Event-related functional magnetic resonance spectroscopy

Abstract

Proton-Magnetic Resonance Spectroscopy (MRS) is a non-invasive brain imaging technique used to measure the concentration of different neurochemicals. "Single-voxel" MRS data is typically acquired across several minutes, before individual transients are averaged through time to give a measurement of neurochemical concentrations. However, this approach is not sensitive to more rapid temporal dynamics of neurochemicals, including those that reflect functional changes in neural computation relevant to perception, cognition, motor control and ultimately behaviour. In this review we discuss recent advances in functional MRS (fMRS) that now allow us to obtain event-related measures of neurochemicals. Event-related fMRS involves presenting different experimental conditions as a series of trials that are intermixed. Critically, this approach allows spectra to be acquired at a time resolution in the order of seconds. Here we provide a comprehensive user guide for event-related task designs, choice of MRS sequence, analysis pipelines, and appropriate interpretation of event-related fMRS data. We raise various technical considerations by examining protocols used to quantify dynamic changes in GABA, the primary inhibitory neurotransmitter in the brain. Overall, we propose that although more data is needed, event-related fMRS can be used to measure dynamic changes in neurochemicals at a temporal resolution relevant to computations that support human cognition and behaviour.

Figure 1. Blocked versus event-related functional MRS task design

a-b. Schematic showing difference between ‘blocked’ and ‘event-related’ MRS task designs. Vertical lines represent trials. Shaded areas indicate time periods assigned to a particular condition. Two task conditions are included, A (blue) and B (green). In the ‘blocked’ design, multiple trials within each condition are presented consecutively, before there is a switch to the alternative condition. In the ‘event-related’ design, trials across the two conditions are presented in a random order. In both blocked and event-related designs, a jitter can be included between trials to optimise task design and minimise expectation effects. c. Schematic showing how MRS data acquired using both ‘blocked’ and ‘event-related’ can be analysed by taking all spectra from a given condition and applying processing to obtain an average spectra per condition (for example, (Koolschijn et al., 2021)). Alternatively, MRS data acquired using event-related design may be analysed using a General Linear Model based analysis applied to the time series (Apšvalka et al., 2015).

Figure 2. Data acquired using event-related fMRS

a. Event-related fMRS used to detect changes in lactate in primary visual cortex in response to 1 s of visual stimulation with acquisition delay of 0 s, 3 s, 5 s, 8 s and 12 s. Plotted as a percentage with respect to rest (mean ± standard deviation). Adapted from (Mangia et al., 2003). b. Event-related fMRS used to detect increase in glutamate in anterior insular cortex in response to acute heat pain (‘Pain’) compared to rest (‘Rest’). Red: median; Upper and lower box plot limits show 75^th and 25^th percentiles, respectively. Adapted from (Gussew et al., 2010). c. Event-related fMRS used to demonstrate repetition suppression effects in glutamate. Upper: Schematic showing event-related fMRS task used to assess difference in neurochemical concentration between novel and repeated stimuli. Lower left: Mean (pink) and median (yellow) MRS voxel location, overlaid on significant BOLD response to stimuli (blue) in lateral occipital complex (LOC). Lower middle: Significant reduction in event-related glutamate response (mean ± 95% CI) to repeated stimuli compared to novel stimuli. Lower right: Significant reduction in event-related BOLD response to repeated stimuli compared to novel stimuli. *p < 0.05. Adapted from (Apšvalka et al., 2015). d. Event-related fMRS used to detect increase in glutamate:GABA ratio in primary visual cortex during recall of a visual cue. Upper left: Auditory-visual associative memory task performed in virtual reality. Upper right: schematic showing interleaved fMRI and fMRS sequence, with each data modality collected within each TR. Lower left: During recall of a visual cue the ratio between glutamate and GABA significantly increased in ‘remembered’ compared to ‘forgotten’ trials. Lower middle: Compared to a null distribution generated by permuting trial labels, the concentration of GABA significantly decreased during recall of a visual cue. Lower right: The increase in glutamate:GABA ratio during recall of a visual cue was predicted by the BOLD response in the hippocampus. Adapted from (Koolschijn et al., 2021).

Figure 3. Monte Carlo simulations to assess the effect of including/removing prior constraints on GABA.

a. Schematic of the simulation procedure. Sets of synthetic spectra were generated with changes in GABA imposed, which included multiples (±0.5, ±1, ±2) of the observed effect size in (Koolschijn et al., 2021). Model fitting was performed in LCModel with ‘constraints on’ (red) and ‘constraints off’ (orange). b. For different imposed changes in GABA, the mean ± standard deviation for the resulting GABA:total Creatine (tCR) concentrations (n = 2000) are shown, as estimated with LCModel using ‘constraints on’ and ‘constraints off’. The dashed line represents agreement between the imposed and estimated GABA:tCr concentration. c. With ‘constraint on’ the measured changes deviated from the imposed changes at both higher and lower SNRs. With ‘constraints off’, the estimated changes closely reflected the imposed changes in GABA at different SNRs. These simulations show that across a range of different SNRs, dynamic changes in GABA are robustly reflected in the measured GABA concentrations using ‘constraints off’. Adapted with permission from Koolschijn et al. (2021).