Toward Mapping Neurobehavioral Heterogeneity of Psychedelic Neurobiology in Humans

,

nature of these substances, a precision medicine approach is essential to map the neural signals related to clinical efficacy to identify patients who can maximally benefit from this treatment. Recent studies have shown that bridging the gap between pharmacology, systems-level neural response in humans, and individual experience is possible for psychedelic substances, therefore paving the way for a precision neuropsychiatric therapeutic development. Specifically, it has been shown that the integration of brain-wide positron emission tomography or transcriptomic data, i.e., receptor distribution for the serotonin 2A receptor, with computational neuroimaging methods can simulate the effect of psychedelics on the human brain. These novel computational psychiatry approaches allow for modeling interindividual differences in neural as well as subjective effects of psychedelic substances. Collectively, this review provides a deep dive into psychedelic pharmaconeuroimaging studies with a core focus on how recent computational psychiatry advances in biophysically based circuit modeling can be leveraged to predict individual responses. Finally, we emphasize the importance of human pharmacological neuroimaging for the continued precision therapeutic development of psychedelics.
https://doi.org/10.1016/j.biopsych.2022. 10.021 Classic psychedelics are substances that acutely induce an altered state of consciousness and act primarily as partial or full agonists on the serotonin 2A receptor (5-HT 2A R) localized on apical dendrites of layers IV-V pyramidal neurons (1)(2)(3). In western medicine, psychedelics were first investigated for their psychotomimetic effects: they were used to experimentally induce a temporary state of consciousness mimicking certain aspects of mental illnesses, in particular psychosis (4). This line of research has greatly advanced our understanding of the neurobiological and neuropharmacological basis of psychiatric disorders and continues to reveal important insights for drug development (4). The discovery of the psychoactive effects of lysergic acid diethylamide (LSD) in 1943 then catalyzed the second branch of psychedelic research, testing these molecules in the treatment of psychiatric disorders (5). While these early clinical trials often suffer from methodological limitations, their results point to clinical efficacy across multiple indications (6).
Psychedelics are being investigated in humans after a hiatus of several decades, which was caused by the decision to place these substances in Schedule I in 1970 in the United States and subsequently in many other western countries (7). Although more data are needed and critical questions regarding dose and administration frequency still need to be answered, an increasing number of modern studies show promising safety, tolerability, and efficacy data in various patient populations (8)(9)(10)(11)(12)(13). This has given rise to the therapeutic framework of psychedelic-assisted therapy. In these studies, psilocybin, LSD, and ayahuasca provide fast-acting and longlasting symptom relief after only 1 or 2 administrations when embedded in a psychotherapeutic program. However, the acute and delayed neural mechanisms conferring this effect remain largely unknown.
The rapid growth of neuroimaging technology has greatly advanced our understanding of psychedelics' acute systemlevel effects and the concurrent changes in behavior (7). However, given the highly psychoactive nature of these substances, a precision approach is badly needed to identify person-specific neural and behavioral responses and, in turn, patients who can maximally benefit. To achieve personalized precision, the key knowledge gap involves mapping withinand between-individual neurobehavioral variation of acute and long-term psychedelic effects to pinpoint the neural targets that reflect specific symptom changes in clinical populations.
To advance psychedelic-assisted precision therapy, this review provides a deep dive into psychedelic pharmacological neuroimaging studies. We focus on recent advances in computational psychiatry and neuroimaging analytics that can inform neural system-level targets for predicting personalized response patterns. We aim to provide an integrative view of how to move the field toward precision pharmaco-functional magnetic resonance imaging (fMRI) for iterative and rational refinement of psychedelic-assisted therapy as a viable tool in psychiatry. We describe the limitations of currently available datasets and provide a strategic roadmap for future study design and analysis that can close current knowledge gaps.

