The system for generating deletions devised at Exelixis involves the use of three FRT-bearing transposon constructs (Figure 1a) 510. Each construct carries the miniwhite marker and at least one FRT sequence. Deletions can be recovered by combining pairs of insertions in trans in the presence of an inducible genomic source of FLP recombinase. FLP-mediated recombination generates a deletion if the FRT sequences are arranged in the same relative orientation. If the miniwhite markers lie internal to the FRT sequences, deletion-bearing progeny will have white eyes from elimination of both copies of miniwhite upon recombination (Figure 1b). If they lie external to the FRTs, deletion-bearing progeny often have darker eyes than flies in either single-FRT progenitor strain from the presence of two copies of miniwhite (Figure 1c). We generated 87% of our deletions using construct combinations eliminating miniwhite markers; the remaining deletions were isolated in progeny bearing two copies of miniwhite. Pairs of insertions can also be used to recover deletions with no net gain or loss of miniwhite markers (Figure 1d), but such screens were not attempted in this project.

Approximately 16,500 FRT-bearing transposon insertions on the X chromosome and autosomes were generated and placed into public distribution by Exelixis 1011. The density of insertions afforded by such a large collection gave us flexibility in designing deletions. As we will describe below, convenient insertions were usually available so that we could flank haploinsufficient genes closely with pairs of deletions in our efforts to maximize deletion coverage and improve the distribution of breakpoints. In some chromosomal regions, however, the scarcity of insertions limited our efforts.

We isolated 793 deletions. They are described in detail in Additional file 1 and their breakpoint positions are shown relative to annotated genes in Additional file 2. Because we were primarily concerned with breakpoint placement, we did not standardize the size of the deletions. We did not generate deletions with heterochromatic breakpoints because most heterochromatic P{XP}, PBac{WH} and PBac{RB} insertions have not been localized unambiguously. In the screens that were successful, one deletion was recovered for every approximately 50 progeny screened with a range of roughly 1 in 10 to 1 in 1,000. We retained only one deletion from each screen for the BDSC collection.

We confirmed the existence of deletions by demonstrating that genes located between the progenitor insertions were missing. We usually designed deletions so they encompassed at least one gene with an existing loss-of-function mutation to use in complementation tests. Only two deletions designed early in the project were small enough to avoid removing vital genes. We also used preexisting deletions in complementation tests with new deletions if the progenitor FRT-bearing insertions were not themselves homozygous lethal to demonstrate that at least one vital gene was deleted. If an appropriate complementation test was not available, we verified the presence of a deletion by polytene cytology or by demonstrating that a particular internal DNA fragment could not be PCR amplified from the deletion chromosome. Eighteen new deletions showed dominant phenotypes expected for deleting known haploinsufficient genes (Additional file 1).

We also verified that the expected deletion event occurred using the 'hybrid PCR' method 5. In brief, PCR primers within each FRT-bearing progenitor insertion can amplify a fragment only if they are juxtaposed by FLP-mediated recombination. We tested 93% of the deletions lacking miniwhite markers by this method and all gave the expected result (Additional file 1). We used a positive hybrid PCR test as the sole criterion for verifying only six deletions; all other deletions were also confirmed on the basis of deleted genes or sequences.

The Exelixis insertions were generated in a uniform genetic background 10 and we maintained this background in most deletion screens as described in Parks et al.5. Noninsertion chromosomes specific to our screens were substituted into the standard background prior to use. Minimizing genetic variability in the deletion stocks makes them more valuable in the analysis of background-sensitive phenotypes such as behavioral traits. It was necessary to establish 27 deletions in stock with nonstandard balancers due to dominant effects of the deletions on viability or fertility or due to noncomplementation with mutations on standard balancers (Additional file 1).

