Adult mosquitoes were collected from Asendabo, Oromia region, Ethiopia (7°40^′31^″ N, 36°52^′56^″ E), where organophosphate and pyrethroid resistance had been previously reported in An. arabiensis populations (Messenger et al., 2017). Mosquitoes were sampled at the end of the long rainy season, between 3rd September-10th October 2017, following IRS with bendiocarb by the National Malaria Control Program (NMCP) in this area in June 2017.

Upon obtaining householder consent, indoor-resting, blood-fed female Anopheles mosquitoes were collected from the walls of 12 houses (situated approximately <5 km apart) between 4:00 and 6:00 a.m. using handheld aspirators. Mosquitoes were held in paper cups with access to 10% sucrose and transported to the Tropical and Infectious Diseases Research Center (TIDRC) in Sekoru, Oromia region (7°54^′50^″ N, 37°25^′23.6^″ E). F₁ progeny were obtained from field-collected mosquitoes using forced-oviposition (Morgan et al., 2010). Blood-fed, field--collected mosquitoes, morphologically identified as An. gambiae s.l. (Gillies and Coetzee, 1987), were maintained for 4–5 days until fully gravid and checked daily for survival. Each fully gravid female was transferred to a 1.5 ml microcentrifuge tube containing damp cotton wool and allowed to lay eggs. Eggs from 246 adult An. gambiae s.l. were transported to the U.S. Centers for Disease Control and Prevention (CDC), Atlanta, USA, and pooled for rearing in the CDC insectary.

An. arabiensis from the insecticide susceptible Dongola reference strain (originating from Sudan, obtained from the Malaria Research and Reference Reagent Resource Center, MR4) and the Sekoru insecticide susceptible laboratory strain (originating from Ethiopia, obtained from the Vector Biology and Control Research Unit, TIDRC, Jimma University) (Balkew et al., 2010), were also reared in the CDC insectaries. All adult mosquitoes were maintained under standard insectary conditions (27±2 °C, 80% relative humidity, light:dark cycles of 14:10 h) with access to 10% sucrose solution ad libitum. F₁ adult females of each strain were randomly mixed in cages for subsequent bioassays.

CDC bottle bioassays for malathion (organophosphate) and permethrin (pyrethroid) were conducted according to published guidelines (Centers for Disease Control and Prevention, 2012). Stock solutions of the diagnostic dose required to kill 100% of susceptible mosquitoes (malathion: 50μg/bottle and permethrin: 21.5μg/bottle), were prepared by diluting technical grade insecticide in 50 ml of acetone. Each Wheaton 250 ml glass bottle along with its cap was coated with 1 ml of the stock solution by rolling and inverting the bottles. In each test, a control bottle was coated with 1 ml of acetone. Bottles were left to dry in the dark for 3 h and were washed thoroughly and re-coated before every test. Following a 2-h acclimatization period in paper cups with access to 10% sucrose, approximately, 20–25 unfed, 3 day-old adult female An. gambiae s.l. were introduced into each bottle using a mouth aspirator and knock-down/mortality was recorded after 30 min of exposure. Additionally, a susceptible reference An. arabiensis strain (Dongola or Sekoru) was assayed in parallel. Bioassays were conducted between 15:00 and 17:00 each day to avoid any bias in RNA transcript expression related to circadian rhythm. Multiple replicates were performed per insecticide to obtain sufficient phenotyped material for RNA-sequencing analysis. A mosquito was defined as ‘alive’ at the diagnostic time if it was capable of standing and flying in a coordinated manner; surviving mosquitoes (defined as resistant) and non-exposed mosquitoes (from acetone-treated bottles) were stored separately at −80 °C. Additionally, non-exposed, unfed, 3 day-old adult female An. arabiensis from the Sekoru and Dongola susceptible laboratory strains were also preserved for analysis at −80 °C.

Additional resistance intensity bioassays were undertaken with F₁ field mosquitoes to characterize susceptibility levels to carbamates (bendiocarb and propoxur) and pyrethroids (alpha-cypermethrin, deltamethrin and permethrin), following exposure to 1, 2, 5 and 10 times the diagnostic doses. Bioassay data were interpreted according to the WHO criteria: mortality of 98% or higher indicates susceptibility, mortality of 90–97% is suggestive of resistance, and mortality of less than 90% indicates resistance (World Health Organization, 2013). Mortality in untreated control bottles was less than 5% in all resistance intensity bioassays. Mean percent mosquito mortality was calculated across all replicates for a given insecticide.

