Selection on Alu sequences?

The draft of the human genome project reveals an unexpected distribution of Alu interspersed repetitive sequences [1]. This has been interpreted as evidence that Alu sequences 'may benefit their human hosts' [1] and that they 'have a positive function' [2]. The implication is that the majority of Alu sequences increase the Darwinian fitness of their bearers. Here I suggest that this conclusion is inconsistent with our knowledge of human population genetics. There is a relationship between the time since an Alu sequence was inserted into the genome and the GC content of its surrounding DNA. Alu sequences inferred to have inserted within the last five million years are slightly more abundant in low-GC, gene-poor regions, whereas older Alus, in classes inserted from 5 to 100 million years ago, are increasingly found in high-GC gene-rich regions. Contrary to the conclusions drawn in the report [1] and the associated News and Views article [2], this observation does not indicate that Alu sequences are advantageous to their human hosts. While one could imagine that Alu sequences active 50 million years ago were targeting GC-rich regions while today's Alus target AT-rich regions, a more parsimonious interpretation is to assume that the sequence's insertion preferences have been constant, and that the change in the sequence's relative abundance reflects a process in which, following insertion, it increases its relative abundance in GC-rich DNA with time. The most likely such process, and one considered and dismissed by the authors [1], is that deletions removing