Deletion stocks were placed into public distribution as soon as the deletions were verified; consequently, many deletions have been in widespread use for years and have contributed to published research (for examples, 12131415161718). Details of each deletion screen have been documented in FlyBase and references to the individual reports are given in Additional file 1. FlyBase developed a graphical interface for viewing molecularly defined deletions during our project 19 and the BDSC website lists available deletions 20.

We conducted 708 screens that produced no deletions and we were often able to determine the cause of the failure. First, the rate of recovering deletions was often low. We were able to generate most deletions with modest-sized screens, but some deletions required larger screens. Four percent of deletions were recovered only after additional rounds of screening. Five percent of the screens failed even though both progenitor insertions had been used in successful screens, there was no haploinsufficient gene in the region and the deletion would not have been particularly large, suggesting small screen size as the most likely explanation for failure. Second, the insertions in many stocks did not map as described. We saw evidence that insertions were mismapped in 12% of the unsuccessful screens. (All eight of the fourth chromosome deletion screens were in this category, suggesting that a high proportion of the fourth chromosome insertion stocks are not as represented. We abandoned plans for additional screens without validating the remaining putative fourth chromosome insertion stocks.) Third, in 6% of the failed screens, putative deletions could not be established in stock due to lethality and/or sterility, even though no haploinsufficient genes existed in the region. We suspect that the abnormal juxtaposition of genic regions upon deletion results in deleterious neomorphic phenotypes more often than is generally appreciated. Fourth, 5% of unsuccessful screens failed because flies in the intermediate crosses were weak, sterile or dead. We attribute these problems to noncomplementation and dominant interactions between mutations in the stocks. Fifth, 10% of unsuccessful screens failed because we intentionally deleted haploinsufficient genes. (We will discuss these experiments below.) The remaining screens probably failed from small screen sizes or mismapped insertions. They all involved at least one insertion stock that had not been used successfully in another screen. From these experiences, we recommend that anyone planning FLP-FRT screens verify the positions of progenitor insertions, plan screens larger than our routine screen sizes and, if possible, attempt to create deletions of a region using more than one combination of progenitor insertions.

False positives were recovered in many screens. For example, white-eyed progeny carrying no deletion were recovered in screens designed to eliminate the miniwhite markers from both progenitor insertions upon FLP-mediated recombination (Figure 1b). While false positives did not prevent the recovery of deletions, they increased the workload in the verification step. At least 9% of screens successfully producing deletions also produced false positives. Inexplicably, X chromosome screens were much more likely to give false positives than autosomal screens (49% versus 4%). At least 23% of the screens producing no deletions gave false positives, again, with a disproportionate number involving the X chromosome rather than an autosome (56% versus 15%). False positives were also seen in Exelixis and DrosDel screens (A Parks, personal communication) 9. Consequently, we recommend testing multiple putative deletions in all screens. Though the false positives have not been characterized in detail, we surmise they originate from the repair of DNA damage associated with abnormal FLP-mediated recombination events.

The first goal of our project was to improve deletion coverage of the genome. As summarized in Table 1 and shown in detail in Additional file 2, 98.4% coverage of euchromatic genes is now available using deletions distributed by the BDSC regardless of their origin. The deletions we generated (the 'BSC' deletions) encompass 81.5% of euchromatic genes as compared to 65.2% for the DrosDel deletions at Bloomington and 47.4% for the Exelixis deletions. Together, the molecularly defined deletions from the three large-scale projects provide 94.9% coverage with 17.9% of euchromatic genes removed exclusively by BSC deletions. The remaining genomic coverage is provided by chemical- or irradiation-induced deletions (3.4%) and FRT-derived deletions from individual investigators (0.1%).