Prior to pooling specimens for RNA extraction, 4–6 legs from each mosquito tested in bioassays were removed and genomic DNA was extracted using the Extracta^™ DNA Prep for PCR-Tissue kit (QuantaBio, USA), according to the manufacturer’s protocol. Molecular identification of An. gambiae s.l was carried out using species-specific PCR with primers for An. gambiae s.s., An. arabiensis and An. quadriannulatus (Wilkins et al., 2006): AR-3T (5^′-GTGTTAAGTGTCCTTCTCCGTC-3’; specific for An. arabiensis), GA-3T (5^′-GCTTACTGGTTTGGTCGGCATGT-3; specific for An. gambiae s.s.), QD-3T (5^′-GCATGTCCACCAACGTAAATCC-3’; specific for An. quadriannulatus) and IMP-UN (5^′-GCTGCGAGTTGTAGAGATGCG-3’; common for all species). Each 25 μl reaction volume contained 20–40 ng of DNA, 5X Green GoTaq^® Reaction Buffer (Promega), 25 mM MgCl₂, 2 mM of each dNTP, 1U GoTaq^® DNA polymerase and 25 pmol/μl of primers AR-3T, GA-3T, QD-3T and IMP-UN. PCR cycling conditions were: 95 °C for 5 min, followed by 30 amplification cycles (95 °C for 30 s, 58 °C for 30 s, 72 °C for 30 s) and a final elongation step at 72 °C for 5 min. Amplified PCR products were visualized on 1.5% agarose gels, stained with GelRed^™ (Biotium, USA). Positive control DNA from An. arabiensis Sekoru, An. gambiae s.s. Kisumu and An. quadriannulatus Sangwe strains and no-template negative controls were included with all reaction runs. PCR products of 387bp, 463bp or 636bp were indicative of An. arabiensis, An. gambiae s.s. or An. quadriannulatus, respectively.

The presence of the G119S Ace-1 mutation was determined using PCR restriction fragment length polymorphism analysis (Weill et al., 2004). Amplifications were performed in 25 μl reactions containing 20–40 ng of DNA, 5X Green GoTaq^® Reaction Buffer (Promega), 2.5 mM of each dNTP, 1U GoTaq^® DNA polymerase, 25 pmol/μl of primers MOUSTDIR1 (5^′-CCGGGNGCSACYATGTGGAA-3^′) and MOUSTREV1 (5^′-ACGATMACGTTCTCYTCCGA-3^′). PCR cycling conditions were 95 °C for 5 min, followed by 35 amplification cycles (95 °C for 30 s, 52 °C for 30 s, 72 °C for 1 min) and a final elongation step at 72 °C for 5 min. PCR products were initially visualized on 2% agarose gels, stained with GelRed^™ (Biotium, USA) before incubation with AluI restriction enzyme (New England Biolabs, USA) at 37 °C for 16 h, followed by 65 °C for 20 min. DNA fragments were visualized on 2% agarose gels, stained with GelRed^™ (Biotium, USA). DNA from An. arabiensis Sekoru was used as a negative control alongside a no-template control. DNA from An. coluzzii AKDR was used as a positive control. Undigested PCR products of 194bp indicated the susceptible allele (wild type) and 120bp and 74bp digested fragments indicated the presence of the resistant allele. The presence of all three bands indicated the sample was a heterozygote.

West African kdr (L1014S) and East African kdr (L1014F) alleles were detected using protocols for allele-specific PCR (AS-PCR) (Martinez-Torres et al., 1998; Ranson et al., 2000). Primers IPCF (5^′-GATAAT GTGGATAGATTCCCCGACCATG-3^′), AltRev (5^′-TGCCGTTGGTGCAGACAAGGATG −3^′), WT-R (5^′-GGTCCATGTTAATTTGCATTACTTACGAATA −3^′) and East-F (5^′-CTTGGCCACTGTAGTGATAGGAAAATC-3^′) were used to detect the L1014S allele (AS-PCR East), whereas primers IPCF, AltRev, WT-R and West-F (5^′-CTTGGCCACTGTAGTGATA GGAAATGTT-3^′) were used to detect the L1014F allele (AS-PCR West). Each 25 μl reaction volume contained 20–40 ng of DNA, 5X Green GoTaq^® Reaction Buffer (Promega), 25 mM MgCl₂, 2 mM of each dNTP, 1U GoTaq^® DNA polymerase, 2.5 pmol/μl of primers IPCF and AltRev and either 5 pmol/μl of primer WT-R and 2.5 pmol/μl of primer East-F to detect the L1014S allele (AS-PCR East), or 25 pmol/μl of primer WT-R and 8.8 pmol/μl of primer West-F to detect the L1014F allele (AS-PCR West). PCR cycling conditions were 95 °C for 5 min, followed by 35 amplification cycles (95 °C for 30 s, 57 °C for East or 59 °C for West for 30 s, 72 °C for 30 s) and a final elongation step at 72 °C for 5 min. Amplified PCR products were visualized on 2% agarose gels, stained with GelRed^™ (Biotium, USA). DNA from An. gambiae Kisumu was used as a negative control alongside a no-template control. DNA from An. coluzzii AKDR and An. gambiae s.s. RSP-ST were used as positive controls for L1014F and L1014S, respectively. Successful amplification was indicated by a PCR product of 314 bp; additional bands of 214bp and 156bp identified susceptible (wild type) and resistant alleles, respectively. Pearson’s Chi squared tests were used to evaluate deviations from Hardy-Weinberg equilibrium at the population-level.

