From Macro to Micro: De Novo Genomes of Aedes Mosquitoes Enable Comparative Genomics Among Close and Distant Relatives

Abstract

The yellow fever mosquito (Aedes aegypti) is an organism of high medical importance because it is the primary vector for diseases such as yellow fever, Zika, dengue, and chikungunya. Its medical importance has made it a subject of numerous efforts to understand its biology. One such effort was the development of a high-quality reference genome (AaegL5). However, this reference genome was sourced from a highly inbred laboratory strain with unknown geographic origin. Thus, the reference is not representative of a wild mosquito, let alone one from its native range in sub-Saharan Africa. To better understand the genetic architecture of Ae. aegypti and their sister species, we developed two de novo chromosome-scale genomes with sequences sourced from single individuals: one of Ae. aegypti formosus (Aaf) from Burkina Faso and one of Aedes mascarensis (Am) from Mauritius. Both genomes exhibit high contiguity and gene completeness, comparable to AaegL5. While Aaf exhibits a high degree of synteny to AaegL5, it also exhibits several large inversions. We also conducted comparative genomic analyses using our genomes and other publicly available culicid reference genomes to find extensive chromosomal rearrangements between major lineages. The expanded gene families common to Aaf, AaegL5, and Am revealed that while the overarching category of genes that have expanded are similar, the specific genes that have expanded differ. Our findings elucidate novel insights into chromosome evolution at both microevolutionary and macroevolutionary scales. The genomic resources we present are additions to a growing arsenal for biologists in understanding mosquito biology and genome evolution.

Significance.

Aedes aegypti is a major arboviral disease vector found throughout the tropics and subtropics. Its subspecies differ ecologically, as the native sub-Saharan African form feeds on mammals generally and inhabits both sylvatic and domestic areas and the global invasive form preferentially feeds on humans and lives primarily domestic areas. Their medical importance has prompted the development of a high-quality reference genome, but it was sourced from an inbred laboratory strain of unknown origin. Here, we leveraged PacBio HiFi sequencing and HiC sequencing to develop the first de novo genome of Ae. aegypti sampled its native range in Burkina Faso. We also present a de novo genome of Aedes mascarensis, its sister species. Our genomes are comparably contiguous and complete to the reference genome. Comparative genomic analysis using our genomes and other culicid reference genomes reveals extensive chromosomal rearrangements.

Introduction

Mosquitoes (family Culicidae) are major vectors of arboviruses and parasites that cause minor to severe illnesses that affect millions of people globally (Yee et al. 2022). To mitigate the impact of mosquitoes on global public health, tremendous efforts have been made to elucidate mosquito biology. One such effort is the development of numerous genomic resources of mosquito species, especially those considered medically important such as Anopheles and Aedes mosquitoes. An example of these resources is the reference genome for Aedes aegypti (henceforth, “AaegL5”; Matthews et al. 2018), the primary vector of arboviruses, including yellow fever, dengue, chikungunya, and Zika viruses (Pierson and Diamond 2020).

At the time of its development, AaegL5 utilized the latest available technologies to achieve an accurate, complete, and contiguous chromosome-level genome assembly. Beyond improving upon the contiguity of previous versions of the reference (Dudchenko et al. 2017), AaegL5 also provided an improved set of gene annotations (Matthews et al. 2018), which allows for finer-scale mapping of genes and gene families. Indeed, the development of this reference has been invaluable—allowing researchers to study in detail Ae. aegypti transcriptomics (Herre et al. 2022), developmental biology (Herre et al. 2022), population genetics (Schmidt et al. 2020; Soghigian et al. 2020; Gloria-Soria et al. 2022), species distribution modeling (Rose et al. 2020), and phylogenomics (Soghigian et al. 2023). This resource is also valuable for elucidating biological differences between the subspecies Ae. aegypti formosus (Aaf) and Ae. aegypti aegypti (Aaa), which differ in their bionomics—Aaf is native to sub-Saharan Africa, inhabits both forested and domestic areas, and takes blood meals from a variety of mammals; Aaa is found in the tropics and sub-tropics outside of Africa, inhabits urban areas, and females have a strong affinity to feed on humans (Powell et al. 2018).

Despite its value as a genomic resource, AaegL5 is not without its shortcomings. The sequencing technology available at the time precluded the authors from using wild mosquitoes for two co-related reasons. First, library construction requires high DNA input—a major hurdle for small insects such as flies and mosquitoes where a single individual may not be sufficient to extract enough source DNA leading to assemblies where multiple individuals are pooled. Second, the long-read sequencing platforms available at the time were error-prone and required polishing using other, more accurate sequencing technologies (ie Illumina short reads), which again would require more input DNA from additional individuals. Matthews et al. (2018) mitigated both challenges by sequencing 80 male siblings from a highly inbred laboratory strain (Liverpool), thereby reducing misassemblies resulting from high levels of heterozygosity (Whibley et al. 2021) while also extracting enough input DNA to meet library construction requirements.

A corollary of using an inbred laboratory colony for a reference genome, of course, is the reduction of heterozygosity and genetic diversity that it represents. This is not problematic per se—a single individual is a mere snapshot of the complex ebbs and flows of population dynamics within a single population, and a laboratory strain attempts to capture and preserve that snapshot. However, laboratory rearing protocols can measurably alter allele frequencies leading to gradual divergence between colony and source and erosion of genetic diversity within the colony. Gloria-Soria et al. (2019) found this to be true in commonly used Ae. aegypti laboratory strains, including Liverpool. In fact, their analyses showed that these strains are more closely related to one another and to Aaa than to Aaf (Gloria-Soria et al. 2019), which Liverpool is supposed to represent (Macdonald 1962). This is not an indictment of AaegL5 as a reference, nor is it criticism of studies that use it—alignment programs are capable of accurately aligning newly sequenced reads to the reference to be used in downstream analyses. It is, however, a cautionary note, as genetic deviations away from source populations may lead to erroneous conclusions and problems with reproducibility in the future (Brekke et al. 2018). Furthermore, a reference-quality genome assembly from a known source (especially from sub-Saharan Africa) is valuable.

Within Culicidae, chromosome-scale genome assemblies are taxonomically concentrated in the Anophelinae, and these assemblies have allowed biologists to uncover some of the intricate evolutionary dynamics that bridge the divide between microevolution and macroevolution (Neafsey et al. 2015; Lukyanchikova et al. 2022). Recent efforts to expand the phylogenetic sample of mosquitos have produced numerous chromosome-scale assemblies in the Culicinae (eg Peng et al. 2021; Ryazansky et al. 2024). This effort to expand the phylogenetic scope beyond Anophelinae allows biologists to further investigate the structural variation between members of major clades (eg subgenera, tribes).

