RNA recognition by the human polyadenylation factor CstF.

Polyadenylation of mammalian mRNA precursors requires at least two signal sequences in the RNA: the nearly invariant AAUAAA, situated 5' to the site of polyadenylation, and a much more variable GU- or U-rich downstream element. At least some downstream sequences are recognized by the heterotrimeric polyadenylation factor CstF, although how, and indeed if, all variations of this diffuse element are bound by a single factor is unknown. Here we show that the RNP-type RNA binding domain of the 64-kDa subunit of CstF (CstF-64) (64K RBD) is sufficient to define a functional downstream element. Selection-amplification (SELEX) experiments employing a glutathione S-transferase (GST)-64K RBD fusion protein selected GU-rich sequences that defined consensus recognition motifs closely matching those present in natural poly(A) sites. Selected sequences were bound specifically, and with surprisingly high affinities, by intact CstF and were functional in reconstituted, CstF-dependent cleavage assays. Our results also indicate that GU- and U-rich sequences are variants of a single CstF recognition motif. For comparison, SELEX was performed with a GST fusion containing the RBD from the apparent yeast homolog of CstF-64, RNA15. Strikingly, although the two RBDs are almost 50% identical and yeast poly(A) signals are at least as degenerate as the mammalian downstream element, a nearly invariant 12-base U-rich sequence distinct from the CstF-64 consensus was identified. We discuss these results in terms of the function and evolution of mRNA 3'-end signals.