Cleavage of highly structured viral RNA molecules by combinatorial libraries of hairpin ribozymes. The most effective ribozymes are not predicted by substrate selection rules.

Combinatorial libraries of hairpin ribozymes representing all possible cleavage specificities (>10(5)) were used to evaluate all ribozyme cleavage sites within a large (4.2-kilobase) and highly structured viral mRNA, the 26 S subgenomic RNA of Sindbis virus. The combinatorial approach simultaneously accounts for target site structure and dynamics, together with ribozyme folding, and the sequences that result in a ribozyme-substrate complex with maximal activity. Primer extension was used to map and rank the relative activities of the ribozyme pool against individual sites and revealed two striking findings. First, only a small fraction of potential recognition sites are effectively cleaved (activity-selected sites). Second, nearly all of the most effectively cleaved sites deviated substantially from the established consensus selection rules for the hairpin ribozyme and were not predicted by examining the sequence, or through the use of computer-assisted predictions of RNA secondary structure. In vitro selection methods were used to isolate ribozymes with increased activity against substrates that deviate from the GUC consensus sequence. trans-Acting ribozymes targeting nine of the activity-selected sites were synthesized, together with ribozymes targeting four sites with a perfect match to the cleavage site consensus (sequence-selected sites). Activity-selected ribozymes have much higher cleavage activity against the long, structured RNA molecules than do sequence-selected ribozymes, although the latter are effective in cleaving oligoribonucleotides, as predicted. These results imply that, for Sindbis virus 26 S RNA, designing ribozymes based on matches to the consensus sequence may be an ineffective strategy.