role in adjusting and re ining the predictions, facilitating the alignment between the internal model and the incoming sensory information. The theory suggests that prediction error signals are propagated hierarchically, with higher-level brain regions receiving feedback about the prediction errors and updating their predictions accordingly. This iterative process allows the brain to continually update its internal representations and improve the accuracy of its predictions. Early evidence supporting the predictive processing framework in neocortex came from studies on classical visual phenomena, such as end-stopping. These phenomena can be explained as prediction errors, where the suppression of neuronal responses in the surround of a classical receptive ield is attributed to top-down inhibition. This concept of top-down prediction inhibiting bottom-up input has been used to explain various classical visual receptive ield properties within the framework of predictive processing. In the mouse visual cortex, SST interneurons, likely driven by lateral projections from neighboring cortical neurons, have been found to play a causal role in surround suppression. Surround suppression involves the inhibition of spiking in response to stimuli presented in the surrounding area of its receptive ield. Traditionally, sensory neurons are characterized by tuning curves, which illustrate how their iring rates change in response to various stimulus attributes. In the case of surround suppression, the neuron's responses to stimuli outside the region that typically triggers iring are suppressed. While the majority of research on surround suppression has focused on visual cortex, it seems to be a widespread computation observed also in olfactory, somatosensoryand auditory cortices. Recent experimental data investigating the effect of visuomotor mismatch seems to give support to the existence of error units in visual cortex. The effects of visual input and locomotionrelated input on the membrane potential of neurons in the neocortex were found to be opposing in layer 2/3 (L2/3) neurons. This observation supports the hypothesis that L2/3 neurons compute a difference between visual (bottom-up) and motor (topdown) inputs. In line with other reports showing prediction error signals in L2/3. In contrast, deep-layer neurons, usually associated with encoding of expectations, did not exhibit this characteristic, showing depolarizing responses to both types of input. Moreover, different L2/3 excitatory neurons responded to the mismatch with signi icantly different change in membrane potential, ranging from strong depolarization to strong hyperpolarization. These responses align with signaling positive or negative prediction errors, which can be attributed to the opposing inputs from visual and locomotion-related sources. Importantly, this computational characteristic appears to be speci ic to L2/3 neurons. According to this predictive coding microcircuit, activity of positive and negative prediction error neurons should be anticorrelated, therefore when positive prediction neurons are excited negative prediction neurons should instead be inhibited. It is plausible to suggest that the differentiation between these two functional types of neurons in L2/3 is determined by variations in circuit wiring. One element of the circuitry associated to negative prediction error L2/3 neurons has been elucidated: the visually driven inhibition onto these neurons originates from SST interneurons, which selectively target the apical dendrites of L2/3 excitatory neurons. Recently, a speci ic subset of SST interneurons, Sst44 cells, in the posterior parietal cortex was observed to activate synchronously during trajectory corrections for successful goal-directed navigation, effectively providing a navigation prediction error signal. Interestingly, while Sst44 cells in the retrosplenial cortex also show synchronous activation, they do not convey navigation error signals. This suggests that the functional relationship between this subset of cells, and behavior or sensory inputs may be region-speci ic. The role of SST interneurons in predictive coding is re lected by their responses to predictable and deviant stimuli. Inhibiting SST interneurons activity speci ically diminishes the detection of deviations in cortical circuits, on the other hand, baseline responses remain largely unaffected at the cellular or circuit level. Conversely, predictable stimuli are associated with reduced activity of SST interneurons. However, contrasting indings have been reported, with some studies indicating that SST interneurons are more reactive to familiar rather than novel stimuli. Extensive recording in visual cortex using longitudinal 2-photon calcium reveals that SST interneurons can respond differently to novelty, either being inhibited or excited, thereby providing a reconciliation of these con licting indings. Stimulus adaptation upon repeated exposure can be understood as an expression of the brain's ef icient strategy for minimizing prediction error by adapting its predictions concerning the content and precision of incoming sensory in-puts. While these indings underscore the importance of SST interneurons in the predictive coding circuitry, the precise relationship between adaptation and different types of prediction errors remains unclear. In summary, cortical SST interneurons are involved both in the modulation of surround suppression and in the computation of prediction errors, highlighting their potential signi icance in predictive coding processing.
