A De Novo Expression Profiling of Anopheles funestus, Malaria Vector in Africa, Using 454 Pyrosequencing

Background Anopheles funestus is one of the major malaria vectors in Africa and yet there are few genomic tools available for this species compared to An. gambiae. To start to close this knowledge gap, we sequenced the An. funestus transcriptome using cDNA libraries developed from a pyrethroid resistant laboratory strain and a pyrethroid susceptible field strain from Mali. Results Using a pool of life stages (pupae, larvae, adults: females and males) for each strain, 454 sequencing generated 375,619 reads (average length of 182 bp). De novo assembly generated 18,103 contigs with average length of 253 bp. The average depth of coverage of these contigs was 8.3. In total 20.8% of all reads were novel when compared to reference databases. The sequencing of the field strain generated 204,758 reads compared to 170,861 from the insecticide resistant laboratory strain. The contigs most differentially represented in the resistant strain belong to the P450 gene family and cuticular genes which correlates with previous studies implicating both of these gene families in pyrethroid resistance. qPCR carried out on six contigs indicates that these ESTs could be suitable for gene expression studies such as microarray. 31,000 sites were estimated to contain Single Nucleotide Polymorphisms (SNPs) and analysis of SNPs from 20 contigs suggested that most of these SNPs are likely to be true SNPs. Gene conservation analysis confirmed the close phylogenetic relationship between An. funestus and An. gambiae. Conclusion This study represents a significant advance for the genetics and genomics of An. funestus since it provides an extensive set of both Expressed Sequence Tags (ESTs) and SNPs which can be readily adopted for the design of new genomic tools such as microarray or SNP platforms.


Introduction
The mosquito Anopheles funestus is a major vector of malaria throughout much of sub-Saharan Africa.Its efficiency as a vector is partially conferred by its highly anthropophilic and endophilic behaviours and in many places, parasite infection rates of An. funestus exceed those of An. gambiae, the other major malaria vector [1].However, it has only recently started to receive the scientific attention such an important public health pest deserves.This is in part due to its relatively intractability to work with and difficulties in establishing laboratory colonies.Hence, there are far less genetic tools available for An.funestus than for An.gambiae.For example only 75 microsatellite loci have been identified in An. funestus compared with 300 in An. gambiae ; 600 Single Nucleotide Polymorphisms (SNPs) have been reported for An.funestus [2] compared to 400,000 in An. gambiae and only 2,846 ESTs (Expressed Sequence Tags) have been reported in An. funestus [3] compared to 153,165 in An. gambiae (http://www.ncbi.nlm.nih.gov).The genome of An. gambiae was sequenced in 2002 [4] while the genome of An. funestus is yet to be sequenced.This lack of genetic information has also resulted in a paucity of genetics studies in An. funestus, limiting the understanding of the biology of this important and neglected malaria vector.As an example, the availability of the high-resolution genetic information in An. gambiae has enabled microarray studies to be carried out in An. gambiae [5,6] in order to elucidate various aspects of the its biology.No such microarrays have been designed for An.funestus.There is therefore an urgent need to design more genetic tools for An.funestus.
Transcriptome sequencing is a suitable alternative to whole genome sequencing, because it is significantly cheaper, but still provides information for the transcribed portions of genes.Indeed, next-generation 454 sequencing allows highly parallel DNA sequencing that is many orders of magnitude faster than standard sequencing methods and increase sequencing depth and coverage while reducing time, labour and cost [7].Additionally these sequencing methods allow de novo sequencing, assembly and annotation of expressed genes [8] which makes it very suitable for a non-model species such as An.funestus.This will enable functional genomic studies in An. funestus using gene expression analysis, genetic and association mapping using the abundant SNPs identified.
454 sequencing could also allow differential gene expression analysis of the whole transcriptome between different phenotypes such as insecticide resistant and susceptible mosquitoes.
Here we report the comparative cDNA sequencing, using the 454 platform, between a pyrethroid resistant laboratory strain of An. funestus and a susceptible field sample.The main aim of this study was to generate functional genomic tools for this species.Additionally, we also attempted to identify potential genes associated with pyrethroid resistance by comparing the expression profiles of the two strains although different mosquito life stages where pooled for the sequencing.

