Complete genome annotation relies on precise identification of transcription units bounded by a transcription initiation site (TIS) and a polyadenylation site (PAS). To facilitate this process, we developed a set of two complementary methods, 5 Long serial analysis of gene expression (LS) and 3LS. These analyses are based on the original SAGE and LS methods coupled with full-length cDNA cloning, and enable the high-throughput extraction of the first and the last 20 bp of each transcript. We demonstrate that the mapping of 5LS and 3LS tags to the genome allows the localization of TIS and PAS. By using 537 tag pairs mapping to the region of known genes, we confirmed that >90% of the tag pairs appropriately assigned to the first and last exons. Moreover, by using tag sequences as primers for RT-PCRs, we were able to recover putative full-length transcripts in 81% of the attempts. This large-scale generation of transcript terminal tags is at least 20 -40 times more efficient than full-length cDNA cloning and sequencing in the identification of complete transcription units. The apparent precision and deep coverage makes 5LS and 3LS an advanced approach for genome annotation through whole-transcriptome characterization.genome annotation ͉ full-length cDNA ͉ transcription analysis T he complete genome sequences of human and other model organisms (1-5) have raised questions as to the accuracy of previously used gene annotation algorithms. Current genome annotations are mostly based on the growing but still limited cDNA sequence databases. Computational gene prediction algorithms, even when trained based on current cDNA data sets, may not provide reliable ab initio predictions of new genes, and all predictions need to be validated by further experimental evidence (6, 7).Efforts in large scale full-length cDNA sequencing (8-11; reviewed in ref. 12) provide not only the complete sequences of expressed genes but also the ability to locate transcription initiation sites (TISs), polyadenylation sites (PASs), as well as splicing junctions. The localization of these functional sites related to transcripts in the context of the genome can greatly help to define the surrounding regulatory elements such as promoters. However, sequencing all possible full-length transcript clones is an expensive and cumbersome process. Despite current progress, complete disclosure of the transcriptome has yet to be achieved. It is clear, therefore, that the precise identification of all genes and their regulatory elements will require more comprehensive and facile technologies with greater throughput to provide the necessary empirical evidence.Serial analysis of gene expression (SAGE) (13, 14) and massively parallel signature sequencing (MPSS) (15) are high-throughput methods that use short tags (14-21 bp of internal transcript signatures) to count transcripts in transcriptomes. When mapped to assembled genome sequences, these short tags help to locate transcription units in the context of the genome. Because concatenation of these short tags and t...