The neurotranscriptome of the Aedes aegypti mosquito

Background A complete genome sequence and the advent of genome editing open up non-traditional model organisms to mechanistic genetic studies. The mosquito Aedes aegypti is an important vector of infectious diseases such as dengue, chikungunya, and yellow fever and has a large and complex genome, which has slowed annotation efforts. We used comprehensive transcriptomic analysis of adult gene expression to improve the genome annotation and to provide a detailed tissue-specific catalogue of neural gene expression at different adult behavioral states. Results We carried out deep RNA sequencing across all major peripheral male and female sensory tissues, the brain and (female) ovary. Furthermore, we examined gene expression across three important phases of the female reproductive cycle, a remarkable example of behavioral switching in which a female mosquito alternates between obtaining blood-meals from humans and laying eggs. Using genome-guided alignments and de novo transcriptome assembly, our re-annotation includes 572 new putative protein-coding genes and updates to 13.5 and 50.3 % of existing transcripts within coding sequences and untranslated regions, respectively. Using this updated annotation, we detail gene expression in each tissue, identifying large numbers of transcripts regulated by blood-feeding and sexually dimorphic transcripts that may provide clues to the biology of male- and female-specific behaviors, such as mating and blood-feeding, which are areas of intensive study for those interested in vector control. Conclusions This neurotranscriptome forms a strong foundation for the study of genes in the mosquito nervous system and investigation of sensory-driven behaviors and their regulation. Furthermore, understanding the molecular genetic basis of mosquito chemosensory behavior has important implications for vector control. Electronic supplementary material The online version of this article (doi:10.1186/s12864-015-2239-0) contains supplementary material, which is available to authorized users.

Our work builds on these efforts by incorporating biological replicates sequenced at greater depth and from many isolated tissues in parallel in both females at several behavioral states, and in males. This large dataset makes it possible to detect genes expressed at low levels or expressed in only a few neurons, and to identify differential gene expression with statistical confidence. Since the anatomical substrate of host-seeking, egglaying, and other mosquito behaviors is likely to be distributed across several tissues, parallel transcriptional profiling of multiple tissues in a single study increases the likelihood of capturing the full repertoire of genes involved in these complex behaviors.
To generate a transcriptome of peripheral and central neural tissues (or "neurotranscriptome") in Ae. aegypti, we performed Illumina mRNA-sequencing (RNA-seq) on RNA isolated from male and female tissues. Tissues sampled included the brain, antenna, maxillary palp, proboscis, abdominal tip, legs, and female ovary. To understand the influence of blood-feeding state on gene expression, we performed RNA-seq on a subset of tissues in female mosquitoes at three time-points: prior to a blood-meal (nonblood-fed), at 48 hours following a blood-meal (blood-fed), and at 96 hours following a blood-meal (gravid).
This project, as part of the NIAID VectorBase Driving Biological Projects Initiative [35], set out to accomplish three major goals: 1) to improve the existing annotation of protein-coding genes in the Ae. aegypti genome and identify genes not found in the current genome assembly; 2) to catalogue gene expression at the resolution of single tissues in host-seeking female and male mosquitoes; and 3) to identify changes in gene expression that are correlated with blood-feeding state and its associated behavioral changes. This neurotranscriptome significantly enhances the Ae. aegypti genome annotation, and identifies a large number of genes whose expression is sexually dimorphic and/or variable across the female gonotrophic cycle. We anticipate that these data will drive studies of the genetic and neural circuit basis of host-seeking and egg-laying behavior in Ae. aegypti.