PSYCHEDELICS INDUCE HETEROGENEOUS INDIVIDUAL RESPONSES
Psychedelics induce a wide range of subjective effects. Participants may experience not only visual alterations, changes in bodily perception, feelings of blissfulness, and insightfulness, but also, at times anxiety (14). This acute subjective response is highly variable between individuals but also across administrations within the same individual (15). This observation is vital, especially for psychedelics' clinical applications, as 1) the acute experience may be related to treatment response and could therefore explain some of the heterogeneity in treatment outcome (16); 2) the mapping of such acute effects onto neural targets, especially in relation to clinical benefits, remains unknown; 3) it is unclear whether the full acute subjective experience is needed in each person to provide neural target engagement, especially because it is challenging to prepare patients for the experience and some patients may only tolerate lower doses (17); and 4) studies have consistently reported that the acute psychedelic experience can be shaped by the individual's mindset and environment (set and setting) and the dose (15,18). That said, it is becoming increasingly clear that variation in individual biological factors contributes significantly to the heterogeneity in pharmacological and subjective responses to psilocybin and LSD (19)(20)(21). In the last decade, neuroimaging studies have shaped our understanding of the neurobiological systems-level mechanism of action. Leveraging these technologies to understand the heterogeneity in subjective and clinical responses to psychedelic substances will pave the way for precision medicine within the framework of psychedelic-assisted therapy.

EMERGING NEUROIMAGING EVIDENCE PINPOINTS NEURAL SYSTEM EFFECTS OF PSYCHEDELICS
The functional neuroimaging literature on psychedelic effects can be broadly grouped into task-evoked studies and restingstate studies that map pharmacological effects on spontaneous fluctuations of the blood oxygen level-dependent signal. Both methods have clear advantages and disadvantages (22). Here we focus on resting-state fMRI (rs-fMRI) as it allows for 1) the quantification of acute neural system effects of psychedelics in the absence of any given task that may directly impact the intrinsic subjective experience and 2) the mapping of both within-and between-subject acute neural effects in a task-unconstrained manner. Such rs-fMRI effects can then be quantitatively related to the acute and delayed subjective experience of psychedelics, which has key implications for understanding therapeutic effects.

INFORMING MECHANISM OF NEURAL SYSTEM EFFECTS OF PSYCHEDELICS
Importantly, neuroimaging maps that capture the effect of pharmacological manipulation in functional neural systems do not inherently provide information on the mechanisms underlying these effects. It is therefore necessary to validate neuroimaging features against mechanistic information at the genetic, molecular, and circuit levels. A powerful approach to bridging the gap is through the complementary use of computational models and mechanistically informative neural features. In particular, whole-brain biophysically based neural circuit models have been used to simulate large-scale neural dynamics and patterns of network-level functional connectivity observed in fMRI, both with and without pharmacological perturbation. These whole-brain models typically rely on the mean activity and variance of smaller local populations of excitatory and inhibitory neurons, which interact via long-range connections constrained by structural connectivity data ( Figure 1A). Such models can account for the impact of spatially selective and differential changes in interareal interactions, structural connectivity, and local physiological circuit properties on emergent functional connectivity to capture the effect of pharmacological manipulation at the whole-brain level (43). Furthermore, the activity of local neuronal populations is strongly affected by neuromodulators such as serotonin (44)(45)(46)(47). Independent molecular-level information, such as gene expression or receptor density maps, can inform the microcircuit properties and large-scale neural dynamics captured by whole-brain models, thus providing insight into the molecular-level generative mechanisms of observed pharmacological neuroimaging effects.