As early as the 1950 s, it was hypothesized that the alteration of subjective experience caused by psychedelics was related to the 5-HT system. This pioneering idea stemmed from the observation that the chemical structures of LSD and 5-HT bear a striking resemblance. Soon after, it was proposed that LSD might act as an agonist of a speci ic class of 5-HTRs. Remarkably, this idea was advanced before the con irmation of the existence of different classes of 5-HTRs. Subsequently, chlorpromazine, a typical antipsychotic, was found to signi icantly dampen the psychological alterations induced by LSD in humans. At the time this result was misattributed to the general sedative effect of the compound. Convincing evidence collected since those early years of psychedelic research have con irmed that psychedelics exert their overt effect on consciousness by interacting with the 5-HT system, and more precisely, by activating the 5-HT R. Chlorpromazine is indeed a potent 5-HT R antagonist. The administration of ketanserin, a selective blocker of 5-HT R, can signi icantly reduce the subjective effects of several psychedelics in humans. Moreover, the "head twitch response" in rodents, a behavioral proxy for the effect of psychedelics, was shown to rely on 5-HT R. Studies in humans also indicate a correlation between the psychedelic effects of psilocybin and the occupancy of 5-HT Rs, as measured through positron emission tomography in cortical regions. Furthermore, hallucinogenic potency is highly correlated to 5-HT R binding af inity. While some psychedelics have a considerably large binding spectrum that includes 5-HT, dopamine, and adrenergic targets, it is clear that 5-HT R plays a mechanistically fundamental role in the psychedelic experience. The activation of the 5-HT R is associated with an elevation in intracellular calcium levels, resulting in the depolarization of the resting potential and possibly increased spiking activity. 5-HT Rs are located throughout the cortical sheet, in the prefrontal, cingulate, visual, temporal and motor areas. They are often found on the apical dendrites of pyramidal neuronsbut they are also present on inhibitory interneurons. This was shown using both immunohistochemistryand electrophysiological methods. Given the localization on both excitatory and inhibitory neurons, the downstream effect of 5-HT R agonism is dif icult to predict. Experimental indings regarding the in luence of psychedelics on brain waves in human present a complex picture, with diverse outcomes reported across different studies (Table). While delta, gamma, and theta bands have shown both an increaseor a decreasein either global or localized patterns, one consistent observation emerges: a pervasive decline in cortical alpha oscillations across psychedelics. In most cases this reduction in alpha power is positively correlated with the intensity of the subjective psychological effect. Functional imaging studies in humans have revealed also that psychedelics signi icantly impact the default mode network (DMN), comprising the medial prefrontal cortex, posterior cingulate cortex, and parietal regions. This network is typically active during rest and internal mental processes. Psychedelics are found to decrease the functional connectivity within the DMN, the amount of covarying neural activity occurring simultaneously across different brain regions. Additionally, they increase connectivity between sensory regions belonging to different networks. Neuroimaging studies combined with pharmacological blockade methods have con irmed that these LSD-induced changes in functional network con iguration are dependent on the activation of 5-HT Rs.
The 5-HT system, comprising a complex network of neurons primarily located in the raphe nuclei on the midline of the brain stem, plays a pivotal role in various neural and cognitive processes, in luencing both brain function and behavior. 5-HT, for example, participates in the regulation of mood, sleep, appetite, memoryand decision-making. The broad in luence of the 5-HT system stems from its extensive projections across various cortical and subcortical regions. Despite the relatively small number of neurons expressing 5-HT, less than 0.1%, projections from these neurons extensively innervate virtually all areas of the brain. There are seven major classes of 5-HTRs, all with the exception of 5-HT R, the only ionotropic receptor, are G-protein coupled receptors with complex intracellular chemical pathways. The heterogeneous distributions of the various 5-HT receptor classes imply that 5-HT has the ability to regulate multiple aspects of animal cognition and behavior by modulating large-scale receptor networks across the entire brain. 5-HTRs have been shown to modulate synaptic plasticity, neuronal excitability, and information processing. The 5-HT system, therefore, seems to fundamentally contribute to the intricate machinery underlying fundamental cognitive activities. Beyond these multifaceted roles, it has been recently proposed that 5-HT might be responsible for the encoding of novelty and cognitive lexibility in dynamic environments. To study the adaptation to an unexpected change in a familiar environmentemployed a reversal learning paradigm in mice in which the association between odor cues (conditioned stimuli, CSs) and different outcomes (unconditioned stimuli, USs), either a reward, a neutral stimuli or 2A a punishment, were suddenly reversed during the execution of the task. Using photometric recordings in the dorsal raphe it was observed that when confronted with reward outcomes that deviated from expectations, whether they were better (positive prediction error) or worse (negative prediction error) than anticipated, 5-HT neurons displayed a similar pattern of transient excitation. There was a distinct activation upon the reversal with a neutral US stimulus. This response contrasts sharply with dopamine neurons, which exhibit an inhibitory response. This suggests that 5-HT neurons are responsive to unsigned violations of expectation in contrast to dopamine neurons that respond with either excitation or inhibition depending on the valence of the reward prediction error. In another study, tetrode recordings from optotagged 5-HT neurons showed that iring was positively correlated to unexpected uncertainty (i.e., surprise) in a dynamic foraging task in which mice are required to choose between two alternative sources of water. Here unexpected uncertainty was de ined as the deviation of the reward prediction error (RPE) from expected uncertainty, its historical weighted average. Interestingly, 5-HT neurons were found to correlate, both positively and negatively, also with expected uncertainty over relatively long periods (tens of seconds to minutes). Expected uncertainty, corresponding to precision, modulates the in luence of prediction errors on updating priors, thereby adjusting the learning rate from current sensory information. These indings suggest a dual involvement of 5-HT in both irst and second order predictions. In accordance with the proposed role of 5-HT in mediating prediction errors, impairments in performance were observed in a reversal learning task in marmosets with selective depletions of 5-HT in the prefrontal cortex. The primary factor contributing to these impairments was the tendency to exhibit perseverative responding towards the stimulus that was previously rewarded. Further research indicated that this de icit speci ically affected reversal learning and not attentional set shifting. Additionally, it was discovered that the impairment in reversal learning was