As expected, our tests revealed high statistical association of VGSC, the target of permethrin, in all three treatment groups (S2 Table). VGSC is located on the q arm of chromosome 3 [21,24] (Fig 3). Intriguingly, some genes in the high association group interact with VGSC in other organisms. For example, CHL (LOC5578279), a neuronal-specific zinc-Cys2His2 type transcriptional regulator [25], genetically interacts with the drosophilid ortholog of VGSC (PARA) [26]. Synaptic control of motoneuron function has been characterized in dipteran models of epilepsy, wherein increased frequency and regularity of calcium signaling waves, synaptic excitation, are indicative of pathology [26–29]. For example, in a drosophilid epilepsy model, reduced expression of CHL further increased sensitivity of the PARA bang-sensitive (PARA ^bss) allele, which has a high basal sensitivity to paralysis. PARA ^bss was also sensitive to alterations in calcium levels [29], which is consistent with a functional interaction between Vgsc and calcium channels. Similarly, deltamethrin has been shown to reduce calcium channel efflux in mammalian and mosquito cell culture [30–32]. Consistent with this evidence, a calcium channel alpha subunit (caα1, LOC5564339) was also within the high association group.

The ankyrin protein domain is present in a large variety of structural proteins that link transmembrane proteins to the actin cytoskeleton [33,34]; they serve a variety of functions, including cytoskeletal anchoring and mechano-transduction of sensory responses [35,36]. A subset of these proteins physically interacts with Vgsc in mammals [28]. Five ankyrin domain-containing proteins were also identified in our high association gene set (S2 Table). Finally, NEDD4, also a high association gene, has predicted physical interactions with N_AV (Vgsc) in mammals (S2 Table) [37]. Additional support for N_AV-Nedd4 interactions was also observed in its regulation of cardiac N_AV [38]. NEDD4 codes for E3 ubiquitin-protein ligase, which regulates degradation of plasma membrane proteins with PY motifs [39]. Aedine Vgsc has a partial PY motif (PPS, amino acids 1170–1172) rather than the canonical PY of mammalian N_AV proteins (PPSYDSV) [40]. Until now, no support for VGSC-NEDD4 interactions has been reported for dipterans.

The importance of each of the associations summarized above was further substantiated by GSEA of each corresponding functional category. We tested each functional group for relative enrichment within the high association set compared to all coding genes. This was done using a hypergeometric test, which models a binomial distribution and tests the probability of a gene being represented more often than expected by chance [41]. Specific results are described below in the context of each functional group.

To further explore the role of selected genes in pyrethroid resistance, we evaluated 2 putative transcription regulators in the high association data set. We hypothesized that when silenced, relevant components would result in increased pyrethroid susceptibility. One gene, NVY (LOC5563881), codes for a predicted central and peripheral nervous system-specific transcriptional regulator [42] with dual functions in axon guidance [43] and transcriptional repression [44]. The other gene, CHL, described above, is genetically associated with VGSC in drosophilids. We used RNA interference-induced gene silencing via dsRNA-injection to temporarily silence each gene in Vergel colony mosquitoes (F₂₄). Silencing was confirmed relative to levels in non-specific dsRNA (β-Gal-dsRNA) treated controls (Fig 4A), and mosquitoes were subjected to a 1.5 μg permethrin discriminating dose in standard CDC bottle assays [45]. Both nvy-dsRNA and chl-dsRNA injected groups were significantly more susceptible to permethrin knockdown than (β-GAL- dsRNA)-injected controls (analysis of variance (ANOVA), p<0.0001) (Fig 4B). In contrast, PBS-injected controls showed sensitivity and viability levels that paralleled that of the β-GAL-injected group, indicating that exogenous dsRNA treatment alone had no effect (S1 Fig). The proportion of mosquitoes with lethal phenotypes at 24 hours post-treatment was also significantly greater in nvy-dsRNA and chl-dsRNA groups than that of β-GAL-dsRNA-injected mosquitoes (Fisher's exact test, p = 0.001 and p<0.0001, respectively) (Fig 4B).

