Hi Yeliz - depending on how exactly you are implementing these ROI analyses, the likeliest explanation is that the ROI analysis adjusts the correction for multiple comparisons when thresholding. Therefore, when running the analysis with all limbic structures, there are more voxels, i.e. more statistical tests, which results in a more stringent threshold (relative to amygdala alone), which no voxels in amygdala surpass; when including only amygdala, there are fewer voxels, a lower threshold, and therefore some above-threshold voxels. Hope this helps.
-Ben Turner