Total RNA was isolated from three pools containing five mosquitoes each from the following groups: mosquitoes phenotyped as resistant following a malathion or permethrin bioassay, non-insecticide exposed mosquitoes and susceptible An. arabiensis colony mosquitoes from Dongola and Sekoru strains. RNA was extracted using the Arcturus^® PicoPure^® RNA isolation kit (Life Technologies, USA) and quantified using the Agilent RNA ScreenTape 4200 assay, according to the manufacturers’ protocols. Two micrograms of starting material were treated with Baseline-ZERO^™ DNase (Lucigen, USA) and ribosomal RNA was removed using the Ribo-Zero^™ Magnetic Core Kit and Ribo-Zero^™ rRNA Removal kit (Illumina, USA), according to the manufacturers’ protocols. Individual RNA-Seq libraries were prepared from each pool of extracted RNA using the ScriptSeq^™ v2 RNA-Seq library preparation kit (Illumina, USA), using 12 cycles of PCR amplification, according to the manufacturer’s protocol. Libraries were purified using Agencourt AMPure XP beads (Beckman Coulter, USA) and assessed for quantity and size distribution using the Agilent DNA ScreenTape D5000 assay.

Two experiments, each comprising nine RNA-Seq libraries, were sequenced as 2 × 125bp paired-end reads, on the Illumina HiSeq platform at the CDC. The first experiment (henceforth “malathion experiment”) contained three biological replicates each of malathion bioassay survivors, non-exposed mosquitoes and the susceptible Dongola strain. The second experiment (henceforth “permethrin experiment”) contained three biological replicates each of permethrin bioassay survivors, non-exposed mosquitoes and the susceptible Sekoru strain. Each experiment was sequenced on two HiSeq lanes to give an estimate of technical variation.

De-multiplexed paired end sequencing reads for each sample were evaluated for quality using FastQC v0.11.5 (Andrews, 2016). Concatenated files for R1 and R2 reads were used for downstream analysis. Initially concatenated files for each sample were trimmed and filtered using fastp v0.21.0 (Chen et al., 2018) to remove adapter and low-quality reads according to the following criteria: minimum base quality score = 20, minimum length required = 25, polyG and poly tail trimming = True. Trimmed and filtered read pairs (R1/R2) were aligned against the reference genome, An. arabiensis Dongola (genome assembly version = AaraD1.11, GeneBank assembly identifier = GCA_000349185.1; GeneBank WGS Project = APCN01), directly downloaded from VectorBase (release 48) (Giraldo-Calderón et al., 2015), using ‘subjunc’ v2.0.1, part of the subread aligner v2.0.1 (Liao et al., 2013), with default parameters. The resulting alignment was filtered to remove reads with low mapping quality (q < 10) and sorted successively using Samtools v1.10 (Li et al., 2009). Descriptive statistics for the malathion and permethrin read libraries and sequencing alignments are shown in Table S1.

Tags (a read pair or single, unpaired read) mapped to the sense orientation of the annotated An. arabiensis Dongola genes (gene set of AaraD1.11 in gff downloaded from release 48 from Vector Base), were quantified using FeatureCounts, as part of the subread-aligner package v2.0.1 (Liao et al., 2013). The tag count with FeatureCount was carried out using the following criteria: 1) count only read pairs that have both ends aligned; 2) count fragment instead of reads; 3) minimum number of overlaps required = 1; 4) feature_type = exon; 5) attribute type = gene_id; and 6) strandness = sense. The FeatureCount analysis generated a tag count matrix table which was inputted to edgeR (Robinson et al., 2010) for differential expression analysis. Metrics describing the transcriptome alignments for the malathion and permethrin experiments are shown in Table S2.

To remove the effect of noise and lowly expressed genes, for each pairwise comparison, genes with a total tag count less than 50 across all libraries (control vs treatment) were filtered out before further analysis. Only genes with a total tag count equal to or higher than 50 were considered. The function calcNormFactors (part of the edgeR package (Robinson et al., 2010)), using the TMM (Trimmed Mean M-values) method, was used to normalize tag count among samples, by finding a set of scaling factors for the library sizes that minimized the log-fold changes between samples for most genes. The tag count was not normalized for gene length and GC content, as these values do not vary from sample to sample, so this would be expected to have little effect on DEGs. The DEGs between control (unexposed) and resistant (exposed) mosquitoes were selected after multiple testing using the decideTests function, part of the limma package (Ritchie et al., 2015). A critical value absolute fold-change = 2 and FDR (False Discovery Rate) ≤ 0.01 was used. Different pairwise comparisons were conducted: 1) between resistant field mosquitoes (treatment) and unexposed field mosquitoes (control): CON-M vs MAL-R and CON–P vs PERM-R; 2) between a susceptible laboratory strain and exposed field mosquitoes: DON vs MAL-R, SEK vs PERM-R and DON vs PERM-R; 3) between the two susceptible laboratory strain: DON vs SEK; and 4) between field mosquitoes exposed to different insecticides: MAL-R vs PERM-R.