The single biological impediment to complete deletion coverage is the existence of genes that must be present in more than one copy for viability and fertility. Deletions of these haplolethal and haplosterile genes cannot be recovered in the absence of a compensating chromosomal duplication. (Obviously, haploinsufficiency pertains to X-linked genes only in females where two gene copies are normally present.) Duplications had been isolated for some haplolethal and haplosterile genes before we began our work, but generating new duplications in order to screen for deletions was beyond the scope of this project. Instead, we generated pairs of deletions to flank haplolethal and haplosterile genes as closely as possible as a way to maximize deletion coverage. This strategy allowed us to improve coverage substantially, but it precluded 100% coverage. The sizes of gaps in coverage were determined by the availability of FRT-bearing transgene insertions; consequently, some genes closely linked to haplolethal and haplosterile genes also remain undeleted. We chose not to use preexisting duplications to screen for deletions of haploinsufficient genes because we wanted to preserve a consistent genetic background in our deletion stocks. We were also aware of the fact that deletions of haploinsufficient genes have limited experimental value: the phenotypes of haploinsufficient genes often confound the interpretation of complementation tests and they enhance so many other mutant phenotypes that they are generally viewed as nuisances in modifier screens.

If we consider the distribution of deletion breakpoints for the DrosDel, Exelixis and BSC deletions, we see there is a total of 81 gaps in deletion coverage across the entire euchromatic genome (Additional file 2). Fifty-five gaps contain haploinsufficient genes, five gaps (including three with haploinsufficient genes) lie adjacent to centric heterochromatin where screening was not possible, three gaps (including one with a haploinsufficient and one adjacent centric heterochromatin) lie on the fourth chromosome where we recovered no deletions and 23 gaps are not covered despite the absence of haploinsufficient genes. The median size of the gaps is only 7 genes and the gaps range in size from 1 to 36 genes (1 to 27 if a large fourth chromosome gap is disregarded). We made multiple attempts to provide deletion coverage for most gaps lacking haplolethal or haplosterile genes, but we were unsuccessful for the reasons outlined above.

Fortunately, many gaps in coverage between BSC, Exelixis and DrosDel deletions are covered by older deletions or FRT-derived deletions from individual investigators. We have no clear explanation for the failure of FLP-FRT screens in regions where older deletions provided coverage. Although the haplosterility or haplolethality of some genes might depend on genetic background, it is unlikely this is the case for a large number of genes. Screens with different progenitor insertions to the ones we used might prove successful. When deletions from all sources are considered - including those maintained in stock with a duplication - only 35 gaps in coverage remain. Thirty-two gaps contain haploinsufficient genes, three gaps (including one with a haploinsufficient gene) lie adjacent to centric heterochromatin, and two gaps (including one adjacent to centric heterochromatin) lie on the fourth chromosome. The median gap size is 7 genes with a range of 1 to 18 genes. We verified that the older deletions span gaps with complementation tests and PCR amplifications. In many cases, these verification experiments also allowed us to map deletion breakpoints of the older deletions more precisely (Additional file 2). In addition, the breakpoints of three gap-covering deletions (Df(2L)C144, Df(2R)CX1 and Df(3L)ZN47) were mapped with comparative genomic hybridization microarrays at the same time breakpoints of X chromosome duplications from another BDSC resource project were localized 21.

Although it was necessary to identify haplolethal and haplosterile genes in order to improve deletion coverage, most were poorly mapped and few had been associated with annotated genes when we began our project. Consequently, the process of planning deletion screens was intertwined with the task of identifying haploinsufficient genes. When there were no hints about the molecular identity of a haploinsufficient gene, we faced the tedious process of iterative deletion screening to narrow its location.