Results
We profiled Ae. aegypti gene expression in tissues of males, and females at three points in their gonotrophic cycle ( Figure 1A). To confirm the distinct behavioral states associated with these time points in the mosquitoes used for our RNA-seq study, we measured responses to live hosts and carbon dioxide, and monitored female egg-laying. A uniport olfactometer was used to measure female mosquito attraction to a human forearm, which was presented along with carbon dioxide (CO 2 ) to activate the mosquitoes [12]. Nonblood-fed females showed strong attraction to these host cues, while blood-fed females did not respond and gravid females showed only modest host attraction ( Figure 1B). Because oviposition is known to release host-seeking suppression [11], egg-laying was prevented prior to behavioral testing or tissue dissection by depriving females of access to water. We confirmed that oviposition behavior was normal in these animals by placing individual blood-fed and gravid female mosquitoes into oviposition vials and scoring the number of eggs laid over an 8 hour period. As expected, blood-fed mosquitoes laid no eggs, but nearly all gravid mosquitoes did ( Figure 1C).
To determine whether the diminished responses to human hosts following blood feeding can be solely attributed to a reduction in sensitivity to CO 2 , we utilized a multi-insect three-dimensional flight-tracking system [8] to assess the response of a group of 20 female mosquitoes to a 40 second pulse of CO 2 ( Figure 1D and 1E). As previously described, nonblood-fed female mosquitoes displayed a robust increase in flight activity in response to CO 2 [8], while blood-fed females showed no increase in activity following administration of CO 2 . Gravid females displayed elevated pre-CO 2 baseline flight activity. Male Ae. aegypti mosquitoes also showed a strong response to CO 2 ( Figure 1D and 1E).
To study tissue-specific mRNA expression in isolated sensory and neural tissues of male and female Ae. aegypti, we dissected individual tissues from pools of non-blood-fed, blood-fed, and gravid female mosquitoes, as well as males. Tissue was flash-frozen and total RNA was extracted and used to generate Illumina RNA-seq libraries from polyAselected mRNA. From female mosquitoes, we generated libraries from brain, antenna, proboscis, maxillary palp, foreleg, midleg, hindleg, and ovary ( Figure 2A). We also dissected the rostrum, a tissue that includes both maxillary palp and proboscis, and the abdominal tip, defined in females as the three terminal abdominal segments, including genitalia and ovipositor (Figure 2A). From male mosquitoes, we generated libraries from brain, antenna, rostrum, foreleg, midleg, hindleg, and abdominal tip, defined in males as the -7 -three terminal abdominal segments, including genitalia ( Figure 2B). We generated at least three biological replicates for each tissue and subjected each to deep sequencing using an Illumina HiSeq instrument (with the exception of four libraries sequenced on the Illumina Genome Analyzer II) ( Figure 2C).
Correctly quantitating transcript expression from RNA-seq experiments depends on accurate gene models. Before analyzing gene expression across tissues, the two sexes, and the female gonotrophic cycle, we utilized our sequencing data from all tissues from non-blood-fed females and males to update the annotation of protein-coding genes in the Ae. aegypti genome ( Figure 2D). The depth, replication, and diversity of our sequencing allowed us to re-evaluate the existing annotation of protein-coding genes in the Ae. aegypti genome using two complementary approaches: de novo transcriptome assembly using Trinity [36] and alignment of sequencing reads directly to the reference genome. By aligning contigs from the de novo assembly back to the genome, we were able to combine data generated from these two approaches and use PASA2 software [37] to update existing gene annotations (AaegL2.1; obtained from VectorBase [35]). Reads were also aligned to the genome using STAR [38], and those aligning to genes were counted using featureCounts [39], allowing us to estimate transcript abundance and calculate differential expression at the gene level using DESeq2 [40].
We first carried out a principal component analysis of male and non-blood-fed female libraries to examine the clustering of data by tissue and sex. Large batch effects from library construction methods or problems with tissue contamination during dissection [41] may be revealed by this process. Virtually all of the biological replicates of the same tissue clustered tightly in principal component space, and for brain and legs across the two sexes ( Figure 2E).
A comparison of the protein-coding transcriptomes from the community annotations AaegL2.1 and AaegL3.3 and our updated transcriptome, termed AaegL.RU, can be found in Figures 3A-3C. In addition to updating existing gene models, analysis identified 403 putative novel protein-coding genes that did not overlap with existing gene annotations and >25,000 new putative alternatively spliced isoforms as compared to the AaegL3.3 geneset ( Figure 3A-B). Furthermore, we identified 169 predicted protein-coding transcripts with orthology to other insect transcriptomes on unaligned contigs from our de novo assembly that likely derive from unsequenced portions of the Ae. aegypti genome ( Figure 3B and Additional file 1). Our data also allowed for the extension of existing gene models, substantially increasing the length of UTRs and coding sequence for many existing genes -8 -as compared to the community reference annotations ( Figures 3B-C). Together, we propose that there are potentially 572 protein-coding genes missing from the currently published annotation, although we note that our data must be integrated with and evaluated in the context of the large body of other genomic, transcriptomic, and bioinformatics evidence available for Ae. aegypti [35]. The results of our reannotation, including novel transcripts, can be downloaded as Additional file 2.
Using our updated geneset annotation, we next classified genes into families related to neuronal function by both incorporating previously published classifications as well as considering their relationship to genes in the well-annotated and -studied vinegar fly Drosophila melanogaster. For pre-existing genes, we identified their closest orthologue in D. melanogaster using pre-calculated orthology calls of OrthoDB [42]. To account for genes added in our geneset re-annotation, and thus not considered by the OrthoDB databases, we additionally performed blastx of the predicted coding sequence of all transcripts against the D. melanogaster proteome (Flybase release 6.06) and report the top BLAST hits with evalues below 0.01 (Additional file 3). Of note, 163 of our 572 proposed novel genes (28.5%) have blastx hits that meet this criterion, as compared to 85.5% of all other annotated genes.