Here, we present the first de novo genome assembly of a wild Aaf specimen from West Africa and a de novo assembly of Aedes mascarensis (Am)—a partially reproductively isolated sister species of Ae. aegypti found on the island of Mauritius in the southwest Indian Ocean (Hartberg and Craig 1970). Aedes mascarensis diverged from Aaf roughly 8 to 10 MY (Soghigian et al. 2020). These two species are members of what is now known as the Aegypti group (Le Goff et al. 2013; Soghigian et al. 2020). Each assembly was built from a single wild-caught mosquito from their respective locales (Aaf: Burkina Faso; Am: Mauritius) using the recently developed Pacific Biosciences high fidelity (PacBio HiFi) sequencing platform (Wenger et al. 2019). The high accuracy (99.999%) of PacBio HiFi reads allows us to sequence a relatively highly heterozygous individual while the length of the reads (>13 kb) helps to span and resolve highly repetitive regions ubiquitous in the Ae. aegypti genome (Matthews et al. 2018). We also compare new genomes with those of other mosquitoes and place them in an evolutionary context to understand how genes and genomic structure have changed across major culicid lineages. Lastly, we present a method for comparing draft assemblies and implement it as a new, publicly available, asmidx package for R.

Results

Identifying the Best Combination of HiFi Assembler and Purging Program

We generated 12 draft assemblies each for Aaf and Am—4 outputs directly from the assemblers and 4 of each output from purging programs ph and pd after taking the outputs from the assemblers as input and removing duplicated contigs and haplotigs. We provide detailed comparisons in the supplement (Supplemental Results; Table S4). In brief, asmidx allowed us to identify HiCanu paired with ph output the best Aaf assembly, and hifiasm paired with ph output the best Am assembly (Fig. 1). While not free of arbitrary decisions, asmidx allows users to compare draft assemblies transparently and flexibly.

Fig. 1.

Lollipop plot ranking each draft assemblies for a) Aaf and b) Am. The two best assemblies for each taxon is indicated by larger, blue circles. The scores based on duplicated (0.15), fragmented (0.125), missing (0.15), collapses (0.10), expansions (0.1), inversions (0.1), N50 (0.1), and relative genome size (0.175) [metric (weight)].

Final Assembly Characteristics

Scaffolding the assemblies using Hi-C, and manually curating the scaffolds using Juicebox (Durand et al. 2016) substantially improved the contiguity of the assemblies. For Aaf, N50 saw an 82-fold improvement and reduced L50 from 82 to 2, yielding an assembly with 706 scaffolds, 1,124 contigs, and total assembly size of 1.24 Gb (Figs. S3 and S4; Table 1). In terms of gene content, we detected 3,127 (95.2%) complete single-copy, 55 (1.7%) duplicated, 45 (1.4%) fragmented, and 58 (1.7%) missing BUSCOs from Diptera orthodb10 ortholog set (n = 3,285). For Am, scaffolding improved N50 by 30-fold and reduced L50 from 22 to 2, outputting an assembly with 74 scaffolds, 269 contigs, and a total assembly size of 1.29 Gb (Figs. S3 and S4; Table 1). We detected 2,990 (91%) single-copy, 179 (5.4%) duplicated, 48 (1.5%) fragmented, and 68 (2.1%) missing BUSCOs. We attempted to fill gaps in both assemblies using the HiFi reads, but neither resulted in dramatic reduction in scaffold count, although we were able to fill some gaps between contigs in both assemblies (Fig. S3; Table 1). We additionally assembled the mitochondrial genome of these individuals, which yielded mitogenomes at roughly 16 kb total size for both species.

Table 1.

Final assembly metrics for Aaf (Ae. aegypti formosus) and Am (Ae. mascarensis) compared with those of AaegL5 (Ae. aegypti)

Metric	AaegL5	Aaf	Am
No. of scaffolds	2,310	706	74
No. of contigs	2,539	1,124	269
Total length (Gb)	1.279	1.239	1.289
Gap %	0.002	0.007	0.003
L50	2	2	2
N50 (Mb)	40.978	39.859	42.609
GC ± SD	0.382 ± 0.029	0.381 ± 0.074	0.381 ± 0.04
Mitogenome length (Mb)	16,790	16,617	16,428
No. of exons	65,598	63,767	62,375
No. of genes	18,580	17,672	17,009

Here, AaegL5 is re-masked and reannotated following the same procedure used to mask and annotated Aaf and Am.

Structural annotation of both assemblies showed their compositions were proportionally similar to one another and to the reannotated AaegL5 (Fig. 2). Repetitive elements comprise the majority (up to 79%) of genomic content in all three genomes. Approximately 20% of the genomic content are classified as long terminal repeat (LTR) retrotransposons in Aaf and Am, while AaegL5 had roughly 30%. Up to 15% of the genomes were attributed to non-LTR retrotransposons (Fig. 2). In all three assemblies, DNA transposons accounted for 6% of the genomic content, and ∼1% of each of the assemblies were classified as helitrons (Fig. 2). Additionally, ∼30% of Aaf and Am were considered repetitive, but unable to be classified, while this category constituted ∼24% of AaegL5 (Fig. 2). For all three assemblies, genomic repeat contents appear to have accumulated recently relative to the consensus repeat sequences with the peak occurring at 1%, 5%, and 3% (AaegL5, Aaf, and Am, respectively; Fig. 2). Nonrepetitive genomic content for both assemblies accounted for 22% of each of the three assemblies (Fig. 2). Exon content accounted for 25 Mb in AaegL5, 28 Mb in Aaf, and 27 Mb in Am (roughly 2% for each assembly). Introns accounted for 185 Mb (14% of total length), 211 Mb (15% of total length), and 230 Mb (16% of total length) for AaegL5, Aaf, and Am, respectively. Lastly, monoexonic genes constituted 18% (3,338/18,580), 17% (3,120/17,009), and 16% (2,764/17,672) of the gene contents of AaegL5, Aaf, and Am, respectively.

Fig. 2.

Repeat landscape plots of a) the unmasked and reannotated Ae. aegypti reference genome (AaegL5; Matthews et al. 2018), b) Aaf, and c) Am. Colored bars show sequence divergence in 1% intervals estimated using Kimura model of sequence evolution modified to account for high mutability of CpG sites. Percentages shown in each figure legend show the percent total of the repeat class in each assembly. Landscape plots do not account for non-repetitive DNA. Note that the percentage of RNA detected in AaegL5 was <0.1%.

Identifying Putative Endosymbiont and Contaminant Sequences

We assessed both the gap-filled assemblies and the alternate assemblies (eg assemblies consisting of haplotigs, duplicates, and “debris”) for any potential endosymbionts using the BlobToolKit pipeline (Challis et al. 2020). Of the sequences in the primary Aaf assembly, we traced 1.18 Gb to Arthropoda, while the rest (54.4 Mb) yielded no hits. In the alternate assembly, we detected two sequences, a contig and a scaffold, totaling 1.59 Mb originating from α-proteobacteria (Fig. S5). We isolated and removed this sequence from the alternate Aaf assembly and queried the first 4,400 bp of the scaffold sequence on BLAST. The highest hit (99.89% identity) was for a Rickettsia endosymbiont found in Cimex lectularius (GenBank accession no.: CP084572.1), while the next two highest hits (88.73%, 88.32%), were similarly for Rickettsia endosymbionts isolated from Oedothorax gibbosus and Culicoides impuctatus, respectively (GenBank accession no.: OW370493.1; CP084573.1). In the primary Am assembly, we traced 1.28 Gb of the sequences to Arthropoda and the rest (7.74 Mb) yielded no hits. Unlike the Aaf alternate assembly, the Am alternate assembly did not contain any sequences that did not originate from either Arthropoda or yielded no hits.

