The DArT technique is based on the analysis of "genomic representations", which are simplified surrogates of the DNA samples of interest. Concretely, a genomic representation is a set of DNA fragments of various sizes and sequences which are characteristic of the studied sample and obtained through highly reproducible (and preferentially technically simple) methods. These methods are usually based on restriction digestion: genomic DNA is digested using one or several restriction enzymes, with simultaneous ligation of appropriate adaptors to the restriction fragments and subsequent amplification of fragments by PCR using the adapter and the restriction site as targets for primer annealing 6. A suitable genomic representation typically includes 5,000–20,000 amplified fragments, i.e. a number low enough to ensure the reproducibility of the PCR reaction, but high enough to yield a reasonable number of polymorphic markers. Fragment sizes are ideally evenly distributed in a 100–1000 bp range, and representations showing distinct bands on agarose gel are avoided because these are presumably derived from repetitive genomic sequences and/or mitochondrial or chloroplast DNA.

In Ae. aegypti, several restriction enzyme combinations were tested (Additional file 1), but all of them gave genomic representations unfavourable to the application of the traditional DArT protocol, with clear repetitive bands and/or an unsuitable range size for the fragments (See Additional file 2 for an example of poor-quality genomic representations). On the basis of these results, a different strategy was thus adopted. The underlying idea was to exploit any kind of motifs occurring frequently in the genome as a second anchor during the PCR reaction, in addition to the adaptor-ligated restriction site. By adjusting PCR conditions, it was possible to preferably amplify fragments with the restriction site on one extremity and the chosen motif on the other one, so that the genomic representations to be obtained were expected to be a mixture of such fragments. Because of their abundance in eukaryote genomes, TEs were good candidates for such a purpose. We selected a particular MITE family named Pony to perform the role of second anchor in the Ae. aegypti genome (Figure 1). Pony TEs have all the characteristics of MITEs, including terminal inverted repeats, A+T richness and a small size 15. Two highly divergent subfamilies, Pony-A and Pony-B, can be distinguished and occur in about 8,400 and 9,900 copies in Ae. aegypti genome, respectively 15. We designed a primer targeting any Pony sequence present in the genome (PonyAll primer; Table 1), as well as one specific to the Pony-B subfamily (PonyB primer; Table 1).

A library comprising 6144 DArT clones was constructed using the approach described in Figure 1 (see Methods for details). Hybridizations were performed for 58 Aedes aegypti individuals with two, three and four replicated representations independently hybridized for 25, 4 and 29 individuals, respectively. These Ae. aegypti individuals belonged either to the Bora-Bora strain (29 individuals), susceptible to all insecticides, or to a strain artificially selected for several generations to develop resistance to the insecticidal Bacillus thuringiensis var israelensis (Bti) toxins (29 individuals; see Methods).

The polymorphism analysis was performed on the obtained images focusing on three parameters which are central for the data quality: (1) the Call Rate, which corresponds to the percentage of successfully scored replicates for a given marker; (2) the P value, which measures the fraction of the total variation across all individuals due to bimodality (i.e. polymorphism) for a particular marker; and (3) the discordance, which measures the overall variation of scores within replicates and is thus an indication of the marker reproducibility. First, the most unreliable markers (discordance > 5%) were discarded from the analysis. The remaining markers were then sorted out by decreasing P values and grouped in bins with an increment of 50 markers between two successive bins. As shown in Figure 2A, the average discordance increased as the average P value of a marker group increased (Pearson correlation coefficient = -0.996, p < 0.01). There was also a quasi-linear relationship between the decrease in average P and the decrease in Call Rate (Pearson correlation coefficient = 0.999, p < 0.01; Figure 2B). Overall, the top 350 markers (P = 80.06%) showed a satisfactory Call Rate (88.90%) while displaying an acceptable level of discordance (1.48%). They met the standard quality thresholds usually applied to DArT data, giving a polymorphism rate of about 5.7% in the whole mosquito dataset.

The primary goal of this study was to characterize loci potentially responsible for resistance to Bti. For this purpose, population genomics was adopted to reveal loci displaying apparent selection footprints, such as an atypical pattern of genetic variability compared to the rest of the genome. To implement this approach, a robust estimation of the overall genetic diversity throughout the genome first had to be obtained. This was achieved by slightly relaxing the quality parameters to include more markers in our analyses and allow a more comprehensive sampling of the genome. A polymorphism analysis was thus performed with a minimum Call Rate, minimum P value, and maximum discordance set at 81%, 71%, and 5%, respectively. It identified a set of 476 markers (polymorphism rate = 7.75%) with a mean Call Rate, P value, and discordance of 87.72 %, 78.34 %, and 1.93%, respectively.