The annotation set of the AraD1.11 reference genome included 13,307 protein-coding genes and 378 additional non-coding genes (Table S3) (https://legacy.vectorbase.org/organisms/dongola/aarad111). However, Gene Ontology (GO) description of only 9074 of these genes was provided in VectorBase (Giraldo-Calderón et al., 2015) (cellular component: 4784; molecular function: 7261; biological processes: 5316). To increase the annotation efficiency, the predicted protein gene set fasta file of AraD1.11 was downloaded from VectorBase (release 48) (Giraldo-Calderón et al., 2015) and was used for functional annotation using Blast2GO (Conesa and Götz, 2008). A Blastp search of the protein fasta file was conducted against the Insecta category of the non-redundant protein NCBI database, with a maximum e-value cut-off of 1e–3. Additionally, the RefSeq protein IDs corresponding to the best blast hits of each query sequence were mapped to the GO database as curated and updated in the last release of Blast2GO database (November 2020). The resulting non-annotated genes from the Blast2GO analysis were mapped to the An. gambiae proteome (AgamP4.13) using a Blastp search with a maximum e-value cut-off of 1e 10 for ortholog inference. The best alignments (based on e-value and alignment score) were considered as orthologous genes, were ID mapped to the GO annotation of AgamP4.13 using the panda’s python library (McKinney, 2011). The newly annotated genes were concatenated with the Blast2GO annotation, which was used as the background for the functional enrichment analysis of the DEGs. From this analysis, 10,456 (78.6%) of 13,307 protein coding genes were GO annotated.

GO term enrichment analysis of up- and down-regulated genes was carried out using Goatools (Klopfenstein et al., 2018) based on the go-basic database (release 2021-02-01). The list of 10,456 annotated genes of An. arabiensis with their associated GO terms was used as the background reference set. The P values used to evaluate significantly enriched GO terms were calculated based on Fisher’s exact test and corrected by Benjamini-Hochberg multiple test correction method. Finally, we used a FDR adjusted P-value <0.05 to tag statistically significant overrepresented GO terms associated with the list of DEGs.

A subset of eleven differentially transcribed genes was selected for quantitative real-time reverse transcription PCR validation (qRT-PCR). One microgram of RNA from three replicates of malathion resistant or permethrin resistant, non-exposed and Dongola strain mosquitoes were used to synthesize cDNA using the High-Capacity cDNA Reverse Transcription kit (Applied Biosystems, USA) with oligo-dT20 (New England Biolabs, USA), according to the manufacturer’s instructions. Primer sequences and efficiencies are detailed in Table S4. Standard curves of Ct values for each gene were generated using a five-fold serial dilution of cDNA to assess PCR efficiency. Reactions were performed using either a QuantStudio 6 Flex Real-Time PCR system (Applied Biosystems, USA) with PowerUp SYBR Green Master Mix (Applied Biosystems, USA) or a Stratagene Mx3005P Real-Time PCR system (Agilent Technologies) with LightCycler^® 480 SYBR Green I Master Mix (Roche, UK). cDNA from each sample was used as a template in a three-step reaction: 50 °C for 2 min, denaturation at 95 °C for 10 min, followed by 40 cycles of 15 s at 95 °C, 1 min at 60 °C and a final step of 15 s at 95 °C, 1 min at 60 °C, and 15 s at 95 °C. The relative expression level and Fold Change (FC) of each target gene from resistant field samples, relative to the susceptible laboratory strain (Dongola), were calculated using the 2^−ΔΔCT method (Rao et al., 2013), incorporating PCR efficiency. Two housekeeping genes, ribosomal protein S7 (RpS7: AARA000046) and ubiquitin (AARA016296), were used for normalisation.

The RNA-Seq reads of all resistant groups and susceptible strains were mined for the prevalence of non-synonymous Single Nucleotide Polymorphisms (SNPs) involved in Anopheles spp resistance to either DDT, organophosphate or pyrethroid insecticides. The primary target of the analysis was the para Voltage-Gated Sodium Channel (VGSC) gene (AARA017729), for which the presence of 21 recently reported non-synonymous SNPs (A1125V, A1746S, A1934V, D466H, E1597G, F1920S, I1527T, I1868T, I1940T, K1603T, L995F, L995S, M490I, N1575Y, P1874L, P1874S, T791M, V1254I, V1853I, V402L, and V1853I) were investigated (Clarkson et al., 2021). Additionally, non-synonymous variants G119S in the acetylcholinesterase (Ace-1) gene (AARA001814), L119F and I114T in GSTe2 (AARA008732) (Mitchell et al., 2014; Lucas et al., 2019), were also investigated. Prevalence of the target site mutations in the RNA-Seq datasets was determined as follows. The coding sequences (CDS) corresponding to VGSC, Ace-1, and GSTe2 from AaraD1.11 were downloaded from VectorBase (Giraldo-Calderón et al., 2015) and were aligned separately with their respective homologous gene retrieved from the AgamP4.4 gene set, using Clustalw Omega (Sievers et al., 2011). Next, the sequence (~30–40 nucleotides) flanking the codon and the site of interest from each gene in An. arabiensis was identified and extracted from the alignment as described here (Lol et al., 2019). The resulting flanking sequence was BLASTn (Chen et al., 2015) searched against the AaraD1.11 reference genome (release 48 in Vectorbase) (Giraldo-Calderón et al., 2015), which gave the exact chromosomal numerical position of the nucleotide. Finally, the sorted bam files, which were previously used as the input featureCount for DEG analysis were separately uploaded to Integrative Genomics Viewer (IGV) (Thorvaldsdótti et al., 2013) and zoomed to the position to the flanking sequence. The allele frequency in the population was calculated as the percentage of RNA-Seq reads spanning the codon with the SNP of interest.