Analysis of VGSC transcript levels showed a ~50% reduction following nvy-dsRNA or chl-dsRNA treatment compared to β-GAL-dsRNA-injected controls, indicating that each transcriptional regulator plays a role in modulating VGSC transcript levels (Fig 4C). The association mapping data was interrogated to identify CHL and NVY non-synonymous mutations in each of the collections. No non-synonymous mutations (LOD >4.0) were identified in either CHL or NVY in the VCD collection, suggestive of cis-regulatory effects. A single common mutation at A2T occurred in VCP and VeP, with a few other isolated amino acid substitutions in CHL (Fig 4D).

Detoxification is a common pyrethroid resistance mechanism [14,15,46–48], and several key enzymes in this category are within the metabolic functional group. Overall, seventy-eight high association genes had predicted activity in metabolic processes, as shown in Fig 2B (S2 Table), (78/1053), hypergeometric test p value = 0.02). Though type I, or non-cyano group bearing pyrethroids, eg., permethrin, and type II, or alpha-cyano group bearing, eg., deltamethrin, have distinct toxicity mechanisms [30], individual detoxification genes may respond to both types I and II [49]. The major pyrethroid detoxification proteins are mixed function oxidases and alpha- and beta-esterases (S2 Table) [50,51].

In anophelines, an acetylcholine esterase resistant allele (ACE-1^R) can partially reduce deltamethrin sensitivity, and ace-1^R works synergistically with VGSC resistant alleles to substantially reduce sensitivity to permethrin [52]. However, there is little data to indicate an association between selection of acetylcholine esterase alleles and permethrin resistance in Ae. aegypti. Nevertheless, two acetylcholine esterases were included in the high association group (S2 Table, LOC5570776 and LOC5578456). Three genes coding for carboxylesterases (COEs, LOC5564561, LOC5576941, LOC5571034) were also in this group. COEs have been tied to organophosphate resistance mechanisms and, to a lesser extent, to deltamethrin resistance [53]. In addition to supporting a role for these genes in pyrethroid detoxification, their presence in the common gene set is consistent with the idea that there may be some overlap between type I and II pyrethroid detoxification mechanisms. In contrast, there were no epilson class glutathione S transferases in the high association group, which is surprising, given that they have been shown to be pertinent to resistant phenotypes [12,14,15,54–56].

Over-representation of the metabolic gene category is consistent with studies in drosophilids wherein amino acid catabolism was tied to permethrin sensitivity [57]. Specifically, Brinzer et al. found a decrease in proline levels in resistant larvae that had been exposed to permethrin, which could be explained by either increased conversion from proline to glutamate or inhibition of the proline precursor ornithine. Conversion of proline to glutamate has also been reported upon exposure to malathion or DDT [58,59]. In the current work, the aedine ortholog of 5-oxoprolinase, responsible for the degradation of glutamate, which would be required if it was in excess, is present in the high association group (S2 Table) [57].

Additional metabolic pathway genes in the high association group included lipid biosynthetic genes (S2 Table), for which transcripts are reportedly enriched in temephos-resistant mosquitoes [60]. Interestingly, there were also multiple genes that participate in lipid metabolic processes, ie., 3 lipases and 4 phospholipases (S2 Table). Increased lipase activity has been tied to insecticide resistance in Tribolium spp [61]. One possible explanation of this is that the catabolism of choline phospholipids is protective against the loss of choline, caused by the release of acetylcholine from synaptosomes [62] and inhibition of acetyl cholinesterase [63] that occurs in pyrethroid-susceptible insects. Similarly, a sphingomyelin phosphodiesterase (SMAse) activator domain protein was also in the high association group. SMAses are hydrolases that break down sphingomyelin into phosphocholine and ceramide (reviewed in [64]).

Some pyrethroid detoxification proteins are in the redox functional category. For our analyses, the redox and mitochondrial-associated gene subsets were combined, as key redox genes are localized to mitochondria. This category was not over-represented compared to all coding genes. Nevertheless, genes within this high association group included eighteen confirmed or putative CYPs, eight of which are CYP4c orthologs. CYP4c is an insect-specific member of the CYP4 family; CYP4 family proteins are best known for oxidation of fatty acids and localize to peroxisomes and the endoplasmic reticulum. Recent work has also reported their involvement in permethrin resistance [16]. Though CYPs do not directly detoxify permethrin, they may function downstream of esterases, which are responsible for much of permethrin catabolism (reviewed in [50]). Moreover, CYP6 and CYP9 (S2 Table) are also well documented for their involvement in resistance to both deltamethrin and permethrin [48,49].