Fortunately, most haploinsufficient genes in Drosophila encode protein components of cytoplasmic ribosomes (Rp genes) or translation initiation factors (eIF genes) and are associated with the Minute syndrome, a characteristic set of phenotypes including short, thin bristles and slower development 22. All Minute genes are associated with some degree of haplolethality and haplosterility, but the phenotypes of many Minute genes are not severe enough to prevent the recovery of deletions in the absence of duplications. We assured that all Minute genes were flanked as closely as possible with deletions and we attempted to delete 34 Minute genes to assess their viability and fertility. Deletions for 7 Minute genes could be established in simple heterozygous stocks, but deletions for the remaining 27 could not be recovered or heterozygous stocks were too weak to maintain long term. We did not attempt to delete Minute genes where evidence already existed for strong lethality or sterility effects. Table 2 shows Minute genes with viability or fertility effects strong enough to prevent the recovery of deletions in the absence of a duplication; Table 3 shows Minute genes with weaker effects. Because Minute phenotypes lie on a continuum and vary by genetic background, genes with intermediate effects might be classified differently in other experiments. Additional files 2 and 3 provide details of deletion coverage near Minute genes. This inventory updates the number of Minute loci given in Marygold et al.22 from 65 to 66 with the reclassification of RpL23.

Nine haplolethal or haplosterile loci do not encode ribosome-associated proteins (Table 2; Additional files 2 and 3). Four of the non-Minute genes (Fs(1)10A, Hdl, the unnamed haplolethal in 67DE and Ms(3)88C) have not been mapped to single transcription units, but the DrosDel Project localized Ms(3)88C to a small region by successive rounds of screens 9 and we localized Fs(1)10A, Hdl and the 67DE haplolethal similarly. The Tpl locus corresponds to a repetitive gene cluster 23. We defined a small region within the Tpl cluster that could not be further narrowed by deleting from the proximal and distal sides using the available FRT insertions, but it is possible that viability simply requires a minimal number of gene repeats. Thirty-five of our failed screens were devoted to mapping these non-Minute haploinsufficient loci.

Tables 2 and 3 present comprehensive inventories of haplolethal, haplosterile and Minute genes in D. melanogaster, but we also compiled a list of haploinsufficient genes associated with developmental or cellular phenotypes that do not affect viability or fertility enough to prevent the recovery of deletions (Table 4; Additional file 2). All were removed by molecularly defined deletions, but we made no attempt to flank these genes closely with deletions. Only the Hup locus has not been mapped to a transcription unit. We did not identify new haploinsufficient genes with obvious external morphological phenotypes and few are likely to exist. Additional haploinsufficient genes with cellular phenotypes may have been described in the literature and others will probably be identified in the future, so Table 4 is undoubtedly incomplete. We do not know how many haploinsufficient genes might exist with moderate viability or fertility effects similar to the Minute genes in Table 3 yet have no obvious external morphological effects.

The non-Minute haploinsufficient genes listed in Tables 2 and 4 implicate no single biological pathway or process, though there are a few discernable themes. The largest subset encodes the muscle components actin (Act88F), myosin (Mhc and Mlc2) and tropomyosin (Tm2) as well as a group of closely linked, muscle-related genes regulated by a haplolethal sequence within an intron of the Troponin I (wupA) gene 24. This subset may also include Hdl, which may correspond to Troponin T (up) 2125. Like the ribosomal protein genes, these genes may be particularly dosage sensitive because muscle assembly requires minimal levels or a particular stoichiometry of component proteins 222426. Other definable subsets encode homeodomain proteins (Abd-B, Dll, Scr and Ubx), Notch pathway components (Dl, H and N), Polycomb group repressor proteins (Pc and Pcl), apoptosis regulators (lok and p53) and melanin biosynthetic enzymes (b and e).

For years, the BDSC has maintained and distributed a set of deletions providing maximal genomic coverage using the fewest deletions deemed practical. The 'Deficiency Kit' was established before FLP-FRT deletions existed and deletions were added to the original deletion set when they provided novel coverage. Consequently, the kit consisted mostly of deletions with breakpoints mapped only at the level of polytene cytology. With large collections of molecularly defined deletions available, the kit had become outdated.