In this working dataset comprising 476 markers, linkage disequilibrium was low with only 0.72% and 1.90% of pairwise linkage indices > 0.95 and < 0.05, respectively. Furthermore, a subset of 70 markers involved in 76.92% of the marker pairs showing high linkage disequilibrium was sequenced. Only two of them (2.86%) turned out to be redundant, with one pair of markers differing by a gap and the second one by a mutation (similarity > 98.5% in BioEdit). After trimming the Pony motif, the 68 unique marker sequences (GenBank accession no. FJ231034–FJ231090; sequences shorter than 50 bp could not be deposited) were blasted against the Ae. aegypti genome. In total, 41 of them could be assigned without ambiguity to single genomic positions distributed on 40 different supercontigs. The two markers situated on the same supercontig were separated by more than 215 kb.

The working dataset was used to assess the genetic diversity observed between and within the two studied mosquito strains. As revealed by an analysis of molecular variance (AMOVA), most of the genetic variation was distributed between the resistant and the susceptible strains (69.6 %, versus 30.4 % within strains; p < 0.0001 in both cases), corroborating the high genetic differentiation observed between the two strains (Fst = 0.556). Likewise, in a principal coordinate analysis (PCO), the first axis unmistakably separated the two strains and explained 56.59 % of the variation (Figure 3). However, when the PCO analysis was carried out on each strain independently, enough genetic diversity seemed to be retained within each strain to clearly differentiate individuals (data not shown), with the first two axes explaining 20.48 % and 26.72 % of the variation for the susceptible strain and the resistant strain, respectively.

Assessment of genetic diversity within strains allowed to complete this picture and revealed a strong difference between the two strains (Table 2). All the diversity indices calculated (Shannon index of phenotypic diversity S, mean Jaccard pair-wise coefficient, Nei's gene diversity and proportion of polymorphic markers at the 5% level, see Methods for more details) give evidence of a high level of genetic diversity within the susceptible strain, whereas these indices were substantially lower for the resistant strain, except the mean pairwise Jaccard coefficient.

Two laboratory strains of Ae. aegypti were used as biological material in this study: the standard Bora-Bora strain, susceptible to all insecticides, and a strain artificially selected for several generations to develop resistance to a decomposed tree leaf litter showing a high toxicity when ingested by mosquito larvae. This toxic litter was proved to contain Bacillus thuringiensis var israelensis (Bti) bacteria strains from commercial origin 28. Bti bacterium produces insecticidal toxins which are widely used for mosquito control, and the toxic litter was collected in a mosquito pond in Eastern France three months after treatment with commercial Bti insecticide (Bactimos, Valent Biosciences Corporation). This experimental design allowed us to study resistance mechanisms to Bti toxins in a situation close to field conditions.

Selection of the toxic leaf litter resistant strain was performed on early fourth-instar larvae of the Bora-Bora strain. At each generation, groups of 200 calibrated larvae were exposed to 30 mg of finely ground toxic leaf litter in 200 ml of tap water. Selection was carried out repeatedly for 18 generations, with an average of 2000 larvae being exposed to toxic litter per generation. At each generation, the experiment was stopped when mortality reached 80% in order to obtain a minimum of 300 adults for the next generation. The survivors were transferred to clean water, fed with hay pellets, and allowed to emerge as adults, reproduce, blood feed and lay eggs for the next generation. The average generation turnover was 30 days.

To monitor the evolution of resistance to toxic leaf litter, bioassays were conducted at each generation in plastic cups containing 20 fourth-instar larvae in 50 ml of tap water and various doses of toxic leaf litter 29. The lethal dose for 50% of individuals after 24 h exposure (24 h-LD₅₀) was determined using the Probit software 30. The resistance ratio (RR) of the selected strain was calculated by dividing the 24 h-LD₅₀ value of the selected strain with the value obtained for the susceptible strain. After 18 generations of selection, the RR of the resistant strain was 4-fold.

Genomic DNA was extracted from fourth-instar larvae using the Qiagen DNeasy Tissue Kit and protocol (Qiagen). To avoid bacterial contamination, the larvae midgut was removed carefully before DNA extraction.