Though CYPs are the best studied of the redox functional group [10,11,13,15], mitochondrial function genes also have been implicated in permethrin resistant anopheline mosquitoes, as indicated by the presence of elevated reactive oxygen species [46]. In addition, key electron transport gene cytochrome B (CYTB) transcripts are elevated in Ae.aegypti following permethrin treatment [65], consistent with a requirement for increased mitochondrial function. Here, there were five genes coding for cytochrome B and C associated genes in the high association subset (S2 Table).

The transport functional category was the most statistically significant of all over-represented categories evaluated (120/1053, hypergeometric test p value = 1.3 x 10⁻⁶). This is intriguing, because one would not necessarily expect the category to be over-represented among all coding genes. Because the major pyrethroid target site, Vgsc, is a sodium transporter [5,66–71], it is possible that off-target receptor/channel functions are perturbed upon pyrethroid exposure and thus selected. Indeed, 55 of 120 genes in this category code for transporter, receptor or channel proteins. An alternate possibility is that selection of transport genes represents compensatory mutations to restore overall membrane homeostasis disrupted by pyrethroid binding to Vgsc in polarized neuronal cells (reviewed in [72]).

The largest specific functional category was for genes that affect DNA replication/ repair/transcription/translation (RRTT) functions. About 12.7% of all common genes were in this category, and though they were not over-represented relative to all coding genes (134/1053), there was a trend toward enrichment (hypergeometric test p value = 0.066). Of these genes, twenty are predicted to regulate transcription, including CHL, described above. Two orthologs of the transcription factor grauzone were also in this group. Grauzone is a Cys2His2 zinc-finger positive regulator of transcription and is required for meiosis in oogenesis [73]. Its possible role in insecticide resistance is unknown.

The effects of changes in transcriptional regulation of pyrethroid resistance are understudied in mosquitoes. Within the RRTT functional group, ribosomal protein subunits, histone proteins and histone methyltransferases were identified, hinting that insecticide resistance may require alterations from chromatin modifications to the initiation of gene expression to translational processes for establishment of the resistant phenotype.

Cytoskeletal components are crucial to proper neuronal function (reviewed in [74]). Genes coding for predicted cytoskeletal/structural function proteins were not over-represented among the subset common to all three collections. However, specific cytoskeletal/structural category genes with ankyrin domains are predicted to interact with Vgsc [27,28,33] and are genetically associated with PARA^bss in Drosophila spp. [27]. In addition, sickie (LOC5564933), a cytoskeletal positive regulator of neuronal axon growth, is also in the high association set [75], as was β-spectrin (β-SPEC) (S2 Table), a cytoskeletal protein important to maintaining neuronal structure in mammals and dipterans [74,76]. In mammals, β-SPEC is a neuronal structural component that physically interacts with N_AV channels [33] and also affects presynaptic stability [33].

Genes with predicted function in cellular signaling were significantly over-represented (31/1053, hypergeometric probability p value = 0.013). G-protein coupled receptors (GPCR) are suspected of activating expression of CYPs in permethrin resistant Culex quinquefasciatus [77]. Twelve of 24 genes in this subset were among the high association set; these included GPCRs, GTP-binding proteins and GTPases. Second messenger signaling components are also important to overall control of cytoskeletal changes required for neural synapse function (reviewed in [74]).

