The use of grammatical metaphor (GM) has been found to be a typical feature of academic writing, but few studies focused on the variations of GM across disciplines. In order to find out how GMs vary across disciplines, we selected four textbooks (linguistics, law, medicine and physics respectively representing arts and humanities, social sciences, life sciences, and physical sciences) which were published by prestigious publishers, each being the sixth edition at least, and then with these textbook contents we built a corpus of 1.1 million tokens. Other corpora used include the British Academic Written English Corpus (BAWE) and the LOB family (LOB, FLOB and CLOB). With Lancsbox 6.0 as the corpus tool, we observed the occurrences of ideational GMs and modality metaphors in the data. Then with Scholar GPT, correlation analyses have been performed. The results show that the use of GM is discipline common, time constrained, and proficiency dependent. The strategy of using GMs varies in texts compiled in different disciplines but GM is a persistent feature in general and there is a trend of increasing GM usages in academic writings.