Raw 454 data and assembly
The Roche 454 FLX platform produced 375,619 reads totalling 68,308,429 bp with an average read length of 182 bp similar to the 197 bp observed for the butterfly Meliteae cinxia [8].One quarter of the reads was between 237 bp and 277 bp in length and 50% of the reads between 185 to 285 bp in length.This dataset was trimmed using in-house tools to 213,410 reads of .30bp in length and totaling 40,078,792 bp with an average read length of 188 bp.The reads were assembled with Mira 2.9.15 [9] to create 18,103 contigs (average length of 253 bp) using 149,406 reads, of which 1,039 contigs were at least 500 bp in length, similar to that obtained by 454 pyrosequencing in other insects such as M. cinxia [8] and the six-spot burnet moth, Zygaena filipendulae [10].The average depth of coverage of these contigs (number of reads assembled into a contig) was 8.3 similar to that of M. cinxia [8].The sequencing reads have been deposited to NCBI's SRA database with the assembly output (Accession number: SRA009034) and contig sequences have been submitted to the Transcriptome Shotgun Assembly sequence database (TSA) of NCBI (accession number: EZ915182 -EZ933284).A comparison of the 18,103 An. funestus 454 contigs to An. gambiae and Aedes aegypti total transcript sets from Vectorbase (with TBLASTX, E = 10 23 ) as well as previously published An. funestus ESTs from Genbank showed a large degree of overlap between these species (Figure 1).41.2% (7,471) of the 18,103 contigs match an EST in these three transcriptomes.65% (1855 ESTs) of the 2,846 already published An. funestus ESTs (using SANGER sequencing) was represented in this 454 transcriptome.As expected, this percentage was lower for An.gambiae (49.7%) and Ae.aegypti (32.2%).The percentage of annotated An. funestus ESTs (41.2%) obtained in this analysis is similar to that observed in other large assembled EST data sets of non-mammalian species suggesting that the un-annotated contigs and even the singletons could represent a substantial portion of An. funestus transcriptome [8,10].
The assembly used 149,406 of the 213,410 reads, making use of 28,853,866 bp (72% of trimmed sequence).The remaining reads were singletons or considered either erroneous or contaminations.To test for these conditions, the unused reads were compared to nr database (the largest nucleotide database available through NCBI BLAST).Of the 64,004 unused trimmed reads, 58,205 were at least 50 bases in length and therefore suitable for BLAST with default parameters.13,716 (23.5%) match a reference sequence in nr (BLASTX, E = 10 25 , or 17,815 matches for E = 1) suggesting that a proportion of these singletons represents genuine ESTs but with probably low level of expression.
The 375,619 reads consisted of two independent 454 runs each covering half a plate.One run was from an insecticide resistant laboratory strain (FUMOZ) and the other from an insecticide susceptible field strain (Kela).The resistant run produced 170,861 raw reads, reducing to 98,021 after trimming and counting only those over 40 bp.The susceptible run produced 204,758 raw reads, reducing to 107,714.Assembly used 73,589 and 75,817 of these reads for resistant and susceptible strains respectively.Of the unused reads which 11,740 match nr, 6,574 were resistant reads and 5,166 were susceptible (BLASTX, E = 1) (Table 1).