Indoor resting F₀ adult An. gambiae s.l. were collected from houses in Asendabo, Oromia region, Ethiopia from July-September 2017 and F₁ progeny were generated by forced-oviposition (Morgan et al., 2010). Susceptibility to the diagnostic doses (1X) of malathion (organophosphate) and permethrin (pyrethroid) was determined for 273 F₁ An. gambiae s.l. mosquitoes, using U.S. Centers for Disease Control and Prevention (CDC) bottle bioassays (Centers for Disease Control and Prevention, 2012). These mosquitoes were subsequently confirmed via species-specific PCR as An. arabiensis (Wilkins et al., 2006). The average mortality to malathion was 71.9% [95% CI: 65.3–78.5] and to permethrin was 77.4% [95% CI: 44.0–100.0%]. Resistance intensity assays, using an additional 1183 PCR-confirmed F₁ An. arabiensis, were conducted with alpha-cypermethrin (1X), bendiocarb (1X), propoxur (1X), deltamethrin (1X, 2X, 5X and 10X) and permethrin (1X, 2X, 5X and 10X) (20). Complete (100%) mortality was observed to the diagnostic doses of alpha-cypermethrin, bendiocarb and propoxur, while moderate to intense resistance was detected to deltamethrin and permethrin, with small proportions of mosquitos capable of surviving five to ten times the diagnostic concentrations (Fig. 1).

Phenotyped individuals were screened for known insecticide resistance target site mutations. The G119S-Ace-1 mutation was not detected in any mosquitoes from the malathion bioassays (n = 173). The L1014F-kdr mutation was identified in 52% (30/58) of An. arabiensis exposed to the diagnostic dose of permethrin, with allele frequencies of 0.65 in surviving mosquitoes and 0.26 in dead mosquitoes. A greater proportion of An. arabiensis surviving permethrin bioassays were homozygous for L1014F-kdr (46%; 6/13) compared to those that died (9%; 4/45), and 38.5% of survivors (5/13) and 33% of dead individuals (15/45) were heterozygous. The L1014S-kdr allele was not detected in any sample tested.

Malathion or permethrin bioassay survivors, field mosquitoes which were not exposed to insecticide, and two An. arabiensis susceptible reference strains (originally from Sudan or Ethiopia – Dongola or Sekoru, respectively) were submitted for transcriptomic analysis. For the malathion experiment, Illumina RNA-sequencing generated more than 620 million raw reads across three biological replicates, sequenced in technical duplicate with an average of 68.9 (±5.1) million reads per group. (Table S1). After filtering and quality trimming, an average of 67.6 (±5.0) million reads were retained per group (98.15%) for subsequent analysis. An average of 51 (±7.8) million quality filtered reads per group (75.40%) were mapped to the whole An. arabiensis Dongola AaraD1.11 reference genome, with around 59% of the counted fragments mapped to all exonic features of the gene set (Table S1). The permethrin experiment generated more than 569 million reads across three biological replicates, sequenced in technical duplicate with an average of 63.3 (±10.9) million reads per group (Table S1). Quality control filtering retained an average of 61.4 (±10.7) million reads per population (97.02%), with an average of 42.6 (±14.3) million total filtered reads aligned to the reference genome (69.48%) and around 64% of the counted fragments successfully assigned to exons of the gene set (Table S1). Full results for the analyses of the malathion and permethrin experiments are presented in Table S5, and results of gene ontology (GO) enrichment analysis for sets of differentially expressed genes (DEGs) are shown in Table S6.

Differential expression analysis was performed on transcripts retained after quality control and removal of genes with low read counts. Aligned reads were mapped to the An. arabiensis genes dataset (AaraD1.11) to quantify levels of gene expression, with between 52 and 69% of alignments successfully assigned to the exonic regions of the reference genome (Table S2). Three pairwise comparisons were conducted for malathion: resistant vs susceptible (R–S; MAL-R vs DON), resistant vs unexposed control (R–C; MAL-R vs CON-M) and unexposed control vs susceptible (C–S; CON-M vs DON). The R–C comparison allowed us to account for induction of transcription during insecticide exposure; genes were filtered by analysing their expression profiles in the susceptible Dongola strain, with the assumption that constitutive resistance genes will be significantly differentially expressed between both bioassay survivors and the non-exposed field mosquitoes, when compared to the susceptible strain.

At the most conservative level (P-values adjusted for multiple testing based on a false discovery rate (FDR) < 0.01 and fold change (FC) > 2), a total of 1212 (12.2%; 872 upregulated and 340 downregulated) genes were significantly differentially expressed in mosquitoes that survived malathion exposure and 598 (6.0%; 398 upregulated and 200 down-regulated) were significantly differentially expressed in non-insecticide exposed field mosquitoes as compared to the susceptible strain (Fig. 2A; Table 1). A total of 170 (1.8%; 137 upregulated and 33 downregulated) genes were significantly differentially expressed in mosquitoes that survived malathion exposure compared to their non-insecticide exposed counterparts (Fig. 2A; Table 1).