A single example of a novel protein-coding gene, RU318, is depicted in Figure 3D. It has high sequence conservation to the D. melanogaster TRP channel waterwitch [43].
Notably, the current Ae. aegypti geneset annotation lacks a predicted orthologue to waterwitch, while both other mosquito genomes (An. gambiae and Cu. quinquefasciatus) contain orthologues. Based on this sequence similarity, we have included RU318 in our revised annotation of TRP channels in Ae. aegypti. Finally, to aid in transgenesis and other genome engineering approaches, we used all predicted coding sequences to generate a consensus Kozak sequence for Ae. aegypti ( Figure 3E).
To describe transcript abundance across tissues, we mapped reads from our tissuespecific RNA-seq libraries to the AaegL3 genome and report transcript abundances in units of transcripts per million (TPM) [44]. Mapping statistics for each library can be found in Additional file 4, and TPM values for each replicate library can be found in Additional files 5 and 6. We first described the expression of genes related to neuronal function in specific tissues of non-blood-fed female and male Ae. aegypti.
Insects use three major neurotransmitters: acetylcholine is the primary excitatory neurotransmitter in the central nervous system, glutamate is used for neuromuscular transmission from motor neurons to muscles, and GABA is generally considered to be the primary inhibitory neurotransmitter in insect central synapses. Additionally, a number of -9 -biogenic amines including serotonin, dopamine, tyramine, octopamine, and histamine function as neurotransmitters or neuromodulators.
We identified acetylcholine receptor subunits by orthology to D. melanogaster and An. gambiae [45]. Expression of most acetylcholine receptor subunits was highest in male and female brain ( Figure 4A). We profiled the expression of three types of glutamate receptors: Kainate, NMDA, and AMPA ( Figure 4B). Overall, we found generally broader expression in peripheral tissues as well as brain, consistent with a role in neuromuscular transmission. Ionotropic GABA-A receptors and metabotropic GABA-B receptors were uniformly expressed broadly, with peaks in central brain ( Figure 4C). We found two histamine receptor orthologues: HisCl1 and HisCL2 ( Figure 4D). Both were expressed in brain at high levels, with HisCl1 (but not HisCl2) also expressed broadly in peripheral tissues. Interestingly, we were unable to identify an orthologue of the glycine receptor Grd Ae. albopictus mosquitoes fed constitutively with L-DOPA exhibit lower levels of hostseeking behavior [50]. Serotonin neurons innervate the antennal lobe of Ae. aegypti and An. gambiae [51], as well as the gut of Ae. aegypti [52] and serotonin has been shown to modulate feeding behavior in larval D. melanogaster [53]. Most serotonin receptors were expressed at appreciable levels in brain and in various peripheral tissues ( Figure 4E). Dopamine receptors had generally variable expression across tissues, including brain, legs, and antennae ( Figure 4F). Octopamine/tyramine receptors had generally lower expression values than other receptors, but were detected in brain as well as peripheral tissues ( Figure   4G). The TyrR orthologue AAEL004396 was highly and selectively expressed in ovary. We generally observed little obvious sexual dimorphism in neurotransmitter receptor expression. This suggests that there is a gross conservation of neuronal cell types and signaling pathways, at the transcriptional level, across male and female tissues.
Neurotransmitter processing enzymes can serve as cell-type specific markers to reveal the major neurotransmitters produced in particular tissues. The orthologue of D.
melanogaster glutamic acid decarboxylase 1, or Gad2, was expressed in brain and all peripheral sensory tissues, but not ovary. In contrast Gad1, was largely restricted to brain, as well as male abdominal tip ( Figure 5A). All dopamine processing enzymes were expressed in male and female brain, including the family of dopamine decarboxylase (Ddc) -10 -genes, as well as tyrosine hydroxylase (Th), tyrosine decarboxylase (Tdc), and Tyramine β hydroxylase (Tbh). We also observed strong peripheral expression of Dth and the two Ddc orthologues ( Figure 5A). Processing enzymes associated with GABA, glutamate, and acetylcholine were expressed in male and female brain, with many also expressed broadly across the majority of tissues sampled ( Figure 5A).
Genes corresponding to Ae. aegypti neuropeptides and neuropeptide receptors were defined by orthology to canonical insect neuropeptides and receptors [12,54,55]. Many neuropeptides were expressed primarily in brain of male and female mosquitoes, while a number had broader expression patterns that included but were not limited to brain ( Figure   5B). This is consistent with reports of the direct detection of neuropeptides in specific regions of the brain, including the antennal lobe [56]. Other peptides, including several predicted orthologues of eclosion hormone (EH), ecdysin-triggering hormone (ETH), and bursicon, were not detected at appreciable levels. We speculate that these genes are expressed selectively in earlier developmental stages that were not sampled in the present study of adult tissues. Neuropeptide receptors had generally broad expression patterns ( Figure 5C), indicating that neuropeptides may be centrally produced while exerting anatomically far-reaching humoral effects.
Insects sense chemical substances such as tastants, odorants, pheromones, noxious chemicals, and CO 2 with an array of chemosensory receptors encoded by large gene families of odorant receptors (ORs), odorant binding proteins (OBPs), ionotropic receptors (IRs), gustatory receptors (GRs), and PPK and TRP ion channels. Other groups have previously used RNA-seq to profile gene expression of a number of these genes in various tissues from several mosquito species [27,28,[30][31][32][33][34]. Our work builds on these efforts by simultaneously profiling expression of all of these chemosensory genes in multiple individual tissues in Ae. aegypti in the same study ( Figures 6 and 7).