In one example, Deco et al. (48) leveraged receptor density maps of the 5-HT 2A R from human positron emission tomography data to mechanistically explain the effects of LSD-an agonist on various serotonin, dopamine, and adrenergic receptors (49,50)-on functional neural connectivity in healthy participants. Specifically, a high-resolution 5-HT 2A R density map was used to modulate the firing rates of neuronal pools in each region of a whole-brain dynamic mean field model. Animal studies have shown that serotonin has both excitatory and inhibitory effects on neuronal activity (51)(52)(53). Importantly, scaling the gain function of the model in each region based on 5-HT 2A R density resulted in an optimal fit with functional connectivity under LSD, but not placebo ( Figure 1B). Furthermore, the model fit using 5-HT 2A R outperforms models Figure 1. Combining pharmaco-fMRI, transcriptomics, and computational modeling approaches to inform the mechanism of psychedelic effects. (A) A biophysical model can be used to simulate the large-scale dynamics of the brain. In the large-scale network model, each node represents a local microcircuit with recurrently coupled excitatory and inhibitory neuronal populations, and nodes interact with each other through structured long-range excitatory projections constrained by diffusion MRI-derived structural connectivity. The effect of a psychedelic (e.g., LSD) can be modeled as a modulation of neural gain due to receptor agonism, with the strength of modulation informed by the regional expression level of the gene encoding the receptor [e.g., the 5-HT 2A Rcoding gene HTR2A (110)] or the PET imaging-derived receptor density [e.g., of 5-HT 2A receptors (50)]. Baseline (i.e., placebo) activity is not modulated by receptor expression/density. The simulated synaptic activity in each node can then be transformed to a simulated blood oxygen level-dependent signal. Adapted from Burt et al. (110). (B) Using such a model, Deco et al. (48) showed that scaling the excitatory gain function in each region with PET-derived 5-HT 2A R density revealed an optimal fit with functional connectivity under the LSD (minimum of blue line), but not placebo (green line), condition (50). (C) In further support of the role of 5-HT 2A Rs in LSD's pharmacology, the change in GBC induced by LSD (vs. Pla) is highly spatially correlated with the expression topography of HTR2A, the gene which encodes the 5-HT 2A R protein (36). Here DGBC denotes the change in GBC between the LSD and Pla conditions. (D) Gene expression maps of HTR1A and HTR2A also correlate with Psi-induced changes in GBC; furthermore, the strength of these correlations increases over the first 70 minutes post-Psi (19). DGBC denotes the change in GBC between the Psi and Pla conditions. similarly modulated by other serotonin receptor density maps. Overall, these results suggest that neural activity under LSD is strongly dependent on 5-HT 2A R, specifically the spatial distribution of 5-HT 2A R in the brain.
More recently, Lawn et al. (54) used a receptor-enriched analysis of functional connectivity by targets analysis (55) to leverage positron emission tomography maps of serotonergic and dopaminergic receptor densities as spatial regressors against subject-specific resting-state data. The resulting time series reflected the dominant fluctuations, which covary with each of the neurotransmitter maps, and were used in a second regression to compute spatial maps of the subject-specific receptor-enriched blood oxygen level-dependent response, for each receptor, under LSD and placebo conditions. Finally, the serotonergic receptor-enriched maps were shown to predominantly relate to LSD's perceptual effects (e.g., hallucinations), while dopaminergic systems were more closely linked to aspects of perceived selfhood and cognition.
Another approach to mechanistically inform neuroimaging results is through the integration of gene expression maps. Recent work used high-throughput transcriptomic data from the Allen Human Brain Atlas to map spatially distributed gene expression patterns in the human brain (56). Importantly, cerebral functional alterations induced by LSD and psilocybin align with the spatial distribution of key receptors targeted by these substances (19,36). Leveraging gene expression maps against pharmaco-fMRI can thus provide critical molecular benchmarking for informing functional neural target engagement.
For instance, relative to placebo, LSD induces reduced resting-state functional global brain connectivity (GBC) in associative regions and elevated GBC in sensorimotor networks (36). Critically, the map of LSD-induced changes in GBC closely resembles the cortical spatial expression pattern of the 5-HT 2A R-coding gene, HTR2A ( Figure 1C). This effect is preferential for HTR2A compared with other candidate genes of interest, including genes coding for other serotonin receptor subtypes HTR1A, HTR2C, and HTR7, and dopamine receptorcoding genes DRD1 and DRD2. Consistent with the modeling results of Deco et al. (48), this implicates the role of 5-HT 2A R in LSD neuropharmacology. Psilocybin induces similar changes in neural dysconnectivity (19), mediated by serotonergic targets, including 5-HT 2A and 5-HT 1A receptors (57)(58)(59). Interestingly, the neural and subjective effects of psilocybin increase linearly over time, peaking at 70 minutes postadministration. Comparing the time-dependent changes in GBC against cortical gene expression maps reveals that psilocybin-induced changes correlate strongly negatively with HTR1A expression and strongly positively with HTR2A expression ( Figure 1D). The strengths of the two correlation coefficients, as well as the difference between them, also increase over time and hold at the single-subject level, indicating that the relationship between gene expression maps and GBC becomes more robust along with the neural and experiential effects. These results collectively implicate the 5-HT 2A and 5-HT 1A receptors in the neuropharmacology of psilocybin and demonstrate that integrating gene expression with pharmaco-fMRI can inform novel targets for developing personalizable psychedelic therapeutics. Of note, care should be taken when inferring receptor density levels based on gene expression levels. RNA expression levels and protein abundance have been shown to poorly correlate in the human brain, and protein levels are heavily influenced by regional differences in cytoarchitecture, development, and function (60). However, the densities of 5-HT 1A and 5-HT 2A receptors in the human cortex, as measured by in vivo molecular imaging, show fair to strong correspondence with HTR1A and HTR2A gene expression levels (61).