Of the genes that were differentially expressed in all treatment groups (n = 9), 2 were upregulated while 7 were downregulated in one or more conditions (Fig. 2A). Five of these genes had retrievable annotations, all of which were molecular functions or cellular components (for R–C/R–S/C–S comparisons: AARA017080 = peptide methionine sulfoxide reductase, FCs = 2.57, 0.43 and 0.18; AARA016556 = sulfotransferase, FCs = 2.23, 23.88 and 9.92; AARA007045 = protease M1 zinc metalloprotease, FCs = 0.40, 0.18 and 0.44; AARA002630 = transient receptor potential protein, FCs = 0.21, 0.49 and 2.37; and AARA002503 = ion binding protein, FCs = 0.37, 0.04 and 0.17, respectively).

A total of 402 genes were differentially expressed commonly in the R–S and C–S groups (Fig. 2A). Among the top 10 over-expressed genes with retrievable annotations were enzymes with structural, cellular or immune functions, including chitinase (AARA007329: FCs = 50.04 and 10.80 for R–S/C–S comparisons, respectively), D7r4 short form salivary protein (AARA016237: FCs = 33.29 and 31.34), cytoplasmic actin (AARA015772: FC = 29.53 and 7.33), cuticular protein CPLCG (AARA011115: FCs = 26.80 and 20.12), alkaline phosphatase (AARA002132: FCs = 26.33 and 11.83), sulfotransferase (AARA016556: FCs = 23.88 and 9.92), serine protease (AARA009441: FCs = 23.73 and 24.43), polyubiquitin (AARA016579: FCs = 21.67 and 31.07), ADP/ATP carrier protein (AARA017958: FCs = 21.15 and 5.23) and deoxyribonuclease (AARA000505: FCs = 17.0 and 12.15). A total of 19 genes were differentially expressed commonly in the R–C and C–S groups (Fig. 2A). Among the top over-expressed genes with retrievable annotations were notably two odorant binding proteins (for R–C/C–S comparisons, respectively: AARA007908: FCs = 5.17 and 0.17; AARA004722: FCs = 3.24 and 0.19).

Significant differential expression of some members of the detoxification gene families associated with metabolic resistance were observed among R–S and C–S comparisons (Table 2; Fig. 3A). These included nine CYP450s (CYP9K1, CYP9J5, CYP6AA1, CYP4C36, CYP6AA1, CYP9L1, CYP6M2, CYP6M3 and CYP6P4), six GSTs (GSTE2, GSTE3, GSTE4, GSTE5, GSTE7 and GSTD3) and two COEs (AARA016305 and AARA016468). With the exception of GSTD3 and GSTE3, the FCs of all of these detoxification enzymes increased in response to malathion exposure (Table 2). Two additional CYP450s were also upregulated between R–C conditions (CYP4G16, FC = 3.40; and CYP4G17, FC = 2.03) (Supplementary Fig. S1; Table 2).

Significant differential expression of eighteen mosquito salivary gland proteins were identified among R–S and C–S comparisons (Table 2; Fig. 3A), most notably D7r4 short form salivary protein (FCs = 33.29 and 31.34 for R–S/C–S, respectively), TRIO salivary gland protein (FCs = 4.26 and 7.16), AARA009957 (FCs = 10.48 and 7.57) and salivary gland protein 7 (FCs = 5.87 and 6.12). Among these salivary gland proteins, twelve were downregulated following malathion exposure (Table 2); one salivary gland protein was significantly overexpressed between R–C conditions (AARA008387, FC = 2.04). Furthermore, fifteen proteins associated with cuticular function were significantly overexpressed in the R–S condition, including chitinase (AARA007329) (FCs = 50.04 and 10.80 for R–S/C–S, respectively), cuticular protein CPLCG family (AARA011115) (FCs = 26.80 and 20.12), cuticular protein RR-2 family (AARA001131) (FCs = 14.39 and 22.16) and cuticular protein RR-1 family (AARA003903) (FCs = 10.06 and 6.09). The majority of these were upregulated after insecticide treatment (Table 2), with an additional cuticular protein RR-2 family member, significantly overexpressed between R–C conditions (AARA017766, FC = 2.45) (Supplementary Fig. S1).

In malathion resistant mosquitoes, several ontologies were enriched in genes overexpressed relative to susceptible mosquitoes (Table S5). In particular, many of these ontologies were associated with metabolic processes, including “cellular metabolic process” (GO:0044237), “catalytic activity” (GO:0003824) and “generation of precursor metabolites and energy” (GO:0006091). Between R–C conditions, additional metabolic ontologies were upregulated, including “generation of precursor metabolites and energy” (GO:0006091) and “cellular metabolic process” (GO:0044237), potentially associated with increased physiological stress in response to insecticide exposure (Adedeji et al., 2020).

