Computational protocol: Cerebral blood flow predicts differential neurotransmitter activity

Similar protocols

Protocol publication

[…] A Brodmann area map as included in the MRIcron tool was normalized into the Montreal Neurological Institute space using the Statistical Parametric Mapping (SPM12) normalize function. For putamen, caudate and fusiform gyrus regions the corresponding automated anatomical labeling atlas regions were used. Mean CBF values were extracted for each subject for drug and placebo images from the 41 regions corresponding to Brodmann and subcortical areas covered by the receptor density maps described above. A delta drug minus placebo (change versus placebo) was then computed for each subject and region. A ∆CBF profile for each drug was calculated as an effect size per region for the whole group by computing the average regional drug-induced change across all subjects divided by the standard deviation of the change in the respective region across all subject (Fig.  and Supplementary Figure 4). Major differences in receptor densities and pharmacological effects across individuals have been reported in earlier animal studies that are also likely to apply to human experiments,. Effect size normalizes the mean signal in each region by the variability of the signal. Therewith, it takes this variability into account whilst providing an index of the drug effect. It was therefore chosen to minimize the potential impact of such between subject and region variability in signal to noise but also due to site, scanner and sequence differences across different studies. For Study 5, test retest data of each subject for visit 1 and 2 were randomly assigned to either drug or placebo condition. All group-level ∆CBF imaging data alongside with receptor density maps and activity estimates are provided in Supplementary material. [...] For interpretation of subsequent correlations between receptor density and ∆CBF maps we first aimed to understand if and how the obtained receptor maps provide similar or differential information regarding spatial receptor density distribution. Pearson correlations and coefficients of determination were computed between each pair of receptor density maps to estimate their association strength. Validity of the parametric statistics described below (parametric and normality assumptions for Pearson correlations and multiple linear regression models) was established using permutation statistics and Shapiro-Wilk tests for normality (s. Supplementary methods and results for description and outcomes of those analyses, Figure ). Additionally, to further ensure that the choice of parametric tests did not bias the results we repeated all analyses using Spearman correlation coefficients (Figure ).We next aimed to understand if and how receptor density maps are linked to observed drug-induced ∆CBF profiles (1) separately for each receptor type and (2) whilst controlling for the variance explained by other receptor density maps. For these we performed two types of analyses: (1) simple pair-wise Pearson correlations between each receptor density map and each ∆CBF map and (2) multiple linear regression models using each ∆CBF map as a dependent variable and including all 13 receptor maps as regressors in a single model. To ensure that the outcomes of analysis (1) are not biased by coarse receptor density scale, outliers or distribution assumptions we recomputed all correlations using Spearman correlation coefficients (Figure ). All analyses were implemented using default parameters in IBM SPSS Statistics (Version 23.0, IBM Corp., Armonk, NY). For analysis (1) we report if correlations survive strict Bonferroni correction (accounting for the number of tests performed per compound) and additionally as exploratory findings all correlations surviving an uncorrected p < 0.05. For analysis (2) to reduce data loss due to missing values the interpolated receptor density maps described in Supplement 1 were used. To ensure that the interpolation did not bias the results, this analysis was repeated using the initial receptor density maps excluding regions with missing values (s. Supplementary results). In analysis (2) we test for significance of each overall model and each single receptor density regressor to predict ∆CBF profiles (p < 0.05) whilst controlling for the effects of all other regressors included in the model.We then tested if continuous in vivo receptor density estimates further improve the associations between receptor densities and ∆CBF. For this we computed for all compounds with known dopaminergic mechanism correlation coefficients with DAT density estimates obtained through DAT-SPECT. Similarly, we computed for midazolam being the only GABAergic compound its correlation with flumazenil-based GABAa receptor density estimates. To ensure that the putative outlier region showing a very low GABAa expression (Fig. ) did not bias the results we further repeated the correlation analysis with GABAa using non-parametric Spearman correlation coefficient with all data and after the removing the outlier.Lastly, as several ASL acquisitions were available for some of the compounds, we used those data to estimate the reliability of the obtained ∆CBF to receptor density and affinity mappings: (1) within session data of ASL run 1 and 2 for haloperidol, olanzapine and risperidone and (2) between session data for low and high dose data of risperidone. For both, within session and between session data we then used intra-class correlation coefficients (ICC(C,k)) to assess reliability of established receptor density to ∆CBF profiles (s. Supplement 1 for detailed outcomes of those analyses). […]

Pipeline specifications

Software tools MRIcron, SPM, AAL, SPSS
Applications Miscellaneous, Magnetic resonance imaging
Organisms Homo sapiens