speci ically associated with 5-HT depletion and not dopamine depletion in the orbitofrontal cortex. More speci ically, selective blocking of 5-HT R signi icantly impaired reversal learning, while low doses of LSD facilitated the same task. Exposure to acute psilocybin was also observed to increase cognitive lexibility in a task requiring switching between previously learned strategies both in human and rodents. The relation between prediction error and 5-HT was also investigated in human using fast-scan cyclic voltammetry in the dorsal striatum during a sequential investment game in which patients were instructed to select an investment level on each trial while real historical inancial market prices unfolded. Here it was observed that 5-HT was signi icantly elevated in response to failed bets, indicating the occurrence of negative reward prediction errors (i.e., when participants predicted a reward that did not materialize). When categorizing the bets into high (wager >60% of current wealth) and low (wager <50% of current wealth), a particularly pronounced response was observed in relation to failed high bets. This outcome is linked to the stronger prediction of reward associated with high bets (high bets are logically associated with increased con idence of the prediction), leading to a more substantial error signal when the expected reward is not obtained. Notably, 5-HT exhibited a transient increase during instances of regret arising from not placing a substantial bet that would have been suc-2A cessful (i.e., positive prediction error, outcome better than expected). Interestingly, the 5-HT system did not show signi icant activation in response to predicted rewards or US, in line with prior research. Consistent with these indings, 5-HT neuromodulatory tone shows a tight relationship with pupil size, a behavioral marker providing an online surprise signal. Both in human and rodent, pupil size showed a remarkable increase upon experiencing prediction errors. Moreover, the association between 5-HT and pupil size is modulated by the amount of uncertainty in the environment, representing the baseline amount of prediction errors. In mice, in situations of low uncertainty, optogenetic stimulation of the 5-HT system consistently leads to signi icant transient pupil dilations. However, when the environment is already uncertain, this effect is diminished. This inding resembles precision weighting, the mechanism that facilitates robust learning signals to drive more substantial hypothesis revision. Conversely, in conditions of heightened uncertainty, the prediction error unit are inhibited, diminishing its impact on the updating of priors. Additionally, 5-HT seems to play a direct role in perception. Experimental data from human recordings using fast-scan cyclic voltammetry showed that sub-second signaling of serotonin in the human striatum is involved in real-time inference about the external world.showed that there is a transient increase in 5-HT levels when coherence in a dot motion paradigm, the fraction of coherently moving dots, is low. Coherence here corresponds to sensory uncertainty. In contrast, there is a transient decrease in 5-HT levels when coherence is high. These observations suggest that 5-HT dynamically tracks the level of uncertainty present in sensory information, displaying opposite modulation depending on the degree of coherence. Collectively, these indings present a captivating portrayal of the involvement of 5-HT in monitoring novelty across multiple levels of abstraction. 5-HT appears to function as a signal for prediction errors and uncertainty, with a possible direct link with perception. Intriguingly, the involvement of 5-HT in indexing uncertainty can be used to explain its known effect on behavioral inhibitionand patience. Using optogenetics it was shown that high reward probability maximizes the ability of 5-HT to promote patience. This result was explained using a Bayesian decision model, suggesting that 5-HT neuron activation increases the subjective con idence of reward delivery. In other terms, we can consider trials with high reward probability to engender a clear expectation of reward. Stimulating 5-HT release and consequently, according to our model, increasing uncertainty in a context where a clear expectation is present (high precision afforded to the prior) can cause a discounting of sensory evidence in favor of the expectation. Moreover, uncertainty in the sensory data, manipulated in the study by changing the reward timing, also increases waiting times, consistent with a further strengthening of the precision associated to the prior. The activation of the 5-HT system is likely to initiate a cascade of effects, engaging various classes of receptors that, in turn, may in luence other neuromodulatory pathways. For example, photostimulation of the raphe nuclei in luences pupil size with a longer latency compared to noradrenergic stimulation, suggesting an indirect effect. Additionally, 5-HT neurons can directly modulate dopaminergic neurons. These intricate interactions highlight the complex nature of the downstream effects of 5-HT.
At a more abstract level, yet fundamentally anchored in a predictive coding paradigm, we can employ the concept of cognitive penetrability to elucidate the psychedelic experience. The central idea of our proposal is inspired by the aforementioned experimental data pointing at the role of 5-HT in signaling prediction error and it can be succinctly conveyed by a simple thought experiment. Let us assume the existence of some chemical compounds able to trigger, in a speci ic and compartmentalized fashion, one emotional state. According to the cognitive penetrability and affective realism hypotheses, such compounds will potentially have a direct effect on sensory perception and cognition. Consequently, the deliberate arti icial induction of fear is expected to impact prior beliefs. This effect can be expected to result in ambiguous stimuli being perceived as more alarming in comparison to a baseline neutral state. If we were to elicit instead anger, we can hypothesize that the subject would be biased towards inferencing a threat in the presence of an ambiguous stimulus. What if we arti icially elicited surprise? Probably everything would look surprising even in absence of any objective novelty in the environment. Moreover, while other emotions can be expected to have an in luence particularly in ambiguous contexts, an arti icially imposed surprise affect would directly increase the uncertainty of the environment. This can inherently increase the perception of illusory patterns. Cognitive penetrability tend to occur when an input is uncertain or ambiguous, especially when a higher level representation can successfully explain away the prediction error. Moreover, hallucinations and delusions have been reported to be in luenced by the emotional state of the subject. We propose that this is what happens upon consumption of psychedelic compounds: a synthetic surprise affect enforced on our inferential machine and cognition. An interesting corollary, deriving from the hypothesis that psychedelics are irst and foremost (Bayesian) surprise-inducing compounds, is that the perceptual alterations are to be considered downhill effects of the overactivation of prediction error signaling. It follows that the sense of profound surprise, also described as awe (the emotional response to something perceived as extraordinary), often reported upon consumption of psychedelics, is not, as intuitively assumed, triggered at least in part by sensory alterationsbut, on the