Assessing Structural Rearrangement Between AaegL5 and Aaf

We found a substantial degree of synteny between AaegL5 and Aaf, amounting to 1.08 Gb or 87% of the total Aaf assembly exhibiting synteny with AaegL5 (Fig. 3). The Aaf assembly also exhibited 175 inversions (most of which are small), totaling 44 Mb or 3.5% of the total assembly length. We found 870 translocations totaling 4.2 Mb in length (0.34% of total assembly length). The Aaf assembly also exhibited 490 instances of duplicated sequences which totaled 1.9 Mb in length (0.15% of total assembly length). Approximately 5% (61 Mb total length) of the Aaf assembly was not syntenic with AaegL5. We note two relatively large inversions on chromosome 1—one located at 1p34 and one at 1q42 (1.63 and 1.84 Mb long, respectively; Fig. 3; Table S5). We also detect a series of smaller inversions on the telomeric end of the q-arm of chromosome 1. Additionally, we detected smaller inversions on chromosome 3 located near 3p44 and two located on the telomeric end of the q-arm. We made no attempt to interpret nucleotide synteny or structural variation at the nucleotide level between Am and AaegL5 because they are too diverged. Nevertheless, we show the synteny map between these assemblies in Fig. S2.

Fig. 3.

Synteny between chromosomes of Ae. aegypti (AaegL5) and Aaf. BAC (Matthews et al. 2018) positions represented as black horizontal bars.

Gene Order Synteny Across the Culicidae

Our analysis of gene order synteny revealed largely conserved patterns of chromosome evolution within mosquito clades, but patterns between clades showed substantial chromosomal rearrangement (Figs. 4 and S6). Within the Aedini (Aedes and Armigeres) and Culicini (Culex), our analysis showed largely syntenic patterns (ie similar gene order) between the assemblies. However, Anopheles mosquitoes showed substantial chromosomal rearrangements between the subgenera that we included (Fig. 4). We found whole-arm translocations between chromosomes 2 and 3. In fact, whole-arm translocations between chromosomes 2 and 3 appear to be pervasive when viewed at greater evolutionary scales. Without accounting for the translocation of chromosome 1 genes in Aedini onto chromosome 3 in Anopheles, we find five distinct arm associations in chromosomes 2 and 3 (Figs. 4 and S6)—one arm association among the Aedini, one for Sabethes, one for Culex, one for Anopheles cruzii and Anopheles darlingi, and one for Anopheles gambiae. Interestingly, our analysis suggests Anopheles ziemanni have the same autosomal arm associations as the Aedini while maintaining the same chromosome composition as other Anopheles (Figs. 4 and S6). Our analysis showed largely syntenic gene order arrangement between AaegL5, Aaf, and Am. This analysis found a similar set of inversions between AaegL5 and Aaf on chromosome 1 (Figs. 4 and S6). The same regions appeared to be inverted between Aaf and Am (Figs. 4 and S6). We detected an additional inversion between Aaf and Am at the distal end of the q-arm of chromosome 1. Lastly, we note that Phlebotomus papatasi has N = 5 chromosomes—three macrochromosomes (>40 Mb) and two microchromosomes (<20 Mb), but only three macrochromosomes are shown in Fig. 5. This was likely because too few orthologs were detected on the microchromosomes to adequately assess synteny. Nevertheless, we detected considerable gene order rearrangement in P. papatasi (Fig. 4)—genes originating from AaegL5 chromosome 2 comprised much of chromosomes 2 and 3 in P. papatasi, while chromosome 1 of P. papatasi was composed of genes originating from chromosomes 1 and 3 of AaegL5.

Fig. 4.

Riparian plot showing gene order synteny between 11 culicid genomes and the outgroup P. papatasi. Synteny is assessed relative to the Ae. aegypti reference genome (AaegL5). Chromosomes with “*” in their names have been inverted to facilitate visualization. Similarly, some chromosomes appear out of order to facilitate visualization. Note that chromosomes 4 and 5 of P. papatasi are microchromosomes and not shown because too few orthologs were detected on them to adequately assess synteny with any of the chromosomes of AaegL5.

Fig. 5.

Maximum likelihood phylogeny with comparative genomic reconstructions of orthogroup expansions, contractions, and their categories. Values shown at all nodes and tips represent the number of rapidly evolving orthogroups (teal) and expansions (+) or contractions (−). The size of the circle represents the magnitude of the ratio between expansions and contractions (either expansions: contractions or contraction: expansion; depending on which is greater), while its color represents the direction (contraction-biased ratios are negative and darker, expansion-biased ratios are lighter and positive). Bold letters denote ancestral nodes for taxonomic groups: Aedini a), Culicini b), Culicinae c), Anophelinae d). Note that a circle is not shown for ratio between expanded and contracted orthogroups for P. papatasi because the ratio far exceeds those in the in-group.

Gene Family Evolution in the Culicidae

We found 5,723 orthogroups common across all 12 assemblies we included in the analysis. Across the Culicidae, there were 1,050 orthogroups detected exclusive to them (Fig. S7). Between the 2 sub-families, we detected 498 exclusive orthogroups in the Culicinae and 238 in Anophelinae (Fig. S7). The Culex species we included in our analysis had 2,606 exclusive orthogroups in common, while the Aedini taxa had 345 orthogroups exclusive to them (Fig. S7). Within the Aegypti group (ie AaegL5, Aaf, and Am) we detected 355 exclusive orthogroups (Fig. S7). The two Culex species had the highest number of orthogroups detected, followed by AaegL5 (Fig. S7). Our assemblies (Aaf and Am) had similar counts to L5, albeit slightly fewer (Fig. S7).

We detected 2,818 single-copy orthologs across the 12 assemblies we included in these analyses. We created a multiple sequence alignment consisting of 1,762,085 amino acid sites and inferred a maximum likelihood phylogeny using IQ-Tree2 which yielded a log-likelihood score of −16,246,570.272 based on the “Q.insect + F + I + R7 model.” The topology of our phylogeny was largely congruent to that of Soghigian et al. (2023) with the exception of the placement of Sabethes (Fig. 5). Our phylogeny placed Sabethes cyaneus (thus, Sabethini) as sharing a more recent common ancestor with the Aedini than the Culicini, contrary to the results of Soghigian et al. (2023), and more similar to those of Reidenbach et al. (2009).