Differential transcription analysis for the permethrin experiment was performed relative to both DON and Sekoru (SEK) susceptible laboratory strains; the latter analysis was performed with the assumption that this more geographically proximate colony from Ethiopia would be a better biologically comparator than DON. However, greater variation in gene expression was observed, with 2183 (23.5%; 1057 upregulated and 1126 downregulated) and 2312 (23.7%; 1153 upregulated and 1159 downregulated) genes significantly differentially expressed between SEK and mosquitoes that survived permethrin exposure and non-exposed field mosquitoes, respectively (Supplementary Fig. S2; Table 1). A multi-dimensional scaling plot revealed significant variation between SEK and all other mosquito populations (Supplementary Fig. S3); downstream analyses for the permethrin experiment were therefore performed relative to DON.

Consistent with the malathion experiment, three pairwise comparisons were conducted for permethrin: resistant vs susceptible (R–S; PERM-R vs DON), resistant vs unexposed control (R–C; PERM-R vs CON–P) and unexposed control vs susceptible (C–S; CON–P vs DON). Among mosquitoes that survived permethrin exposure and non-exposed field mosquitoes, 1074 (10.9%; 673 upregulated and 401 down-regulated) and 889 (8.9%; 594 upregulated and 295 downregulated) genes were significantly differentially expressed (at P-values adjusted for multiple testing based on a FDR<0.01 and FC > 2), respectively, when compared to the susceptible Dongola strain (Fig. 2B; Table 1). A total of 334 (3.5%; 179 upregulated and 155 downregulated) genes were significantly differentially expressed in permethrin survivors as compared to their non-exposed counterparts (Fig. 2B; Table 1).

Of the genes that were differentially expressed in all treatment groups (n = 35), 3 were upregulated while 32 were downregulated in one or more conditions (Fig. 2B). Eleven had retrievable annotations, which were mostly molecular functions or biological processes (for R–C/R–S/C–S comparisons: AARA015710 = CLIP-domain serine protease, FCs = 2.21, 4.35 and 1.97; AARA015772 = cytoplasmic actin, FCs = 4.24, 51.86 and 12.20; AARA016057 = ATP binding cassette transporter, FCs = 0.41, 2.39 and 5.82; AARA016221 = salivary gland protein 1-like, FCs = 9.47, 2.25 and 4.73; AARA002374 = MIP18 family protein CG7949, FCs = 2.38, 4.01 and 1.67; AARA003468 = peptide methionine sulfoxide reductase, FCs = 3.63, 0.42 and 0.11; AARA003599 = TRPL translocation defect protein 14 isoform, FCs = 2.20, 3.47 and 1.57; AARA009096 = diacylglycerol kinase 1 isoform, FCs = 0.41, 0.22 and 0.53; AARA016129 = sorbitol dehydrogenase, FCs = 0.04, 0.35 and 7.99; AARA017544 = serine protease 7-like, FCs = 2.58, 4.70 and 1.82; and AARA018460 = lysosomal alpha-mannosidase, FCs = 0.42, 4.24 and 10.04, respectively).

A total of 500 genes were differentially expressed commonly in the R–S and C–S groups (Fig. 2B). The top 10 over-expressed genes with retrievable annotations were similar to the malathion experiment, including chitinase (AARA007329: FCs = 93.30 and 16.76 for R–S/C–S comparisons, respectively), D7r4 short form salivary protein (AARA016237: FCs = 20.84 and 43.89), cytoplasmic actin (AARA015772: FCs = 51.86 and 12.20), alkaline phosphatase (AARA002132: FCs = 29.70 and 13.74), sulfotransferase (AARA016556: FCs = 33.61 and 16.61), polyubiquitin (AARA016579: FCs = 21.57 and 67.65) and ADP/ATP carrier protein (AARA017958: FCs = 25.15 and 10.50). Cuticular protein RR-1 (AARA003903: FCs = 19.34 and 15.68) and hexamerin (AARA016988: FCs = 15.78 and 7.50) were also highly upregulated.

Consistent with the malathion experiment, key metabolic enzymes were significantly differentially expressed between R–S and C–S comparisons (Table 2; Fig. 3B), including eight CYP450s (CYP6M2, CYP4C36, CYP6AA1, CYP9K1, CYP6M3, CYP6P4, CYP325C2 and CYP9L1), six GSTs (GSTE2, GSTE3, GSTE4, GSTE5, GSTE7 and GSTD3) and three COEs (AARA016468, AARA001582 and AARA004790). Six of these detoxification genes were downregulated following permethrin exposure, including CYP6AA1, CYP9L1, GSTD3, GSTE3 and two COEs (AARA016468 and AARA001582).

One additional CYP450 was also significantly overexpressed between R–C conditions (CYP6Z3, FC = 2.02). A further GST (GSTD10) was highly overexpressed in both R–S and R–C conditions (FCs = 28.25 and 5.94, respectively), but was not present at sufficient sequence coverage in the C–S comparison. In addition, six mosquito salivary gland proteins were identified among R–S and C–S comparisons (Table 2), most notably D7r4 short form salivary protein (FCs = 20.84 and 43.89 for R–S/C–S, respectively), and salivary gland protein 7 (FCs = 4.80 and 11.35), which in contrast to the malathion experiment, were both downregulated in response to permethrin exposure. A further eleven proteins associated with cuticular function displayed differential expression patterns (Table 2; Fig. 3B), including chitinase (AARA007329; FCs = 93.30 and 16.76, for R–S/C–S, respectively), cuticular protein RR-1 family (FCs = 19.34 and 15.68) and cuticular protein (FCs = 6.53 and 11.34, for R–S/C–S). An additional chitinase was significantly overexpressed between R–C conditions (AARA007329, FC = 5.56) (Supplementary Fig. S1).

