Its short generation time and large number of progeny have made Drosophila melanogaster one of the best model systems for performing large-scale genetic screens for mutations that affect a given process. For example, the classic screens for mutations that affect the patterning of the embryo led to the discovery of almost all of the genes that control segmentation [1]. Similar screens have identified many of the factors that mediate the development of the nervous system [2], and enhancer and suppressor screens have proved a valuable approach for finding novel components of signal transduction pathways [3,4]. The range of forward genetic screens has been significantly advanced by the adaptation of the yeast flp/FRT recombination system to generate mitotic clones of cells that are homozygous for a particular mutagenized chromosome arm [5]. This technique makes Drosophila the only multicellular organism in which it is possible to perform phenotypic screens for mutations that affect the behavior of almost any cell at any stage of development [6,7,8].

The most common mutagens used in Drosophila are P elements, which generate mutations by inserting into genes, and chemicals, such as ethyl methyl sulpfnate (EMS), which modifies bases in the DNA to cause mainly single base substitutions, also known as point mutations [9]. The main advantage of P elements is that it is very straightforward to identify the gene that has been mutated, but P elements are relatively inefficient mutagens. The majority of Drosophila genes are predicted to be 'cold spots' for P-element insertion [10], and P-element insertions have been recovered in only a fifth of the genes within a 2.9 megabase (Mb) region that has been thoroughly characterized [11]. This makes P elements a poor mutagen for saturation screens that set out to identify most of the genes required for a particular process. In contrast, chemical mutagens cause a much higher mutation rate, and have less bias in their ability to cause mutations in different loci. The main drawback of these mutagens, however, is that it is not trivial to map point mutations to specific genes, and this is often the rate-limiting step in the analysis.

Two complementary approaches have been standard in mapping point mutations: mapping by meiotic recombination between visible genetic markers, and deletion mapping, in which the mutation is positioned by its failure to complement deficiencies. However, mapping a mutation between widely spaced markers can only give a statistical estimate of its position, and existing collections of deficiencies do not cover the whole genome. Furthermore, the exact positions of many of the visible markers and the breakpoints of the deletions have often only been inferred from cytological and genetic data, and this can make it difficult to link the genetic and molecular maps [12]. Various other approaches have been used to refine the location of the gene, such as P-mediated male recombination [13], or meiotic recombination between two P elements that flank the region containing the mutation. However, several rounds of P-mediated recombination are often necessary to narrow down the region sufficiently for candidate genes to be tested.

In other organisms, which lack the many visible genetic markers and deletions of Drosophila, mutations have been commonly mapped using molecular polymorphisms, such as restriction fragment length polymorphisms (RFLPs) or SNPs [14,15]. These have the advantage that they directly link the genetic and physical maps of the chromosome, and can provide a very high density of markers that allow the precise mapping of mutations. For example, the sequencing of the human genome has revealed a very large number of SNPs that cover the genome at a density of one per 1.9 kb, and this will greatly facilitate the mapping of both single-gene and polygenic disease loci [16]. This approach has also been employed to a limited extent in Drosophila, but up till now the difficulty and expense of discovering polymorphisms at the DNA level has confined its use to the mapping of mutations within small regions between closely linked markers [17,18].

The recent completion of the Drosophila genome sequence [19] makes it possible to search for molecular polymorphisms much more efficiently. It should now be possible to discover enough SNPs to allow the rapid mapping of point mutations within entire chromosome arms. Teeter et al. [20] and Hoskins et al. [21] have already generated a collection of SNPs that are polymorphic between different inbred wild-type strains. Here we report a high-density map for chromosome arm 3R that has been specifically designed for the rapid mapping of mutations from clonal screens that use the standard FRT chromosome. This approach has two advantages over the use of SNP maps derived from wild-type lines. First, most genetic screens are performed in more complex genetic backgrounds that cannot be traced back to a specific wild-type isolate, and it is therefore more convenient to have a collection of SNPs that can be used directly on the mutagenized chromosomes. Second, by screening for polymorphisms between this chromosome and a standard third chromosome that carries four visible mutations on 3R, we have developed a hybrid mapping strategy that exploits both traditional and molecular markers. This reduces the cost of SNP mapping by a factor of four, and makes it affordable for even small Drosophila laboratories. Using this strategy, a mutant can be mapped within two months to a region of about 50 kb with a single meiotic recombination cross.

In this paper, we describe the development of a SNP map of chromosome arm 3R between the widely used Drosophila chromosomes FRT 82B and rucuca, and demonstrate that this map can be used to map mutations rapidly and accurately. One essential requirement for efficient mapping is the availability of a high enough density of polymorphic markers. Although these stocks were not chosen for their diversity, the average frequency of SNPs between them is 2.3 per kb, and even the least polymorphic region has more than one SNP every 2 kb. As the Drosophila euchromatin contains one gene every 9 kb on average, it should therefore be possible to find more than enough markers in any region of the chromosome to map a mutation to the level of a single gene. The frequency of SNPs that we obtained is slightly less than half that observed by Teeter et al. between pairs of wild-type chromosomes [20], but within the range observed by Hoskins et al. [21]. Although the full pedigrees of these two chromosomes are unknown, the recessive visible mutations on the rucuca chromosome were all discovered in Thomas Hunt Morgan's laboratory in the first quarter of the twentieth century, whereas the FRT 82B is inserted on a ry⁵⁰⁶ chromosome that was derived from an unspecified wild-type stock in 1972 [27,28]. Thus, these stocks probably originate from different wild-type isolates, and have been separated in the laboratory for at least 80 years. Preliminary analysis suggests that there is a similar rate of polymorphism between other well-separated pairs of laboratory chromosomes, such as the standard mapping stock for the second chromosome, al dp b pr c px sp and the FRT 40A chromosome used in flp/FRT screens on 2L, and between this mapping stock and a cn chromosome that we have used in an enhancer screen (J.-R. Huynh, I. Torres and D.St J., personal communication). Thus, it should be possible to discover sufficient SNPs between other pairs of laboratory chromosomes to use this strategy for mapping mutations that have been induced in different genetic backgrounds.

