Background

Aedes mosquitoes are a primary vector for dengue (DENV) and other arboviruses, including Zika, chikungunya, and yellow fever [1]. Amongst these, DENV transmission poses a particularly severe risk to human health as it is responsible for 50–100 million infections annually across the approximately 125 countries in which it is endemic [2,3,4,5]. Above and beyond existing strategies for Aedes population control (e.g., insecticides and other chemical or physical management mechanisms) alternative biological approaches can contribute to sustained disease reduction [6]. For example, substantial reductions in DENV transmission have been achieved through the release of mosquitoes stably infected with the bacterial symbiont Wolbachia [7,8,9,10,11], which confers resistance to DENV and other arboviruses (commonly referred to as “pathogen blocking”) [12,13,14,15].

Wolbachia pipientis is a maternally inherited gram-negative endosymbiotic bacterium that is estimated to naturally infect at least two-thirds of arthropod species [16]. Wolbachia facilitates its spread within populations through a variety of mechanisms that manipulate host reproduction, including cytoplasmic incompatibility, male feminization, male killing, and the induction of parthenogenesis [17]. The Wolbachia strain wMel, which naturally infects Drosophila melanogaster, has been developed to stably transinfect Ae. aegypti. In this host, wMel induces cytoplasmic incompatibly, whereby uninfected males can produce viable offspring with wMel-infected females but matings between wMel-infected males with uninfected females result in embryonic death [18]. Thus, wMel can spread rapidly and then become stably maintained in populations [19, 20]. However, recent studies have documented fitness reductions in individuals harboring wMel that have the potential to slow the spread or reduce the persistence of transinfected individuals [21, 22]. This, in turn, has led to the identification of locally adapted wMel strains with higher fitness that can be strategically deployed to improve biocontrol outcomes [23,24,25].

Two primary mechanisms are believed to form the basis of wMel induced pathogen blocking: (1) induction of innate host immunity and (2) resource competition between microbes [26]. Pathogen blocking due to wMel-induced immunity is supported by several observations. First, wMel confers a partial antiviral effect against RNA viruses in D. melanogaster and viral inhibition in mosquito cell lines [27, 28]. Second, wMel infection in Ae. aegypti induces the upregulation of the Toll and Imd immunity pathways that target pathogens for removal through the production of antimicrobial peptides [29,30,31]. Third, a mechanistic link between immunity induction and pathogen blocking is further supported by the fact that DENV infection in Ae. aegypti triggers a Toll pathway-mediated response and that Wolbachia density itself decreases during DENV infection [32, 33]. The second mechanism implicated in pathogen blocking relates to competition for host resources [34,35,36]. As an obligate endosymbiont, Wolbachia (much like viruses) relies on a suite of host resources. For example, neither Wolbachia nor DENV are able to independently synthesize cholesterol [35]. wMel infection in Ae. aegypti increases cholesterol storage in lipid droplets and inhibits viral replication [36]. Release of cholesterol back into the cytosol reverses this inhibition, and pathogen blocking is also reduced in other insects when raised on a high-cholesterol diet [34]. It is also important to note that the strength of pathogen blocking is variable and has been found to positively correlate with variation of Wolbachia infection density across strains [37, 38]. For example, the higher-density infections characteristic of wMelPop, relative to wMel, is particularly effective in reducing Ae. aegypti DENV vectoring capacity across DENV serotypes [39].

The density and distribution of Wolbachia infection across tissues may also be dependent on interactions with the Aedes microbiome, which in turn may impact the efficacy of Wolbachia as a biological control agent. In fact, several commensal microbes, such as species in the genera Serratia and Asaia, have already been demonstrated to inhibit stable Wolbachia transinfection [40,41,42]. As such, a more wholistic understanding of the tripartite relationships between the Ae. aegypti host, Wolbachia and the microbiome is therefore a priority. A previous investigation of wMel infection on the Ae. aegypti microbiome found no decrease in compositional richness, but did reveal a reduction in a suite of low abundance bacterial taxa [43]. Furthermore, no changes were observed among species with known wMel incompatibilities (e.g., Serratia and Asaia). It is important to note, however, that this study utilized laboratory-raised Ae. aegypti and it is unclear to what extent these results apply to wild populations. Here, we aim to address this uncertainty through the establishment of a refined understanding of Ae. aegypti microbiome variation in populations subject to previous releases of wMel infected individuals.

