Emily,
You might try warping the original brain to MNI space and then using adwarp to apply that warp to the GM mask. Given that the MNI152 is fairly blurry and segmentations can be a bit blurry, this might be some of the problem. I modified your code bit below, also I treated them as if they are AFNI datasets and not NIFTI simply because I like having the view immediately apparent.
@auto_tlrc -base MNI_avg152T1.nii -input ${subject}+orig -no_ss -init_xform AUTO_CENTER
adwarp -apar ${subject}+tlrc -dpar ${subject}.GM+orig -dxyz 2 -resam NN
PM