Similar to the malathion experiment, ontologies enriched in the permethrin experiment also included terms related to “metabolic process” (GO:0008152), “generation of precursor metabolites and energy” (GO:0006091), “oxidoreductase activity” (GO:0016491) and “carbohydrate metabolic process” (GO:0005975).

A total of 717 (45.7%; 512 upregulated and 205 downregulated) transcripts were significantly differentially expressed in mosquitoes that survived either malathion or permethrin exposure, compared to the susceptible strain (Table S7). Eight key upregulated metabolic enzymes were shared between both resistant groups (MAL-R vs DON and PERM-R vs DON), including six CYP450s (CYP6P4, CYP4C36, CYP4G16, CYP6M3 and CYP9K1 and CYP9L1), six GSTs (GSTD3, GSTE2, GSTE3, GSTE4, GSTE5 and GSTE7) and three COEs (AARA004790, AARA016468 and AARA001582) (Table 2; Fig. 4); two additional CYP450s (CYP9M2 and CYP304B1) were both downregulated. Unique detoxification DEGs to the malathion resistant group were CYP9J5 (FCs = 2.34 and 2.11 for R–S/C–S, respectively), CYP6P3 (FCs = 2.29 and 2.09) and one COE (AARA016305: FCs = 3.56 and 4.04 for R–S/C–S, respectively). One detoxification DEG was unique to the permethrin resistant population, CYP325C2 (FCs = 4.04 and 3.49, for R–S/C–S, respectively), but was not present at sufficient sequence coverage in the malathion resistant population.

Among salivary gland DEGs, six were shared between both resistant populations (Table 2; Fig. 4): D7r4 short form salivary protein (AARA016237), D7 long form salivary gland protein (AARA011280), salivary gland protein 1-like members (AARA016223 and AARA016221), TRIO salivary gland protein (AARA001829) and salivary gland protein 7-like members (AARA016088). Twelve additional salivary gland proteins were exclusive to the malathion resistant population and none to the permethrin resistant population (Table 2; Fig. 4).

Among cuticular DEGs, ten were shared between both resistant populations: cuticular protein RR-1 family members (AARA002622, AARA003897, AARA003903 and AARA016147), chitinases (AARA007329 and AARA009226), a cuticular protein CPLCG family member (AARA011120), a cuticular protein RR-2 family 16 member (AARA002342), cuticular proteins (AARA016553 and AARA016552). There were eight and one DEGs which were unique to the malathion and permethrin resistant populations, respectively (Table 2; Fig. 4).

Finally, we mined the RNA-seq data to investigate expression patterns of other recently described resistance mechanisms in An. gambiae complex members (Ingham et al., 2018, 2019) (Table 3). We identified orthologues in An. arabiensis of four α-crystallins, two hexamerins, ATPase subunit e and SAP2 which were significantly differentially expressed between R–S/C–S conditions.

RNA-Seq reads from the malathion and permethrin experiments were screened for target site mutations associated with DDT, pyrethroid, organophosphate or carbamate resistance and known voltage-gated sodium channel (VGSC) mutations in An. gambiae s.l. (Tables S8 and S9). Consistent with the target site PCR data generated in this study, we did not detect the presence of either L1014S kdr or G119S Ace-1 mutations in any populations. The L1014F-kdr mutation was detected in all groups except DON, with average population allele frequencies of CON-M = 27%; CON–P = 24%; MAL-R = 31%; PERM-R = 79%; and SEK = 55% (Table S8). None of the previously described GSTe2 target site mutations (L119F and I114T) (23,24) were present in our dataset, nor was N1575Y, which is linked to L1014F-kdr and found at variable frequencies in parts of West and Central Africa (Jones et al., 2012; Collins et al., 2019; Lynd et al., 2018). Of 20 recently described non-synonymous VGSC mutations from West and Central Africa (Clarkson et al., 2021), we detected the presence of seven (R254K, A1125V, I1868T, P1874L, F1920S, A1934V and I1940T) across the Asendabo field population at very low frequencies (range of 1–7%); 2 of these were also found in SEK (I1868T and I1940T).

Quantitative RT-PCR was used to validate the FCs of eleven genes (CYP4G16, CYP4G17, GSTM3, CPR130, GSTE7, CYP6M2, D7r4 short form salivary protein, chitinase, cuticular protein RR-1 family, CYP6M3 and GSTE3), relative to two housekeeping genes (40S ribosomal protein S7; RPS7 and ubiquitin) (Fig. 5). The majority of the qRT-PCR results supported the directionality of the changes in expression levels as estimated by RNA-Seq.