other hand, is the upstream cause of those alterations. The awe and novelty experienced under psychedelics are proposed to be the primary driver, setting off a cascade of perceptual changes, rather than a result of these changes. In a normal state, surprise arises from the interplay of sensory inputs and internally generated predictions. Its magnitude is directly proportional to the extent of disparity between predictions and actual sensory input and is associated with physiological changes, including an increased heart rate, elevated blood pressure, heightened skin conductance response, and enlarged pupil size. Studies indicate that substances such as LSD and psilocybin can trigger these identical physiological responses. Speci ically, LSD and psilocybin have been observed to increase blood pres-sure, heart rate, and pupil size. Additionally, LSD has been found to elicit a clear skin conductance response. These indings suggest that the psychedelic experience evokes an interoceptive state resembling the one elicited by natural surprise. Following both historical and contemporary theories of emotion, which emphasize the integral role of physiological responses in shaping our emotional experiences, we propose that this altered interoceptive state might have an important emotional in luence. The psychedelic experience itself is so foreign compared to the normal palette of conscious states that it seems to naturally require an extraordinary explanation, a fundamental change in brain function. This assumption is likely driven by the inherent novelty of the psychedelic experience, activating brain mechanisms that are usually triggered by unexpected stimuli in normal physiological conditions. Consequently, according to our model no fundamental alteration of brain functioning should be observed. The psychedelic state should, instead, bear resemblance to a native surprised state. In accordance with this view the one neural signature common to all psychedelics, a reduction in alpha power, is also observed naturally in result of exposure to novelty. Using multilaminar electrodes in monkey during a delayed match to sample task, alpha power of cortical areas was decreased when the task was less predictable. In humans EEG recordings during a novelty detection task showed that alpha power was decreased upon the detection of a temporary unpredictable stimulus. Furthermore, the anterior cingulate cortex (ACC), a region implicated in encoding surprise, showed increased spiking and altered neuronal ensemble activity in mice in response to psilocybin. A state of enhanced surprise has been previously associated with schizophrenia. Trait abnormalities in these patients have been linked to dif iculties in accurately predicting sensory input, leading to a state where all perceptions are experienced as surprising. This has downstream effects on event-related potentials (ERP) measurable via electrophysiological methods. The mismatch negativity (MMN) is an ERP component primarily elicited in response to rare or unexpected stimuli in a sequence of standard stimuli. However, in schizophrenic patients, the MMN responses to such oddball stimuli are diminished. The predictive coding framework offers an explanation for these de icits. This framework posits that the fundamental issue in schizophrenia might not be the prediction of sensory input per se, but rather lies in the balance between the precision assigned to prior beliefs and sensory evidence. Simulations can be employed to demonstrate how abnormal levels of sensory precision, whether increased or attenuated, can account for atypical responses to surprising events. In accordance with our hypothesis, some reports showed that psychedelics induced alterations in ERP responses that are in line with a state of surprisal. Speci ically, both LSD and psilocybin could reduce signi icantly the MMN amplitude in response to deviant stimuli in human. Other studies, however, have failed to observe differences. In considering the consequences of inducing a state of synthetic surprise, that is an increase in prediction error and uncertainty, within the predictive coding framework, we can explore the possible outcomes and examine their alignment to the observed effects of psychedelics. These outcomes are shaped by the precision associated to the prediction error. Precision is the inverse of the expected uncertaintyand can be modeled as the weighted historical average of the prediction errors. A consistently enhanced prediction error is expected to increase the expected uncertainty, therefore, reducing the precision afforded to sensory data. The complexity of the situation, however, is further compounded when considering the roles of other neuromodulator in precision weighting. Speci ically, in addition to 5-HT, precision weighting has been linked also to acetylcholine and dopamine. It is conceivable that the precision of the prediction error (PE) might be enhanced via 5-HT independent routes. Precision is anticipated to dynamically affect information processing as follows: Considering the enhanced belief updating, the synthetic surprise model can explain the heightened openness, ego dissolution and some psychosis like symptoms observed in subjects under the in luence of psychedelics. In a normal state of consciousness, our awareness is limited to a speci ic set of accessible priors that aid in explaining incoming sensory information. These priors are continuously updated to align with the sensory input, with each prediction error signaling the need for a new version (prior n → prior n + 1) to be considered. At a theoretical onset of perception, each prior can be associated with a likelihood, and when a prediction error signal reaches the higher layers of processing, new priors are sequentially explored until the most suitable one is identi ied. Under the in luence of psychedelics, the heightened activation of prediction error signals pushes the brain into a more malleable state. This plastic state can access priors that would typically be inaccessible in a normal state. Interestingly, a similar mechanism has been previously proposed to explain the genesis of delusions. This expanded range of accessible priors, upon a persistent prediction error signal, contributes to an increased openness to external ideas and concepts, even those that may appear implausible in a default state. Consistent with this idea, both psilocybin and LSD have been found to enhance the accessibility of remote associations, thereby facilitating the activation of cognitive contents that typically remain latent under normal circumstances. This suggests that psychedelics promote a broader range of cognitive associations and allow for the exploration of novel connections and perspectives. The increased number of accessible priors might underlie the noetic quality characteristic of psychedelic experiences, the profound sensation of accessing direct insights and revelations. Such moments of insight are believed to play a foundational role in how psychedelics reshape beliefs in pathological situations. High PE precision → Enhanced belief updating: arti icially increasing the prediction error would lead to more substantial updates in beliefs, allowing for greater lexibility in incorporating new information and revising prior assumptions. Low PE precision → Strong priors: the brain relies more heavily on existing beliefs due to the reduced reliability of sensory input. This could result in a tendency to perceive and interpret experiences in accordance with pre-existing expectations, possibly leading to perceptual biases or hallucinations.