Our analysis of gene family expansion and contraction showed substantial variability in gene family gains and losses, the majority of which occurred at the species level (Fig. 5). In addition to raw numbers of gains and losses of gene families, rapid gene family evolution (ie gene families with statistically significant changes in count AND categorized to have higher than average rates—reflected by blue numbers to the right of nodes in Fig. 5) appears to have happened at or near the tips rather than toward the root (Fig. 5). Indeed, our analysis showed that deeper internal nodes tended to have very few quickly evolving gene families than more relatively shallow nodes, such as the nodes leading to Aedes and the ancestor of Culex (Fig. 5). Despite the lack of quickly evolving gene families, our analysis did suggest an overall gene family expansion in the subfamily Culicinae, and contraction in the subfamily Anophelinae (Fig. 5). Between tribes in the Culicinae, our results indicated that the Aedini saw nearly double the number of gene family expansion than contractions (Fig. 5). Of the three Aedes genomes, AaegL5 had the greatest number of expansions with nearly double the number of expansions (698) to contractions (312) (Fig. 5). Am had 493 expanded orthogroups and 392 contractions, which was the most of the Aedes genomes (Fig. 5). Aaf had the least number of expansions (363) and a roughly similar number of contractions (330) (Fig. 5). Among tips, Armigeres subalbatus had the highest number of rapidly evolving orthogroups (143) (Fig. 5). The distribution of these appears to differ between assemblies, though in general, rapid changes in orthogroup counts appear to be gains (Fig. S8).

We assigned functional annotation to the orthogroups we detected, and highlighted the most reoccurring, significantly expanded or contracted gene families by total number of copies represented among the species included. In general, while many gene families have expanded and contracted since diverging from each taxon's recent common ancestor, we did not detect any changes that would be indicative of a pattern particular to any group of taxa (Fig. S9), reflecting that most differences appeared to be between tips, rather than between genera or higher taxonomic rankings. In our analysis, we found three instances of α-tocopherol transfer protein-related proteins, Krueppel-like transcription factors, sodium-coupled monocarboxylate transporters, and defective proboscis extension response (DPR)-related proteins each across two orthogroups (Fig. S9). We also found three instances of EG0003.4 protein-related proteins and lipases across three orthogroups (Fig. S9). Lastly, we found five instances of protease family M14 carboxypeptidase A,B across three orthogroups (Fig. S9).

Comparison of Gene Ontology Between Aaf, Am, and AaegL5

We assessed gene function at a finer scale across the three Aedes genomes by assessing the biological process Gene Ontology (GO) terms associated with the expanded orthogroups in these genomes. There were 640 unique biological process GO terms across the three genomes. We detected 635 of these GO terms in AaegL5 of which 233 were unique to it. Aaf had 299 GO terms of which 84 were unique to it. Lastly, Am had 402 GO terms of which 144 were unique to it. We used K-means clustering to group the 646 GO terms into 11 clusters (Fig. 6). Of these clusters, three clusters pertained to RNA transcription, and represented the most common set of GO terms (Fig. 6). Vesicle-mediated transport and signal transduction related terms each formed relatively large clusters (Fig. 6).

Fig. 6.

Heatmap showing presence/absence of GO terms associated with expanded orthogroups and similarity of these terms for the Ae. aegypti reference genome (AaegL5), Aaf, and Am. Similarity of 646 GO terms were assessed using semantic similarity outlined by Mazandu and Mulder (2013) and used K-means clustering to form 11 clusters of GO terms. The heat map on the left shows the presence or absence of GO term in each of the assemblies we analyzed. The heat map on the right shows similarity of individual GO terms, wherein GO terms are clustered together by similarity and divided by horizontal and vertical lines. The most commonly occurring descriptive terms for each cluster is shown to the right, where more frequent terms are shown in larger text.

Gene Families Located in Putative Inversions in Aaf

We extracted the coding sequences in the putatively inverted regions (relative to AaegL5) of Aaf chromosomes, of which 354 were also identified in AaegL5 (Table S7). Notable genes among those found (Fig. S10; Table S8) were different odorant receptors (Or4) odorant binding proteins (Obp 56a and d), and ion channels (Shaker, NaCh) [important for signal transduction (Bohbot et al. 2007; McBride et al. 2014; Matthews et al. 2016)]; E3 ubiquitin ligases (RNF 19B and RNF 126) [similar E3 ubiquitin ligases are implicated in susceptibility to flavivirus infection (Giraldo et al. 2020; Dubey et al. 2022)]; heat shock proteins [important for heat and dehydration tolerance (Zhao et al. 2009; Benoit et al. 2010, 2011)]; cytochrome P450; and adult and larval cuticle proteins [important for insecticide resistance (Poupardin et al. 2010)].

Discussion

De Novo Assembly of Wild, Individual Aedes Mosquitoes

Small body size and a high input DNA requirement have been major hurdles to producing high-quality, chromosome-scale genome assemblies from many wild insects. Recent advances in sequencing technologies that generate highly accurate long reads, like PacBio HiFi (Wenger et al. 2019), allowed us to obtain enough high-quality reads from a single mosquito to use for the first de novo genome assembly of Aaf and Am, avoiding the need of rearing many generations of inbred colonies in the laboratory to obtain sufficient material. One additional roadblock, albeit minor compared with issues such as DNA input requirements, is determining which combinations of varied software tools produce the best assembly, particularly when considering numerous genome assembly metrics. asmidx allowed us to overcome this hurdle, choosing the best assembly from a range of excellent assemblies produced by a variety of genomic tools. Combined with Hi-C aided scaffolding (Burton et al. 2013; Dudchenko et al. 2017), the resulting genome assemblies from our pipelines are both highly contiguous and highly complete (Figs. S3 and S4). Using these chromosome-level assemblies together with other high-quality Culicid reference genomes, we conducted a series of phylogenomic and comparative genomic analyses. The phylogenomic analysis (Fig. 5) showed minor differences to those recently published (Soghigian et al. 2023), likely due to substantial differences in taxonomic (ie number of taxa and lineages) and genetic sampling (ie whole genomes vs. sequence capture). The comparative genomic analysis revealed notable structural differences across large phylogenetic distances (Fig. 4) and numerous insights on the evolution gene families in the Culicidae (Figs. 5 and 6 and S8 and S9). We detail the implications of these findings below.

Comparing Aaf, Am, and AaegL5

The assemblies of Aaf and Am had genome sizes comparable to AaegL5 assembly size (Fig. S3; Table 1) and within the range of genome size estimates from flow cytometry (Matthews et al. 2018). Both assemblies exhibited comparable gene content and accuracy to the Ae. aegypti reference genome, AaegL5 (Fig. S3; Table 1) with a high degree of synteny between homologous chromosomes (Fig. 3). However, we also found evidence of inversions on Aaf chromosomes relative to AaegL5, though they require further testing via PCR for veracity. Recent studies (Redmond et al. 2020; Liang et al. 2025) have described numerous inversions in each of the chromosomes of African and global populations of Ae. aegypti. In our Aaf assembly, we detected several of the same relatively large inversions found on 1p34, 1q42, and 3q43 (Fig. 3; Liang et al. 2025). The inversions on 1p34 and 3q43 are common among African populations of Ae. aegypti (referred to as 1pA and 3qG, respectively by Liang et al. 2025). Furthermore, the inversion we detected on 1q42 is found near both 1qF and 1qG detected in Burkina Faso populations Ouahigouya and Ouagadougou (Liang et al. 2025). Structural variations, such as inversions (Table S5), rearrange gene order, which in turn can lead to adaptive phenotypes that are shielded from recombination (reviewed in: Wellenreuther and Bernatchez 2018; Wellenreuther et al. 2019). This phenomenon is well-documented among Anopheles mosquitoes, wherein inversions are associated with numerous local adaptations (Powell et al. 1999; Ayala et al. 2014; Cheng et al. 2018) and genomic diversity (The Anopheles gambiae 1000 Genomes Consortium 2020). Inversions in Ae. aegypti have a long history of interest (Macdonald and Sheppard 1965) with recent research focusing on identifying inversions in different populations in both subspecies (Dickson et al. 2016; Redmond et al. 2020; Liang et al. 2025). The adaptive effect of these inversions remains unclear; however some genes that have been found in these regions have been functionally characterized. For example, odorant receptor 4 (Or4; McBride et al. 2014) located near 1q42 has been linked to preference for human odor. Other genes may have implications for vector management and adaptation. For example, overexpression and diversification of cuticle proteins are implicated in insecticide resistance in many insects (reviewed in: Balabanidou et al. 2018). Similarly, upregulation or increased copy number of heat shock proteins may contribute to more readily adaptable populations under increasing global temperatures (but see: Ware-Gilmore et al. 2023). A targeted, transcriptomic approach is necessary to further interrogate how genes in these inverted regions are expressed and their phenotypic consequences.