ORs are an insect-specific family of divergent seven transmembrane domain chemoreceptors that sense volatile odors, including pheromones [57]. The majority of ORs are expressed in the antenna, with restricted subsets expressed in either proboscis or maxillary palp ( Figure 6A), consistent with previous reports of OR-expressing sensory neurons in these tissues in An. gambiae [58]. In contrast to the ORs, OBPs are expressed widely in the tissues profiled here, and vary greatly in their transcript abundance ( Figure 6B; note expanded TPM scale relative to Figure 6A). Similar results were found in an analysis of OBP expression in An. gambiae mosquitoes [31] and ants [59].
-11 -Expression (TPM) values for ORs were broadly elevated in female as compared to male antenna. Antennae of male Ae. aegypti are specialized for audition and contain an exaggerated pedicel at their base when compared to female antenna [60]. We speculate that extra cell numbers associated with this enlarged pedicel would effectively dilute mRNA coming from other cells in male antenna, thus reducing the tissue-wide abundance of odorant receptors and other genes expressed in olfactory sensory neurons. To account for these putative differences, we plotted the expression of ORs in male and female antenna ( Figure 6C) normalized to the olfactory co-receptor orco to reflect the approximate number of olfactory sensory neurons. This normalization depends on the assumption that orco expression is not sexually dimorphic, and therefore a reasonable proxy for olfactory sensory neuron number across sexes. Even after accounting for this normalization, we identified 12 OR genes with apparent enrichment in female antenna, with 10-to 35-fold increase in raw expression values as compared to male antennae ( Figure 6C and 6D). Because a number of untested assumptions were the basis of these conclusions, we note that these results would need to be validated independently, perhaps by RNA in situ hybridization, to gain cellular resolution of gene expression. Increases in mRNA expression could arise either by selective upregulation of gene expression in females, or through developmental changes that would lead to an increase in the number of neurons expressing these receptors in females. Because female mosquitoes show sex-specific chemosensory behavioral responses to odors associated with hosts and oviposition sites, it is not unreasonable to expect sex-specific differences in ORs tuned to these specific odors.
IRs are ligand-gated ion channels derived from variant ionotropic glutamate receptors that tend to be gated by ligands such as acids, aldehydes, and amines [61]. The IR family of Ae. aegypti has been previously described [62], and we identified predicted gene models for 2 additional IRs (RU164 and RU199) using blastx against the D. melanogaster proteome [Additional files 2-3]. We found three patterns of IR gene expression: those generally restricted to antenna; others selectively expressed in proboscis, rostrum, and maxillary palp; and a small number of IRs expressed across many different tissues examined ( Figure 7A). Similar results were previously reported in D. melanogaster, Apis mellifera, An. gambiae, and Culex quinquefasciatus [30,31,62].
GRs are a family of transmembrane receptors distantly related to ORs [57] that mediate detection of pheromones, tastants, CO 2 [63], and in the case of D. melanogaster Gr28b (Ae. aegypti Gr19), light and heat [64,65]. The annotation of the GR family of Ae.
Pickpocket (PPK) channels are a family of amiloride-sensitive DEG/ENaC sodium channels that are involved in the transduction of a number of sensory modalities, including mechanosensation, hygrosensation, and pheromone sensing [67]. We first identified Ae.
aegypti PPK channels by searching for orthologues to previously described PPKs in D.
melanogaster and An. gambiae [67]. Gene expression profiles of the PPK channels reveal broad expression of most genes across several peripheral tissues ( Figure 7C), including proboscis and legs, consistent with a role in various forms of contact chemosensation.
Transient receptor potential (TRP) channels have been implicated in diverse sensory modalities, including heat, light, and chemosensation [68]. Ae. aegypti TRP channels were identified by conducting orthologue searches against the 13 identified D. melanogaster TRP channels [68]. We identified orthologues to all 13, and two additional genes predicted to be orthologues of D. melanogaster painless. Expression of TRP channels was generally broad ( Figure 7D), with several interesting tissue-specific expression patterns, most notably in brain and antenna.
In addition to profiling specific gene families, we examined transcripts with sexually dimorphic expression ( Figure 8, Additional file 7). We compared expression in 6 tissues between non-blood-fed females and males ( Figure 8A-F). Differences may arise from sexually dimorphic expression within individual cell-types or differences in tissue-specific cell-type composition between sexes. To identify a set of broadly dimorphic transcripts, we imposed a more conservative threshold of a fold-change greater than 8 in any tissues, and examined them for overlapping dimorphism across different tissues ( Figure 8G and H).
Expression patterns of genes that were determined to be dimorphic in at least 3 tissues are indicated in gray in the Venn diagrams in Figure 8G and 8H, and displayed as heat maps in Figure 8I and 8J. We note the enrichment of newly annotated RU genes in the malespecific transcripts, including myo-sex (RU529) [69] and nix (RU468) [70]. This likely represents the ability of our RNA-seq data to capture transcripts produced from the Ae.
aegypti male-specific locus on chromosome 1 that have eluded classical genome sequencing, and therefore annotation, due to large repetitive regions.
Finally, we examined changes in gene expression across the female gonotrophic cycle. After locating and biting a host, female mosquitoes become engorged on a blood--13 -meal that can exceed their unfed body weight. Over the next few days, they must digest this blood and use its nutrients to mature a batch of eggs. During the first 48 hours following a blood-meal, mosquitoes are less responsive to host cues and demonstrate very little locomotion overall ( Figure 1B and 1D-1E) [11,12]. Dramatic changes in gene expression in An. gambiae [32,71], and olfactory function in An. gambiae and Ae. aegypti [72,73] after a blood-meal have been documented. Despite these interesting observations, the mechanisms of host-seeking suppression following a blood-meal in Ae. aegypti are not well understood. To examine the transcriptional changes associated with this behavioral shift, and begin to approach possible mechanisms, we asked whether there were significant changes in gene expression between non blood-fed, blood-fed, and gravid mosquitoes in