Assembly quality
To assess the quality of the assembly, the 18,103 contigs were compared to themselves and to the unused reads of over 50 bp.A good assembly of the 454 reads should normally produce contigs with minimal sequence similarity (apart from conserved regions) to other 454 contigs.A successful comparison required the subject to be a full match to any part of the query.6,269 contigs showed variable degree of similarity with other contigs which is higher than that observed in a similar analysis for M. cinxia where only 8% of contigs had a match another contigs [8].Comparing the unassembled reads to the contigs, 12,775 reads over 50 bp in length matched the contigs.Considering the 6,269 contig-tocontig matches, the average redundant contig length is 151 bp, this ranged from 40 to 285 bp.The average target contig is 206 bp, this ranged from 40 to 2,126 bp.Considering the 12,775 read-to-contig matches, the average improperly unused read has a length of 171 bp, ranging from 50 (the minimum cut-off) to 285 bp.The average target contig is 216 bp, ranging from 50 to 2,126 bp.
The 6,269 contigs that match a query (i.e.exactly or partially match another contig) can only be assembly error, presumably cases where initially different contigs are built up using different reads with the same sequence, leading to the contigs independently converging to the same sequence.Contig diversity varies considerably, 6,269 of the 18,103 contigs share some similarity with other contigs when tested with E = 1.Consequently, the assembly demonstrates both error and sequence conservation in gene families, with 6,269 contigs showing some similarity and 11,834 unique contigs.These matches between contigs could also be due to conserved motifs in different genes, or the presence of variants of single genes created by alternative splicing or allelic differences within the population of individuals sequenced [8].
The failure of 12,775 reads to join a matching contig can be put down to basic assembler error or low quality sequences [7].454 sequencing produces a quality score for each base of sequence, which can be used by the assembler to compare likely reliability of key reads.Quality comparison of the assembled reads, unused reads and improperly unused reads shows the quality of the improperly unused reads was higher than other reads.
For most of the reads, the best hits matched a mosquito species in the nr database, most frequently An. gambiae.Figure 2 shows the number of best hits for each read matching any of the species represented in the nr database.Key species have been highlighted in Figure 2 with the wide variation in the number of genes in reference species to be noted.Although at a lower frequency, An. funestus reads also matched genes from distant species such as human or mouse probably for conserved genes.Analysis of the contig length distribution resulting from the assembly of reads shows a large peak at 240 bp (Figure 3a).The longest contig is 2,226 bp long and made use of 224 reads, coverage varied from 1 to 53 supporting reads.
As expected, the average coverage for short contigs (average with 100 base cut-off) was higher than for longer contigs greater than 500  bases (average with 500 base cut-off) (Figure 3b).Indeed, shorter contigs are more likely to be fully covered by single read which significantly increases their average coverage.In contrary for longer contigs, because there are no reads which fully cover the full length of the contig, their average coverage is lower as shown in Figure 3b.
The coverage based on contig length compared to reference length (An.gambiae) is shown in figure 3c.The reference length was normalised to unity and the average (blue) and maximum (black) depth of contigs was counted for each reference sequence.The maximum is effectively the longest contig matching a reference sequence.In each case, the number of reference sequences with the given coverage is summed.This figure shows that some references are very well covered.
The analysis was repeated (Figure 3d) but this time using absolute reference sequence length, showing the contig coverage of some of the longer genes.As can be seen, even the maximum (i.e.most optimistic) measure of reference coverage falters when considering large proteins.
The effect of the coverage depth of contigs in the BLASTX analysis is shown in Figure 4 through the ratio of contig and reference lengths.Contigs were compared to An. gambiae, Ae. aegypti and An.funestus using TBLASTX with an expectation of E = 10 23 and raw score of .45.Average contig coverage is the average number of reads that make up the individual contigs.Ratios well below unity can be considered hits inside gene families or low scoring false positives, only as the minimum alignment length increases can the validity of the assignment be reasonably certain.Each point is coloured according to reference sequence length, and it can be seen there is a general trend for greater than unity assignments to be to short references and less than unity assignments to longer references.The ratio of the length of individual contigs to the length of their reference orthologue coding region (An.gambiae, Ae. aegypti and An.funestus) increased with the depth of the coverage of the contig as observed previously in another 454 de novo assembly [8].Also included are three lines following three reference cut-off lengths of all, 0.8 to 1.8 K and .1.8K bp.In all three cases this