Aedes aegypti and Ae. mascarensis diverged ∼8 to 10 MY (Soghigian et al. 2020), thus we expected overall similarity between genome structure, especially in genic regions between the three Aedes assemblies. Indeed, our results showed a high degree of gene order synteny between these assemblies (Fig. 5) and numerous sets of GO term clusters key to protein synthesis (de la Cruz et al. 2015) and signaling cascades likely necessary for sensory perception (Bohbot et al. 2007). However, an equally intriguing observation is that while the clusters represent similar overarching biological functions, the specific set of genes with these particular GO terms appears to vary in the Aegypti group (Fig. 6). These differences may be manifestations of local adaptations—to environmental conditions where the mosquitoes were sampled (for Aaf and Am) or to the laboratory (for AaegL5). Indeed, numerous studies have documented local adaptations in Aaa to climatic/environmental (Soudi et al. 2023), altitude (Kramer et al. 2023), and vector competency for dengue serotypes (Lambrechts et al. 2009; Gloria-Soria et al. 2017). Another possibility is that the source population for AaegL5 is not necessarily representative of wild Ae. aegypti (Gloria-Soria et al. 2019). Expanding the taxonomic sampling to other members of the Aegypti group may shed light on what functional genetic differences exist within the group.

We found that the repetitive contents of our genomes are similar to those of the reannotated AaegL5 (Fig. 2), and similar to another recent re-examination of the repeat content in AaegL5 (Ryazansky et al. 2024). The largest discrepancy between the repeat contents of our assemblies and those in the reannotated AaegL5 was the LTR retrotransposons. Other Ae. aegypti assemblies whose repeat content is characterized (Aag2 cell line: Whitfield et al. 2017; ROCK chromosome-scale assembly: Fisher et al. 2022) reported levels similar to the original repeat content reported in Matthews et al. (2018). The repeat landscapes for both Aaf and Am assemblies (Fig. 2a and b) are similar to those reported by Whitfield et al. (2017), with sequence divergence peaking close to zero relative to the consensus sequences. These landscapes, particularly that of LTR retrotransposons, represent recent activation and thus may represent recent infection from an RNA virus (Whitfield et al. 2017).

Sex determination in culicine mosquitoes is controlled by the dominant M-locus gene Nix (Hall et al. 2015), located on the P-arm of chromosome 1, near the centromere (Matthews et al. 2018). Unlike in Anophelines, which have a sex chromosome (Biedler and Tu 2016), chromosome 1 is an autosome and homomorphic between males and females, save for this locus, where males are heterozygous at this locus (M/m) and females are homozygous (m/m) (Biedler and Tu 2016; Matthews et al. 2018). This locus has been of keen interest to biologists as it can be informative for better understanding chromosome and sex determination evolution. Furthermore, understanding how sex determination can be modified in Ae. aegypti has great value from the perspective of vector control because ectopic expression of Nix in genetically female mosquitoes can lead to testes development and masculinization (Hall et al. 2015). While this locus has been well characterized in Ae. aegypti, it remains to be studied extensively in Am (but see: Liang et al. 2024). Unfortunately, our assemblies consisted of only female mosquitoes, and thus we are unable to contribute to this dearth in knowledge at this time. However, Am may prove to be useful in the future, when male Am are sequenced and aligned to Am.

A Rickettsia Endosymbiont Found in Aaf

We discovered an endosymbiont in our Aaf assembly (Fig. S5). The genome size of this endosymbiont (1.59 Mb) was similar to the Torix-group Rickettsia endosymbiont first detected in Culicoides impunctatus (Davison et al. 2022). While Rickettsia are better known for causing typhus fever and spotted fever (Raoult and Roux 1997), work conducted in the past 20+ years has revealed their extensive association with invertebrates and their tendency to manipulate host reproduction (see Perlman et al. 2006 and sources cited therein). This body of work has revealed their taxonomic diversity (Perlman et al. 2006; Weinert et al. 2009; Davison et al. 2022) and the diversity of their host range (eg Kikuchi et al. 2002; Thongprem et al. 2021), but their effects on hosts are still understudied. The most documented effect of many invertebrate-affecting Rickettsia is host reproductive manipulation, similar to those of Wolbachia (reviewed in: Werren 1997; Perlman et al. 2006). These endosymbionts are vertically transmitted from infected mother to her offspring, thus hijacking host reproduction to benefit themselves. These effects generally lead to a female-biased sex ratio either by killing males (Werren et al. 1994) or inducing parthenogenesis (Hagimori et al. 2006; Aguin-Pombo et al. 2021). Torix-group Rickettsia have a direct effect on body size of host leech species, wherein infected individuals exhibit larger body sizes (Kikuchi et al. 2002), and low dispersal among infected spiders (Goodacre et al. 2006). The prevalence or the effect of this Rickettsia in Aaf from Burkina Faso is unknown.