Discussion
Here we present the "neurotranscriptome" of brain and peripheral nervous system tissues in female and male Ae. aegypti mosquitoes. We used both genome-guided mapping of RNA-seq reads as well as de novo transcriptome reconstruction to improve the annotation of existing protein-coding gene models as well as identify 572 putative novel protein-coding genes. By mapping tissue-specific RNA-seq libraries to transcripts generated by these updated gene models, we examined gene expression in 10 female and 6 male tissues from non-blood-fed animals, as well as a subset of tissues from blood-fed and gravid female mosquitoes, representing two important and distinct behavioral states following a human blood-meal.
-14 -Given the fragmented state of the Ae. aegypti genome and gene annotations, it was important to include a de novo assembly approach in our analysis. This allowed us to examine the expression pattern of genes derived from unassembled regions of the genome. For example, a myosin heavy chain gene, RU529, is identical in sequence to Ae.
aegypti myo-sex [69], a gene linked to the sex-determining M-locus of Ae. aegypti in a region absent from the current genome assembly. A targeted search for genes in our dataset with similar expression patterns revealed additional genes with male-specific expression. Interestingly, novel unmapped genes from our de novo assembly are overrepresented in this list and suggest that these may also derive from unassembled genomic loci similar to the M-locus. A recent study confirmed the presence of such a factor, nix (identified in our study as RU426), and demonstrated its critical role in sex-determination in Ae. aegypti [70].
The present study is a valuable dataset, presenting a comprehensive view of proteincoding gene expression in adult tissues, and yet it remains incomplete. We only sequenced polyadenylated RNA derived from 10 adult tissues. We have not explored the repertoire of small RNAs including microRNAs, non-coding RNAs, or the regulation of alternative splicing. Our re-annotation approach relied on alignments of short reads and de novo transcripts back to the current draft of the genome, meaning that gene models residing on misassembled genomic contigs might be incorrectly represented. Indeed, a recent effort to generate a physical map for the Ae. aegypti genome found a misassembly rate of approximately 14%, including 6 of the 10 largest supercontigs [23], making it likely that many gene models that rely on the present assembly remain incorrect. Ultimately, a comprehensive annotation of protein-coding genes and non-coding loci within the Ae. aegypti genome will require the incorporation of additional genomic sequencing and transcriptomic data derived from distinct developmental stages and tissues [74].
Whole tissue RNA-seq can identify genes differentially expressed across male and female tissues. However, further work will be required to resolve gene expression profiles in individual cells and cell-types. For the purposes of this study, sexually dimorphic transcripts were conservatively defined as those for which the fold-change observed was greater than 8, though we note that there are many more transcripts with less extreme sex-biased expression. Anatomical differences will make it difficult to determine whether observed differences in transcript abundance represent differential regulation within shared cell-types or variation in the cell-type composition of male and female tissues. Interestingly, we describe relatively few examples of sexually dimorphic expression within the chemosensory gene families examined, suggesting that the striking behavioral differences seen between male and female Ae. aegypti may be encoded in the neural circuits responsible for the processing of sensory stimuli as opposed to gene expression differences at the sensory periphery.
We do note the statistically significant up-and down-regulation of a handful of olfactory receptors in antenna from gravid females. This is similar to an observed shift in OR expression in the antenna of An. gambiae following a blood-meal [32,71], and suggests that a behavioral shift from host-seeking to oviposition site selection may involve the increased expression of particular ORs tuned to ligands associated with oviposition sites and a concomitant decrease in expression of ORs tuned to host odor. With few exceptions [8,75], the ligand tuning of specific chemoreceptors has not been determined in Ae.
aegypti. A systematic effort to de-orphanize Ae. aegypti chemoreceptors will be required to address the functional relevance of these observed gene expression changes.
A major goal of this work was to identify gene expression changes correlated with blood-feeding state to gain insight into possible mechanisms by which a blood-meal might influence behavior. We describe many genes that change expression in tissues from bloodfed and gravid mosquitoes, including chemoreceptors, neuropeptides, neuropeptide receptors, and neurotransmitter receptors and processing enzymes, all of which might play important roles in the regulation of behavior and physiology. However, genes from these classes comprise a small minority of all regulated genes, and thus, are unlikely to alone explain the marked shifts in behavior as female mosquitoes transition from host-seeking to oviposition.
We envision this dataset as a resource to guide the selection of candidate genes involved in mosquito behavior as well as providing insight into the principles of gene expression regulation by blood-feeding. Transgenesis of mosquitoes [16] and precisely targeted mutagenesis with tools such as zinc-finger nucleases [3,8,12], TALENs [18,19], homing endonucleases [17], and RNA-guided nucleases [20][21][22] now allow for the generation of stable mutant lines and other genetic reagents to test the function of candidate genes in mosquito behavior.