To identify the highest resistance-associated genes among all treatment groups (VCP, VCD and VeP), genes among the top 10% of–log₁₀(p value) values in the high association set were identified. Values within this extreme group were all greater than 9.02, 7.95 and 9.85 for VCP, VCD and VeP, respectively. Thirteen genes, including VGSC, were within this subset, and eight of these were located on the p arm of chromosome 3, in the same region as VGSC. Therefore, extreme association on the 3p arm may be due to a selective sweep of VGSC, resulting in incidental selection of genes that have no functional association. However, if this was the case, we would expect a higher percentage of the 737 high association genes located on the chromosome 3p arm to be included in the extreme group. Specifically, the extreme association group represents just 1.7% of the VGSC -proximal high association cluster. This is consistent with the hypothesis that key genes within this subset are specifically selected due to functional attributes rather than by proximity to VGSC alone. Importantly, these observations do not negate the possibility of a selective sweep in this region.

In addition to VGSC, the extreme association group included two transcriptional regulators (NERVY (NVY) LOC5563881 and LOC5579659), a Gr14 gustatory receptor (LOC5575007), a putative lipid binding protein, neural LAZARILLO (LOC5572156), CINGULIN (LOC5577979), a structural component of gap junctions at neuronal synapses [78] and three hypothetical genes. LAZARILLO, a lipid-binding apolipoprotein, though not studied in mosquitoes, is a biomarker for deltamethrin liver toxicity in mammals [79]. For all of these gene products, the mechanisms of association with pyrethroid resistance in mosquitoes remain to be explored.

Permethrin binding to the neuronal-specific transmembrane protein, Vgsc, could result in synaptic injury. The resulting toxic effects may select for mutations in genes that control pre- and post- synaptic processes. To explore this hypothesis, the high association group was interrogated for the presence of genes with known association to pre- and post-synaptic function. Twenty-one genes with demonstrated associations to this process were identified (S2 Table) [74,80,81]. Calcium signaling is requisite to motor neuron function [82,83], and the characteristic phenotype of pyrethroid treatment is the loss of motor control, due to insecticide binding to Vgsc. Therefore, the genetic association of both CAa1 and VGSC to pyrethroid resistance is consistent with complementary contributions of each protein toward motor synaptic function. In the drosophilid model, PARA expression is stimulated upon limiting synaptic activity levels [83]. Consistent with these observations, the translational repressor, PUM, helps regulate motoneuronal activity, a function that is conserved in mammals and dipterans [84,85]. It does this via control of VGSC (PARA) translation, thus regulating synaptic excitation [83]. Of note, PUM was also present in our high association set, consistent with a role in the selection of resistance (S2 Table). Synaptotagmins were another gene cluster associated with synaptic function; these genes code for synaptic vesicle calcium sensors (S2 Table) [86] that interact with N_AV in mammals [87].

Neuronal cytoskeletal genes are crucial to maintenance of synaptic control (reviewed in [74]). Several cytoskeletal and associated regulatory genes important to synaptic activity were in the high association set (S2 Table), including an atypical protein kinase C (aPKC) ortholog (LOC5576493). aPKC stabilizes the microtubule cytoskeleton for proper synaptic function (reviewed in [74]). β-SPEC is an important structural feature of synaptic connections [35] and interacts with N_AV channels in mammals [88]. Moreover, neuroligin-1 was also present in the high association group, also consistent with selection at excitatory synapses [89].

The analytical methods reported here have been used previously to identify polymorphic differences between aedine subspecies and sex-specific polymorphisms [19,93]. Though the published versions of the papers reported F_ST values rather than contingency χ² –log₁₀(p values), preliminary analyses of–log₁₀(p values) showed trends identical to those in the published reports. Importantly, the sex-specific polymorphisms we described were also replicated independently by others [94]. All raw reads were trimmed of adapters and filtered using cutadapt [95]. The AaegL5 genome build [21] of 18,081 coding and non-coding genes was used, including all 5’UTRs, exons, introns, 3’UTRs. The 5’ and 3’ non-transcribed regions in previously reported alignments were excluded [19]. Individual replicate trimmed fastq files were aligned to a custom reference file using GSNAP (version 2013-10-28), allowing 10% divergence [96]. GSNAP outputs were converted to *.mpileup files, using SAMtools [97]. The “readcounts” command in Varscan2 (v2.3.5) [98] was used to convert *.mpileup files to readcounts output, using the following options:—min-coverage 15—min-base-qual 30. FORTRAN programs used in this analysis are available upon request. The number of aligned reads was determined in .bam files, using SAMtools flagstat and multiplied by the read-length (100 nt) to achieve the total nucleotides aligned. The ratio of variant sites per chromosome was multiplied by 1000 and divided by the total aligned nucleotides (S1 Table) to obtain the “Ratio sites/aligned nucleotide * 1000.”