Coverage of An funestus transcriptome to Gene ontology (GO)
The GO coverage of the An.funestus transcriptome 454 contigs was assessed with comparison to that of An. gambiae transcriptome reference.For this purpose a Directed Acyclic Graph (DAG) rooted at ''biological process'' (one of the 3 GO categories with ''cellular component'' and ''molecular function'') was produced for transcriptomes of both species (Figure S1).When ''cellular component'' was used as the root, this produced a sparsely populated graph for both An.funestus and An.gambiae transcriptomes, while molecular function produced no graph at all for either species which is different from the GO pattern in Z. filipendulae where both ''cellular component'' and ''molecular function'' produced significant hits [10].Sequences from both An.funestus 454 contigs and the An.gambiae transcriptome reference were assigned to GO reference sequences using BLASTX.The DAG was followed to find all paths from the GO reference sequences to the root.A total of 45,021 paths to the GO category ''biological process'' were found for An.funestus and 88,647 paths for An.gambiae.The difference in the number of paths between the two species is probably due to a higher coverage of An. gambiae transcriptome than that of An. funestus.An analysis of the GO terms found for biological process indicated that An. funestus transcriptome as well as the An.gambiae reference transcriptome covers 23 terms (Figure S1).The proportion of each of the 23 terms was very similar between these two Anopheles species with the dominant term being ''biological regulation'' found at 42% in An. funestus and 40% in An. gambiae.Other major GO terms observed in the graph are ''metabolic process'' at 29% in An. funestus and 33% in An. gambiae; ''cellular process'' at 15% in An. funestus and 16% in An. gambiae.When moving to a level of two nodes from the root, 197 GO terms were covered by both transcriptomes.Most of these 197 GO terms are associated with only one of the 23 terms (found at one node from root) while few are found in two of the 23 terms (Figure S1).The distribution of these GO terms that were one edge away from the root (''biological process'') is also summarized in a pie chart for both species (Figure 5).The similarity of GO coverage observed between An. funestus and An.gambiae indicates that the 454 contigs obtained in this study represents a good coverage of An. funestus transcriptome.Although most of the GO terms covered by An. funestus 454 contigs are general biological terms, some are associated to insect species such as larval behavior, hatching behavior, mating and response to stimulus or stress.

Transcriptome profiling between resistant and susceptible samples
The most differentially expressed contigs were selected after firstly normalising the reads of each strain to the total reads number.The contigs were then ranked according to the read ratio between resistant and susceptible strains.The most differentially up-regulated contigs in the resistant strain were those with the highest ratio of (r2s)/(r+s) while the top up-regulated in susceptible were those with the highest ratio of (s2r)/(r+s), r and s being the number of resistant and susceptible strain reads in a contig.The analysis of the reads ratio between the resistant strain (FUMOZ-R) and the susceptible strain (Kela) showed a significant differential expression of some ESTs between these samples (Table 2 and Table S1).BLASTX results strongly indicate that the Each point is coloured according to reference sequence length, and it can be seen there is a general trend for greater than unity assignments to be to short references and less than unity assignments to longer references.Also included are three lines following three reference cut-off lengths of all, 0.8 to 1.8 K and .1.8K bp.doi:10.1371/journal.pone.0017418.g004 EST (EZ919527) with the highest FUMOZ-R/Kela read ratio (241 reads in FUMOZ-R and zero in Kela) corresponds to a cytochrome P450 gene with the closest match to the CYP4H18 P450 gene in An. gambiae (89% similarity of amino acid sequence).This result correlates with previous findings showing that P450s are involved in pyrethroid resistance in the FUMOZ-R An. funestus [11,12].It is also located on Chromosome 2 L where a QTL (named rp 2 ) associated with pyrethroid resistance in the FUMOZ  strain was detected [12].So far, no candidate gene for this rp 2 QTL has been identified and CYP4H18 should be further investigated to assess whether it is associated with rp 2 QTL.An EST matching a cuticular protein gene is also highly expressed in the resistant samples suggesting that this enzyme family could potentially be involved in pyrethroid resistance as already been suggested in An. gambiae and An.stephensi [13,14].Some ESTs were more highly abundant in the susceptible strain (Table 2).An analysis of these contigs revealed that they belong to genes with a possible role in larval ecology or environmental adaptation.This could be explained by the fact that this is a field strain facing a different ecological environment than that of the laboratory resistant strain.However, this difference could also result from the different geographic origin of the two strains (one from West Africa, Mali and another from Southern Africa in Mozambique).Additionally, the laboratory colonization of the FUMOZ-R strain from Mozambique could have led to significant bottleneck and genetic drift in this strain inducing an artificial selection that may have increased the difference in the transcription profile between the two strains.We cannot also rule out that bias due to incorrect assembly of reads in both strains have not impacted this transcription profile.