To assess the influence of wMel infection on microbiome composition, we analyzed available short-read genomic DNA sequence data from a previous study of 172 Ae. aegypti females reared in standard conditions from eggs collected in the wild from six geographic locations [20]. These locations included four sites with historic releases of wMel-infected females (Bungalow, Cairns North West, Parramatta Park North and Westcourt) and two sites without historic releases (Cairns North East and Parramatta Park South). Our approach (1) achieved microbiome classifications at the species-level for 94.6% of the nearly 21 million reads that mapped to bacterial genomes and greatly expanded upon previous microbiome characterizations in this species, (2) quantified microbiome variation while also identifying core bacterial species at the population-level, (3) identified substantial variation in the density of wMel infection across individuals and (4) revealed that wMel was a central hub of negative interactions within a larger Ae. aegypti microbiome interaction network.

Methods

Ae. aegypti genome sequencing data

Double digest restriction-site associated DNA sequencing data (ddRAD-seq) were utilized from a study of Ae. aegypti populations from six geographic locations near Cairns, Australia (NCBI BioProject PRJNA412140) [20]. Releases of Ae. aegypti transinfected with the D. melanogaster strain of Wolbachia (wMel) were conducted in four of these locations: Parramatta Park North and Westcourt (2013 releases); Bungalow and Cairns North West (2014 releases). Schmidt et al. [20] conducted Illumina HiSeq2500 (100 bp reads) paired-end ddRADseq library sequencing of DNA isolated from 172 female mosquitoes reared using standardized conditions from eggs collected in the wild. We have taken advantage of the fact that whole mosquito genomic DNA isolation also captures microbiome DNA. Additional information about the samples can be found in the metadata for NCBI BioProject PRJNA412140.

Taxonomic classification Ae. aegypti microbiome

The Kraken2 pipeline [44] was used to assign taxonomic classifications to all raw ddRAD-seq reads that met standard quality metrics. The Kraken2 search database included all available RefSeq genomes from bacteria (n = 21,333), fungi (n = 60), protozoa (n = 40), viruses (n = 10,388), archaea (n = 390) (NCBI RefSeq Release Number 200 on August 26, 2020) and the recently reannotated Ae. aegypti reference genome (AaegL5 [45]). The database was built with 512 GB of available RAM on 4 threads using the default Kraken2 parameters for minimizer length and spaces, and a conservative k-mer length of 31. Low complexity regions of genomes were masked using the DustMasker program [46]. As expected, the vast majority of reads (93.6% average across 172 individuals; SD = 2.6%) mapped to the Ae. aegypti genome. One percent of reads (185,811 reads on average per individual; range: 29,951 – 4,350,613) mapped to bacterial, viral, archaeal, fungal, or protozoan genomes.

Quantification of Ae. aegypti microbiome composition

After taxonomic classification at the species level for bacterial, viral, archaeal, fungal, or protozoan reads, Bracken [47] was used quantify species abundances per individual. The Bracken database was built using default parameters and a read length of 100. Following Bracken quantification, species abundance estimates were standardized to account for variation in the overall sequencing depth per individual. Additionally, we controlled for the number of restriction enzyme cut sites per genome using a script that calculated the total number of NlaIII (CATG/GTAC) and MluCI (AATT/TTAA) enzyme recognition sites in each reference genome (code is provided in Additional file 1). Standardizations were conducted separately in each major taxonomical group (e.g., bacterial, fungi, etc.). An independent verification of quantitative accuracy was conducted using the original wMel qPCR abundance estimates from Schmidt et al. [20]. For the full set of 172 individuals, abundance estimates were highly reproducible (r = 0.87; p < 0.0001). Additionally, we detected wMel, albeit at low average abundance levels, in individuals that were previously deemed to be uninfected by qPCR. After removing these datapoints, the quantitative reproducibility between studies increased (r = 0.92; p < 0.0001).

For subsequent analyses, rare species were removed from the dataset if they (1) did not constitute > 1% of the total bacterial microbiome in any individual and (2) were present in fewer than 25% of individuals. We note that our analysis was based on DNA isolated from whole mosquitos. It therefore reflects the sum of DNA from microbial populations per individual and it does not provide direct information about tissue-specific microbiome composition.

