Based on the information provided, I don't know what really happened.
If you just want to strict the insignificant areas of A and B, you don't have to run this 2nd ANOVA. Instead you could either run step functions directly on (A-B) subbrik, or in the first ANOVA add an option -acontr 1 1 to generate a subbrik for double thresholding with the one for (A-B).
Gang