As the second goal of our project, we created a new Deficiency Kit and began distributing it in July 2009 while we were still screening for new deletions. We chose a tiling path of FRT-derived deletions from the three deletion projects and from individual investigators to provide maximal coverage. The remaining gaps in coverage were filled with older deletions where possible. In many cases, the older deletions were maintained in stock with duplications, allowing us to delete many regions containing haploinsufficient loci. To create a kit better suited for genetic modifier screens, we included the deletion pairs most closely flanking haploinsufficient genes even when they were removed completely by older deletions. Older deletions also allowed us to cover heterochromatic regions where FLP-FRT deletions currently do not exist and the fourth chromosome where few molecularly defined deletions exist. As we generated new deletions, we updated and improved the kit. Table 5 shows a summary of genomic coverage provided by the current kit. Additional file 4 lists the contents of the Deficiency Kit and Additional file 2 shows the deletion breakpoints relative to annotated genes. Information on the Deficiency Kit is also provided on the BDSC website 27.

Because the average size of the DrosDel, Exelixis and BSC deletions is smaller than the average size of the deletions in the previous Deficiency Kit, the new kit is larger (468 versus 271 deletions). Nevertheless, the advantages of molecularly mapped breakpoints and consistent genetic backgrounds offset the inconvenience of handling more stocks in experiments. As we will show in the next section, once a phenotype is mapped to a chromosomal interval using the new Deficiency Kit, its position can be narrowed to a very small interval or even a single transcription unit using additional molecularly defined deletions.

When a gene is mapped with deletions, it is usually localized to a chromosomal interval defined by adjacent breakpoints of different deletions rather than the breakpoints of a single deletion. For example, a mutation would map between the distal breakpoints of two overlapping deletions if it complemented the proximal deletion but not the distal deletion. Consequently, the critical factor determining the resolution of deletion mapping is the density of deletion breakpoints rather than deletion size.

The third goal of our project was to improve the distribution of deletion breakpoints across the genome and our target was to generate deletions so that the intervals between adjacent breakpoints would contain no more than a dozen protein-coding genes. We examined the distribution of breakpoints provided by the DrosDel and Exelixis deletions and undertook screens to interpose new breakpoints between them (Additional file 2). Figure 2 shows the distribution of interval sizes provided by deletions from the three projects. To construct this distribution, we assigned every euchromatic gene a position between adjacent deletion breakpoints based on where it would map uniquely if a mutation in the gene were complementation tested with the deletions. These hypothetical complementation tests allowed us to assign a gene to a single interval even when a deletion breakpoint fell within the gene. The median interval size for the entire genome is nine genes. The median interval size for chromosome arms X, 2R, 3L and 3R is also nine genes, but the median size for 2L, where there are more FRT-derived deletions, is only seven genes.

Many intervals exceeded our target size of 12 genes, though the largest interval contains only 36 genes (it lies on the fourth chromosome, where our screens were unsuccessful; the largest interval elsewhere contains 27 genes). The largest intervals could not be subdivided because no FRT-bearing transposon insertions existed in the interval or repeated attempts to subdivide the region with existing insertions failed. Other intervals exceeded our target size because gene annotations changed in the course of the project or because we counted genes for noncoding RNAs such as microRNAs and small nucleolar RNAs, which we did not consider in planning the deletion screens due to their poor annotation at the time.

As Figure 2 shows, it is now possible to map most mutations and genetic modifiers to very small groups of candidate transcription units with simple crosses. The deletion breakpoints even define 377 single-gene intervals. This degree of genomic subdivision makes deletion mapping the method of choice for most gene localization experiments in D. melanogaster.

Most P{XP}, PBac{WH} and PBac{RB} insertions used in deletion screens were obtained directly from Exelixis, Inc. The remainder came from the BDSC and the Exelixis collection at Harvard University. Other stocks came from Exelixis, the BDSC, the Harvard collection, the Szeged Stock Centre, the Drosophila Genetic Resource Center at the Kyoto Institute of Technology and the laboratory of Michael Ashburner at the University of Cambridge.