Statistical analysis of bacterial microbiome composition

Microbiome complexity was assessed in R [48] using two alpha diversity metrics: observed species richness and Shannon’s diversity index [49]. A linear model was used to assess relationships between alpha diversity measures and total wMel abundance per individual. An analysis of variance (ANOVA) and Tukey’s Post Hoc test was used to compare the density of wMel across sites with historical releases. Co-occurrence and mutually-exclusive interactions between bacterial species were assessed using Pearson’s correlations as implemented by CoNet [50] in Cytoscape [51] on a standardized bacterial species abundance matrix. The matrix was standardized using the Wisconsin function from the R package vegan (version 2.5–7 [52]). Statistical significance of each pairwise comparison was determined using 1000 bootstrap replications and Benjamini–Hochberg multiple testing correction. The network was visualized using the Cytoscape GUI software [51]. Additional analysis of network composition was based on delineating two subnetworks. The first (hereafter referred to as the “wMel subnetwork”) includes those species with a significant edge interaction directly with wMel and the second (hereafter referred to as the “non-wMel subnetwork”) as those species without a direct edge interaction with wMel.

Principal Coordinate Analysis (PCoA) was then used to further characterize bacterial microbiome variation. A distance matrix was calculated by applying the Bray–Curtis dissimilarity index [53] implemented by the vegdist function from the R package vegan (version 2.5–7 [52]) to the same standardized matrix used the CoNet analysis (see above). The PCoA was performed using the pcoa function from the R package ape (version 5.5 [54]). Axes of variation were further investigated using linear models to assess the relationship between Principal Coordinate (PC) axis loading values and standardized species abundance estimates.

Results

Ae. aegypti microbiome composition

The overarching goal of this study was to bioinformatically characterize microbiome variation, including that of wMel, in a wild population of Ae. aegypti. To this end, DNA-seq reads from 172 Ae. aegypti females were mapped to a database containing the Ae. aegypti genome and the full set of RefSeq genomes from bacteria, fungi, viruses, archaea and protozoans (27,435 genomes in total). Data from each individual resulted in an average of 18.7 million mapped reads (SD = 220,395), of which the vast majority mapped to the Ae. aegypti reference genome (93.6%; SD = 2.6%). On average, 5.4% (SD = 1.1%) of reads remained unclassified because they were either low quality reads or derived from species not included in our database. The remaining reads (185,811 reads per individual on average) mapped to bacterial, viral, archaeal, fungal, or protozoan genomes. Of these, the vast majority (90.6%; SD = 5.9%) were derived from bacteria (Fig. 1A). Fungi, protozoa, and archaea were consistently identified at low abundances, while viral load was more variable. As expected, given that our analysis was based on DNA sequencing data, 99.8% of all viral reads identified from bacteriophage (i.e., DNA viruses) and the observed viral load variation was due to the high abundance of Escherichia bacteriophages in a limited number of individuals. Full information regarding fungal, viral, archaeal, and protozoan identifications can be found in Additional file 2.

Fig. 1
figure 1

Population-level Ae. Aegypti microbiome characterization. A Taxonomical distribution of reads mapping to non-Ae. Aegypti genomes. B Taxonomical distribution of the major phyla in the bacterial microbiome. C Bacterial microbiome composition by phyla for all 172 individuals. The eight most abundant phyla are shown and the remaining are grouped as ‘Other’

Full size image

Bacterial microbiome composition

After removing low abundance and rare species, 843 bacterial species remained from 23 phyla and 396 genera (Additional file 3; note that wMel is not included in the following analysis of microbiome composition). These results represent a substantial increase in the size and taxonomical specificity of the characterized Ae. aegypti microbiome. Eight phyla comprised > 99.0% of total bacterial composition (Fig. 1B). Consistent with previous studies, all of which relied on 16S rRNA sequencing [43, 55,56,57,58,59], Proteobacteria (49.0% of bacterial reads) was identified as the predominant phylum. Firmicutes (36.6%), Tenericutes (5.9%), Bacteroidetes (4.7%), Actinobacteria (1.5%), Cyanobacteria (0.6%), Spirochaetes (0.5%), and Fusobacteria (0.5%) were also identified as substantive contributors to the microbiome (Fig. 1C). At the genus level, our analysis confirmed the presence of several previously reported groups, including (1) Acinetobacter and Stenotrophomonas, which have been identified in reproductive tissues, (2) Bacillus, Chryseobacterium, Enterobacter, Klebsiella and Serratia, which are commonly identified in the midgut, (3) Burkholderia, which has been identified specifically in the salivary glands and (4) a suite of six genera with more complex spatial distributions across multiple tissues [43, 55,56,57,58, 60,61,62,63]. We also identified several novel species within the Mycoplasma, Salmonella, and Mannheimia genera that were either ubiquitous among individuals or highly abundant. Finally, several taxa (e.g., species in the Asaia, Spiroplasma and Serratia genera) have been demonstrated to negatively influence Wolbachia density or transmission [40, 41, 64]. Our analysis did not identify any species in the genus Asaia, consistent with Audsley et al. [65], but did identify 17 species in the genera Spiroplasma and three in the genera Serratia. Among these species, a negative abundance relationship was observed between wMel and S. marcescens (see below).