Strong priors, on the other hand, can explain the hallucinations typical of the psychedelic experience. In this context hallucinations, de ined as perceptions without identi iable external stimuli, are not mere errors in sensory processing but rather a result of the predictive machinery operating on overly precise priors. It is posited that hallucinations arise when priors excessively in luence sensory interpretation, overshadowing actual sensory evidence. This theory can explain how conditioned hallucinations can be elicited by associating stimuli of different modality, a long known phenomenon observed in various studies.
Although many aspects of the implementation of the predictive coding hypothesis are still being investigated, recent years have seen progress in unraveling the underlying circuitry. Within this framework, prediction errors play a fundamental role. They serve as crucial signals that trigger updates in the internal models, allowing the brain to re ine its predictions and minimize the mismatch between expectations and reality. Assuming the existence of prediction error neurons, their fundamental function would involve comparing topdown inputs, which provide predictions of sensory input, with bottom-up sensory-driven inputs in order to discern the difference. One potential mechanism for achieving this comparison is through a subtractive process, which would entail prediction error neurons exhibiting balanced and opposing weights for the two types of input. Consequently, if the bottom-up sensory input is excitatory, the in luence of the top-down input should be inhibitory, and vice versa. This bidirectional interaction ensures a dynamic interplay between top-down predictions and bottom-up sensory signals within prediction error neurons. The predictive processing framework entails two distinct types of prediction error neurons: positive and negative. Positive prediction error neurons perform the subtraction of top-down predictions from sensory input, while negative prediction error neurons subtract sensory input from top-down predictions. If the relative strengths of top-down and bottom-up inputs are opposing within individual neurons, any temporary imbalance between these two sources would give rise to prediction error responses. The suggested mechanism explaining the impact of psychedelics on predictive coding circuitry, though speculative, is based on two recent lines of experimental evidence and an analysis of recent transcriptomic data, which support the presence of 5-HT Rs in SST interneurons. First, studies have underscored the signi icant role played by SST inhibitory interneurons in computing prediction errors. In the context of expected visual low resulting from locomotion, it was shown that SST interneurons in V1 exert inhibitory control over neurons responsible for detecting negative prediction errors. These negative prediction errors occur when the expected visual low, anticipated because of self-initiated locomotion, fails to materialize. Moreover, Sst44 positive SST neurons in the mouse parietal cortex provide an error signal in goal-directed navigationand a subgroup of SST interneurons in mice was also showed to be activated by novelty using 2-photon calcium imaging. Second, there is evidence pointing to the possible activation of SST interneurons by 5-HT Rs. While numerous studies have observed responses to 5-HT R activation consistent with the stimulation of inhibitory cells (as discussed in the Introduction), only a few have precisely identi ied the class of interneurons involved. Notably, a recent study focused on the mouse entorhinal cortex found that SST interneurons are activated by 5-HT, speci ically through the activation of 5-HT R (De. More data is necessary to establish whether SST interneurons in other brain areas are also activates by 5-HT R. Third, analysis of the mouse whole-brain transcriptomics cell type atlas provided the Allen Institute for Brain Science, showed enrichment of 5-HT R RNA in SST cortical neurons. The atlas categorizes neurons in a hierarchical system (cell>cluster>supertype>subclass>class). Four classes account for the vast majority of neurons present in the isocortex (99.62%): 01 IT-ET Glut, 02 NP-CT-L6b Glut, 06 CTX-CGE GABA and 07 CTX-MGE GABA. Medial ganglionic eminence (MGE)-derived GABAergic neurons, including SST interneurons, showed the higher percentage of clusters showing signi icant enrichment in 5-HT R RNA (Fig.). A cell was considered to express 5-HT R RNA if log(CPM)> 3.5 (CPM, counts per million reads), the same stringent threshold used in the original report to classify by neurotransmitter type. We focused on the main cortical inhibitory subclasses by number of cells: Vip, Lamp5, PV and SST (Fig.). Except for Vip neurons, signi icant 5-HT R RNA expression was observed in supertypes of all other subclasses. A 5-HT R enriched supertype (5-HT R+) was de ined as having at least 50% 5-HT R prevalence, i.e. half of the cells belonging to the supertype express 5-HT R RNA. SST 5-HT R+ supertypes were distributed across the entire cortex (Fig.) with a majority of SST 5-HT R+ cells present in the isocortex (Fig.). 5-HT R+ cells were present in all cortical layers (Fig.). L1 shows high proportion of Lamp5 5-HT R+ cells. PV and SST 5-HT R+ cells are present across all other layers, particularly prominently in L2/3 and L5. A stable fraction of SST interneurons across layers expressed 5-HT R RNA (63.76 ± 0.98%). Sst44, the marker used to identify SST interneurons involved in error detection during navigation, was found in a signi icant number (≈40%) of Calb2+ and Hpse+ neurons. Notably, cortical SST 5-HT R+ were also found to express a substantial amount of these two genes (Calb2:35.68%, Hpse:34.80%, Fig.right). We also provide an online visualization tool for the exploration of 5-HT R RNA expression based on the MERFISH dataset of. Considering the variations in 5-HT levels across the sleep-wake cycle, it is reasonable to speculate that the activation of SST interneurons might be in luenced by these luctuations. Hence, we would anticipate a positive correlation between the spiking activity of SST interneurons and the levels of 5-HT across different states of the sleep-wake cycle. Speci ically, we would expect the highest activation of SST interneurons during wakefulness, lower during slow-wave sleep, and lowest during rapid eye movement (REM) sleep. Supporting this notion, a study that directly measured the activity of various neuronal classes across different states found that SST interneurons in the dorsal cortical surface exhibited exactly the expected pattern of activation. Notably, this pattern was not observed in PV interneurons, which displayed similar levels of activation during both REM sleep and wakefulness. These indings align with a positive correlation between the activity of SST interneurons and 5-HT release. Within the previously described predictive coding circuitry, it is hypothesized that the prolonged activation of SST interneurons mediated by 5-HT Rs leads to a paradoxical over-inhibition of negative prediction error neurons. Experimental data shows that L2/3 excitatory neurons, supposed