The accuracy of SNP mapping depends on the number of recombinants analyzed. We performed our mapping crosses in 12-15 bottles, and recovered about 1,200 recombinant males of the right genotype in each case; this should allow mapping to a resolution of about 0.1 centimorgans (cM). As the right arm of the third chromosome spans 60 cM, and contains 28 Mb of DNA, 0.1 cM corresponds to an average of 46 kb. Although this will vary along the length of the chromosome with the rate of recombination, our results from the mapping of two mutations in different regions are consistent with this estimate. In each case, we were able to map the mutation in a single mapping cross to a interval of approximately 60 kb. This level of resolution will usually be insufficient to map the mutation to a single gene, and a second round of mapping is often required. One efficient way of doing this is to select for recombinants between closely linked P elements, but one still has to screen at least an order of magnitude more recombinants to narrow the region down to around the size of the average locus. As a region of this size should contain only a small number of genes (typically fewer than ten), in most cases, it is more efficient to test them individually by other means. For example, if the mutation produces a clear embryonic phenotype, the genes can be screened by RNA interference to determine which double-stranded RNA phenocopies the mutant phenotype [29,30]. An alternative, which we used successfully for our first complementation group, is to sequence candidate genes to search for molecular lesions in mutant alleles.

The mapping method described here has two major advantages. First, it is very quick, as the whole process takes only two months, and a precise map position can be obtained within days of determining which recombinant chromosomes carry the original mutation. More importantly, the method is also inexpensive. In the example outlined in detail above, the initial mapping of the mutation to a 400 kb region required 210 PCR reactions, 207 RFLP digests and 3 SSCP reactions, and cost less than £100 (£0.185 per PCR reaction, £1 per SSCP lane and £0.22 per RFLP detection). Even though the more precise mapping required the discovery of six further SNPs, this only doubles the cost per mutant. Furthermore, the SNP map becomes denser each time a mutant is mapped in this way, and the cost and time of mapping subsequent mutations in the same region is therefore reduced.

Most SNP maps that are being developed in Drosophila are between chromosomes from wild-type stocks that have been chosen because of their diversity. Although these chromosomes will be invaluable for mapping quantitative trait loci in wild-type populations, they do not carry any visible markers, and this significantly increases the time and cost of using them for mapping mutations. For example, a mapping experiment with 1,000 chromosomes would require more than 2,300 SNP detection reactions, because every chromosome from the cross needs to be tested for at least two SNPs to determine where recombination occurred. In contrast, only about 300 SNP detections are required to perform the same experiment with the multiply marked chromosome. This also reduces by a factor of four the number of single fly DNA preparations, one of the most tedious and time-consuming parts of the procedure.

In a few cases, the use of visible markers will not help in the mapping of the mutation. If the mutation lies distal to the last visible marker on the chromosome (in this case ca), there is no way to select visually for recombination events between this marker and the end of the chromosome. In this case, one has to screen for recombinants between ca and the tip of the chromosome, using a very distal SNP, as one would do with a non-marked chromosome. Based on the size of this distal region, this should occur less than 10% of the time.

This mapping method with marked chromosomes can be used in any laboratory without the need for a specialized infrastructure. Using the chromosomes and the map described here, RFLPs can be used immediately and the remaining SNPs can either be detected on an SSCP system or by sequencing. It should also be straightforward to develop a SNP map for other pairs of chromosomes, which can then be used to map multiple mutants from screens that have been performed in different genetic backgrounds. The construction of the map presented here required 300 pairs of primers and SSCP detections, and 150 sequencing reactions, and took about three months at a cost of about €4,000. An alternative strategy, which does not involve the use of an SSCP detection system, is to discover the SNPs by sequencing. With a SNP frequency of 2 per kb, it is very likely that every kilobase of DNA sequenced will reveal at least one polymorphism, and this will reduce the number of primers required. As over half of the SNPs actually create an RFLP, one cost-effective strategy would be to ignore the other SNPs and use only the RFLPs for mapping mutants.

In some cases, such as screens for mutants with behavioral phenotypes, it is too time-consuming and expensive to perform large-scale saturation mutagenesis, and this results in the recovery of many mutations that do not fall into complementation groups. Because the mutant chromosomes recovered from screens are often heavily mutated, there is always a risk that the phenotype of a single mutant chromosome is synthetic, that is, the result of several interacting mutations. Similarly, when the mutant chromosome is lethal, it is very difficult to find out whether the lethality is due to the same mutation that causes the phenotype of interest, or to an unrelated hit on the chromosome. SNP mapping provides a simple way to resolve these issues, because mutations can be rapidly mapped to a region of approximately 60 kb. If a mutation that causes the phenotype can be mapped to such a small region, one can almost completely exclude the possibility that the phenotype is the result of two independent mutations. In the same way, any lethal mutations on the chromosome can be mapped and tested for the original phenotype. If the lethality and the phenotype map to the same region, one can be confident that they are caused by the same mutation. The convenience and speed of this approach therefore makes it particularly suitable for mapping single alleles.

Given how valuable such a tool would be for the whole Drosophila community, ideas are being discussed amongst Drosophila workers on how to develop freely available SNP databases that contain a much higher density of SNPs between these chromosomes and several others. Access to such databases, along with the availability of primers or a mapping service, would greatly facilitate the mapping and cloning of mutations, and would make Drosophila an even more attractive model system for forward genetic screens for studying many aspects of biology.