In this meta-analytic synthesis, we examined the acute effect of CBD on human brain activation signal (Crippa et al., 2004, 2011; Borgwardt et al., 2008; Bhattacharyya et al., 2009, 2012, 2018; Fusar-Poli et al., 2009; Winton-Brown et al., 2011; Wilson et al., 2019; Davies et al., 2020; Lawn et al., 2020; O’Neill et al., 2021). Our key analysis included 7 studies in healthy participants (Crippa et al., 2004; Borgwardt et al., 2008; Bhattacharyya et al., 2009, 2012; Fusar-Poli et al., 2009; Winton-Brown et al., 2011; Lawn et al., 2020), 4 in psychosis (Bhattacharyya et al., 2018; Wilson et al., 2019; Davies et al., 2020; O’Neill et al., 2021), and 1 in social anxiety disorder (Crippa et al., 2011). Of these manuscripts, all but two (which used SPECT (Crippa et al., 2004, 2011)) used fMRI. We investigated the effect of a single dose of oral CBD administration (ranging from 400 – 600 mg), compared with placebo, under experimental conditions (1 to 3 hours after administration) on brain activation during an array of cognitive processes using pooled summary data.
When combining data from all studies, we found that CBD modulated the function of 10 (peak) brain regions, with clusters extending to a number of other regions. Within our predicted network of regions (based on the taxonomic definitions proposed by Uddin et al. (2019)), CBD modulated the activation signal relative to placebo in the medial frontoparietal network (attenuation of the hippocampus/ parahippocampal gyrus, augmentation of the superior frontal gyrus, and both attenuation and augmentation of different parts of the middle temporal gyrus), midcingulo-insular network (attenuation of the amygdala) and the pericentral network (attenuation of the supplementary motor area and augmentation of the superior temporal gyrus). Increases in brain signal were also observed in the striatum and cerebellum. Furthermore, we found that CBD modulated activation signal in networks that we had not predicted, including the lateral frontoparietal network (augmentation of the middle frontal gyrus and the inferior temporal gyrus, as well as attenuation in a spatially distinct region of the inferior temporal gyrus) and the dorsal frontoparietal network (attenuation of the post central gyrus) (see Table 2 for coordinates). Contrary to our initial hypothesis, within our main results we found no significant effects in occipital network regions. Our second prediction was that the acute effect of CBD on activation signal across different brain regions will be directly associated with pooled FAAH gene expression data from a set of 6 unrelated healthy volunteers (who did not take part in the neuroimaging studies reported here), as obtained from the Allen Human Brain atlas. We observed an inverse relationship between the effect of CBD on brain activation signal from our main findings with FAAH gene expression –-a proxy measure of local FAAH availability–-but not the other genes of interest.
The findings of this meta-analytic synthesis highlight a general pharmacological effect of CBD in the human brain, which we localised primarily to the medial frontoparietal network. This macro-scale network is proposed to encompass the commonly termed “default mode network” and subsumes the “limbic network”, and while at a more granular level, a mediotemporal subsystem (involved in associative processing and recall) has been identified (Uddin, Yeo and Spreng, 2019). Although there is no current consensus on the broad central functions of the medial frontoparietal network, it has been associated with constructing, phasic binding, and continuous updating of associative representations obtained from goal-states (Uddin, Yeo and Spreng, 2019). It has also been proposed that this network is involved in the generation of predictions (predictive coding) and semantic associations via internal and external salience processing to provide value coding and goal-directed cognition (Bar et al., 2007; Roy, Shohamy and Wager, 2012; Nathan Spreng et al., 2014; Dohmatob, Dumas and Bzdok, 2017). Given that the tasks included in our meta-analysis broadly overlap with the aforenamed processes, such as processing salient stimuli in the MIDT (Wilson et al., 2019) and visual oddball detection (Bhattacharyya et al., 2012) tasks, it makes sense that brain regions within the medial frontoparietal network would be engaged. Therefore, this raises challenges when assessing whether the effects seen here are due to the pharmacological effects of CBD, or simply reflect a task-based neurophysiological response. It is also worth noting the possibility that the engagement of specific brain regions seen within this network, such as the hippocampus/ parahippocampal gyrus, may also reflect the limited cognitive paradigms employed. Although we attempted to observe the domain-general—rather than task-specific—effects of CBD on the brain activation signal, the most common task included in the meta-analysis was the verbal paired associates learning and memory task, which is well known to engage hubs of the medial frontoparietal network, such as the mediotemporal and frontal cortices (Bhattacharyya et al., 2009, 2018; O’Neill et al., 2021). This makes it difficult to distinguish whether the effects of CBD in the brain regions reported here are truly a result of its pharmacological effects, or rather a product of the types of cognitive paradigms employed (which each restrict findings to those spatial regions engaged by the task).
