Search

The quantification of big pools of diverse molecules provides important insights on brain function, but is often restricted to a limited number of observations, which impairs integration with other modalities. To resolve this issue, a method allowing for the prediction of mRNA expression in the entire brain based on microarray data provided in the Allen Human Brain Atlas was developed. Microarray data of 3702 samples from 6 brain donors was registered to MNI and cortical surface space using FreeSurfer. For each of 18,686 genes, spatial dependence of transcription was assessed using variogram modelling. Variogram models were employed in Gaussian process regression to calculate best linear unbiased predictions for gene expression at all locations represented in well-established imaging atlases for cortex, subcortical structures and cerebellum. For validation, predicted whole-brain transcription of the HTR1A gene was correlated with [carbonyl-11C]WAY-100635 positron emission tomography data collected from 30 healthy subjects. Prediction results showed minimal bias ranging within ±0.016 (cortical surface), ±0.12 (subcortical regions) and ±0.14 (cerebellum) in units of log2 expression intensity for all genes. Across genes, the correlation of predicted and observed mRNA expression in leave-one-out cross-validation correlated with the strength of spatial dependence (cortical surface: r = 0.91, subcortical regions: r = 0.85, cerebellum: r = 0.84). 816 out of 18,686 genes exhibited a high spatial dependence accounting for more than 50% of variance in the difference of gene expression on the cortical surface. In subcortical regions and cerebellum, different sets of genes were implicated by high spatially structured variability. For the serotonin 1A receptor, correlation between PET binding potentials and predicted comprehensive mRNA expression was markedly higher (Spearman ρ = 0.72 for cortical surface, ρ = 0.84 for subcortical regions) than correlation of PET and discrete samples only (ρ = 0.55 and ρ = 0.63, respectively). Prediction of mRNA expression in the entire human brain allows for intuitive visualization of gene transcription and seamless integration in multimodal analysis without bias arising from non-uniform distribution of available samples. Extension of this methodology promises to facilitate translation of omics research and enable investigation of human brain function at a systems level.

Neural plasticity is a complex process dependent on neurochemical underpinnings. Next to the glutamatergic system which contributes to memory formation via long-term potentiation (LTP) and long-term depression (LTD), the main inhibitory neurotransmitter, GABA is crucially involved in neuroplastic processes. Hence, we investigated changes in glutamate and GABA levels in the brain in healthy participants performing an associative learning paradigm. Twenty healthy participants (10 female, 25 ± 5 years) underwent paired multi-voxel magnetic resonance spectroscopy imaging before and after completing 21 days of a facial associative learning paradigm in a longitudinal study design. Changes of GABA and glutamate were compared to retrieval success in the hippocampus, insula and thalamus. No changes in GABA and glutamate concentration were found after 21 days of associative learning. However, baseline hippocampal GABA levels were significantly correlated with initial retrieval success (pcor = 0.013, r = 0.690). In contrast to the thalamus and insula (pcor>0.1), higher baseline GABA levels in the hippocampus were associated with better retrieval performance in an associative learning paradigm. Therefore, our findings support the importance of hippocampal GABA levels in memory formation in the human brain in vivo.