Towards a core Ae. aegypti bacterial microbiome

Our analysis provided a unique opportunity to leverage intraspecific variation to assess the community composition of the core Ae. aegypti microbiome at a population level. First, we investigated core microbiome composition as a function of species presence across the population. This analysis revealed that 164 species were present in 95% of individuals and 54 species (hereafter referred to as the “core microbiome”) were present consistently across the entire population (Fig. 2A; Additional file 4). Species present in large numbers of individuals also constituted a high proportion of total microbiome composition (Fig. 2B). Specifically, the 517 bacterial species identified in at least 75% of individuals comprised 93.0% of the total bacterial microbiome. The remaining 7.0% of the microbiome was a combination of 326 additional species that were of far lower abundance on average. Among the 54 species comprising the core microbiome, five species constituted more than 5% of the core microbiome on average (Fig. 2C). Bacillus subtilis (42.4%) had the highest average abundance by far, followed by Salmonella enterica (7.9%), Staphylococcus aureus (6.2%), Escherichia coli (6.1%), and Enterococcus faecium (5.5%). The high abundance of each of these species was likely due to their contribution to the midgut microbiome (Bacillus subtilis, Salmonella sp., Staphylococcus sp., Escherichia sp., and Enterococcus sp. have all been previously identified in the Ae. aegypti midgut microbiome [60, 63, 66,67,68,69]). Although it is noteworthy that several of these species are mammalian pathogens and we cannot rule out possible contamination. Species in the genera Acinetobacter, Clostridium, Cupriavidus, Klebsiella, Serratia were also identified as members of the core microbiome. These genera have all been previously identified in mosquitoes but were not associated with a particular tissue [43, 65, 70, 71]. Unexpectedly, 62.3% (or 34 species) of the core microbiome was comprised of species previously unidentified in the Ae. aegypti microbiome (Additional file 4).

Fig. 2
figure 2

Ae. Aegypti core microbiome. A Number of bacterial species is plotted in relation to the proportion of individuals in which each species was identified. B Percent of microbiome composition is plotted in relation to the proportion of individuals in which each species was identified. Those species present in 75% or more individuals comprised 93% of total microbiome composition. Fifty-four species were identified in all 172 individuals. C Contribution of the 54 ubiquitously present microbiome species to core microbiome composition. Species that do not constitute > 5% of the core microbiome have been grouped as ‘Other.’

Full size image

Population distribution of wMel

Historic releases of wMel-infected females were conducted in four of the six populations analyzed here, within which stable wMel inheritance has already been documented [20]. Based on recent observations from controlled releases, which revealed very limited geographic dispersal of transinfected individuals [8], we predicted a substantively higher prevalence of wMel in locations with historic releases. This was confirmed by an ANOVA that showed significant variation between the wMel across all sites (p = 0.0006; Fig. 3A), including significantly higher abundance levels in the populations with historic releases relative to those without historic releases (p = 1.45 × 10–5; Fig. 3B). Nevertheless, a substantial amount of intraspecific variation in wMel abundance was also observed in populations with historic releases. Overall, average wMel abundance across all populations was 460,458 (coefficient of variance = 1.05) and ranking individuals by wMel abundance revealed a largely continuous gradient of infection density with the highest density infections exceeding 1.5 million reads (Fig. 3C). Given that the strength of pathogen blocking is dependent on the density of Wolbachia infection [38], we next sought to assess the relationship between such variation and bacterial microbiome composition.