In addition, combining pharmacological neuroimaging and transcriptomics-informed computational modeling can serve to capture psychedelic-induced experientially relevant modes of neural variation at the individual level. Burt et al. (43) combined the HTR2A gene expression map with biophysically based circuit modeling to simulate the acute neuromodulatory effects of LSD on large-scale cortical spatiotemporal dynamics. Similarly to Deco et al. (48), the effect of LSD on microcircuits was modeled as a gain modulation of 5-HT 2A R agonism on excitatory neurons. Here, the level of 5-HT 2A R modulation in each region was scaled by the expression level of HTR2A in that region. The model was used to simulate GBC maps both with and without HTR2A modulation, for the LSD and placebo conditions, and the difference was taken to compute a simulated DGBC map which was quantitatively similar to the empirical LSD versus placebo map computed from resting-state pharmaco-fMRI (36) ( Figure 1E). These findings further indicate that the functional neural effects of LSD are due to 5-HT 2A R-mediated modulation of pyramidal neuron activity, consistent with the literature (62,63).
Furthermore, the HTR2A-modulated model captured patterns of individual differences in DGBC that tracked the degree of altered states of consciousness experienced by the participants. The authors fit the model to each of the 24 subjects' individual-level data and demonstrated that the principal component of both the empirical and model individual-level DGBC maps were quantitatively similar to experientially relevant modes of neural variation. This suggests that the HTR2A expression-modulated model can capture DGBC patterns that predict an individual participant's experience in response to LSD. By combining transcriptomic maps with biophysical modeling and pharmaco-fMRI, these results demonstrate the ability to link specific molecular-level mechanisms induced by acute psychedelic agents (i.e., 5-HT 2A R modulation of excitatory pyramidal cells via LSD) to observable systems-level changes in neural function. Even more importantly, they illustrate how the integration of these methods can inform precision medicine approaches within the novel framework of psychedelics-assisted precision therapy. Given that these models predict individual-level psychedelic-induced changes, in the future they may be extended to provide information regarding likelihood of treatment response prior to substance administration. This also holds promise for advancing drug development and precision medicine in psychiatry beyond psychedelics. Future biophysically based neural circuit models may also benefit from concurrent multimodal imaging strategies, such as simultaneous electroencephalography (EEG)/ fMRI or magnetic resonance spectroscopy/fMRI (29,38,(64)(65)(66). Simultaneous EEG/fMRI may allow for the improved modulation of biophysically based neural circuit models: EEG's high temporal resolution combined with fMRI's high spatial resolution may better capture the rapid timescale involved in neuronal excitation and inhibition. Additionally, magnetic resonance spectroscopy can inform changes in neurotransmitter levels in vivo and has been shown to be sensitive to psychedelic-induced changes.

DEVELOPING REPRODUCIBLE NEUROBEHAVIORAL MODELS OF PSYCHEDELIC EFFECTS
The individual variation in both neural and experiential psychoactive responses shown by Burt et al. (43) further highlights the need to stably and reproducibly map subjective behavioral features to neural measures at the single-subject level. Mapping the relationships between behavioral and neural changes in individual subjects is critical for understanding how psychedelics affect different neural mechanisms in different people, which may in turn underlie their specific symptoms (Figure 2), and ultimately for the development of novel psychiatric treatments.