Validation of a set of contigs for gene expression studies by qPCR
qPCR reactions were carried out with a set of six contigs in samples from different life-stages and both sexes used for the transcriptome sequencing (female, male, pupae and 4 th instar larvae).These six contigs were chosen mainly because they belong to gene families known to be involved in insecticide resistance such as Glutathione-S-Transferases (GSTe2 and GSTd1-3), P450 genes (CYP6AH1 and CYP9J14) and the cuticular genes 76 and 125.
All the six ESTs were successfully amplified with qPCR products at the expected size in both FUMOZ and Kela strains.They are also expressed in both females and males, in pupae and larvae as well.Although based on a limited number of contigs, this result is an indication that contigs generated from this study could be suitable for gene expression studies in An. funestus particularly for microarray study.A custom-designed microarray chip could now be constructed using oligonucleotides generated from the 18,103 contigs of this study.Knowing that no microarray studies using specific An.funestus probes have been published so far, this will represent a significant progress for functional genomics in this species.
The expression levels of the 6 contigs were generally comparable between FUMOZ and Kela [for 20 out of the 24 comparisons (females, males, larvae and pupae for the 6 contigs)].However, a differential expression between the strains was observed for the contig corresponding to GSTe2 with a 2.5-fold change (P,0.05) for females and 2-fold change in pupae (Figure 6).GSTe2, which belongs to the glutathione transferase family, has been found to be involved in detoxification and in insecticide resistance in An. gambiae, the other malaria vector [5].However, although there is a significant difference in expression level for GSTe2, between resistant and susceptible, the fold change of the expression is significantly lower than the 18-25 fold change observed for previously identified genes conferring pyrethroid resistance in An. funestus, such as the duplicated P450 genes CYP6P9 and CYP6P4 [12].This suggests that if GSTe2 is playing a role in this resistance, it is probably a minor role.GSTe2 overexpression was higher in females than in males, a pattern previously observed for other genes related to pyrethroid resistance (CYP6P9 and CYP6P4) in An. funestus, where the resistance level is higher in females than in males [12,15].A 2.2-fold change overexpression of GSTd1-3 was also observed in Larvae for FUMOZ compared to Kela while a 4.1-fold change overexpression of the cuticular gene 76 was observed in Kela pupae compared to FUMOZ.The contigs corresponding to the P450s CYP6AH1 and CYP9J14 and the cuticular gene 125 did not show a significant difference.

SNP discovery
The use of two strains in this 454 sequencing, one from the field and the other from the laboratory, offers a good opportunity to identify additional SNPs in An. funestus.The MIRA assembler used for the reads assembly identified a total of 31,000 possible SNPs over 4.579 Mb of sequence.To assess the reliability of these SNPs, we carried out a visual inspection of 20 contigs coding for genes from three different biological functions; immunity, detoxification and housekeeping.We used the same criteria as done previously by [8] for SNP calling (at least two occurrence of the minority allele at sites covered by at least five sequences).The results are summarized in Table 3.A total of 14,295 bp of coding sequences was covered revealing 203 SNPs, an average of 1 SNP every 70 bp which is less polymorphic than what was observed for 50 genes previously sequenced using a conventional (SANGER) sequencing method, where 1 SNP was found in every 48 bp [2].This difference could be due to the fact that genes analysed here were selected from only three biological functions while the previous study was done with a set of genes randomly selected across the genome.This difference could also be due to the more stringent SNP selection used in this study and SNPs of low frequency could have been missed.More SNPs are found as expected in the non coding 39UTR sequences with 1 SNP every 49 bp but still lower than what was observed with SANGER sequencing [2] of 1SNP every 31 bp for non-coding regions.The 31,000 SNPs are indicated in the contig sequences submitted in TSA by degenerate letters such as Y for C/T, M for A/C etc.The frequency of transition and transversion substitutions was very similar between the twenty contigs and the previously fifty SANGER sequenced genes (Figure 7).Transition substitutions were more predominant than transversions at the same ratio as previously observed with SANGER sequenced genes (62% vs 38%) [2].Transitions C«T and A«G are over-represented with 35.9 and 27.0% of the total substitutions respectively once again very similar to the previous result (35.4 and 27.2% respectively).The four transversion classes also occurred at similar levels as seen previously (Figure 7).There was a higher frequency of SNPs occurring at the third coding position than in the first and second position as previously observed (Table 3).The difference in the frequency of transversion substitutions previously observed between non-coding regions (39UTR), coding regions and third coding position was also observed in this analysis (Table 4).These strong correlations suggest that most SNPs identified in this study are likely to be true SNPs.A detailed analysis showed a significant differences in the polymorphism level between genes from the three functional groups, with more polymorphisms observed for immunity genes (1SNP every 51 bp on average) followed by detoxification genes (1SNP every 65 bp) and housekeeping genes (1SNP every 96 bp) (Table 3).This is in accordance with previous studies in An. funestus or other mosquitoes indicating a higher genetic variability for genes associated with adaptation such as immunity genes [2,3,16].The average length of 39UTR in the 20 An. funestus contigs was 160 bp compared to 122 bp for An.gambiae.