Fig. 3
figure 3

wMel variation across population and individuals. A wMel abundance (normalized read count) per individual for each collection site (status and date of previous wMel release is indicated): Bungalow (BN), Cairns North West (CNW), Cairns North East (CNE), Parramatta Park North (PPN), Parramatta Park South (PPS) and Westcourt (WC) [20]. ANOVA revealed significant variation between the wMel abundance across sites, p = 0.0006. A post hoc Tukey test revealed that wMel in CNE was significantly less than those from PPN (p = 0.005) and CNW (p = 0.001) and in PPS was significantly less than those from PPN (p = 0.049). B wMel abundance (normalized read count) per individual comparing sites with and without historic releases. Wilcoxon test revealed significantly higher wMel abundance in sites with historic releases (p = 1.45e-05). C Individuals from all populations ranked by total wMel abundance

Full size image

wMel was a hub of mutually exclusive relationships with other microbiome species

wMel abundance did not correlate with either total bacterial abundance (p = 0.66) or observed species richness (p = 0.89). However, there was a significant inverse relationship between wMel abundance and Shannon’s Index (p = 0.02). Thus, as wMel abundance increased, the relative abundance of other bacterial species in the microbiome became less uniform. A previous investigation of wMel-infected Ae. aegypti revealed a reduction in several low abundance genera across the microbiome, but did not report information about wMel infection variation [43]. To specifically explore the effect of wMel abundance, we constructed an interaction network to assess patterns of co-occurrence and mutual exclusion within the microbiome. This approach revealed a network of 71 species, which was significantly enriched for core microbiome species (obs. = 42; exp. = 4.6; χ2 = 355.7; p = 2.4 × 10–79) and comprised two predominant subnetworks (Fig. 4a; Additional file 5). The first subnetwork (hereafter referred to as the “wMel subnetwork”) comprised 35 mutually-exclusive bacterial relationships specifically with wMel and had a low relative level of connectivity (average of 1.5 edges per node). The second, larger subnetwork (hereafter referred to as the “non-wMel subnetwork”) comprised 56 species and was notable for a higher average level of connectivity (2.2 edges per node), and both co-occurrence (n = 66) and mutually exclusive (n = 58) interaction edges. Three additional characteristics further distinguished these subnetworks. First, the wMel subnetwork comprised species identified at significantly lower abundances than those in the non-wMel subnetwork (Fig. 4B) despite the proportion of core microbiome species being indistinguishable between the two (χ2 = 0.45; p = 0.50). Second, interaction edges in the wMel subnetwork were significantly weaker in their absolute magnitude than those in the larger network (Fig. 4C; p = 8.38 × 10–6). Third, the larger network included B. subtilis, a confirmed Ae. aegypti midgut microbe and the highest abundance species in this study, as well as numerous other high abundance taxa previously identified in the Ae. aegypti midgut (e.g., B. cereus/thuringiensis, E. faecium, E. coli, S. aureus/simulans/cohnii, S. enterica). Thus, the larger subnetwork was more robustly interconnected and contained a diversity of abundant taxa from the midgut microbiome. In contrast, wMel serves as a hub of negative pairwise relationships with relatively low abundance species, even though many of these are part of the core microbiome. We propose that this structure may derive from more direct interactions between wMel and members of the reproductive microbiome, given the higher abundance of wMel in reproductive tissues relative to somatic tissues [72].

Fig. 4
figure 4

wMel forms a negative interaction hub within the Ae. Aegypti microbiome network. A Microbiome interaction network analysis using Wisconsin normalized read counts resulted in a network of 71 species and 159 edges. Red edges represent mutual exclusion interactions (significant negative Pearson’s correlation coefficient) and green edges represent co-occurrence interactions (significant positive Pearson’s correlation coefficient). Core microbiome species are highlighted in purple. wMel was involved exclusively in negative pairwise interactions with 35 species. B Density plot displaying the strength of interactions edges with wMel (orange) and the remaining edges in the network not involving wMel (grey). C Density plot displaying the mean read count for species included in the network. Species with mutually exclusive interactions with wMel are shown in orange and have a significantly lower average read count (t = 2.43; p-value = 0.02) than all other species in the network (grey)