A further consideration for interpreting the main results of this meta-analysis is the heterogenous study group examined, including both healthy participants and those with psychiatric disorders such as psychosis (primarily) and social anxiety disorder. We opted for this analytical approach in an effort to boost power, given the limited number of studies that have so far examined the acute effects of CBD using neuroimaging. However, a key concern is that the effects of CBD may differ in patients relative to healthy control groups, which may be driven by differences in the neural pathology of patient cohorts. Contemporary preclinical models of psychosis suggest that alterations of brain regions within the medial temporal lobe (including the hippocampus, parahippocampus, and amygdala (Cutsuridis and Yoshida, 2017)) may drive subcortical dopamine dysfunction through projections to the striatum and midbrain (Modinos et al., 2015). Furthermore, neuroimaging studies in individuals at clinical high risk for psychosis suggests a relationship between the later onset of psychosis with greater alterations in parahippocampal structure (Mechelli et al., 2011) and function (Allen, Chaddock, et al., 2012; Allen et al., 2016, 2018) and to elevated striatal and midbrain dopamine activity (O. D. Howes et al., 2011; O. Howes et al., 2011; Allen, Luigjes, et al., 2012). This suggests large variation in the regional effects, and concurrent effect-size estimates, of studies included for analysis driven the different population samples (disease vs healthy). Given that the effects of CBD may differ in different population samples this increase in noise may have decreased our sensitivity to detect significant meta-analytic effects due to the thresholding set. To evaluate the extent to which our results may be influenced by psychiatric group differences, we visually compared the overlap between our main findings and those restricted to healthy participants following subgroup analysis. We found an overlap of brain regions including hippocampus and amygdala, suggesting that effects in these regions were not driven exclusively by the psychosis group. Overlap was also identified within the inferior and middle temporal lobes, further highlighting that these effects are likely associated with a pharmacological effect of CBD rather than group variances between healthy and neuropsychiatric subjects.
Our second major finding was the observation of a negative linear relationship between the effect of CBD on brain signal (as indexed by the pooled effect-size estimate) and FAAH gene expression levels (as estimated on the basis of an average from 6 post-mortem brains of healthy individuals obtained from the Allen Human Brain Atlas). This is of interest as previous studies have shown that CBD may enhance endogenous anandamide signalling indirectly, by inhibiting the intracellular degradation of anandamide (Bisogno et al., 2001; Leweke et al., 2012) catalysed by FAAH in rodents (Bisogno et al., 2001; Ligresti et al., 2006; Petrocellis et al., 2011; Leweke et al., 2012; Elmes et al., 2015). However, in contrast to robust findings of FAAH inhibition by CBD in rodents, one study has reported that CBD does not inhibit the enzymatic actions of human FAAH (Elmes et al., 2015). Elmes et al. transfected human FAAH into HeLa cells (Landry et al., 2013), with FABP5 knocked out (Berger et al., 2012; Kaczocha et al., 2012), and measured FAAH hydrolytic activity and [14C]anandamide uptake inhibition using enzyme assays. CBD had no significant effect on anandamide levels and did not modulate the proportion of intracellular anandamide that is hydrolysed following uptake. Elmes and colleagues further identified that CBD did not inhibit anandamide hydrolysis by human FAAH in cell homogenates. These findings suggest that CBD may function by blocking the delivery of anandamide to FAAH but does not affect anandamide hydrolysis by FAAH (Elmes et al., 2015). The conflicting findings considering mode of action of CBD on FAAH may be attributed to rodent and human species specificity (Elmes et al., 2015). Nevertheless, the findings reported here, that the effects of CBD on human brain function are in part inversely related to local FAAH availability, complement independent experimental evidence that CBD has some effect on FAAH across both rodents and humans. Moreover, our findings are consistent with indirect human evidence that CBD significantly increases serum anandamide levels in people with psychosis, which was associated with its concomitant reduction of psychotic symptoms (Leweke et al., 2012). Taken together, while the results of this meta-analysis cannot provide direct evidence on the underlying molecular mechanisms by which CBD exerts its effects, they may suggest that there is a strong case for investigating FAAH as a potential mechanism of action of CBD. It is noteworthy that the absence of a significant relationship between the effect-size estimates and the other genes in the regression model (DRD2, HTR1A and CNR1) should be interpreted with caution primarily because t-statistics offer limited insight into the predictive ability of a variable.
