Hi Sam,
I can't find the original thread using just this number. Could you send the permalink (shown at the bottom of the messages on Jiscmail)?
Could you also show the plot you've made?
At any rate, it seems the plot is for the interaction, so in fact, the groups may have no significant association with GM, but when this (presumably weak) association is compared between groups, it becomes significant. You'd have to make two scatter plots, one for each of the two groups represented in the contrast. They may show opposite trends (most likely from the description), or at least different slopes. Also, this plot needs to use data from voxels or regions where the interaction was found significant. Finally, a it seems there is a 7th EV for some other nuisance, and that needs to be taken into account too.
All the best,
Anderson