Full size image

Microbiome independent and dependent components of wMel variation

To assess overall patterns of variation within the microbiome, we utilized a Principal Coordinate Analysis and observed three main axes that captured 76.34% of the total variation (Fig. 5). The first axis (58.73% of variation) included wMel as the eigenvector with the largest magnitude (Fig. 5A). The contribution of wMel variation to this axis was supported by the fact that Axis 1 loadings exhibited a significant correlation with wMel abundance across individuals (r = -0.90; t = -26.78, p = 4.60 × 10–63; Fig. 5B). Thus, Axis 1 largely captures variation in wMel and this variation was largely orthogonal to the other main eigenvectors. However, we note that, of the 10 species with the highest positive loadings on Axis 1 (i.e., those with loadings opposite to wMel), five participate in direct negative interactions with wMel in the wMel subnetwork (Fig. 4A). Axis 2 also identified wMel as a large negative eigenvector but, in contrast to Axis 1, it included two species with large positive eigenvectors (i.e., M. haemolytica and B. subtilis; Fig. 5C). The relationship with these species was further supported by a significant positive correlation between Axis 2 loadings and the abundance of wMel (r = -0.24; t = -3.21, p = 0.0016), and a significant negative correlation with M. haemolytica (r = 0.72; t = 13.49, p < 2.2 × 10–16) and B. subtilis (r = 0.48; t = 7.08, p = 3.61 × 10–11; Fig. 5D). Notably, both species were present in our network analysis (Fig. 4A) and their interaction connections to wMel were consistent with the identified PCoA relationships. In summary, Axis 1 captured wMel variation and its weak, mutually-exclusive interaction with a suite of species identified in the wMel network, and Axis 2 captured stronger, but indirect, wMel covariation with two of the most abundant species in the microbiome, including the midgut microbe B. subtilis. Lastly, we explored whether microbiome composition varied by collection site by PCoA and observed limited evidence of clustering above and beyond the variation between sites with and without releases associated with wMel (Additional file 6).

Fig. 5
figure 5

Covariance between microbiome composition and wMel abundance. A PCoA analysis of microbiome variation identified a primary axis of variation (PC1: 58.73%) that corresponded with the eigenvector for wMel which was the largest observed vector. Other large eigenvectors representing highly abundant taxa were largely orthogonal to Axis 1. B Axis 1 loadings were significantly correlated with wMel abundance across individuals but not other microbiome species. C PCoA analysis of microbiome variation identified a secondary axis of variation (PC2: 10.19%) which corresponded with the eigenvectors for B. subtilis, M. haemolytica, and to a lesser extent wMel. D Axis 2 loadings were significantly correlated with B. subtilis and M. haemolytica abundance and anticorrelated with wMel abundance across individuals

Full size image

Discussion

The release of Ae. aegypti mosquitoes transinfected with the wMel strain of Wolbachia has proven to be an effective strategy to limit the spread of DENV [12, 73]. However, less is known about the long-term dynamics of infection within populations and how these dynamics might influence the multivalent utility of wMel more broadly as a biological control agent for arboviruses. Given the mechanisms responsible for pathogen blocking (i.e., immune priming and resource competition), it is reasonable to assume that wMel infection stability and density may be dependent on interactions with the remainder of the microbiome. Here, we explore this possibility through a population-level Ae. aegypti microbiome survey using a bioinformatic approach that leverages available genomic sequencing data. In addition to achieving a species-level Ae. aegypti microbiome characterization from a wild population, our analyses revealed substantial intraspecific variation in wMel density that covaried with microbiome composition. This variation is of potential relevance to the efficacy of pathogen blocking and the implementation of Wolbachia-based pathogen control strategies against DENV and other arboviruses.

Towards an Ae. aegypti population-level core microbiome

Interactions between bacterial components of the microbiome and their mosquito host impact important aspects of fitness, including fertility, longevity, and immunity [14, 40, 59]. As such, the identification of a population-level, core microbiome – those bacterial species present across many or all individuals – has the potential to provide unique insights into taxa that support host development and function [74]. Importantly, and unlike previous studies that have relied on laboratory populations [56, 62, 75, 76] or utilized 16S sequencing [56, 57, 65, 71, 75] on wild populations, our survey provides a unique species-level perspective into microbiome structure in an Ae. aegypti population. Our analysis was also successful in identifying a relatively large repertoire of core microbiome species. Whereas functional analyses will ultimately be required for a refined understanding of core microbiome-host dynamics [77], several observations are worthy of discussion.