Gene conservation between species
The twenty contigs analysed above provided an opportunity to investigate gene conservation between An. funestus and related mosquito species such as An.gambiae, Aedes aegypti and Culex quinquefasciatus.Orthologs or similar genes to the An.funestus ESTs in the three species were identified by a BLASTX search in vectorbase (http://www.vectorbase.org/index.php) and nr.A ClustalW alignment of all amino acid sequences was carried out and the percentage of similarity estimated.As expected, the highest similarity was always observed between An. funestus and An.gambiae irrespective of the functional group analysed, with an average of 79.35% similarity between the two anopheline species (Table 5).As reported in previous studies [3], a significant decrease of similarity percentage was observed when compared to Ae. aegypti and Cx quinquefasciatus (57.75 and 56.5% similarity respectively) which reflects the greater phylogenetic distance between An. funestus and these two culicines species.Overall, An. funestus ESTs are slightly closer to Ae. aegypti than Cx quinquefasciatus as also reported previously [3].This similarity pattern remains the same regardless of functional groups.However, the mean percentage of similarity differs significantly between functional groups with the lowest similarity observed for immune genes, while the housekeeping genes shown the highest similarity between species (Table 5).Genes involved in immunity are known to evolve rapidly, presumably in response to selective pressures imposed by parasites and this could explain both the highest genetic variability and the lowest conservation of these genes in An. funestus compared to the other two categories analysed.

Conclusion
This study has generated a large set of Expressed Sequence Tags (ESTs) and SNPs that are publicly available for the design of new genomic tools such as microarray or SNP platforms.This will facilitate further functional genomic and genetic studies in this important African malaria vector.In addition it has provided further support for the role of P450 genes in pyrethroid resistant while suggesting that other gene families such as cuticular genes could also be involved.

Mosquito strains, RNA extraction and synthesis of double-stranded cDNA
Total RNA was extracted from different life-stages of two strains of An. funestus, FUMOZ-R a laboratory strain resistant to pyrethroids [15] and a field collected pyrethroid susceptible sample from Kela, a village in Mali (11u889N, 8u459W).In order to obtain a large and broad transcriptome data set, RNA was extracted from 10 fourth-instar larvae, 10 pupae, 10 adult females and 10 males from each strain.The use of a field sample was thought likely to increase the genetic diversity of An. funestus covered by this study which is particularly useful for SNP discovery.
Total RNA was extracted using the PicoPure RNA isolation kit (Acturus, Mountain View, CA, USA) following the manufacturer's protocol.Total RNA quantity and quality were assessed on a 2% agarose gel and using the Agilent 2100 Bioanalyzer (Agilent technologies, Santa Clara, CA, USA). 10 ml RNA (at around 950 ng/ml) from each of the four life stages were combined to form a RNA pool for each strain.Double stranded cDNA was synthesized from each of these RNA pools using the SMART TM PCR cDNA synthesis kit (Clontech, Mountain View, CA, USA).Briefly, the first strand of the cDNA was synthesized using a mixture of 0.5 mg of total RNA for each pooled sample, 1 ml of 39SMART CDS primer II A at 12 mM (59-AAGCAGTGGTAT-CAACGCAGAGTACT(30)V N-39), 1 ml of SMART II A oligonucleotide at 12 mM (59-AAGCAGTGGTATCAACGCA-GAGTACGCGGG-39).After incubation and centrifugation, this mixture was completed with 2 ml of 56 first strand buffer, 1 ml of DTT at 20 mM, 1 ml of dNTPs (10 mM each) and 1 ml of MMLV reverse transcriptase (Clontech, Mountain View, CA, USA).This mixture was incubated for 1 h at 42uC.The double-stranded Quantitative PCR (qPCR) with highly differentially expressed ESTs We compared the expression pattern of the contigs of the most differentially expressed contigs between the resistant strain FUMOZ-R and the susceptible strain Kela using the GenomeLab GeXP genetic analysis system from Beckman Coulter.Total RNA was extracted from three batches (biological replicates) of fifteen mosquitoes from each life stage (1 day old female and male adults, pupae and fourth-instar larvae) using a PicoPure TM RNA isolation kit (Arkturis) according to manufacturer's instructions.Total RNA quantity and quality were assessed using Nanodrop spectrophotometer (Nanodrop Technologies, Oxfordshire, UK).Primers used are presented in Table 6.
The quantitative PCR reaction was carried out using the GenomeLabTM GeXP Start Kit (Beckman and Coulter) according to the protocol provided as described by [12].The expression level of the RSP7 ribosomal gene was used to normalise for variation in total cDNA concentration.