Because the probes were designed using genome build L1.3, and the current reference genome is L5, we checked one of the libraries (VCDA-rep1) to determine the percentage and depth of sequencing for capture target sequences that aligned to L5. Coverage depth files of .bam files were generated using SAMtools, then the analogous L5 capture coordinate intervals were located. About 15.3% (chr 1), 6.5% (chr 2), and 11.7% (chr 3) of L5 capture targets were not either not sequenced or had coverage less than the 15 read coverage cut-off. In addition, a blastn search of L5 transcripts using the capture probes indicated that they cover (13,942/18,081) 77% of the genes in L5.

All sequencing data generated under this project are available at the National Center for Biotechnology Information (NCBI) Sequence Read Archive, Bioproject accession number PRJNA393171. VCPA permethrin-treated resistant replicates 1 and 2 are SRR5805471 and SRR5805472, respectively. VCPD permethrin-treated susceptible replicates 1 and 2 are SRR5805473 and SRR5805470, respectively. VCDA deltamethrin-treated resistant replicates 1 and 2 are SRR5805467 and SRR5805466, respectively. VCDD deltamethrin-treated susceptible replicates 1 and 2 are SRR5805469 and SRR5805468, respectively. VePA permethrin-treated resistant replicates 1 and 2 are SRR5805465 and SRR5805464, respectively. VePD permethrin-treated susceptible replicates 1 and 2 are SRR5805475 and SRR5805474, respectively.

In-house FORTRAN scripts were used to calculate heterogeneity χ² values with degrees of freedom equal to the number of alternate nucleotides minus 1. The contingency χ² calculation tested for the presence of an alternate nucleotide at each position along the genome and computed a -log₁₀(p value) at each SNP [19]. For the purpose of highlighting genes with high statistical association to pyrethroid resistance, calculations of weighted gene-wise average scores were made by determining the mean of the top 5% -log₁₀(p value) values per gene. Our rationale for this calculation was as follows. Along the length of a given coding sequence, many nucleotide sites showed little change relative to the reference sequence, thus resulting in many -log₁₀(p value)at or near zero for a given SNP. Coding changes indicative of selection toward pyrethroid resistance may be present in a small subset of sites relative to the full length gene. Therefore, a weighted average allowed us to identify those genes with localized high value SNPs. Statistical significance was determined using the Benjamini-Hochberg False Discovery Rate [22] (FDR, alpha = 0.01) and a cut-off of 4.0 was established for gene-wise averages.

Gene annotations were obtained from Vectorbase [23] using BioMart. Functional groups were assigned to the following categories: ‘apoptosis’(APOP); ‘chemosensory response‘ (CSR); cytoskeletal/structural (CYT/STR); ‘diverse’ (DIV) for genes with multiple or less clearly defined function(s); ‘mitochondrial’ (MIT) for gene products localized to mitochondrial compartment, regardless of function; (LIPID) for genes with predicted function in lipid processing or biosynthesis; (ReDox) for genes with predicted function in oxidation/reduction processes; proteolysis or proteosomal activity (PROT); replication/(DNA)repair/transcription /translation’ (RRTT); signal transduction (SigT); metabolism (MET); ‘unknown’ (UNK) for uncharacterized genes. Finally, the ‘transport’ (TRP) category included all gene products predicted to be involved in moving molecules across membranes, including receptors, exclusive of secondary messengers and signaling receptors. For GSEA, the mitochondrial and redox categories were combined, as several key redox activities occur in proteins localized to mitochondria.

Descriptive statistics were calculated in R (version 3.0.2). The ratio of variant sites per nucleotide of aligned reads was calculated as follows: binomial probability distributions were performed and hypergeometric tests were calculated; each functional subset was compared to the corresponding group within the set of all coding genes (n = 14,626). This analysis allowed us to identify the gene sets that were enriched compared to similar categories among all coding genes.