Repeated Genomic Rearrangement in Culicidae

Recent advances in sequencing technologies, physical mapping, and three dimensional chromosomal structure inference have made evident the extensive evolutionary rearrangement of chromosomes in Culicidae (Sharakhov et al. 2002; Neafsey et al. 2015; Palatini et al. 2020; Lukyanchikova et al. 2022; Yurchenko et al. 2023; Ryazansky et al. 2024). Microscale structural variants that confer local adaptations (Powell et al. 1999; Ayala et al. 2014; Cheng et al. 2018) and macroscale whole-arm translocations detected between species are well-studied in Anopheles (Sharakhov et al. 2002, 2016; Neafsey et al. 2015; Wei et al. 2017; Artemov et al. 2018). In the Culicinae, Arensburger et al. (2010) detected whole-arm translocations between Ae. aegypti and Culex quinquefasciatus, which Ryazansky et al. (2024) recently confirmed. Our analysis expands on the scope of these studies by including more taxa from across the Culicidae and shows several intriguing trends. First, associations of chromosome arms have repeatedly changed throughout the evolutionary history of the Culicidae (Figs. 4 and S6). Indeed, relative to AaegL5, all non-Aedini genomes we investigated showed whole-arm translocations between chromosomes 2 and 3 (Figs. 4 and S6; also see: Arensburger et al. 2010; Neafsey et al. 2015; Ryazansky et al. 2024). Second, in stark contrast to our first point, chromosomes 2 and 3 showed stability in the Aedini, as none of the assemblies in the tribe showed whole-arm translocations (Figs. 4 and S6). Our taxonomic sample covers roughly 50 MY of evolutionary history in the Aedini, and in a similar timeframe, each of the anophelines in our dataset evolved to exhibit unique chromosomes 2 and 3 arm associations, again in line with the findings of Neafsey et al. (2015), wherein syntenic blocks rapidly decayed in that timeframe. Lastly, despite multiple major rearrangements, most culicids exhibit a karyotype 2N = 6, with the sole exception being Chagasia bathana (subfamily Anophelinae), where 2N = 8 (Rai and Black 1999). This level of conservation is remarkable in the context of other arthropods such as Coleoptera (Blackmon et al. 2024), Lepidoptera (Wright et al. 2024), and within Diptera (Morelli et al. 2022). To better interrogate chromosome evolution within Culicidae, more high-quality, chromosome-scale genome assemblies are required, especially within the Culicinae.

Evolutionary Changes in Distribution and Copy Number of Gene Families

Changes in copy number of key gene families may play key adaptive roles in mosquitoes leading to differences in vector effectiveness between species (Arcà et al. 2017; Palatini et al. 2017; Catapano et al. 2023) and populations (Lambrechts et al. 2009; Bennett et al. 2021). We compared our Aaf and Am genomes to high-quality mosquito genome assemblies that were publicly available and found striking differences both in the distribution of rapidly evolving orthogroups (Fig. S8) and copy number change (Fig. S9).

Rapid changes in copy number may also be an indication of adaptation (Simon et al. 2015; Xie et al. 2018). Rapid gains, in particular have been attributed to adaptative evolutionary changes as duplicated gene copies are “released” from stabilizing selective pressures and may respond adaptively as the environment or the context in which they are expressed changes (Guo and Kim 2007; Vieira et al. 2007). However, other works have also shown rapid losses to also lead to adaptation (McBride and Arguello 2007; Goldman-Huertas et al. 2015). Rapid gains appear in all mosquito assemblies in our data set (Fig. S8), with only few instances of rapid losses. At a more granular, functional level, we detected seven reoccurring rapidly evolving orthogroups in our data set (Fig. S9). However, these genes, nor their distribution within our dataset does not exhibit any discernible biological pattern with regard to phenotype.

Conclusion

Here, we presented a de novo genome assembly of a wild-caught Aaf and Am each derived from a single, wild-caught individual. Our assemblies are comparable to the reference Ae. aegypti assembly (AaegL5; Matthews et al. 2018) in terms of contiguity and gene content but differ in that the Aaf assembly exhibits genomic structural variation particular to West Africa (Liang et al. 2025), and notable differences in ortholog counts and their functions. At 8 to 10 MY diverged, we view the nucleotide synteny between Am and AaegL5 as unreliable, but gene order in Am is highly conserved and shows a high degree of synteny with Ae. aegypti. With the three Aedes genome assemblies we also showed variation in gene family expansion, and the function of those expanded gene families, showing population- (between AaegL5 and Aaf) and species- (between Ae. aegypti and Am) wide differences. These assemblies will be valuable assets for future studies to understand the biology and evolution of Ae. aegypti—the Aaf assembly more closely reflects natural populations of Ae. aegypti in the ancestral range, and the Am assembly provides the closest genome output to date for this species. We used our newly assembled genomes along with ten other reference genomes evolutionary changes within the Culicidae and found repeated bouts of major chromosomal rearrangements, particularly between chromosomes 2 and 3. The genomes we present here represent initial steps toward the development genomic resources for all of the currently described taxa in the Aegypti group (Soghigian et al. 2020). Further development within this group can elucidate genomic architecture that differentiates the ecology and behavior of the African and the global invasive subspecies.

Materials and Methods

Sample Collection

We sampled for wild mosquitoes in two locations: Ouagadougou, Burkina Faso, for Aaf (October 2021); and Chamarel, Mauritius for Am (May 2022). In both locations, we collected mosquito eggs by placing ovitraps lined up with seed-germination paper. These eggs were shipped to The Connecticut Agricultural Experiment Station (CAES) for rearing. We reared mosquitoes from eggs at CAES by submerging the paper containing the eggs in deionized water and provided TetraMin Tropical Flakes ad libitum as food source for larvae. Once mosquitoes pupated, we transferred the pupae into medicine cups filled with larval water and placed them into insect rearing cages (12 × 12 × 12 inches/28 L) where the adults emerged. Larval trays and cages were maintained in an incubator at 27 °C with a 12:8 light/dark cycle throughout the rearing process. We provided adults with ad libitum sugar water for 3 to 5 d until they were collected in dry ice for the DNA extraction protocol.

DNA Sequencing for Genome Assembly and Hi-C Genome-wide DNA Cross-linking

To generate DNA sequences for PacBio HiFi for both Aaf and Am, we collected and froze adult female mosquitoes on dry ice, then homogenized individuals with a sterile DNAse/RNAse free plastic pestle. We eluted the homogenate using 180 µL of PBS and processed the solution using the MagAttract HMW DNA Kit (Qiagen, Germantown, MD, USA) following the frozen tissue protocol from the manufacturer. We then sent the extracted DNA to the Maryland Institute of Genome Sciences of the University of Maryland for low input library preparation, where two Sequel II 8M SMRT Cell runs (CCS/HiFi mode—30 h movie) were used to obtain sequences.

For Hi-C sequencing, we pooled multiple individuals (four females for Aaf and eight females for Am) together and pulverized them in dry ice. For Aaf, we prepared the DNA library using the Swift Accel NGS 2S Plus DNA Library kit (IDT, San Diego, CA, USA) following protocol outlined by the Arima Hi-C 2.0 kit (Arima Genomics, San Diego, CA, USA) for cross-linking. For Am, we prepared the DNA library using the Arima High Coverage Hi-C kit followed by the Arima Library Prep kit following manufacturer's protocol. We then sequenced both Hi-C libraries at the Yale Center for Genome Analysis on an Illumina NovaSeq 6000 to achieve 100 million 150 bp paired-end reads.

Genome Assembly

The PacBio HiFi sequencing platform and the programs (or the specific modes) built to handle these data are still nascent, so we used four different assemblers and compared the outputted draft assembly from each program. We used HiCanu (Nurk et al. 2020), flye (Kolmogorov et al. 2019), hifiasm (Cheng et al. 2021), and IPA (available at: https://github.com/PacificBiosciences/pbipa). For all programs, we specified an estimated haploid genome size of 1.3 Gb and used default settings and set flags necessary to assemble PacBio HiFi reads (Table S1).

We also compared the performance of two purging programs designed to identify and remove duplicate haplotypes from the draft assemblies—purge_haplotigs (Roach et al. 2018) and purge_dups (Guan et al. 2020) (henceforth, ph and pd respectively). Note that hifiasm and IPA employ a purging step as part of their respective assembly pipelines by default—hifiasm employs a variant of pd with a different algorithm for haplotype identification, and IPA simply uses pd. Thus, in our workflow, assemblies output by these programs were purged twice, which allowed us to assess “out-of-the-box” performance of all programs.