to compute prediction errors, react to visuomotor mismatches exhibiting membrane potential responses ranging from strongly depolarizing to strongly hyperpolarizing. A blanket SST interneurons activation by 5-HT Rs is expected to inluence also L2/3 neurons that physiologically would receive only weak inhibitory input. Consequently, a discrepancy arises between the signaling of positive and negative prediction errors, even in the absence of a signi icant mismatch with sensory input, ultimately resulting in an unbalanced positive prediction error signal (Fig.). In the context of locomotion, a positive prediction error is triggered when an unexpected visual low is experienced in the absence of self-generated movement. Intriguingly, this description resonates with commonly reported visual distortions induced by psychedelics, such as the perception of movement in stationary objects. Other circuit implementation of the predictive coding hypothesis exists, for example predictions can be theoretically implemented via lateral inhibition instead of feedback connections, in another variant the entire computation can be performed by local voltage dynamics in individual neurons. In both cases SST interneurons, being involved in lateral inhibitionand in luencing dendritic compartments of excitatory cells, are expected to have major impact. Furthermore, we cannot disregard the possibility of additional direct effects in primary cortices, potentially involving a modi ication of surround suppression within the cortex through the activation of SST interneurons. In exploring the effects of psychedelics through the lens of the predictive coding framework, we recognize that our hypothesis is constrained by the current limited understanding of the biological mechanisms underlying this framework. In accordance with these limitations, we concentrate on SST neurons, as emerging evidence suggests their involvement in prediction error signaling. It's important to note, however, that 5-HT Rs potentially in luence a broad spectrum of neuronal classes. For instance, cortical Lamp5 and PV neurons exhibit a high concentration of 5-HT R RNA (Fig.). This is supported by studies in mice showing that PV neurons can be activated by 5-HT R. Some excitatory subclasses also show considerable expression, subclasses localized in the claustrum (001 CLA-EPd-CTX Car3 Glut), L5 (005 L5 IT CTX Glut, 006 L4/5 IT CTX Glut), L6 (004 L6 IT CTX Glut) and L6b (027 L6b EPd Glut, 028 L6b/CT ENT Glut, 029 6b CTX Glut) exhibit a prevalence of > 50% (see online visualizer). Another complication not addressed in this review is the functional selectivity of 5-HT Rs.
In recent years, there has been a resurgence of interest in re-evaluating the therapeutic potential of psychedelics as groundbreaking treatments for psychiatric disorders. While this concept may seem novel, it builds upon earlier explorations conducted during the 1950 s and 1960 s, when psychedelics were investigated as potential therapies for depression and alcoholism. Recent experimental results are encouraging towards a potential role for psychedelics in clinical settings, speci ically for the treatment of depression and anxiety-related disorders. Interestingly, the therapeutic effect of psilocybin, in humans, has been found to correlate with various subjective qualities of the psychedelic experience, pointing at the importance of the subjective experience. It is worth mentioning, however, that some studies performed in rodents, have presented contrasting indings. This divergence highlights the necessity for additional research in this ield, particularly considering that animal models of major depression display substantial shortcomings in fully replicating the human condition. The clinical use of psychedelics presents a fascinating avenue for exploring the implications of the synthetic surprise model. The predictive coding framework offers valuable insights into the underlying mechanisms of depression, emphasizing the role of excessively strong and narrow predictions within speci ic neural networks. This framework suggests that depression arises from a "locked-in" brain state, characterized by an insensitivity to prediction errors. This state hinders the ability of the brain to update its internal models, leading to a breakdown in adaptive processing. The model proposes that individuals with depression have a propensity to predominantly expect negative events or experiences, which they subjectively validate by reappraising disconirming evidence, ultimately leading to the establishment of a self-reinforcing negative feedback loop. The underlying cause can be identi ied in the excessive precision assigned to prior beliefs with negative valence as a key contributor to the pathology, accompanied by a decrease in prediction errors. Intriguingly, a theoretical compound that enhances prediction errors seems to logically be a itting candidate for therapeutical interventions in such context. According to our hypothesis psychedelics ful ill exactly this role by enforcing a synthetic surprise affect. We propose that the clinical value of psychedelics might be strictly related to their surprise-inducing property. By extending this logic, other psychoactive compounds might also offer bene icial effects. In fact, when individuals encounter for the irst time the effects of a psychoactive substance, they are likely to be surprised. Accordingly, throughout history a remarkable spectrum of substances have showcased promising therapeutic outcomes: anxiolytics like benzodiazepinesand barbiturates, empathogens like MDMA, psychostimulants like methylphenidate, theophylline, moda inil (Price andand cocaine, depressants like opioids, esmethadone, GHBand cannabis, dissociatives like ketamine, Dextromethorphan, nitrous oxide, Salvia divinorum, and methoxetamine, deliriants like scopolamineand also anesthetics like propofol, sevo luoraneand iso luorane. Notably, these compounds affect virtually all known existing neurotransmitter systems, including acetylcholine, dopamine, GABA, noradrenaline, adenosine, 5-HT, glutamate, GHB, sigma, opioid, and histamine receptors. Particularly puzzling is the recent report of an absence of difference between general anesthesia and ketamine for the treatment of depression in a triple-masked, randomized, human placebo-controlled trial. We propose that the potential antidepressant properties of these compounds may not be attributed primarily to a speci ic pharmacological action, but rather to the subjective experience of encountering something profoundly novel, a phenomenon achievable through alterations in various neuromodulatory pathways. Despite the promising results seen with the aforemen-tioned array of compounds, we are still in search of a truly effective pharmacological solution to depression. One common limitation predicted by our hypothesis is that upon repeated administrations, the novelty effect is expected to wane. In contrast, psychedelics are likely to have a more enduring effect due to their intrinsic ability to directly induce surprise. Nonetheless, the possibility of negative outcomes due to intake of psychedelics should still be kept in mind.
