The Whitelaw et al. paper 10 compared genome filtration with random genomic shotgun sequencing. From the random library, 73% of 34,074 sequences were identified as repetitive. In contrast, 35% of the 95,233 methylation-filtered and 21% of 100,000 high-CoT sequences were repetitive. Over 900,000 sequence reads of the latter two libraries have now been completed and deposited in a public database 11. The high-CoT and methylation-filtered clone sequences were found to be enriched for sequences related to known plant genes. For example, 13% of methylation-filtered and 11% of high-CoT sequences were similar to known plant expressed sequence tags (ESTs), whereas only 4% of sequences from random libraries were similar. Palmer and coworkers 9 developed an independent set of approximately 100,000 methylation-filtered sequences, and found that 8.6% of these exhibited sequence similarity to their gene database, while 24% of them matched a known repetitive sequence. They additionally showed that rates of new gene discovery per sequence read were similar for EST and methylation-filtration libraries 9.

An earlier study suggested that methylation-filtration can detect 95% of maize exons 12, and analyses in the two recent papers 910 suggest that most maize genes may be captured by filtration. These predictions are, however, based on detecting typical polypeptide-encoding genes. Will enrichment techniques capture genes encoding very small proteins or small RNAs? Tandem duplications, which are common in plant genomes, are another concern 46. Will filtration be able to distinguish between copies, including those that have evolved distinct functions? It is possible that genome filtration could miss a number of genes.

There are, however, reasons for optimism. First, sequences for genes encoding small polypetides or RNAs could be among the uncharacterized sequences found in the filtered libraries. After sequencing reads were assembled into contigs, 63% of high-CoT assemblies and 39% of methylation-filtration assemblies had no significant matches to a gene or repeat sequence in the database at The Institute for Genomic Research 1011. Second, the methylation-filtration and high-CoT techniques sample from partially different fractions of the maize genome. It was estimated that of all the sequences sampled in the methylation-filtration and high-CoT libraries, approximately one-third were recovered by both approaches 10. Using both techniques thus samples a greater fraction of the genome, and it seems possible that genes encoding microRNAs and small polypeptides will be captured by one or other technique.

The application of genome filtration for sequencing the maize genome would require the mapping of sequences onto physical or genetic maps, as noted at the NSF workshop 1. How this mapping step is carried out will be a critical decision. As positional cloning is likely to be a major use of the mapped sequences, high-resolution map data are desirable. Placing sequences onto maps derived from bacterial artificial chromosome (BAC) contigs by hybridization or low-pass sequencing, would be appropriate. Genome filtration may prove to be most effective when a closely related species has already been sequenced, because synteny between species can then provide the positional information. Studies of cereal genomes suggest that rice is not sufficiently related to maize to adequately fill this gap in genome information 1314. In this light, synteny to important crops, in addition to genome size, may be an important criterion for selecting model species to sequence in the future.

Enrichment may not be an appropriate approach for all species. Methylation filtration has worked well in maize because plant genes are largely unmethylated 12. Furthermore, there is little repetitive sequence within plant genes themselves that could interfere with high-CoT selection, the exception being MITES (miniature inverted-repeat transposable elements), which are very small and usually poorly conserved 15. Plant transcription units tend to be small 456, and their regulatory regions are compact. A wealth of experience with transgene constructs in plants demonstrates that in general only a few kilobases of flanking sequence are required for tissue and developmental regulation, although exceptions do exist. For instance, the maize P1 gene promoter is unusually large, extending 5 kilobases upstream of the transcription start site 16. Gene and genome organization must be considered before applying genome-filtration techniques to other organisms.

If funding becomes available, there are strong reasons for sequencing the entire maize genome. Access to the hundreds of mutations isolated over the past 75 years is one compelling reason. The agronomic importance of maize, in the United States and other countries, is another. A complete sequence of the maize genome would provide researchers with gene sequences, regulatory sequences, precise positional information, and markers for high-resolution mapping. These are the obvious reasons for whole-genome sequencing, but others may in fact prove more rewarding. We now know that different maize lines do not have identical complements of genes. In one region sequenced from two lines, four of the ten genes present in one line were absent from the other 17. Tandem duplications provide an opportunity for gene number to increase or decrease within pedigrees 1819, and duplication allows epigenetic regulation of gene expression 1920. Perhaps epigenetic interactions and variation in gene content underlie heterosis, whereby hybrids show increased vigor compared to their parents. This, together with the long breeding records and extraordinary genetic variation in maize, provides very special opportunities. Genome filtration coupled with mapping relatively inexpensively provides much of the same information that can be found in a complete genome sequence. But a full genome sequence provides a much broader foundation for exploring the complete genome.