To assemble the mitochondrial genomes, we used the program mitohifi (Uliano-Silva et al. 2023) in “reads” mode and input the PacBio HiFi reads. This program was reference guided, so for both species we used the Ae. aegypti complete mitochondrion found on GenBank (OR544945.1).

asmidx: A Holistic Approach to Assessing Genome Quality based on User Input

To quantify assembly metrics, we fed each intermediate assembly to the program Inspector ver. 1.0.1 (Chen et al. 2021). This program quantifies basic assembly metrics, detects assembly errors at the structural (expansion, collapses, inversions, and haplotype switches ≥50 bp) and small scales (base substitutions, expansions, collapses <50 bp), and attempts to correct them. We also assessed gene content completeness for each assembly generated using BUSCO ver. 5.2.2 and the Diptera OrthoDB data set ver. 10 (n = 3,285 single-copy orthologs).

Common genomic metrics (eg N50) and gene content of an assembly can be effective indicators of assembly quality, but there is no consensus on which of these characteristics (or what set of them) best characterizes a “good” genome assembly. Furthermore, it is unknown whether each assembly program used in tandem with a purging program outputs similar quality genome assemblies. To facilitate identifying the “best” assembly derived from the same set of reads, we wrote an application using the shiny package ver. 1.7.4 for the R statistical programming language ver. 4.2.1, which we call asmidx (available at https://github.com/genmor/asmidx). The application takes as input a user-generated data set in tabular format, with column headers where each row contains an assembly, and each column contains a metric describing that assembly. The user can then select assembly metrics that should be maximized and minimized to assess quality. Additionally, the user can identify a genome size column and input a known genome size (eg from a reference genome) which will be used to compute relative genome size differences, where smaller differences in relative genome size will be considered better. Each chosen metric is then feature-normalized to be between 0 (the worst) and 1 (the best). A row sum is then computed and multiplied by 100 resulting in normalized scores which are used to rank each assembly by quality. We also allow users to differentially weight each of the selected characters. To make this process intuitive, we allow for any positive value for weighting. The weights are multiplied to their respective feature-normalized columns, row sums recomputed and multiplied by 100 to output a weighted score. For both weighted and unweighted scores, higher values will be associated with better assemblies based on the quality metrics supplied (and weighted) by the user. The application outputs these rankings in tabular format and visually represents them using a lollipop plot, both of which the user can download. In addition to this shiny application, we wrote helper functions to convert output from Inspector into tabular format.

To rank our assemblies, we considered four sets of metrics: gene content (duplicated, fragmented, and missing BUSCOs), structural errors in the assembly (total number of expanded bases, collapsed bases, and inverted bases), contiguity (N50), and genome size difference relative to the reference. Our rankings maximized N50 and minimized all other metrics. We weighted all structural errors and N50 by 0.1, duplicated and missing BUSCOs by 0.15, fragmented BUSCOs by 0.125, and relative genome size difference by 0.175. For each specimen, we used the assembly with the highest weighted score for scaffolding using Hi-C.

Scaffolding Contigs to Chromosomes via Hi-C and Post-Processing

To prepare cross-linked, paired-end Illumina short reads for use in scaffolding, we trimmed the first five bases from the reads using the program fastp (ver. 0.23.4; Chen et al. 2018). We then followed the Arima Genomics mapping pipeline (available at https://github.com/ArimaGenomics/mapping_pipeline). This pipeline relies on BWA (ver. 0.7.17-r1198-dirty; Li and Durbin 2009), SAMtools (ver. 1.15; Danecek et al. 2021), picard (ver. 2.2.4; Broad Institute 2019), and custom Perl scripts for aligning the short reads to a draft contig-level genome assembly and preparing it for scaffolding (Table Program details). To scaffold, we used the program Yahs (ver. 1.2a.2; Zhou et al. 2023). We manually curated the outputted scaffold-level assembly using the programs Juicer tools and Juicebox (ver. 1.11.08; Durand et al. 2016) to further remove duplicate contigs and correct misassemblies (ie inverted and misjoined contigs and scaffolds) and generated finalized contact maps for visualization purposes using HapHiC (ver. 1.0.5; Zeng et al. 2024). After manually curating the scaffolds, we used TGS-gapcloser (ver. 1.2.1; Xu et al. 2020) to close gaps in the assembly. We inspected gene content of this assembly using BUSCO and then finalized the gap-filled assemblies by using a custom script (available at https://github.com/genmor/CSBfilter) to remove “debris” sequences (ie contigs and scaffolds with duplicate Hi-C signal) and scaffolds/contigs containing only duplicate BUSCOs that were not located on the chromosomal scaffolds. We concatenated these “debris” sequences together with the duplicates detected by the purging program. We then fed these assemblies to the BlobToolKit (ver. 4.2.1; Challis et al. 2020) suite to determine whether our assemblies contained sequences from contaminants or endosymbionts and to output final summary statistics.

Validating Structural Variation and Departures From AaegL5

To ascertain the validity of the structural variations we observed in Aaf and Am (relative to AaegL5), we used NCBI BLAST (ver. 2.12.0+; Camacho et al. 2009) to create databases from our scaffolded Aaf and Am assemblies to find positional hits of 88 Ae. aegypti bacterial artificial chromosome (BAC) clone sequences (Matthews et al. 2018; Table S2). We retained only the best hits (ie highest bit score) for each BAC clone and visualized their positional order in both assemblies using the R package ChromoMap (ver. 4.1.1; Anand and Rodriguez Lopez 2022; Fig. S1). After this validation step, we aligned the scaffolded Aaf assembly to AaegL5 using minimap2 (Li 2018), and the resultant alignment file fed into SyRI (ver. 1.6.3; Goel et al. 2019) which identified nucleotide synteny and structural variation (ie duplications, translocations, and inversions). We performed the same analysis for Am but chose not to interpret nucleotide synteny because it is too divergent from Ae. aegypti, even after using less stringent minimap2 settings (ie -asm20) and found the resulting output uninterpretable (see Fig. S2).

Genome Structural Annotation

We used the RepeatModeler pipeline (ver. 2.0.4; Flynn et al. 2020) to model and identify repetitive elements in the genomes. After generating a de novo repeat database from our draft assemblies, we soft-masked them using RepeatMasker (Smit et al. 2013) in four iterations, passing each outputted soft-masked FASTA file to the subsequent step: (i) mask only simple repeats; (ii) mask repeats using the -species flag with “diptera” which queries the Dfam database (ver. 3.7; Storer et al. 2021) for dipteran repeat sequences; (iii) mask repeats identified in Ae. aegypti (Nene et al. 2007) from TEfam repeat database; (iv) mask repeats based on the de novo repeat database created from RepeatModeler. We then quantified the diversity and divergence relative to the consensus sequences of the repetitive content in Aaf and Am using the “calcDivergenceFromAlign.pl” script included with RepeatMasker. This script estimated divergence using the Kimura (K81) model of sequence evolution modified to account for the high mutability of “CG” sites (Tsunoyama et al. 2001).