Limitations
Certain limitations should be considered when interpreting our results. The principal limitation is perhaps the heterogeneity across the included studies, particularly in combining healthy, psychosis, and social anxiety disorder participants. We sought to mitigate this limitation by performing subgroup analyses and assessed the influence of individual studies on the main findings by conducting jack-knife leave-one-out sensitivity analysis. This procedure involved looping the analysis each time excluding 1 single study to investigate whether each cluster reported in the main analysis remained significant. Therefore, this step allowed identification of unduly influential studies. Of the 12 studies included in this analysis, 80% of all clusters survived the jack-knife, suggesting stability in the results. To further investigate the influence of heterogeneity, QH statistics were assessed in terms of a chi-square distribution and reported after conversion to standard z values to create a map. The QH map was overlayed on top of the map of the main findings for visual inspection. There were no areas of overlap which suggests that the brain regions reported in the main findings were not affected by heterogeneity.
A further limitation is that, although we included studies that employed distinct fMRI paradigms, significant overlap was present in the participants who completed them. While solutions have been proposed for non-neuroimaging meta-analyses, such as the application of a generalised-weights meta-estimator (Bom and Rachinger, 2020), correction for correlated data in image-based meta-analyses is not trivial. Some have proposed that our approach is appropriate (Turkeltaub et al., 2012) given that the tasks were completely independent in their outcome measure from one another (such as reward and memory processing). Nevertheless, the use of overlapping participants may have increased rates of false positive findings and inflated effect-sizes. A separate study, using the same dataset, compared findings from an activation likelihood estimate meta-analysis with one using a modified algorithm to correct for within-group effects (Turkeltaub et al., 2012). While usual activation likelihood estimate functions by summing the probabilities within a given activation peak, per study, to produce an activation map, the modified algorithm considered only the maximum probability associated with an activation peak reported by each study. Turkeltaub et al. (Turkeltaub et al., 2012) report that although correlated datasets can influence an activation map, there was negligible difference in comparison to the modified algorithm to control for these effects.
Furthermore, there are certain limitations inherent to meta-analytic integration of neuroimaging data which we have divided into two categories, (1) design and (2) analysis. When considering our meta-analytic design, the results are based on summary data from individual studies, as opposed to imaging data obtained from individual participants. Acquiring data from individual participants would involve the same participants conducting multiple cognitive and emotional processing tasks in addition to obtaining baseline receptor data quantified using PET imaging. Although this idealised design would have allowed more direct testing of our hypotheses, collecting this type of dataset in a comparable number of participants as reported here would be both logistically and financially challenging. Therefore, the present meta-analysis provides insight into these questions by capitalising on existing available data, albeit using a less than perfect approach. When considering the second type of neuroimaging limitation related to the analysis technique, our approach used both t-maps and coordinates. The use of coordinates in the analysis, as opposed to t-maps alone, may have increased the risk of bias in our results as coordinates are reported using a family-wise error correction threshold of p<0.05. This stringent threshold is likely to have excluded clusters which, when pooled with other results from other studies, may have produced a significant difference in activity driven by the pharmacological effect of CBD. Nevertheless, we attempted to mitigate this issue by including as many t-maps as possible.
A further limitation of this study is that we used mRNA expression as a way to indirectly estimate receptor density. The Allen Human Brian Atlas provides an indirect measure of receptor density through an index of gene transcriptional activity which is governed by gene translation. This is notable as previous reports have highlighted difference between tissue mRNA and protein levels (Futcher et al., 1999; Gygi et al., 1999; Greenbaum et al., 2003). Moreover, it has been reported that gene expression (transcriptional activity) and protein abundance (translational activity) do not always have a positive correlation (Margineantu et al., 2007; Schwanhüusser et al., 2011). Furthermore, the relationship found between mRNA expression and the effect-size estimate as reported here is only an indirect evidence that complements independent evidence indicating FAAH as a molecular target for CBD and may not reflect a causal association.
Notwithstanding these limitations, the major finding of the current study extends previous evidence on the haemodynamic effects of CBD on regional brain activation signal at the meta-analytic level. We also provide preliminary evidence that suggests a negative relationship between the effect of CBD on brain signal and local FAAH expression. Together, by examining the effects of CBD on brain regions engaged across diverse cognitive and emotional processes (Shine et al., 2019), where its effects may also be related to FAAH availability across the brain, these findings highlight not only the neuropharmacological profile of CBD’s effects across the brain but also link these (albeit indirectly) to one of the key underlying mechanisms by which CBD is proposed to exert its effects. Future studies should combine experimental CBD challenge with fMRI and PET imaging to index its effects on brain function and FAAH respectively in the same individuals to directly examine whether the effects of CBD on FAAH underlie its effects on brain function and behaviour.