Recently, with the help of large-scale, transdiagnostic datasets from multisite consortia (67-73), promising efforts have been made to derive reliable and reproducible frameworks of neurobehavioral variation in health and across psychiatric spectra. This has enabled researchers to explore both univariate and multivariate brain-behavioral relationships across multiple symptom dimensions (74)(75)(76)(77)(78)(79)(80)(81). These studies differentiate from brain-wide associations using traditional composite clinical scores, which may not optimally parse neurobehavioral heterogeneity in psychiatric conditions (82)(83)(84), and move toward quantitatively informed precision treatment frameworks.
In one study, the authors derived a stable and replicable symptom-neural mapping using dimensionality-reduced behavioral axes in a transdiagnostic psychosis spectrum sample (67,78). They then quantified the derived symptom-neural maps against independent pharmacological response GBC maps from pharmaco-fMRI studies using LSD (36) and ketamine (85). Lastly, they demonstrated a proof-of-principle framework for individualized treatment selection. By projecting neural and symptom data from an independent replication dataset onto the dimensionality-reduced symptom-neural space and using the relationships with LSD and ketamine response maps, the authors differentially selected between two candidate mechanisms (i.e., serotoninergic vs. glutamatergic) for individual patients that may target their specific neural and symptom changes (78) ( Figure 2B). This illustrates a path toward personalized clinical treatment indication, benchmarking selected single-subject data against independent pharmaco-fMRI maps via a stable neurobehavioral mapping.
Critically, multivariate approaches such as canonical correlation analysis can simultaneously solve for maximal covariation between neural and behavioral feature sets, thus  to study the neural mechanisms underlying subjective behavioral and experiential effects of psychedelic drugs such as LSD. Here, we highlight 2 participants from a recent study (36) who displayed highly heterogeneous subjective experiential effects of LSD [as measured by the 5D-ASC scale (115)]. These can be related to the individual neural effects (e.g., global brain connectivity) of the individual participants. In turn, the neural effects can be related to mechanistic information, such as gene expression maps of specific receptors, which can identify candidate mechanisms to target for novel drug development. (B) Pharmaco-fMRI can also be used to inform clinical decisions (85). In this proof-of-principle workflow, the distinct symptoms of these 2 patients can be linked to their individual neural presentations and then related to targetable mechanistic maps [such as maps LSD (36) and ketamine response], which can in turn be used to inform the mechanisms that may be optimal therapeutic targets for each of these patients. 5D-ASC, 5-Dimensional Altered States of Consciousness Questionnaire; fMRI, functional magnetic resonance imaging; LSD, lysergic acid diethylamide. testing the possibility that neural alterations in psychiatric conditions may be better captured by a weighted combination of symptoms. However, such approaches are also prone to overfitting and thus may fail to replicate (86,87). In most cases of canonical correlation analysis, the number of observations (i.e., subjects) needs to be at least one order of magnitude greater than the total number of variables to detect a true effect (88). Furthermore, reproducible brain-wide symptom-neural association results may require thousands of individuals for sufficiently powered analyses, in both univariate and multivariate approaches (87), even with a typical prior dimensionality reduction of the total number of features into the hundreds (77,78,82). Given that psychedelic pharmaco-fMRI studies currently tend to have limited sample sizes (23), larger studies are needed for the reproducible mapping of brain-behavioral associations and individual differences. While these results emphasize the importance of establishing reproducibility in neurobehavioral models, these approaches remain promising and powerful for deriving maximally covarying neurobehavioral relationships.