For each sample, digestion and ligation reactions were carried out simultaneously at 37°C for 3 hours on 50 ng of genomic DNA, using 2 units of restriction enzyme Bsp1286I (New England Biolabs, NEB), 80 units of T4 DNA ligase (NEB) and 0.05 μM Bsp1286I adaptors I (Table 1), in a buffer with final concentrations of 10 mM Tris-OAc, 50 mM KOAc, 10 mM Mg(OAc)_2, 5 mM DTT (pH 7.8), 1 mM ATP and 100 ng/ml Bovine Serum Albumin (NEB). The obtained ligated products served as template in a first round of PCR amplification. For this purpose, ligated products were diluted five times with sterile water and 2.5 μl of the diluted product were added to a 22.5-μl PCR reaction mix leading to final concentrations of 0.04 μM of Bsp1286I primers I (Table 1), 0.4 μM of PonyB primers (Table 1), 10 mM Tris-Cl (pH 8.3), 50 mM KCl, 1.5 mM MgCl₂, 0.1 mM of each dNTP and 1 unit of RedTaq DNA Polymerase (Sigma). The amplification reaction was performed with the following conditions: 94°C for 1 min; 20 cycles of 94°C for 30 sec, 50°C for 40 sec and 72°C for 1 min; followed by a final 7-min extension step at 72°C. The resulting PCR product was diluted 5 times with sterile water and 2.5 μl of the diluted product served as a template for a second round of amplification performed exactly as the first round except that the final volume was 50 μl, the final concentration of both primers and of each dNTP was 0.2 μM and 0.05 mM, respectively, and 2 units of RedTaq DNA polymerase were used.

A preliminary 1536-clone library was first constructed based on genomic representations prepared for 29 individuals of each strain. These representations were obtained as described above, except the sequences of the adaptors and primers were slightly different (Bsp1286I adaptors II, Bsp1286I primer II and PonyAll primer; see Table 1), allowing the amplification from any type of Pony element (subfamily A or B). Representations were mixed according to the origin of the individuals to form a "susceptible pool" and a "resistant pool", and these pools were cloned separately in the PCR2.1 TOPO vector (TOPO TA Cloning kit, Invitrogen) following the manufacturer's instructions. Individual clones were grown overnight in 384-well plates containing LB medium with 100 μg/ml ampicillin and 4.4 % glycerol. Small aliquots of the cultures were used as templates for insert amplification in a 25-μl reaction containing 0.2 μM of each M13 forward and M13 reverse primers (Invitrogen), 50 μM of each dNTP, 50 mM Tris, 6 mM HCl, 16 mM (NH₄)₂SO₄, 1.5 mM MgCl₂ and 2 units of Taq Polymerase. The cycling conditions were as follows: 95°C for 4 min, 57°C for 35s, 72°C for 1 min followed by 35 cycles of 94°C for 35s, 52°C for 35s and 72°C for 1 min and final 72°C for 7 min. After amplification, PCR products were dried, washed with 70% ethanol and resuspended in a spotting buffer developed for poly-L-lysine coated microarray slides (Wenzl et al., in prep.). The final library contained 1536 clones, half of them originating from the "susceptible pool" and half of them from the "resistant pool", so that each pool of genetic diversity was equally represented.

The first genotyping experiments carried out with this preliminary library resulted in a low number of reliable polymorphic clones due to high levels of background noise in signal intensities (data not shown). One likely explanation was that the genomic representations hybridized against the library included too many fragments so that some of them were amplified stochastically during the two rounds of PCRs and/or did not hybridize specifically. To solve this problem, a second library was built, which was based on genomic representations with fewer fragments. It relied on the use of the PonyB primer designed to anneal only to the Pony-B sequences. Except for this different primer, the protocol was identical to that detailed above and lead to the production of a 4608-clone library. Clones from both libraries, i.e. 6144 in total, were printed in duplicates on poly-L-lysine-coated slides (Erie Scientific) using a MicroGridII arrayer (Biorobotics). Printed DNA spots were denatured and fixed on the surface of the slide by incubation in hot water (95°C) for 2 min, followed by dipping in 0.1 mM DTT, 0.1 mM EDTA solution and drying by centrifugation at 500 × g for 7 min.