We input the final, soft-masked assemblies into the BRAKER2 genome annotation pipeline (ver. 3.0.3; Stanke et al. 2006, 2008; Hoff et al. 2016, 2019; Brůna et al. 2021) with the Arthropoda protein data set obtained from OrthoDB (ver. 11; Kuznetsov et al. 2023). Braker uses GeneMark-ES (Lomsadze et al. 2005) and ProtHint (Brůna et al. 2020) to predict protein coding genes in the assembly, then aligns these predicted proteins and regions using DIAMOND (ver. 0.9.24.125; Buchfink et al. 2015) and SPALN (Iwata and Gotoh 2012). High-confidence hits output by these programs are then fed into GeneMark-EP+ (Brůna et al. 2020) and Augustus (ver. 3.4.0; Stanke et al. 2006) to output gene predictions. Following recommendations from the program authors, we used the final output from Augustus. We then used a Python script “selectSupportedSubsets.py” included with Braker to exclude genes predicted without any external support (ie no support from OrthoDB). We used this output for all analysis that involved the proteome. We assessed the quality of these annotations using BUSCO in protein mode and AGAT (ver. 1.2.0; Dainat et al. 2023) to quantify annotation metrics after we subset the output from BRAKER2 to exclude genes supported only through computational predictions and keep only the longest isoforms.

We sought to make direct (ie “apples-to-apples”) comparisons in the repeat and genic contents of our genome assemblies to AaegL5. To do this, we unmasked the nucleotides of the AaegL5 fasta file, then ran it through the process we described above for modeling and masking the repeat content, then passing the newly re-masked assembly to Braker 2 to perform structural annotations.

Comparative Genomic Analysis

We compared the Aaf and Am assemblies with other Culicidae, including AaegL5, and nine (eight culicid; one outgroup) other annotated, high-quality, chromosome-level reference assemblies available on NCBI that represent the breadth of phylogenetic diversity of the family. These assemblies were: An. cruzii (subgenus Kerteszia), An. darlingi (subgenus Nyssorhynchus), An. gambiae (subgenus Cellia), An. ziemanni (subgenus Anopheles), Ar. subalbatus (tribe Aedini), S. cyaneus (tribe Sabethini), Culex pipiens pallens (tribe Culicini), Cx. quinquefasciatus (tribe Culicini), and the sandfly, P. papatasi (Table S3). Again, because we sought to directly compare genic content between these reference genomes and our genome assemblies, we reannotated each reference assembly using Braker2 as described above, then used the gene predictions from it in all subsequently described comparative genomic analyses.

We assessed gene order synteny using the R package GENESPACE (ver. 1.2.3; Lovell et al. 2022). GENESPACE uses Orthofinder 2 (Emms and Kelly 2019) to identify orthologous genes across a set of species and assesses synteny and collinearity of the orthologs between all pairwise combinations of species using MCScanX (Wang et al. 2012). We set AaegL5 as the reference for the riparian plot output by GENESPACE.

We also assessed the evolution of gene families across these genomes using methods described by compare_genomes pipeline (Paril et al. 2023), but rather than running the pipeline itself, we ran the programs or similar methods independently. In brief, the pipeline identifies orthogroups from protein FASTA files of the genomes of interest. This information is used to infer a dated phylogeny. This inferred phylogeny and the orthology information can then be used to functionally annotate orthogroups and reconstruct along the phylogeny where genes may have expanded or contracted. In our workflow, we ran Orthofinder 2 to identify orthogroups (with the -M flag set to “msa” and -A flag set to “mafft”). These settings isolated the single-copy orthogroups present across all genomes and aligned them using MAFFT 7 (Katoh and Standley 2013). We fed this alignment into IQ-Tree2 (Minh et al. 2020) with ModelFinder (ie -m flag set to “MFP”; Kalyaanamoorthy et al. 2017), the –alrt flags set to “1000” and the ultrafast bootstrap approximation (Hoang et al. 2018) to estimate branch support and a least squares algorithm to date the inferred phylogeny (To et al. 2016). We assessed gene family expansion and contraction using Cafe 5 (Mendes et al. 2021). We set the γ-parameter to K = 4 categories, which allowed the model to group different orthogroups into four rate categories. We assigned functional annotations to all identified orthogroups using InterProScan 5 (Jones et al. 2014) with the -dp and -goterms flags while querying the PANTHER classification system (--appl PANTHER; Mi et al. 2013, 2019). Since each ortholog sequence can be assigned a different function, we assumed the ortholog with the lowest E-value for each orthogroup to be the representative function. Using the outputs from Cafe 5 and InterProScan, we examined what biological functions are associated with rapidly expanding orthogroups of AaegL5, Aaf, and Am. We did this by isolating the rapidly expanding orthogroups in each of these genomes and assessed the biological process set (GO:0008150) of GO terms associated with each orthogroup. We used the simplifyEnrichment (ver. 2.2.0; Gu and Hübschmann 2023) package for R to simplify the combined list of GO terms associated with the expanded orthogroups in the three Aedes genomes into larger clusters with a common, underlying category of biological functions. simplifyEnrichment achieves this by measuring pairwise semantic similarity of the GO terms (Mazandu and Mulder 2013). These pairwise similarity metrics are then clustered using K-means clustering to create sets of GO terms that have similar biological functions.

We took advantage of the fact that we identified gene families and their positions in the Aaf genome, so we mapped the ones specifically located in putatively inverted regions (relative to AaegL5). We matched gene identity and names by creating a local BLAST database from the NCBI AaegL5 proteins, and queried the orthologs in the inverted regions, limiting output to a single alignment with e-values <1e−60.

Supplementary Material

Supplementary material is available at Genome Biology and Evolution  online.

Funding

Financial support for this project was provided by NIAID R01 AI101112 awarded to J.R.P. We also acknowledge the support of the Natural Sciences and Engineering Research Council of Canada (NSERC), funding reference number RMS21-73779779 [Cette recherche a été financée par le Conseil de recherches en sciences naturelles et en génie du Canada (CRSNG), numéro de référence RMS21-73779779]. G.M. was supported by NIAID R01 AI155562 awarded to J.R.P., A.G.-S., and J.S.

Data Availability

All read data and final assemblies have been deposited to NCBI's GenBank. The PacBio HiFi reads generated for this project have been deposited in Sequence Read Archive under accessions: SRR33810827, SRR33810828 (Aedes aegypti formosus) and SRR33832123, and SRR33832122 (Aedes mascarensis). The HiC reads used to scaffold the assemblies are accessioned under: SRR33810826 (Aedes aegypti formosus), and SRR33832121 and SRR33832120 (Aedes mascarensis). The final assemblies are accessioned under: JBNQXC000000000, JBNQXD000000000 (principal and alternate assemblies, respectively; Aedes aegypti formosus), and JBNQXA000000000, JBNQXB000000000 (principal and alternate assemblies respectively; Aedes mascarensis). Code used to assemble and analyze the genomes can be found at https://github.com/genmor/GBE_Mosquito_Genome_Assembly_Code.