IMPLICATIONS OF INDIVIDUALLY PRECISE THERAPEUTIC EFFECTS OF PSYCHEDELICS
Symptom-neural mapping is particularly important for understanding the clinical response to psychedelic-assisted therapy. Several studies show preliminary evidence that psilocybin, LSD, and ayahuasca have therapeutic potential in psychiatric disorders such as anxiety associated with life-threatening diseases (12,16,89,90), depression (8,10,13,91), and addiction (9,92,93). So far, most clinical trials have tested the therapeutic effects of psilocybin, while evidence for LSD and ayahuasca is still scarce. While the results of these initial modern studies are promising, larger, well-controlled clinical trials still need to be conducted. Furthermore, indications, molecules, doses, administration frequencies, and psychotherapeutic approaches differ between these studies, and the influence of these parameters has not yet been explored.
Moreover, the precise dose-dependent therapeutic effects still need to be separated from acute subjective effects. Microdosing-the repeated use of very low subhallucinogenic doses of psychedelics-provides a use case where acute subjective effects could perhaps be dissociated from neural changes. To date, studies investigating the beneficial effects of microdoses have found mixed results (94): some studies found no significant benefit relative to placebo or effects that could be related to breaking the blind in the placebo condition (95-100), while others suggest dose-dependent effects on mood, cognition, and pain perception (101,102). However, microdose studies have found significant changes at the brain level (35,103,104) and indirect markers of neuroplastic potential (105), which may confer therapeutic benefits. Crucially, the neural mechanisms of the potential therapeutic effects of psychedelics remain unknown. Further research is needed to 1) distinguish alterations in circuits that contribute to clinical improvements from those associated with side effects; 2) identify whether key individual differences impact treatment response and whether biobehavioral markers could thus facilitate treatment selection; and 3) disentangle the role of expectancy effects and of psychotherapy from the pure pharmacological effect of psychedelics (106).
To date, only 3 cohorts of patients with depression treated with psilocybin have been explored using fMRI (26,66,(107)(108)(109). It is therefore unclear whether effects observed in healthy participants translate to clinical populations. Doss et al. (66) showed that psilocybin-assisted therapy increased dynamic functional connectivity between the anterior cingulate cortex and the posterior cingulate cortex. Meanwhile, Daws et al. (107) found that psilocybin-assisted therapy decreased brain modularity, increased integration between the default mode network and both the salience and the executive networks, and decreased default mode network recruitment. However, a previously published analysis of the same data has reported conflicting results regarding withindefault mode network connectivity, highlighting that results depend on the analytic methods employed and that a clearer mechanistic understanding is needed (26).
Pharmaco-fMRI studies in clinical populations can address these questions and improve comprehension of psychiatric illnesses themselves by generating symptom-neural maps of pathologies for which psychedelic-assisted therapy may be indicated. Indeed, mechanistic studies in healthy populations have provided substantial information on the neural systemlevel effects of psychedelics (19,36,48,110,111). Therefore, a key next step is studying how the neural effects of psychedelics in clinical populations may be linked to 5-HT receptor expression and functioning (112,113), clinical features, and therapeutic responses (78). Such studies may also help to understand the role of psychotherapy, molecule specificities, and how their effects relate to dose and time (19,36,110,103,114). Indeed, the use of titrated doses of psychedelics and the study of their time-dependent effects may provide the means to disentangle psychotherapeutic from pharmacological neuronal changes. Finally, treatments could be optimized in terms of molecule, dose, and administration frequency to target individualized clinical characteristics. Importantly, the effects of concomitant medications, especially those interacting with the serotonergic system, and comorbid conditions also need to be integrated. Such an optimized strategy may tremendously enhance the efficacy of psychedelic-assisted treatments and increase its accessibility (e.g., by providing an option for patients who may not be able or want to be treated with high doses).

STRATEGIC ROADMAP FOR FUTURE OF PSYCHEDELIC NEUROBEHAVIORAL MAPPING
While our understanding of the neurobiological mechanisms of psychedelics has evolved in recent years, most pharmacological neuroimaging studies still suffer from limitations that leave important knowledge gaps for the development of a precision medicine framework. Here, we outline these open questions and how they can be addressed.