The 54 bacterial species ubiquitously present across all individuals comprised only 24% of the total microbiome. However, when the criteria for inclusion was relaxed (to presence in ≥ 75% of individuals), this group of common species comprised 93% of the total microbiome. As such, the vast majority of the microbiome is comprised of common species at the population level. This observation cannot be accounted for by a possible identification bias towards highly abundant species. Whereas many core species were indeed highly abundant, such as B. subtilis that comprises 11.2% of the microbiome on average, the abundance of core species varied by over 1,100-fold (Fig. 4C). Thus, many core species were quite low in relative abundance. This variation is almost certainly explained, at least in part, by the absolute size of tissue-specific microbiomes. We therefore predict that the low average abundance of species in the wMel subnetwork is likely due to their restricted presence in the reproductive microbiome. Similarly, the high average abundance of species in the non-wMel subnetwork, which includes B. subtilis, suggests that they may be representatives of a larger midgut microbiome. Overall, the significant enrichment of core microbiome species (of both high and low abundance) in our network analysis supports the presence of concerted functional assemblages within the microbial community. Furthermore, the topology of the network, including the presence of delineated subnetworks, suggests compartmentalization between tissues that may inform tissue-specific microbiome functionality. Some of core species identified here have also been identified in laboratory colonies derived from widespread global populations (e.g., Brazil, Grenada, and India [62, 68, 69, 78]). Surveys across addition natural and lab populations will further refine our understanding of the functional core microbiome and, additionally, the process by which natural microbiomes experience compositional shifts or diversity-decay when reared in laboratory settings.

Variation in wMel infection density and the microbiome

As a reproductive endosymbiont, Wolbachia was conventionally believed to be concentrated within the host’s reproductive tissues. However, it is now well-established that Wolbachia can be present at variable levels across many non-reproductive tissues [13, 37]. The density and spatiotemporal distribution of Wolbachia varies among host species and strains of Wolbachia [37, 79]. However, there is limited information about the spatiotemporal dynamics of wMel in wild Ae. aegypti populations despite its relevance to pathogen blocking [38]. Despite wMel being stably established in the populations surveyed here [20], we observed a high level of variation in infection density across individuals and a robust covariation between wMel abundance and specific sets of species with the microbiome. Several facets of these relationships are worth highlighting. First, we observed a significant negative relationship between wMel density and Shannon’s Index, indicating a decrease in the compositional evenness in the microbiome. Previous studies comparing alpha diversity metrics in Wolbachia-infected hosts have reported contradicting results [43, 80,81,82]. Our results, based on a large survey of individuals, suggest that the impact of wMel on the Ae. aegypti microbiome is more nuanced than a simple reduction in species richness or load, as previously suggested [43, 80, 81]. Second, wMel served as a hub of consistently mutually-exclusive, but weak, interactions and these interactions were generally with lower abundance species. This pattern is consistent with results reported by Audsley et al. [65]. Based on the composition of the subnetworks revealed in our analysis, we speculate that this pattern may reflect a suite of direct interactions with other members of the reproductive microbiome. It is also important to emphasize that, although these species were generally low in abundance, they were significantly enriched for members of the core microbiome. Third, our PCoA analysis revealed robust covariation between wMel and several of the most abundant species in the microbiome, including the midgut microbe B. subtilis. Given the importance of the midgut microbiome on immune system function [83], the effect of wMel (either direct or indirect) on the midgut microbiome is worthy of further investigation.

Conclusion

We demonstrate the underappreciated value of existing whole-organism DNA-seq data in relation to microbiome characterization. Our analysis achieved a high coverage, species-level characterization of the Ae. aegypti microbiome, including the delineation of a population-level, core microbiome. Despite the absence of information about tissue-specific microbiome composition, network analyses revealed wMel as a hub of interactions with species likely to be of the reproductive microbiome and a small set of robust, indirect interactions with likely members of the midgut microbiome. Intraspecific variation in wMel infection density, in conjunction with diverse interactions with the native microbiome, may have the potential to impact vector competency. The specific interactions identified here could be leveraged to potentially enhance wMel infection density and identify species that may be antagonistic or incompatible with Wolbachia or third-party players that mediate pathways underlying pathogen blocking.