Supporting Information
Figure S1 GO coverage of An. funestus 454 contigs compared to An. gambiae transcriptome.The GO DAG has no single distance from leaves to root as multiple paths can exist towards the root that are of different lengths.Consequently, depth is based on node to node traversal depth from a root node.All paths from leaf to root were used in the analysis.Node usage was counted by incrementing a node use counter from leaf to root, for all connected nodes between leaf and root, for each blast hit between contig and GO reference sequence.Separate per node counts were maintained for An.gambiae and An.funestus 454 contigs, making it possible to compare relative node usage and hence relative GO term usage.(PDF) Table S1 List of all the contigs successfully annotated by BLASTX against the nr database.Further details are provided such as the number of reads in both strains, read ratio between the two strains, E-values, best hits in An. gambiae.(XLS)

Figure 1 .
Figure 1.Venn diagram showing distribution of similarity search results.Numbers are the sum of unique contigs matching An. gambiae, Ae. aegypti and An.funestus and given with their relative percentage.doi:10.1371/journal.pone.0017418.g001

Figure 2 .
Figure 2. Number of hits in reference species in the nr database using BLASTX (E = 1) of the unassembled reads.Corresponding species have been labelled.doi:10.1371/journal.pone.0017418.g002

Figure 3 .
Figure 3. Overview of assembly quality.Contigs were assigned to An. gambiae peptide reference using BLASTX with maximum expectation of 10.The cut-off was reduced to realistic levels using the stated minimum alignment lengths.(a) Histogram of contig lengths, smoothed by assigning contigs to partitions, each 10 bp in length.(b) Histogram of read alignment length compared to the contig it was assigned to.Read depth has been calculated from the assembly coordinates of each read, rounded to partitions of 1%.In effect, this shows coverage averaged across all contigs of lengths 100 and 500 bases.100 bases is inclusive, whereas 500 bases only considers contigs at least twice the length of the original reads.(c) Contig coverage (percentage) of BLAST assigned references, counting references, with minimum contig length of 500 bp.Shown is both the average contig coverage of references, and the maximum coverage, demonstrating some references are covered very well.(d) Repeating Figure c but using amino acid identity length rather than percent identity length.doi:10.1371/journal.pone.0017418.g003

Figure 4 .
Figure 4. Ratio of contigs and reference length, using TBLASTX against An.gambiae, Ae. aegypti and An.funestus, expectation of E = 10 23 and raw score of .45,against average contig coverage (number of reads).Ratios away from unity can be considered hits inside gene families or low scoring false positives, only as the minimum alignment length increases can the validity of the assignment be reasonably certain.Each point is coloured according to reference sequence length, and it can be seen there is a general trend for greater than unity assignments to be to short references and less than unity assignments to longer references.Also included are three lines following three reference cut-off lengths of all, 0.8 to 1.8 K and .1.8K bp.doi:10.1371/journal.pone.0017418.g004

Figure 5 .
Figure 5. GO pie chart: Pie chart summarizing the distribution of the 23 GO terms covered by An. funestus transcriptomes with comparison to An. gambiae.These GO terms are one edge away from the root (biological process).doi:10.1371/journal.pone.0017418.g005

Figure 6 .
Figure 6.qPCR gene expression profiles of six contigs between the susceptible strain [Kela (field, Mali) and a laboratory pyrethroid-resistant strain (FUMOZ-R).The normalised expression ratio of each contig against the RSP7 gene is represented on the primary vertical axis while the secondary vertical axis represents FUMOZ-R/Kela fold change of each contig.doi:10.1371/journal.pone.0017418.g006

Figure 7 .
Figure 7. Distribution of transitions and transversions among SNPs.In brackets are the estimates from the SANGER sequencing.doi:10.1371/journal.pone.0017418.g007

Table 1 .
Summary of reads number between samples.