It is of interest to ind an experimental paradigm able to discern the validity of our model. It would be useful to determine whether under psychedelics priors gain the capacity to shape perception, aligning with the strong prior theory and the synthetic surprise model. The tasks developed previously byand, associated with psychedelics could offer valuable glimpses into the interplay between priors and perception under the effect of psychedelics. If the activation of 5-HT R truly plays a role in indexing uncertainty and signaling prediction error, its antagonism is expected to trigger notable psychological alterations. Initial evidence points in this direction, with some data highlighting its involvement in reversal learning in rodents. Prediction errors are essential for the successful execution of a reversal learning task. In line with our hypothesis 5-HT R antagonism signi icantly increases the number of trials necessary to learn the reversal association, causing a pattern of perseverative responding. Interestingly, applying psilocybin in conjunction with ketanserin was showed to further reduce vigilance compared to psilocybin alone. Moreover, blocking 5-HT R did not rescue performance de iciencies in attention induced by psilocybin. It is essential to further investigate this topic in humans to achieve a deeper understanding of the role of 5-HT R in decision-making and cognitive lexibility. Regarding the circuit implementation, it is necessary to precisely map the effects of 5-HT R activation across different classes of excitatory and inhibitory neurons across different cortical areas. While recent data have shown that SST interneurons are activated by 5-HT Rs in the entorhinal cortex, it remains an open question whether this speci ic class of interneurons is activated by 5-HT R across the cortex at a broader scale. Analysis of transcriptomic data points in this direction (Fig.). By systematically mapping the impact of 5-HT Rs activation on diverse neuronal populations within the cortex, we can gain a more comprehensive understanding of the underlying neural mechanisms that mediate the effects of psychedelics and their relevance in shaping cortical activity and cognitive functions. If the therapeutic bene its of psychedelics truly hinge on their ability to induce surprise, it suggests that other interventions inducing unexpected uncertainty might similarly yield positive effects. This hypothesis prompts a deeper exploration into non-pharmacological methods that can elicit similar states. We propose that psychedelics exert their in luence by inducing a state of synthetic surprise, achieved through the arti icial generation of prediction error. Foundation of this model is the proposed role of 5-HT in signaling prediction error. The downstream effect of this increase in prediction error depends on its precision weighting, a process likely mediated by various neuromodulators. In this way, priors can dynamically be strengthened or weakened. Within this model, hallucinations can be explained by the strong prior hypothesis. By incorporating recent discoveries about 5-HT and predictive coding, we offer a comprehensive perspective on how these substances profoundly alter perception. More experiments are necessary to test the proposed circuit implementation based on activation of interneurons by 5-HT Rs. Nevertheless, the synthetic surprise hypothesis puts forth a falsi iable theory that strives to elucidate the effects of these fascinating substances. CRediT authorship contribution statement The table presents a comparison of various studies examining the effects of psychedelic substances on brain activity. EEG (Electroencephalogram) and MEG (Magnetoencephalography) measures were utilized to assess changes in brain oscillatory bands, including Alpha, Beta, Delta, Gamma, and Theta. The 'Effect' column indicates whether the study reported an increase or decrease in the power of the speci ic band, while the 'Location' column denotes whether the effect was widespread across the brain or localized to speci ic regions.
Create a free account to open full-text PDFs.