All genomic coordinates and gene counts are based on Genome Release 5.16. Cytological band positions were predicted from Release 5 coordinates using FlyBase map conversion tables 3940. We artificially defined the euchromatin/heterochromatin boundary as the most proximal extent of the contiguous genomic assembly for each chromosome arm in Genome Release 5.

The genomic coordinates of P{XP}, PBac{WH} and PBac{RB} insertions were initially determined at Exelixis and most were published in Thibault et al.10. Coordinates of unpublished insertions were obtained directly from Exelixis. Flanking sequences were reanalyzed by Roger Hoskins and Joe Carlson of Lawrence Berkeley National Laboratory and they submitted revised coordinates for many insertions to FlyBase. We used these reanalyzed coordinates where available. Otherwise, we used coordinates from a preliminary reanalysis (R Hoskins, personal communication) or the original Exelixis coordinates in that order of preference. The origins of coordinates are recorded in Additional file 1.

Parks et al.5 described methods for using P{XP}, PBac{WH} and PBac{RB} insertions to isolate chromosomal deletions in a standard genetic background. We followed the basic schemes developed at Exelixis for deletion screening, but we introduced some alternative stocks to improve the vigor of flies in the crosses and final deletion stocks. We built these stocks by carefully introgressing balancer and marker chromosomes into the standard background using chromosome substitution schemes patterned on Craymer 41. The source of the Y and fourth chromosomes was not controlled. (Details of stock constructions will be provided upon request.) Here we present the crosses we eventually found to work best for the isolation of X, second and third chromosome deletions. We identified deletions based on loss of miniwhite or the presence of two copies of miniwhite; see Park et al.5 and Results for the rationale. Tn{w⁺}1 and Tn{w⁺}2 are used generically to denote different miniwhite-marked transposon (Tn) insertions of P{XP}, PBac{WH} and PBac{RB}. Deletions were named with a BSC (Bloomington Stock Center) prefix followed by a number in the order they were placed into public distribution.

Four approaches were used to verify that the planned deletion was recovered from a screen: complementation tests, polytene chromosome cytology, 'hybrid PCR' and PCR with primers flanking transposable element insertions. The complementation tests were routine and require no elaboration. Polytene chromosomes were analyzed in standard lacto-aceto-orcein preparations 42 using the maps of Lefevre 43 and Sorsa et al.4445.

The 'hybrid PCR' approach was described in Parks et al.5. Briefly, primers were used to amplify a DNA fragment from the recombinant transposable element formed upon FLP-mediated recombination between two progenitor FRT-bearing insertions. We used the primers described in Parks et al.5 with the substitution of a new primer (5'-GCTTCTAAACGCTTACGCATAAAC GATG-3') for the RB3' plus and RB3' minus primer. We were unable to establish Hybrid PCR primers and conditions to verify deletions identified from an increase in miniwhite copy number; consequently, we tested only deletions detected by loss of miniwhite markers.

To verify that a particular chromosomal region was removed by a deletion, we designed PCR primers flanking the insertion site of a transposon located within the deleted region. From crosses, we isolated progeny carrying both the transposon and the putative deficiency and isolated DNA. With short extension times, a PCR fragment is amplified only when there is no transposon between the primer sites; consequently, a fragment is recovered if the noninsertion chromosome lacks a deletion. Since this approach verifies a deletion by a negative result, we repeated the test with three to five independent DNA preparations and appropriate positive controls. Transposon insertions and primer sequences used in these tests are given in Additional file 1. Additional details are available from Cook et al.21 where similar tests were used to confirm chromosomal duplications.

Stocks were placed into public distribution by the BDSC as soon as we verified deletions. We submitted a description of each deletion to FlyBase and Additional file 1 provides citations to these reports.

Stocks described in this paper may be obtained from the BDSC. Information on ordering, web pages devoted to deletion collections and the Deficiency Kit and lists of other stocks are available on the BDSC website 46. FlyBase 39 provides information on individual genetic components with cross references to BDSC stocks.