Conclusions
We present a broad view of gene expression in non-blood-fed male and female tissues, focusing particularly on gene families related to neuronal function and chemosensation. We -16 -demonstrate that the effects of blood-feeding on gene expression are broad. This study represents the most comprehensive, tissue-specific survey of gene expression in adult Ae.
aegypti to date and will be foundational in our understanding of the molecular genetic basis of behavior in this important disease vector.

Mosquito rearing
Mosquitoes used in this study were from the genome reference Liverpool strain (LVPIB12) offered a human arm and allowed to feed to completion. Blood-feeding was verified by separating female mosquitoes with engorged abdomens 24 to 48 hours following bloodfeeding. At least 16 hours prior to dissections, mosquitoes were separated under coldanesthesia into groups of the appropriate size for a given library.

Mosquito behavior
Uniport experiments were carried out as described [12], with the exception of the stimulus, which was a 11 cm 2 circle of exposed skin created by cutting a hole in an elbow-length latex glove. CO 2 concentration in the airstream was measured at 5% with a Carbocap Hand-Held CO 2 Meter (model GM70, Vaisala Inc.). SciTrackS experiments were carried out as described [8]. Groups of 20 mosquitoes were placed into the flight arena, allowed to acclimate for 15 minutes, and then presented with a 40 second pulse of CO 2 .
Ethics, consent, and permissions All blood-feeding procedures and behavioral testing with human subjects were approved and monitored by The Rockefeller University Institutional Review Board (IRB; protocol LVO-0652). Subjects gave their written informed consent to participate in these experiments.