1. Most studies have focused on investigating the effects of a single psychedelic substance. We therefore lack crosspharmacology comparisons in the same individuals, and the specificity of drug-induced effects remains unclear. In addition, it is unlikely participants are blinded to the pharmacological condition in placebo-controlled trials, and so expectancy may influence the observed effects. To address this, future studies are needed which include one or more psychoactive control substances. 2. The first proof-of-concept trials in patient populations lacked any mechanistic outcomes (9,16,92). While most trials being conducted now include pre-and postadministration neuroimaging, they still focus on one specific indication. The lack of direct comparison between patient populations is hindering the search for transdiagnostic clinical mechanisms. Future studies should consider including multiple indication groups treated and assessed with the same protocol. 3. Dose-response studies are rare with respect to biomarkers as well as clinical outcomes. Additionally, the potential therapeutic effects of repeated low doses (so-called microdoses) in patients have not yet been investigated. It is conceivable that different patients may benefit from different doses and dosing regimens. To map this, future studies need to start incorporating multiple doses in combination with mechanistic outcome measures. 4. Most mechanistic studies have acquired fMRI data only during the acute effects of psychedelics. This makes it impossible to study long-term changes in the brain's response. As these long-term changes and the variability thereof is important for clinical outcomes, future studies should include follow-up measurements. In addition, the effects of repeated administrations-a common feature in clinical trials-on the brain are currently unknown. Investigating the acute and delayed effects of multiple administrations using longitudinal studies may help to inform the benefit and optimal delay between administrations and shed light on intraindividual variability in response to psychedelics. 5. Pharmaco-fMRI studies with psychedelics suffer from low sample sizes. Most studies have been conducted with 10-20 participants; the largest original within-subjects dataset includes 25 (114). This severely limits individual difference analyses, and larger studies are greatly needed. 6. There is great variability in data collection and analysis methods, limiting the comparability between datasets. Studies have used different in-scanner paradigms (e.g., collecting resting-state scans with eyes open or eyes closed), scanning sequences, and analysis strategies. In addition, measures to assess subjective effects are often crude, lack temporal resolution, and vary between studies. A unified approach to data collection and analysis will facilitate the development of predictive biomarkers (23). 7. Few studies have taken advantage of concurrent multimodal imaging strategies such as simultaneous EEG/fMRI or magnetic resonance spectroscopy/fMRI (29,38,(64)(65)(66). Given the different spatial and temporal resolutions of these techniques, concurrent data collection can reveal response patterns that are not detectable by a single modality alone. In addition, peripheral biomarkers, such as blood drug concentration levels, may explain interindividual heterogeneity and should be taken into consideration.

CONCLUSIONS
Investigating the effects of psychedelics using neuroimaging has substantially advanced our understanding of their neurobiological mechanism of action. However, major knowledge gaps remain. In particular, it is important to understand variation in subjective as well as neurobiological effects to successfully develop a psychedelics-assisted precision therapy framework. As outlined in this article, recent studies have shown that bridging the gap between systems-level neuronal effects and subjective experiences is possible using biophysically based circuit modeling and pharmacological neuroimaging. Leveraging these methodological advances in combination with strategically closing current knowledge gaps may therefore pave the way for the development of predictive biomarkers that allow the treatment of the right patient with the right substance and the right dose. All authors wrote and revised the manuscript. KHP is currently an employee of Boehringer Ingelheim GmbH & Co KG. AA holds equity and is a member of the Technology Advisory Board for Neumora Therapeutics, Inc. AA is a cofounder, serves as a member of the Board of Directors, as a scientific adviser, and holds equity in Manifest Technologies, Inc. JDM holds equity and is a member of the Technology Advisory Board for Neumora Therapeutics, Inc. JDM is a cofounder, serves as a member of the Board of Directors, as a scientific adviser, and holds equity in Manifest Technologies, Inc. LB has participated in a Medical Education Steering Committee for Janssen and has received compensation. JLJ has previously worked for Neumora (formerly Black-Thorn Therapeutics), is currently an employee of Manifest Technologies, and is a coinventor on the following patent: Anticevic A, Murray JD, Ji JL: Systems and Methods for NeuroBehavioral Relationships in Dimensional Geometric Embedding, PCT International Application No. PCT/US2119/ 022110, filed Mar 13, 2019. All other authors report no biomedical financial interests or potential conflicts of interest.