Point mutants induced with a variety of mutagens at the dihydrofolate reductase (dhfr) locus in Chinese hamster ovary (CHO) cells were screened for aberrantly spliced dhfr mRNA by RNase protection and/or reverse transcriptase coupled with cDNA amplification by the polymerase chain reaction (PCR). Of 115 mutants screened, 28 were found to be affected in splicing. All exhibited less than 1% correct splicing, probably because the selection procedure was stringent. All 26 unique mutations were located within the consensus splice sequences; changes were found at 9 of 10 possible sites in this 25-kb six-exon gene. Mutations at the sites flanking the first and last exons resulted in the efficient recruitment of a cryptic site within each exon. In contrast, mutations bordering internal exons caused predominantly exon skipping. In many cases, multiple exons were skipped, suggesting the clustering of adjacent exons prior to actual splicing. Six mutations fell outside the well-conserved GU and AG dinucleotides. All but one were donor site single-base substitutions that decreased the agreement with the consensus and resulted in little or no correct splicing. Starting with five of these donor site mutants, we isolated 31 DHFR+ revertants. Most revertants carried a single-base substitution at a site other than that of the original mutation, and most had only partially regained the ability to splice correctly. The second-site suppression occurred through a variety of mechanisms: (i) a second change within the consensus sequence that produced a better agreement with the consensus; (ii) a change close to but beyond the consensus boundaries, as far as 8 bases upstream in the exon or 28 bases downstream in the intron; (iii) mutations in an apparent pseudo 5' site in the intron, 84 and 88 bases downstream of a donor site; and (iv) mutations that improved the upstream acceptor site of the affected exon. Taken together, these second-site suppressor mutations extend the definition of a splice site beyond the consensus sequence.The correct splicing of exons in pre-mRNA presents two problems of molecular recognition. First, 5' (donor) and 3' (acceptor) splice sites must be identified. In a typical mammalian pre-mRNA, a few bona fide sites must be found among tens of thousands of nucleotides in the primary transcript. Examination of the sequences common to all splice sites has revealed a nonrandom occupancy of nucleotide positions straddling the cleavage site (31). Aside from the nearly invariant GU at the donor site and AG at the acceptor site, the two splice site consensus sequences are quite variable (see Table 3 for the consensus sequences used here). This variability is extensive enough to preclude the definition of a splice site on the basis of a unique sequence or even a best fit to the consensus sequence. That is, in a pre-mRNA molecule there typically exist many sequences in exons and introns that present a good fit to a consensus, but these sites are not used in splicing (45). It is reasonable therefore to consider secondary or...