Reassessing associations between white matter and behaviour with multimodal microstructural imaging

Abstract

Several studies have established specific relationships between White Matter (WM) and behaviour. However, these studies have typically focussed on fractional anisotropy (FA), a neuroimaging metric that is sensitive to multiple tissue properties, making it difficult to identify what biological aspects of WM may drive such relationships. Here, we carry out a pre-registered assessment of WM-behaviour relationships in 50 healthy individuals across multiple behavioural and anatomical domains, and complementing FA with myelin-sensitive quantitative MR modalities (MT, R1, R2∗). Surprisingly, we only find support for predicted relationships between FA and behaviour in one of three pre-registered tests. For one behavioural domain, where we failed to detect an FA-behaviour correlation, we instead find evidence for a correlation between behaviour and R1. This hints that multimodal approaches are able to identify a wider range of WM-behaviour relationships than focusing on FA alone. To test whether a common biological substrate such as myelin underlies WM-behaviour relationships, we then ran joint multimodal analyses, combining across all MRI parameters considered. No significant multimodal signatures were found and power analyses suggested that sample sizes of 40-200 may be required to detect such joint multimodal effects, depending on the task being considered. These results demonstrate that FA-behaviour relationships from the literature can be replicated, but may not be easily generalisable across domains. Instead, multimodal microstructural imaging may be best placed to detect a wider range of WM-behaviour relationships, as different MRI modalities provide distinct biological sensitivities. Our findings highlight a broad heterogeneity in WM's relationship with behaviour, suggesting that variable biological effects may be shaping their interaction.

Keywords: Bimanual coordination; Cognitive control; Microstructural imaging; Multimodal imaging; Myelin; White matter.

Fig. 1

Each neuroimaging modality is sensitive, but not specific, to different features of the biological tissue. This study aimed to use multiple MR modalities that are sensitive to myelin, but measure different biophysical properties of white matter.

Fig. 2

Study design and summary of MRI and behavioural data acquired.

Fig. 3

FA and behaviour. Unimodal relationships between FA and behaviour were tested across anatomical masks (shown in green) that were selected for each task. Results highlight that the Alternating Finger Tapping task (AFT), but not Temporal Order Judgement task (TOJ) and Digit Symbol Substitution Test (DSST) has a significant relationship with FA (red cluster shows voxels with corrected p-values below .05). Within that cluster, mean FA is extracted for each subject and plotted against performance in the scatterplot (with line of best fit and 95% confidence bands), that is for visual assessment of the correlation, rather than for statistical inference.

Fig. 4

Multimodal microstructural imaging and behaviour. Multimodal relationships between behaviour and individual MRI metrics (FA, MT, R1 and R2∗) across Digit Symbol Substitution Test (DSST), Alternating Finger Tapping task (AFT) and Temporal Order Judgement task (TOJ). Only the DSST has a significant relationship with cingulum WM, driven by R1, when considering FWER-corrected p-values (red cluster shows voxels with corrected p-values below .05). Within that cluster, mean R1 is extracted for each subject and plotted against performance in the scatterplot (with line of best fit and 95% confidence bands), that is for visual assessment of the correlation, rather than for statistical inference.

Fig. 5

Lack of evidence for combined multimodal signatures. A Fisher test was used to search for multimodal microstructural signatures relating WM to behavior, but no significant effects were found (2nd column). Effect sizes are reported for each modality-behaviour correlation, as measured by the top 5% t-statistic within peak Fisher clusters. This analysis was carried out to provide a clear visualisation of peak effect size for each pair of MR modality and behaviour, rather than for statistical inference (3rd column). For each WM-behaviour correlation, we used a simulation-based approach to calculate sample sizes needed to reach 80% power (red line), given the observed effect sizes found in our pre-registered tests. Sample sizes needed to detect a combined multimodal effect vary from 190 to 200 participants for DSST, to 40–50 for AFT, to 60–70 for TOJ (4th column).