Cao, D., Yu, J., Wang, H. et al. · Science (2022)
Carhart-Harris, R. L., Giribaldi, B., Watts, R. et al. · New England Journal of Medicine (2021)
Carhart-Harris, R. L. · Current Opinion in Psychiatry (2019)
Carhart-Harris, R. L., Erritzoe, D., Williams, T. et al. · PNAS (2012)
Carhart-Harris, R. L., Kaelen, M., Whalley, M. G. et al. · Psychopharmacology (2014)
Carhart-Harris, R. L., Kaelen, M., Bolstridge, M. et al. · Psychological Medicine (2016)
Carhart-Harris, R. L., Muthukumaraswamy, S., Roseman, L. et al. · PNAS (2016)
Carhart-Harris, R. L., Bolstridge, &. M., Day, C. M. J. et al. · Psychopharmacology (2017)
Carter, O., Burr, D. C., Pettigrew, J. D. et al. · Journal of Cognitive Neuroscience (2006)
Corlett, P. R., Honey, G. D., Krystal, J. H. et al. · Neuropsychopharmacology (2010)
Daws, R. E., Timmermann, C., Giribaldi, B. et al. · Nature Medicine (2022)
Dong, C., Ly, C., Dunlap, L. E. et al. · Cell (2021)
Doss, M. K., Madden, M. B., Gaddis, A. et al. · Brain (2021)
Duerler, P., Brem, S., Fraga-González, G. et al. · Cerebral Cortex (2021)
Dupuis, D. · Frontiers in Psychology (2021)
Dupuis, D. · Transcultural Psychiatry (2021)
Family, N., Vinson, D., Vigliocco, G. et al. · Language Cognition and Neuroscience (2016)
Garcia-Romeu, A., Davis, A. K., Fire Erowid et al. · Journal of Psychopharmacology (2019)
Gattuso, J. J., Perkins, D., Ruffell, S. G. D. et al. · International Journal of Neuropsychopharmacology (2022)
Goodwin, G. M., Aaronson, S. T., Alvarez, O. et al. · New England Journal of Medicine (2022)
Halberstadt, A. L. · Behavioural Brain Research (2014)
Halberstadt, A. L., Geyer, M. A. · Neuropharmacology (2011)
Halberstadt, A. L., Chatha, M., Klein, A. K. et al. · Neuropharmacology (2020)
Hartogsohn, I. · Journal of Psychopharmacology (2016)
Hartogsohn, I. · Drug Science Policy and Law (2017)
Hesselgrave, N., Troppoli, T. A., Wulff, A. B. et al. · PNAS (2021)
Holze, F., Ley, L., Müller, F. et al. · Neuropsychopharmacology (2022)
Holze, F., Gasser, P., Müller, F. et al. · Biological Psychiatry (2023)
Johnson, M. W., Richards, W. A., Griffiths, R. R. · Journal of Psychopharmacology (2008)
Kanen, J. W., Luo, Q., Kandroodi, M. R. et al. · Psychological Medicine (2020)
Kaplan, A. L., Confair, D. N., Kim, K. et al. · Nature (2022)
Kopra, E., Ferris, J. A., Winstock, A. R. et al. · Journal of Psychopharmacology (2022)
Sumiyoshi, T., Kraehenmann, R., Pokorny, D. et al. · Frontiers in Pharmacology (2017)
Kraehenmann, R. ;., Pokorny, D. ;., Vollenweider, L. ;. et al. · Psychopharmacology (2017)
Kuypers, K. P. C., Riba, &. J., De La Fuente Revenga, &. M. et al. · Psychopharmacology (2016)
Kwan, A. C., Olson, D. E., Preller, K. H. et al. · Nature Medicine (2022)
Letheby, C., Gerrans, P. · Neuroscience of Consciousness (2017)
López-Giménez, J. F., González-Maeso, J. · Current Topics in Behavioral Neurosciences (2017)
Maclean, K. A., Johnson, M. W., Griffiths, R. R. · Journal of Psychopharmacology (2011)
Madsen, M. K., Fisher, P. M., Burmester, D. et al. · Neuropsychopharmacology (2019)
Madsen, M. K., Stenbaek, D. S., Arvidsson, A. et al. · European Neuropsychopharmacology (2021)
Moliner, R., Girych, M., Brunello, C. A. et al. · Nature Neuroscience (2023)
Müller, F., Dolder, P. C., Schmidt, A. et al. · NeuroImage (2018)
Perry, C. M., Malina, M. · Psychopharmacology (2021)
Muthukumaraswamy, S. D., Carhart-Harris, R. L., Moran, R. J. et al. · Journal of Neuroscience (2013)
Nagele, P., Duma, A., Kopec, M. et al. · Biological Psychiatry (2015)
Nagele, P., Palanca, B. J., Gott, B. et al. · Science Translational Medicine (2021)
Nichols, D. E., Walter, H. · Pharmacopsychiatry (2020)
Noorani, T., Garcia-Romeu, A., Swift, T. C. et al. · Journal of Psychopharmacology (2018)
Nour, M. R., Evans, J., Nutt, D. J. et al. · Frontiers in Human Neuroscience (2016)
Pallavicini, C., Cavanna, F., Zamberlan, F. et al. · Journal of Psychopharmacology (2021)
Preller, K. H., Herdener, M., Pokorny, T. et al. · Current Biology (2017)
Preller, K. H., Burt, J. B., Adkinson, B. et al. · eLife (2018)
Preller, K. H., Schilbach, L., Pokorny, T. et al. · Journal of Neuroscience (2018)
Schenberg, E. E., Alexandre, J. F. M., Filev, R. et al. · PLOS ONE (2015)
Schmid, Y., Enzler, F., Gasser, P. et al. · Biological Psychiatry (2015)
Spitzer, M., Thimm, M., Hermle, L. et al. · Biological Psychiatry (1996)
Tagliazucchi, E., Roseman, L., Kaelen, M. et al. · Current Biology (2016)
Timmermann, C., Spriggs, M. J., Kaelen, M. et al. · Neuropharmacology (2018)
Timmermann, C., Roseman, L., Schartner, M. et al. · Scientific Reports (2019)
Timmermann, C., Kettner, H., Letheby, C. et al. · Scientific Reports (2021)
Timmermann, C., Watts, R., Dupuis, D. · Transcultural Psychiatry (2022)
Timmermann, C., Roseman, L., Haridas, S. et al. · PNAS (2023)
Ekins, T. G., Brooks, I., Kailasa, S. et al. · Biorxiv (2023)
Vollenweider, F. X., Vollenweider-Scherpenhuyzen, M. F. I., Bäbler, A. et al. · NeuroReport (1998)
Wacker, D., Wang, S., Mccorvy, J. D. et al. · Cell (2017)
Williams, T. M., Davis, A. K., Xin, Y. et al. · Drugs Education Prevention and Policy (2020)