Tissue dissection and RNA extraction
Mosquitoes were cold-anesthetized and kept on ice until dissections were complete.   Transcriptome generation: reference-based mapping All reads from all libraries were aligned to the AaegL2 reference genome obtained from VectorBase [35] using Cufflinks2, Tophat2, and Bowtie2 software packages [76]. Reads were aligned without respect to existing annotations with the following settings: minimum -19 -intron length of 40 bp, maximum intron length of 500 MB. Cufflinks was run on reads from individual conditions and tissues to identify all putative splice junctions, and then combined using cuffcompare to identify a consensus set of putative splice junctions identified in our sequencing reads.
Transcriptome generation: de novo assembly We performed de novo assembly as a second approach to reconstruct transcripts from our data. All reads from all libraries were assembled into a genome-free de novo assembly VectorBase [35]) using the short read aligners BLAT and GMAP. These alignments were combined with the combined cufflinks output from genome-guided mapping to create assemblies of spliced alignments. These assemblies were compared to reference annotations (AaegL2.1; VectorBase [35]) and used to extend, update, or merge reference annotations. Additionally, this analysis identified 403 putative protein-coding genes not covered by the current annotation (see "Geneset annotation: naming of genes and geneset comparisons" below). Default PASA2 parameters were used with the exception of the number of allowed exons in 5' or 3' UTRs (--MAX_UTR_EXONS=3). Due to a technical oversight, 5 genes were added manually after the PASA2 run, using previously published coordinates: AaegGr27, AaegOr54, and AaegIr41d.2, AaegIr75k.4, and AaegIr7h.2.

Identifying novel unmapped genes
To identify novel transcripts that do not map to the current genome assembly, we filtered -20 -our de novo assembly as follows. First, we excluded all contigs that mapped to the genome or to cDNA from an existing transcript using GMAP. Next there was a match with an e-value of less than 0.01. 232 contigs that passed these conservative filters were considered to be high-confidence novel genes derived from portions of the genome that have not been sequenced or had assembly problems. These 232 contigs were collapsed using CD-HIT [78,79] resulting in 169 novel transcripts that were included in downstream analysis.

Geneset annotation: naming of genes and geneset comparisons
To name each gene in our updated geneset, we first compared them to existing annotations in AaegL3.3 using cuffcompare [76] and carried over accession numbers for those loci that were highly similar to existing annotations. Genes that did not match existing loci in these cuffcompare analyses are numbered sequentially as RU1-RU572 (Additional File 2). For genes with a VectorBase accession number, orthology to Drosophila melanogaster was While all sequencing reads were used for genome reannotation, aberrant clustering of transcriptome-wide expression patterns from two non-blood-fed female brain libraries resulted in their exclusion pool and the DESeq2 model was re-run. Additionally, signs of contamination of male rostrum libraries resulted in their removal from expression analysis (note that these are retained in the PCA plot in Figure 2E). Expression data and differential expression analysis were generated using the 'scale' function of R, and clustered using the hclust(method = 'complete') and dist('method = euclidean') functions in R. (Figure 9G).

Competing interests
The authors declare that they have no competing interests.

Authors' contributions
BJM, CSM, and MJD dissected tissues and made libraries and together with LBV conceived and planned the study; BJM carried out the experiments in Figure 1; BJM, CSM, and OD carried out bioinformatic analysis. BJM and LBV wrote the paper and produced the figures, with input from the other authors.

Description of additional data files
The following additional data are available with the online version of this paper.