For each individual, at least two genomic representations were obtained independently as reported in the section "Preparation of genomic representations". Each representation was subsequently precipitated with one volume of isopropanol, washed with 70% ethanol and resuspended in 3.5 μl of sterile water. After a 3-min denaturation at 95°C, the representation was fluorescently-labelled for 3 hours at 37°C with 250 units of Klenow exo^- fragment of E. coli Polymerase I (NEB), 2.5 nmoles of either Cy3-dUTP or Cy5-dUTP (Amersham Bioscience) and 25 μM random decamers. A Cy3- and a Cy5-labelled samples (thereafter called targets) were combined one after the other to 60 μl of a hybridization buffer containing a 50:5:1 mixture of ExpressHyb (Clonetech), herring sperm DNA (Promega), FAM-labelled polylinker of the PCR2.1 vector (Invitrogen) used for library preparation, and 2 mM EDTA (pH 8.0). This mix was denaturated at 95°C for 3 min, deposited onto microarray slides and covered with a glass coverslip. Slides were incubated for 16 h in a humid chamber at 65°C. Following hybridization, coverslips were removed and slides were washed in 1 × SSC + 0.1% SDS for 5 min, 1 × SSC for 5 min, 0.2 × SSC for 2 min, and 0.02 × SSC for 30 sec, before being dried by centrifugation at 500 g for 7 min.

A Tecan LS300 confocal laser scanner was used to scan the hybridized slides and generate three different TIF images per slide, one per type of hybridized dye (Cy3, Cy5 and FAM). Image and polymorphism analyses were performed with DArTSoft version 7.4.3, a software especially developed for this purpose by Diversity Arrays Technology Pty. Ltd. (Cayla et al., in prep.). Briefly, DArTSoft automatically localizes the arrayed spots on the images using a seeded-region-growth algorithm, rejects those with a weak reference signal, computes and normalizes background-subtracted relative hybridization intensities [e.g. log(cy3-target/FAM-reference)], and calculates the median value for replicate spots. Then, polymorphic clones are identified by means of a combination of ANOVA and fuzzy K-means clustering at a fuzziness level of 1.5 before being assigned as 'present' or 'absent' in each representation hybridized to the array.

Independence and uniqueness of DArT markers were evaluated by calculating the linkage index I_{k, l} for each possible pair of markers k and l, according to the following formula:

I k , l = 1 n ∑ | m k i − m l i | MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xI8qiVKYPFjYdHaVhbbf9v8qqaqFr0xc9vqFj0dXdbba91qpepeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaemysaK0aaSbaaSqaaiabdUgaRjabcYcaSiabdYgaSbqabaGccqGH9aqpjuaGdaWcaaqaaiabigdaXaqaaiabd6gaUbaakmaaqaeabaWaaqWaaeaacqWGTbqBdaWgaaWcbaGaem4AaSMaemyAaKgabeaakiabgkHiTiabd2gaTnaaBaaaleaacqWGSbaBcqWGPbqAaeqaaaGccaGLhWUaayjcSdaaleqabeqdcqGHris5aaaa@43E2@

where n is the number of individuals, m_ki the score (0/1) of individual i at marker k, and m_li the score of individual i at marker l. Values of I < 0.05 or I > 0.95 are indicative of a statistical linkage disequilibrium between the two markers under consideration.

A subset of markers involved in pairs showing high linkage disequilibrium was selected and sequenced to assess the level of marker redundancy in the dataset. For these markers, bacterial cultures were sent to Genome Express^{® (}http://www.genome-express.com) for insert amplification and sequencing with M13 forward and M13 reverse primers. Raw sequence files were trimmed and aligned using Bioedit 7.0.9 (http://www.mbio.ncsu.edu/BioEdit/BioEdit.html). Marker sequences were also blasted against the full genomic sequence of Ae. aegypti (available for download at http://aaegypti.vectorbase.org/GetData/Downloads?type=Genome and comprising 4758 supercontigs in total).

Genetic variation was assessed within each mosquito strain by computing the Shannon index of phenotypic diversity S 31 with Popgene v1.32 (http://www.ualberta.ca/~fyeh/pr01.htm) as well as the pair-wise Jaccard coefficients 32 with the vegdist function of the vegan R package (http://cc.oulu.fi/~jarioksa/softhelp/vegan.html). These two diversity indices do not rely on the estimation of allelic frequencies, which for dominant data such as DArT data requires additional assumptions (e.g. Hardy-Weinberg equilibrium). In addition, estimates of allelic frequencies were obtained with the Bayesian method with non-uniform prior distribution 33 implemented in AFLP-SURV v1.0 (http://www.ulb.ac.be/sciences/lagev/aflp-surv.html) 34, and used to calculate Nei's gene diversity 35 and the proportion of polymorphic markers at the 5% level within each strain. Genetic differentiation between strains was estimated by performing an analysis of molecular variance (AMOVA) using Arlequin v3.11 (http://cmpg.unibe.ch/software/arlequin3/) 36 and by calculating the Fst index with AFLP-SURV v1.0. Principal coordinate analyses (PCO) were carried out with PCO v1.0 (http://www.stat.auckland.ac.nz/~mja/Programs.htm).